{
  "schema_version": "1.1",
  "generated_at": "2026-11-17T16:44:56.450811+00:00",
  "benchmark": {
    "source_url": "https://github.com/onyx-dot-app/EnterpriseRAG-Bench",
    "id": "enterpriserag-small-v1",
    "freeze_sha256": "f83ccf8b13723815b884acb9c6c060e49b6df947a19f6ea6a631c56fd994aa16",
    "protocol_sha256": "b4ba877a11159aa99e312e359e2f5d71ed790213608d47110eca0b0083cfb281",
    "documents": 20000,
    "questions": 250,
    "publishable": true,
    "limitations": [
      "Four colliding upstream IDs use unique physical IDs and an official relevance_id equivalence group; qst_0413 has a duplicated gold ID and contradictory versions.",
      "Synthetic English corpus; 20k-document task differs from full EnterpriseRAG.",
      "High-level questions retain original references despite reduced global context."
    ],
    "publication_redaction": {
      "credentials": "removed",
      "machine_paths": "templated or removed",
      "metrics_changed": false
    },
    "website_import": {
      "source_sha256": "c54ad89edb2d34e836b842d3827faecd7d01641c4f6b197b33e0872e55bfedf8",
      "review_status": "owner-approved",
      "metrics_modified": false
    },
    "repository_visibility": "private"
  },
  "protocol": {
    "harness_python": "3.14.4 (main, Aug 20 2026, 10:41:58) [GCC 15.2.0]",
    "measurement_files_sha256": {
      "protocol/ingest-probes.json": "eed8e3cbec2818cc2166b17c682e1ff1ad40b548b7808145653553dd65179bd5",
      "protocol/voyage-probes.jsonl": "459e3a9d74f6f5355d40fd923e0bdaba5cc6820511bdfdf7c61dcaed16226d79",
      "scripts/budget.py": "8ea4d3ccfaccb70f001e99a180d3d1d334a96ed168fa1084ae549113b985f34a",
      "scripts/footprint.py": "8eb5b0fb1078dad737cdf5ed8bcd74e077c704c40326ea6526455881a7cc24ea",
      "scripts/metrics.py": "7018fbddd9f2254a3b6ed10c4d78894679b3f7d8fe0293a13628e8dd0f41e400",
      "scripts/native_reports.py": "f9997c775bf87bcd44daf4515d9b5d940a2247de3711c8ee024e6fd283d5a145",
      "scripts/run.py": "3e72203ca76f56a78f852de3cae06076aee79c7eaf857288fbc483339c83e9ca",
      "scripts/voyage_probe.py": "78b3f32a8ce80ac2a555a357647018901b62b8188f3cad8d87a18138eb867411"
    },
    "reporter_sha256": "e2545ddeea7e2553d52d4184bd29cf64b75879e2344f46861389fe7391b3199c",
    "latency": {
      "headline": "first-ask",
      "requests_per_question_per_block": 1,
      "concurrency": [
        1,
        8
      ],
      "restart_between_blocks": true,
      "index_warmup": "one unrelated query",
      "repeat_secondary": true,
      "query_embedding_cache": {
        "fyrr": "in-process, reset on restart",
        "onyx": "Redis tenant/settings/query hash TTL900s, ephemeral Redis restarted",
        "pipeshub": "No cross-request query-vector cache found in selected search path"
      },
      "provider_probe_sha256": "459e3a9d74f6f5355d40fd923e0bdaba5cc6820511bdfdf7c61dcaed16226d79",
      "provider_probe_schedule": "Separate persistent HTTP client at first-ask start and each25 completed questions, including final; queued tail outside product timing"
    },
    "ingest": {
      "clock": "first submission through native full inventory and late-document probe",
      "probe_sha256": "eed8e3cbec2818cc2166b17c682e1ff1ad40b548b7808145653553dd65179bd5",
      "setup_excluded": true
    },
    "embedding": {
      "model": "voyage-4",
      "dimensions": 1024,
      "document_input_type": "document",
      "query_input_type": "query",
      "truncation": "provider default true",
      "pricing_usd_per_million_tokens": 0.06,
      "pricing_source": "https://docs.voyageai.com/docs/pricing",
      "pricing_date": "2026-11-16"
    },
    "limitations": []
  },
  "variants": {
    "default": {
      "label": "Product defaults",
      "headline": true,
      "reranker": null
    },
    "concurrency-8": {
      "label": "fyrr sized for 8 clients",
      "description": "C8 only; reader.concurrency=8; same index, no re-ingest",
      "metric_prefixes": [
        "first_ask_c8_",
        "warm_c8_"
      ],
      "reranker": null,
      "headline": false,
      "systems": [
        "fyrr"
      ],
      "changes": {
        "reader.concurrency": 8
      },
      "measured_concurrency": [
        8
      ],
      "index_reused": true,
      "retrieval": "unchanged product defaults",
      "approval": "b0d130e4-8488-4f92-9dba-ca6b73793815"
    },
    "lambda-reader": {
      "label": "fyrr on AWS Lambda (4096 MiB)",
      "reranker": null,
      "headline": false,
      "systems": [
        "fyrr"
      ],
      "experimental": true,
      "description": "Experimental 4096 MiB Lambda reader; real 20k namespace, native release / Voyage 4, default concurrency 2 per execution environment; cold starts retained",
      "changes": {
        "deployment": "AWS Lambda",
        "lambda_memory_mb": 4096
      },
      "source_url": "https://github.com/fyrr-ai/fyrr"
    },
    "lambda-reader-2048": {
      "label": "fyrr on AWS Lambda (2048 MiB)",
      "reranker": null,
      "headline": false,
      "systems": [
        "fyrr"
      ],
      "experimental": true,
      "description": "Experimental 2048 MiB Lambda reader; documented 1536 MiB reader memory budget; default concurrency 2 per execution environment; cold starts retained",
      "changes": {
        "deployment": "AWS Lambda",
        "lambda_memory_mb": 2048,
        "reader_memory_budget_bytes": 1610612736
      },
      "source_url": "https://github.com/fyrr-ai/fyrr"
    },
    "parallel-partitions": {
      "label": "PipesHub with parallel partition scheduling",
      "reranker": null,
      "description": "FAIR_SCHEDULING_PARALLEL_PARTITIONS=true, non-default, source-documented; default 20k not completed. Default measured sample rate and projected time are retained separately.",
      "headline": true,
      "systems": [
        "pipeshub"
      ],
      "changes": {
        "FAIR_SCHEDULING_PARALLEL_PARTITIONS": true,
        "indexing_llm_model": "gpt-6-luna"
      },
      "source_url": "https://github.com/pipeshub-ai/pipeshub-ai/blob/v1.0.0/backend/python/app/services/messaging/config.py#L533",
      "approval": "54c6b86a-4995-4550-b136-5347f37d3167",
      "replaces_headline_variant": "default"
    }
  },
  "website": {
    "headline_metric": "recall_at_10",
    "groups": [
      {
        "id": "retrieval",
        "number": "01",
        "label": "Retrieval quality",
        "title": "Find the right evidence.",
        "description": "Primary measure: the right documents in the top 10. Recall measures coverage of gold documents; nDCG also rewards their order. Recall@50 shows deeper coverage."
      },
      {
        "id": "speed",
        "number": "02",
        "label": "Search latency",
        "title": "The first question counts.",
        "description": "First-ask latency includes a fresh embedding-provider call path. Repeated-query timings are secondary and preserve native cache behavior."
      },
      {
        "id": "footprint",
        "number": "03",
        "label": "Footprint & cost",
        "title": "What stays running?",
        "description": "Service counts, measured resources and an explicit monthly estimate. Missing inputs stay n/e, including cost."
      },
      {
        "id": "ingest",
        "number": "04",
        "label": "Ingest & embeddings",
        "title": "Before the first question.",
        "description": "Fresh index and vector cache. Submission-to-searchable wall time ends when every accepted document is searchable. Coverage and native rejections are disclosed; throughput uses accepted documents. Setup and installation are excluded."
      }
    ],
    "metrics": {
      "recall_at_1": {
        "key": "recall_at_1",
        "label": "Recall@1",
        "unit": "fraction",
        "group": "retrieval",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile question bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "recall_at_5": {
        "key": "recall_at_5",
        "label": "Recall@5",
        "unit": "fraction",
        "group": "retrieval",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile question bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "recall_at_10": {
        "key": "recall_at_10",
        "label": "Recall@10",
        "unit": "fraction",
        "group": "retrieval",
        "direction": "higher_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile question bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "recall_at_20": {
        "key": "recall_at_20",
        "label": "Recall@20",
        "unit": "fraction",
        "group": "retrieval",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile question bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "ndcg_at_10": {
        "key": "ndcg_at_10",
        "label": "nDCG@10",
        "unit": "fraction",
        "group": "retrieval",
        "direction": "higher_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile question bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c1_p50_ms": {
        "key": "warm_c1_p50_ms",
        "label": "Warm search C1 p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c1_p95_ms": {
        "key": "warm_c1_p95_ms",
        "label": "Warm search C1 p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c8_p50_ms": {
        "key": "warm_c8_p50_ms",
        "label": "Warm search C8 p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c8_p95_ms": {
        "key": "warm_c8_p95_ms",
        "label": "Warm search C8 p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "ingest_wall_s": {
        "key": "ingest_wall_s",
        "label": "Time to searchable index",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "embedding_cost_usd": {
        "key": "embedding_cost_usd",
        "label": "Embedding cost to ingest",
        "unit": "USD",
        "group": "cost",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "service_count": {
        "key": "service_count",
        "label": "Resident services (measured)",
        "unit": "services",
        "group": "footprint",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null,
        "description": "Raw count of processes/containers resident during the measured local query window. The website displays always-on product-service counts using documented deployment roles; these differ from the local PipesHub measurement.",
        "display_label": "Always-on product services",
        "display_description": "The Services column counts always-on services the product ships and requires: fyrr 1 reader, Onyx 12, PipesHub 15. The fyrr writer runs as a scheduled job; its index is on S3. Generic TLS termination and managed infrastructure are listed separately, and the TLS-proxy cost allowance is not counted as a fyrr service. Raw local resident-process counts remain in the run JSON; the documented production PipesHub profile differs from its measured local stack."
      },
      "idle_rss_bytes": {
        "key": "idle_rss_bytes",
        "label": "Memory at idle",
        "unit": "bytes",
        "group": "footprint",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "monthly_cost_usd": {
        "key": "monthly_cost_usd",
        "label": "Estimated production cost",
        "unit": "USD/month",
        "group": "cost",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "recall_at_50": {
        "key": "recall_at_50",
        "label": "Recall@50",
        "unit": "fraction",
        "group": "retrieval",
        "direction": "higher_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile question bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "load_rss_peak_bytes": {
        "key": "load_rss_peak_bytes",
        "label": "Peak memory during search",
        "unit": "bytes",
        "group": "footprint",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "index_and_data_bytes": {
        "key": "index_and_data_bytes",
        "label": "Index, data & cache on disk",
        "unit": "bytes",
        "group": "footprint",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "warm_c1_embedding_p50_ms": {
        "key": "warm_c1_embedding_p50_ms",
        "label": "Warm C1 embedding stage p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c1_embedding_p95_ms": {
        "key": "warm_c1_embedding_p95_ms",
        "label": "Warm C1 embedding stage p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c8_embedding_p50_ms": {
        "key": "warm_c8_embedding_p50_ms",
        "label": "Warm C8 embedding stage p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "warm_c8_embedding_p95_ms": {
        "key": "warm_c8_embedding_p95_ms",
        "label": "Warm C8 embedding stage p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c1_p50_ms": {
        "key": "first_ask_c1_p50_ms",
        "label": "First-ask p50 · concurrency 1",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c1_p95_ms": {
        "key": "first_ask_c1_p95_ms",
        "label": "First-ask p95 · concurrency 1",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c8_p50_ms": {
        "key": "first_ask_c8_p50_ms",
        "label": "First-ask p50 · concurrency 8",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c8_p95_ms": {
        "key": "first_ask_c8_p95_ms",
        "label": "First-ask p95 · concurrency 8",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "ingest_docs_per_min": {
        "key": "ingest_docs_per_min",
        "label": "Ingest throughput",
        "unit": "docs/min",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "primary",
        "ci": null
      },
      "default_sampled_ingest_docs_per_min": {
        "key": "default_sampled_ingest_docs_per_min",
        "label": "Default configuration sampled ingest rate (incomplete run)",
        "unit": "docs/min",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": null,
        "scope": "pipeshub-only",
        "comparable": false,
        "description": "Observed default baseline interval, not a completed 20,000-document ingest. Projected full time is separate metadata and is not a measurement."
      },
      "parallel_sampled_ingest_docs_per_min": {
        "key": "parallel_sampled_ingest_docs_per_min",
        "label": "Parallel configuration sampled ingest rate (incomplete run)",
        "unit": "docs/min",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": null,
        "scope": "pipeshub-only",
        "comparable": false,
        "description": "Observed interval during the interrupted parallel import, not a completed20k ingest; projection remains separate."
      },
      "ingest_peak_rss_bytes": {
        "key": "ingest_peak_rss_bytes",
        "label": "Peak native RSS during ingest",
        "unit": "bytes",
        "group": "footprint",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "embedding_request_count": {
        "key": "embedding_request_count",
        "label": "Exposed document embedding HTTP attempts",
        "unit": "requests",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "first_ask_c1_failure_rate": {
        "key": "first_ask_c1_failure_rate",
        "label": "First-ask C1 search failure rate",
        "unit": "fraction",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "first_ask_c1_embedding_p50_ms": {
        "key": "first_ask_c1_embedding_p50_ms",
        "label": "First-ask C1 native embedding stage p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c1_engine_p50_ms": {
        "key": "first_ask_c1_engine_p50_ms",
        "label": "First-ask C1 engine estimate p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "first_ask_c1_embedding_p95_ms": {
        "key": "first_ask_c1_embedding_p95_ms",
        "label": "First-ask C1 native embedding stage p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c1_engine_p95_ms": {
        "key": "first_ask_c1_engine_p95_ms",
        "label": "First-ask C1 engine estimate p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "warm_c1_failure_rate": {
        "key": "warm_c1_failure_rate",
        "label": "Repeat C1 search failure rate",
        "unit": "fraction",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "warm_c1_engine_p50_ms": {
        "key": "warm_c1_engine_p50_ms",
        "label": "Repeat C1 engine estimate p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "warm_c1_engine_p95_ms": {
        "key": "warm_c1_engine_p95_ms",
        "label": "Repeat C1 engine estimate p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "voyage_probe_c1_p50_ms": {
        "key": "voyage_probe_c1_p50_ms",
        "label": "Direct Voyage probe C1 p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "comparable": false,
        "description": "Separate held-out embedding requests interleaved during this product window; provider/network diagnostic, not engine performance."
      },
      "voyage_probe_c1_p95_ms": {
        "key": "voyage_probe_c1_p95_ms",
        "label": "Direct Voyage probe C1 p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "comparable": false,
        "description": "Separate held-out embedding requests interleaved during this product window; provider/network diagnostic, not engine performance."
      },
      "first_ask_c8_failure_rate": {
        "key": "first_ask_c8_failure_rate",
        "label": "First-ask C8 search failure rate",
        "unit": "fraction",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "primary",
        "ci": null
      },
      "first_ask_c8_embedding_p50_ms": {
        "key": "first_ask_c8_embedding_p50_ms",
        "label": "First-ask C8 native embedding stage p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c8_engine_p50_ms": {
        "key": "first_ask_c8_engine_p50_ms",
        "label": "First-ask C8 engine estimate p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "first_ask_c8_embedding_p95_ms": {
        "key": "first_ask_c8_embedding_p95_ms",
        "label": "First-ask C8 native embedding stage p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        }
      },
      "first_ask_c8_engine_p95_ms": {
        "key": "first_ask_c8_engine_p95_ms",
        "label": "First-ask C8 engine estimate p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "warm_c8_failure_rate": {
        "key": "warm_c8_failure_rate",
        "label": "Repeat C8 search failure rate",
        "unit": "fraction",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "warm_c8_engine_p50_ms": {
        "key": "warm_c8_engine_p50_ms",
        "label": "Repeat C8 engine estimate p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "warm_c8_engine_p95_ms": {
        "key": "warm_c8_engine_p95_ms",
        "label": "Repeat C8 engine estimate p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "scope": "fyrr-only",
        "description": "Fyrr-only diagnostic, n/e when native provider-wait timings are unavailable; MCP privacy protection suppresses explicit search durations. No engine estimate is inferred from client latency or separate provider probes."
      },
      "voyage_probe_c8_p50_ms": {
        "key": "voyage_probe_c8_p50_ms",
        "label": "Direct Voyage probe C8 p50",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "comparable": false,
        "description": "Separate held-out embedding requests interleaved during this product window; provider/network diagnostic, not engine performance."
      },
      "voyage_probe_c8_p95_ms": {
        "key": "voyage_probe_c8_p95_ms",
        "label": "Direct Voyage probe C8 p95",
        "unit": "ms",
        "group": "speed",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": {
          "confidence": 0.95,
          "method": "percentile request bootstrap",
          "resamples": 10000,
          "seed": 1729
        },
        "comparable": false,
        "description": "Separate held-out embedding requests interleaved during this product window; provider/network diagnostic, not engine performance."
      },
      "ingest_submitted_docs": {
        "key": "ingest_submitted_docs",
        "label": "Ingest submitted documents",
        "unit": "documents",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_accepted_docs": {
        "key": "ingest_accepted_docs",
        "label": "Ingest accepted documents",
        "unit": "documents",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_rejected_docs": {
        "key": "ingest_rejected_docs",
        "label": "Ingest rejected documents",
        "unit": "documents",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_coverage_fraction": {
        "key": "ingest_coverage_fraction",
        "label": "Accepted ingest coverage",
        "unit": "fraction",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "primary",
        "ci": null
      },
      "ingest_stage_writer_sync_s": {
        "key": "ingest_stage_writer_sync_s",
        "label": "writer sync",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null,
        "stage_of": "ingest_wall_s",
        "stage_order": 0
      },
      "ingest_stage_writer_commit_s": {
        "key": "ingest_stage_writer_commit_s",
        "label": "writer commit",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null,
        "stage_of": "ingest_wall_s",
        "stage_order": 1
      },
      "ingest_stage_reader_check_s": {
        "key": "ingest_stage_reader_check_s",
        "label": "reader check",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null,
        "stage_of": "ingest_wall_s",
        "stage_order": 2
      },
      "ingest_stage_mcp_inventory_readiness_s": {
        "key": "ingest_stage_mcp_inventory_readiness_s",
        "label": "mcp inventory readiness",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null,
        "stage_of": "ingest_wall_s",
        "stage_order": 3
      },
      "ingest_stage_late_document_probe_s": {
        "key": "ingest_stage_late_document_probe_s",
        "label": "late document probe",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null,
        "stage_of": "ingest_wall_s",
        "stage_order": 4
      },
      "ingest_stage_native_archive_upload_and_file_store_s": {
        "key": "ingest_stage_native_archive_upload_and_file_store_s",
        "label": "native_archive_upload_and_file_store",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_stage_native_connector_creation_and_start_s": {
        "key": "ingest_stage_native_connector_creation_and_start_s",
        "label": "native_connector_creation_and_start",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_stage_background_indexing_and_native_inventory_s": {
        "key": "ingest_stage_background_indexing_and_native_inventory_s",
        "label": "background_indexing_and_native_inventory",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_stage_late_document_search_probe_s": {
        "key": "ingest_stage_late_document_search_probe_s",
        "label": "late_document_search_probe",
        "unit": "seconds",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_searchable_docs": {
        "key": "ingest_searchable_docs",
        "label": "Searchable documents",
        "unit": "documents",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "secondary",
        "ci": null
      },
      "ingest_pending_docs": {
        "key": "ingest_pending_docs",
        "label": "Documents still pending",
        "unit": "documents",
        "group": "ingest",
        "direction": "lower_is_better",
        "tier": "secondary",
        "ci": null
      },
      "searchable_coverage_fraction": {
        "key": "searchable_coverage_fraction",
        "label": "Searchable share of the submitted corpus",
        "unit": "fraction",
        "group": "ingest",
        "direction": "higher_is_better",
        "tier": "primary",
        "ci": null
      }
    },
    "products": {
      "fyrr": {
        "type": "Permission-aware retrieval over MCP",
        "pros": [
          "Permission-aware search over index files on S3; no database or cluster to run.",
          "Choose an OpenAI-compatible embedding endpoint and an optional configured reranker.",
          "No answer-model API key in fyrr; your existing MCP app writes the answer."
        ],
        "cons": [
          "Returns evidence to your agent; does not supply an answer-generation UI.",
          "Cloud connectors have only been tested against synthetic APIs, not real workspaces.",
          "Readers still need compute, a private disk mirror and an endpoint for live query embeddings."
        ],
        "sources": [
          {
            "label": "Architecture & deployment",
            "url": "/docs/deployment"
          },
          {
            "label": "Connector maturity",
            "url": "/docs/connectors"
          },
          {
            "label": "Model configuration",
            "url": "/docs/configuration"
          },
          {
            "label": "Reranker configuration",
            "url": "/docs/tuning"
          },
          {
            "label": "Existing AI app",
            "url": "/docs/quickstart"
          }
        ]
      },
      "onyx": {
        "type": "AI workspace with search & agents",
        "pros": [
          "Built-in chat, custom agents and deep research.",
          "A broad connector catalog and self-hosted deployment options."
        ],
        "cons": [
          "Self-hosting the full search deployment means operating its application and data services.",
          "Its full workspace includes features beyond this retrieval-only test.",
          "The measured local stack has 12 resident services. The production cost model prices its documented cloud deployment, which has different availability and topology from the local benchmark."
        ],
        "sources": [
          {
            "label": "Product & deployment source",
            "url": "https://github.com/onyx-dot-app/onyx"
          },
          {
            "label": "Deployment services",
            "url": "https://github.com/onyx-dot-app/onyx/blob/main/deployment/docker_compose/docker-compose.yml"
          }
        ]
      },
      "pipeshub": {
        "type": "Enterprise search & context platform",
        "pros": [
          "Cited answers, knowledge-graph navigation and agent tools.",
          "Connects business sources and exposes context through MCP."
        ],
        "cons": [
          "Self-hosting includes vector, graph and application data stores, alongside the application services.",
          "Its agent features are outside this retrieval and search-latency comparison.",
          "The measured non-default configuration kept background enrichment active, with two of 20,000 accepted documents still pending. Search quality and latency are measured; canonical ingest time and complete operating cost remain n/e. Sample rates and infrastructure subtotals are qualified separately."
        ],
        "sources": [
          {
            "label": "Product & deployment source",
            "url": "https://github.com/pipeshub-ai/pipeshub-ai"
          },
          {
            "label": "System architecture",
            "url": "https://docs.pipeshub.com/system-overview"
          }
        ]
      }
    },
    "methodology": [
      {
        "label": "Status & scope",
        "text": "Measured results, approved for publication. Fyrr and Onyx completed their accepted-document ingest protocols. PipesHub retrieval and latency use 19,998 searchable of 20,000 accepted documents, with two duplicate copies pending; all 250 questions and original gold labels are retained. Its canonical ingest clock is unavailable after a harness stop/resume, and its full production cost is n/e. This synthetic English subset has incomplete relevance labels and reduced context; it cannot establish production performance."
      },
      {
        "label": "Subset derivation",
        "text": "Frozen: 250 of the core 500 questions, stratified by official type using Hamilton allocation and stable SHA-256 selection. All selected gold documents plus source-proportional distractors make 20,000 documents. 379 gold documents; 96,402,403 corpus bytes. Seed: fyrr-compare-v1-2026-11-16. The separate 10-question / 1,000-document smoke test is never a published benchmark. Exact freeze and protocol hashes are recorded below."
      },
      {
        "label": "Embeddings",
        "text": "Voyage voyage-4 with 1,024 dimensions for the measured systems, using document/query input types. Native chunking, request prefixes, deduplication and caches are retained and recorded per run. Native provider-billed embedding spend is n/e where unavailable; canonical-input API cost estimates are separate."
      },
      {
        "label": "Uncertainty & missing results",
        "text": "95% percentile bootstrap intervals: 10,000 resamples, seed 1729, paired question IDs. Expected is the frozen denominator; attempted counts actual attempts; n counts valid results; missing = expected minus n, including unattempted rows. Retrieval excludes 15 empty-gold questions, leaving 235. Incomplete aggregates are partial. Unequal successful subsets cannot establish a product win."
      },
      {
        "label": "Speed protocol",
        "text": "Primary: ask each of the 250 questions once per fresh service at concurrency 1 and 8, after one unrelated index warm-up. Fresh query caches; embedding-provider time included; percentile bootstrap intervals. Repeated-query warm timings remain a separate secondary track. Timed runs require a quiet host."
      },
      {
        "label": "Hardware, versions & measurement date",
        "text": "Measured 17 November 2026. Native host: Intel Core i9-12900H, 14 physical cores / 20 logical CPUs, 30.576 GiB kernel-reported RAM, Linux 7.0.0-42-generic / glibc 2.43. Captured CPU, memory and kernel invariants match the native fyrr and Onyx windows. Experimental Lambda readers ran separately on AWS eu-central-1, x86_64, at 4096 or 2048 MiB. Exact commits, build/configuration hashes, phase windows and hardware-source hashes are in the JSON provenance."
      },
      {
        "label": "Cost assumptions",
        "text": "Main monthly cost prices each product’s documented production deployment on AWS Frankfurt for 20,000 documents, 100,000 searches per month and 1% changed daily. Required replicas and managed services are itemized with deployment and price sources. Unknown required costs leave the total n/e, with a known subtotal shown separately. The measured single-host minimum uses matched light/C1 and team/C8 capacity references; the Lambda trial is an independent projection. These price models do not establish an AWS capacity or availability guarantee. Service counts exclude scheduled writer jobs and generic TLS termination; the cost model may still budget for them."
      },
      {
        "label": "Headline wording",
        "text": "Each headline clause compares all representative products with complete measured values for that metric. n/e and partial rows cannot establish a win and are disclosed separately. Similar search quality uses at most a 2 pp gap or overlapping 95% intervals for recall@10 and nDCG@10 when both are measured; this is a descriptive rule, not an equivalence test. Faster queries require separated C1 median intervals; indexing uses accepted-document wall time. A fraction of cost requires complete stated production models and at most half the least expensive measured alternative. Overview numbers are rounded; details and JSON retain full precision."
      },
      {
        "label": "Reproducibility artifacts",
        "text": "Raw evidence bundle (43.6 MB, SHA-256 dd0bf0f454acb5fc617a716cbc819eca6e85ef5639e6f6ee1d2c4fd26bfdca66) available on request; public hosting pending. Kept in a private fyrr-ai repository. Independent replay of the archived scorer reproduced all 30 aggregate means and sample counts (n=235 across five quality rows)."
      },
      {
        "label": "Service count definition",
        "text": "The Services column counts always-on services the product ships and requires: fyrr 1 reader, Onyx 12, PipesHub 15. The fyrr writer runs as a scheduled job; its index is on S3. Generic TLS termination and managed infrastructure are listed separately, and the TLS-proxy cost allowance is not counted as a fyrr service. Raw local resident-process counts remain in the run JSON; the documented production PipesHub profile differs from its measured local stack."
      }
    ],
    "features": [
      {
        "label": "Embedding model",
        "cells": {
          "fyrr": {
            "text": "Choose the endpoint and model in [embedding]. The provider must implement the OpenAI-compatible embeddings protocol.",
            "sources": [
              {
                "label": "Configuration",
                "url": "/docs/configuration"
              }
            ]
          },
          "onyx": {
            "text": "Self-hosted installs choose from a local/cloud model registry. In v5.1.2, new Voyage selections use the voyage-4 family.",
            "sources": [
              {
                "label": "Release model registry",
                "url": "https://github.com/onyx-dot-app/onyx/blob/v5.1.2/backend/shared_configs/embedding_models.py"
              },
              {
                "label": "Index settings",
                "url": "https://docs.onyx.app/admins/advanced_configs/search_configs"
              }
            ]
          },
          "pipeshub": {
            "text": "Choose a supported embedding provider and model, including local or cloud models; Voyage is supported.",
            "sources": [
              {
                "label": "AI models",
                "url": "https://docs.pipeshub.com/ai-models/overview"
              }
            ]
          }
        }
      },
      {
        "label": "Reranker on the tested search path",
        "cells": {
          "fyrr": {
            "text": "Optional endpoint and model in [namespace.rerank], with candidate, deadline and spend limits. No reranker calls when absent.",
            "sources": [
              {
                "label": "Reranker settings",
                "url": "/docs/tuning"
              }
            ]
          },
          "onyx": {
            "text": "v5.1.2: the selected Search UI endpoint uses native hybrid ranking without a neural reranker. Unused reranker helpers do not establish support on this path.",
            "sources": [
              {
                "label": "Release search pipeline",
                "url": "https://github.com/onyx-dot-app/onyx/blob/v5.1.2/backend/onyx/context/search/pipeline.py"
              }
            ]
          },
          "pipeshub": {
            "text": "v1.0.0: /api/v1/retrieve retains native hybrid/RRF ranking. Its request exposes no hosted reranker selection; the local reranker helper is not invoked on this path.",
            "sources": [
              {
                "label": "Release retrieval endpoint",
                "url": "https://github.com/pipeshub-ai/pipeshub-ai/blob/v1.0.0/backend/python/app/api/routes/search.py"
              }
            ]
          }
        }
      },
      {
        "label": "Answer model & API key",
        "cells": {
          "fyrr": {
            "text": "No answer-LLM API key is configured in fyrr. Search and fetch supply evidence to your existing MCP app; its plan and limits still apply. Embeddings and optional reranking may need separate credentials.",
            "sources": [
              {
                "label": "Claude Code example",
                "url": "/docs/quickstart"
              },
              {
                "label": "MCP tools",
                "url": "/docs/mcp-tools"
              }
            ]
          },
          "onyx": {
            "text": "Built-in answers use a configured LLM. External AI apps can also access Onyx through MCP.",
            "sources": [
              {
                "label": "MCP access",
                "url": "https://github.com/onyx-dot-app/onyx/blob/v5.1.2/documentation/overview/onyx_anywhere/mcp_server.mdx"
              }
            ]
          },
          "pipeshub": {
            "text": "Built-in answers use the configured language model. MCP also exposes context to external agents.",
            "sources": [
              {
                "label": "System overview",
                "url": "https://docs.pipeshub.com/system-overview"
              },
              {
                "label": "MCP overview",
                "url": "https://docs.pipeshub.com/mcp/overview"
              }
            ]
          }
        }
      },
      {
        "label": "Programmatic access",
        "cells": {
          "fyrr": {
            "text": "Stdio for a local client; Streamable HTTP authenticated with OIDC for shared access.",
            "sources": [
              {
                "label": "MCP transports",
                "url": "/docs/mcp-tools"
              },
              {
                "label": "HTTP users",
                "url": "/docs/http-users"
              }
            ]
          },
          "onyx": {
            "text": "Service-account API keys require Business, Enterprise or a paid Cloud plan. User tokens/session auth are separate options; the benchmark uses a normal user session.",
            "sources": [
              {
                "label": "v5.1.2 authentication",
                "url": "https://github.com/onyx-dot-app/onyx/blob/v5.1.2/documentation/admins/user_management/api_keys.mdx"
              }
            ]
          },
          "pipeshub": {
            "text": "MCP supports personal access tokens or OAuth. Access follows the authenticated user and their permissions.",
            "sources": [
              {
                "label": "MCP authentication",
                "url": "https://docs.pipeshub.com/mcp/overview"
              }
            ]
          }
        }
      }
    ],
    "variant_note": "Overview: default fyrr and Onyx; PipesHub uses source-documented, non-default parallel partition scheduling and owner-selected gpt-6-luna for native enrichment. Its retrieval and latency are measured with 19,998 / 20,000 documents searchable, two pending duplicates and background enrichment active; canonical ingest time remains n/e. Optional details: fyrr reader.concurrency=8 on the same index, and two experimental AWS Lambda deployments of the real 20k corpus (4096 MiB and 2048 MiB). Neither changes headline wins. Lambda cold starts remain in their timings; their complete operating cost is n/e. The older 300-document synthetic trial is a separate projection. Neither real-20k Lambda deployment met the predeclared C8 latency selector; no Lambda winner is selected.",
    "overview_metrics": [
      "monthly_cost_usd",
      "service_count",
      "idle_rss_bytes",
      "first_ask_c1_p50_ms",
      "ingest_wall_s",
      "recall_at_10"
    ],
    "overview_labels": {
      "recall_at_10": "Quality",
      "monthly_cost_usd": "Monthly cost",
      "service_count": "Services",
      "idle_rss_bytes": "Memory",
      "first_ask_c1_p50_ms": "Query speed",
      "ingest_wall_s": "Ingest"
    },
    "stories": [
      {
        "id": "story-footprint",
        "title": "Costs a fraction to run.",
        "description": "Documented production deployments, priced on AWS Frankfurt for the same reference workload. Resources, replicas, storage, managed services and API usage are explicit. The measured single-host minimum and experimental Lambda projection stay separate.",
        "group": "footprint",
        "metrics": [
          "monthly_cost_usd",
          "service_count",
          "idle_rss_bytes",
          "index_and_data_bytes"
        ]
      },
      {
        "id": "story-speed",
        "title": "Answers fast.",
        "description": "Median (p50) bars and p95 ticks for new and repeated questions. Lower is better. Switch between one and eight simultaneous requests.",
        "group": "speed",
        "metrics": [
          "first_ask_c1_p50_ms",
          "first_ask_c1_p95_ms",
          "warm_c1_p50_ms",
          "warm_c1_p95_ms",
          "first_ask_c8_p50_ms",
          "first_ask_c8_p95_ms",
          "warm_c8_p50_ms",
          "warm_c8_p95_ms"
        ]
      },
      {
        "id": "story-ingest",
        "title": "Indexes fast.",
        "description": "Submit the same 20,000 documents to each product; time until every accepted document is searchable. Accepted and rejected counts stay visible. Lower wall time is better; setup and installation are excluded. Embedding cost is reported separately.",
        "group": "ingest",
        "metrics": [
          "ingest_wall_s"
        ]
      },
      {
        "id": "story-quality",
        "title": "Finds the right documents.",
        "description": "Recall@10 measures whether the evidence your agent needs appears in its first ten documents. Higher is better.",
        "group": "retrieval",
        "metrics": [
          "recall_at_10"
        ]
      }
    ],
    "lambda_projection": {
      "id": "aws-lambda-300-docs-2026-11-17",
      "status": "projection",
      "experimental": true,
      "region": "eu-central-1",
      "architecture": "x86_64",
      "date": "2026-11-17",
      "docs_path": "/docs/lambda",
      "reader_commit": "c2a34a3db86a9568e4e9754c86a4da46f100cdae",
      "documents": 300,
      "memory_mib": 1024,
      "index_bytes": 1076222,
      "warm_execution_median_ms": 134.42000000000002,
      "warm_billed_ms": 135,
      "warm_n": 30,
      "warm_client_median_ms": 289.0088378917426,
      "warm_client_n": 15,
      "cold_client_ms": [
        35756.33890600875,
        2197.629000991583,
        2304.662272101268
      ],
      "cold_billed_allowance_ms": 5494,
      "cold_fraction": 0.01,
      "control_reads": 2,
      "cold_extra_reads": 50,
      "request_volumes": [
        10000,
        100000,
        1000000
      ],
      "prices": {
        "lambda_gb_second_usd": 0.0000166667,
        "lambda_request_usd": 2e-7,
        "s3_get_usd": 4.3e-7,
        "s3_storage_gb_month_usd": 0.0245
      },
      "sources": [
        {
          "label": "AWS Lambda pricing",
          "url": "https://aws.amazon.com/lambda/pricing/",
          "date": "2026-11-17"
        },
        {
          "label": "Amazon S3 pricing",
          "url": "https://aws.amazon.com/s3/pricing/",
          "date": "2026-11-17"
        }
      ],
      "evidence_sha256": {
        "REPORT.md": "e530264479048ba350337d50dc376cdc3c882b812d600441254fefc51b8ba406",
        "artifacts/summary.json": "3b094d51be29da6fc7609b802024027c743191e2bd6de1d0dde047a13eedb885",
        "artifacts/s3-final-inventory.txt": "6e050f1068279910cf0e5cebea9bf48246a0b07bc1f3bb970e76b5e78d08d859",
        "artifacts/lambda-pricing-api.json": "d3065b5c2944c0264d511dd7a19ef7b057d8daed9252132b8255b846da2a9afe",
        "artifacts/support-pricing-api.json": "0d44e533a503fada05cfa4e66c115e97ac66be4ee3d8c1bba24a32aff1e84b87"
      },
      "notes": [
        "Separate synthetic 300-document trial, with deterministic in-container embeddings and a synthetic identity service. Real provider and identity latency are additional.",
        "Warm execution and billed duration are Lambda-reported medians over 30 searches; client-observed warm latency is the varied-query median over 15 requests.",
        "All three cold client observations are retained, including the 35.756-second outlier. Its Lambda-reported billed duration was 5.494 seconds; the additional client delay was not isolated.",
        "The warm-only model is a billing projection. The alternative assumes 1% cold invocations, each billed for 5.494 seconds (the maximum billed duration among the three cold trials), plus 50 additional S3 reads. Neither cold fraction nor higher-volume capacity was measured.",
        "Index storage uses the captured trial-prefix inventory: 60 objects, 1,076,222 bytes. Storage is modelled in 1,024^3-byte GB-months.",
        "No free tier, discount or provisioned concurrency. The trial used 512 MiB ephemeral storage and reserved concurrency two; a four-request cold burst produced two successes and two throttles."
      ],
      "excluded": [
        "scheduled writer",
        "real identity service",
        "embedding and reranking calls",
        "logs",
        "container-image storage",
        "secret retrieval",
        "network egress",
        "gateway or custom domain",
        "taxes"
      ]
    },
    "hardware": {
      "cpu_model": "12th Gen Intel(R) Core(TM) i9-12900H",
      "logical_cpus": 20,
      "physical_cores": 14,
      "sockets": 1,
      "architecture": "x86_64",
      "mem_total_bytes": 32830681088,
      "swap_total_bytes": 17179860992,
      "platform": "Linux-7.0.0-42-generic-x86_64-with-glibc2.43",
      "python": "3.14.4 (main, Aug 20 2026, 10:41:58) [GCC 15.2.0]",
      "schema_version": "1",
      "scope": "Shared local host for native fyrr, Onyx and PipesHub; AWS Lambda is a separate deployment",
      "memory_gib": 30.57595443725586,
      "cpu_note": "Hybrid CPU:14 physical cores /20 logical CPUs. Earlier20-core shorthand means20 logical CPUs, not20 physical cores. No GPU used for retrieval; embeddings hosted by Voyage.",
      "captured_sources": [
        {
          "path": "runtime/fyrr-release-full/host-latency.json",
          "captured_at": "2026-11-17T05:56:34.538589+00:00",
          "original_sha256": "6875be05a8e34e4c3bff598a6e89482d7cacc25394bd6ffb8452d3b717755c20",
          "archive_member_sha256": "fd004f877b65ad8e6902d3f694e2dcb72ae06641dddb0bb53a006fffbb508fa1"
        },
        {
          "path": "runtime/onyx-full/host-latency.json",
          "captured_at": "2026-11-17T03:14:18.427139+00:00",
          "original_sha256": "563aa5d6d499ead0235590cb4d3a6d3e29ad838cb9dfb9731980637901d8a26f",
          "archive_member_sha256": "1e428bfe4f4388717d6ce6b50cff55bc37d64d90fdad538a0246212d015bdc1c"
        }
      ],
      "handoff_observations": {
        "observed_at": "2026-11-17T10:09:30.882370+00:00",
        "os_release_pretty": "Ubuntu 26.04.1 LTS",
        "workspace_filesystem": "ext4 on LVM",
        "storage_device_observation": "Non-rotating EXT X200E SSD exposed as NVMe; host observation at handoff, not a timed storage benchmark",
        "source_commands": [
          "cat /etc/os-release",
          "findmnt -n -T benchmark-repository -o FSTYPE,SOURCE",
          "lsblk -dn -o NAME,MODEL,ROTA,SIZE,TYPE"
        ]
      },
      "notes": [
        "Captured CPU/memory/kernel/Python invariants match across fyrr and Onyx timed windows.",
        "OS release and storage description are supplementary handoff observations, not historical timed samples.",
        "Primary timings use the host; cloud cost profiles were not measured on their priced instances.",
        "Pinned2/4/8 CPU sets exclude SMT siblings; cloud vCPU/core performance equivalence is not established.",
        "Hostnames, usernames, mount paths, device serials and unowned process names are excluded."
      ],
      "aws_lambda": {
        "region": "eu-central-1",
        "architecture": "x86_64",
        "memory_mb": [
          4096,
          2048
        ],
        "details": "Per-system Lambda metadata and results/metadata/lambda/deployment-evidence.json"
      },
      "metadata_sha256": "03da50bd0f7d26abd83b640fbc8d58aad9ab8ae1fae51103d57adfb6a3dd1fc5"
    },
    "dataset_license": {
      "label": "EnterpriseRAG dataset attribution · MIT license",
      "notice": "MIT License\n\nCopyright (c) 2026 DanswerAI, Inc.\n\nPermission is hereby granted, free of charge, to any person obtaining a copy\nof this software and associated documentation files (the \"Software\"), to deal\nin the Software without restriction, including without limitation the rights\nto use, copy, modify, merge, publish, distribute, sublicense, and/or sell\ncopies of the Software, and to permit persons to whom the Software is\nfurnished to do so, subject to the following conditions:\n\nThe above copyright notice and this permission notice shall be included in all\ncopies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\nIMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\nFITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\nAUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\nLIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\nOUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\nSOFTWARE.\n",
      "source_url": "https://huggingface.co/datasets/onyx-dot-app/EnterpriseRAG-Bench/blob/main/README.md"
    }
  },
  "systems": [
    {
      "id": "fyrr",
      "variant": "default",
      "name": "fyrr",
      "version": {
        "version": "0.1.0",
        "commit": "38e7df7198c52f340da1737dca6af665acc69b96",
        "binary_sha256": "9d9c76a6d0fe902535d7f66e5f387dd857a2de6be4f2edc3af4aa6629167640b",
        "build_profile": {
          "name": "release",
          "strip": "symbols",
          "debug": false,
          "lto": "fat",
          "codegen-units": 1
        },
        "cargo_lock_sha256": "242eee8b42e224135af6d096d7bc82a7523cb68efbec9ab17d38441df6757a1d",
        "reader_config_sha256": "f3da6403e6a57e7e2d7b3b5105be61771d176fe0ecb5c643629e3d28ad7a6855",
        "writer_config_sha256": "ba9ebf532f14939ca8d5bdfc64534ccdaf18031c7864739a8f766a74f065143c",
        "source_metadata": {
          "source_url": "https://github.com/fyrr-ai/fyrr",
          "commit": "38e7df7198c52f340da1737dca6af665acc69b96",
          "release_branch": "main",
          "integration_branch": "ci/faster",
          "config_schema_version": "4.0",
          "profile": {
            "strip": "symbols",
            "debug": false,
            "lto": "fat",
            "codegen-units": 1
          },
          "coordinator_go": "5b1d1452-5fef-452f-8c67-cc92d9ca2c3b",
          "upstream_validation": "Coordinator reports quick/full CI green; benchmark agent source CI n/e (unmodified checkout)",
          "cargo_lock_sha256": "242eee8b42e224135af6d096d7bc82a7523cb68efbec9ab17d38441df6757a1d",
          "build_environment": {
            "compiler_profile_overrides": {
              "RUSTFLAGS": false,
              "CARGO_ENCODED_RUSTFLAGS": false,
              "CARGO_PROFILE_RELEASE_LTO": false,
              "CARGO_PROFILE_RELEASE_CODEGEN_UNITS": false
            },
            "rustup_cargo": "${EXTERNAL_PATH}",
            "cargo_version": "cargo 1.97.1 (c980f4866 2026-06-30)",
            "rustc_version": "rustc 1.97.1 (8bab26f4f 2026-07-14)",
            "rustc_wrapper_path": "${EXTERNAL_PATH}",
            "rustc_wrapper_sha256": "311701adf0fd55e78ccd418b63290ce858f2bf515d4d766de9e42e125a908165",
            "wrapper_behavior": "Previously inspected host wrapper delegates unchanged for this /home checkout"
          }
        }
      },
      "config": {
        "binary": "${BENCH_ROOT}/fyrr-src/target/release/fyrr",
        "fyrr_config": "${BENCH_ROOT}/runtime/fyrr-release-full/reader.toml",
        "writer_config": "${BENCH_ROOT}/runtime/fyrr-release-full/fyrr.toml",
        "runtime_dir": "${BENCH_ROOT}/runtime/fyrr-release-full",
        "source_dir": "${BENCH_ROOT}/fyrr-src",
        "python_library_dir": "${EXTERNAL_PATH}",
        "smoke_only": false,
        "namespace": "erag-compare-release",
        "writer_sync_timeout": "60m",
        "model": "voyage-4",
        "dimensions": 1024,
        "document_input_type": "document",
        "query_input_type": "query",
        "truncation": "provider default true",
        "prefixes": "none",
        "chunk_max_utf8_bytes": 16384,
        "overlap_bytes": 0,
        "reranker": null,
        "reader_concurrency": 2,
        "search_tuning": "documented defaults",
        "embedding_batch_size": 32,
        "embedding_timeout": "30s",
        "embedding_max_retries": 3,
        "embedding_writer_max_call_microusd": 32000,
        "embedding_writer_reservation_limit_microusd": 25000000,
        "embedding_reader_max_call_microusd": 100,
        "embedding_reader_reservation_limit_microusd": 100000,
        "budget": {
          "product": "fyrr",
          "ledger": "${BENCH_ROOT}/runtime/budgets/voyage-fyrr.json",
          "ingest_microusd": 25000000,
          "query_microusd": 200000
        }
      },
      "date": "2026-11-17T05:43:10.578697+00:00",
      "measurement_windows": {
        "phase_host_dates": {
          "ingest": "2026-11-17T05:43:10.578697+00:00",
          "latency": "2026-11-17T05:56:34.538589+00:00",
          "quality": "2026-11-17T05:54:26.061375+00:00"
        },
        "first_ask": {
          "1": {
            "seconds": 73.38086427212693,
            "attempted": 250,
            "concurrency": 1,
            "started_at": 1794895026.181897,
            "ended_at": 1794895099.5627606,
            "provider_probe_scheduled": 11,
            "probe_method": "Separate single provider client; at start and each25 completions; pending probes drain outside product timing"
          },
          "8": {
            "seconds": 0.731339514022693,
            "attempted": 250,
            "concurrency": 8,
            "started_at": 1794895136.4269497,
            "ended_at": 1794895137.1582878,
            "provider_probe_scheduled": 11,
            "probe_method": "Separate single provider client; at start and each25 completions; pending probes drain outside product timing"
          }
        },
        "repeat": {
          "1": {
            "seconds": 3.872576412046328,
            "attempted": 250,
            "concurrency": 1,
            "started_at": 1794895099.827459,
            "ended_at": 1794895103.700036,
            "provider_probe_scheduled": 0,
            "probe_method": "Separate single provider client; at start and each25 completions; pending probes drain outside product timing"
          },
          "8": {
            "seconds": 0.7487898361869156,
            "attempted": 250,
            "concurrency": 8,
            "started_at": 1794895141.906068,
            "ended_at": 1794895142.6548584,
            "provider_probe_scheduled": 0,
            "probe_method": "Separate single provider client; at start and each25 completions; pending probes drain outside product timing"
          }
        },
        "ingest_clock": {
          "first_submission_monotonic": 1276690.605401865,
          "first_submission_wall": 1794894192.7234895,
          "completed_monotonic": 1277360.701620344,
          "completed_wall": 1794894862.8197064,
          "wall_seconds": 670.0962180199567
        }
      },
      "status": "partial",
      "metrics": {
        "recall_at_1": {
          "status": "measured",
          "value": 0.7052600472813239,
          "unit": "fraction",
          "n": 235,
          "attempted": 235,
          "expected": 235,
          "missing": 0,
          "ci95": [
            0.6505667848699763,
            0.7571997635933806
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175",
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299",
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340",
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377",
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410",
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428",
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449",
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ],
          "artifact": "runtime/fyrr-release-full/quality.jsonl"
        },
        "recall_at_5": {
          "status": "measured",
          "value": 0.8851654846335697,
          "unit": "fraction",
          "n": 235,
          "attempted": 235,
          "expected": 235,
          "missing": 0,
          "ci95": [
            0.8470555555555556,
            0.9195035460992909
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175",
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299",
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340",
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377",
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410",
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428",
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449",
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ],
          "artifact": "runtime/fyrr-release-full/quality.jsonl"
        },
        "recall_at_10": {
          "status": "measured",
          "value": 0.9478723404255319,
          "unit": "fraction",
          "n": 235,
          "attempted": 235,
          "expected": 235,
          "missing": 0,
          "ci95": [
            0.9231906028368794,
            0.9699175531914893
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175",
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299",
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340",
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377",
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410",
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428",
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449",
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ],
          "artifact": "runtime/fyrr-release-full/quality.jsonl"
        },
        "recall_at_20": {
          "status": "measured",
          "value": 0.9614775413711584,
          "unit": "fraction",
          "n": 235,
          "attempted": 235,
          "expected": 235,
          "missing": 0,
          "ci95": [
            0.939442671394799,
            0.9808882978723403
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175",
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299",
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340",
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377",
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410",
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428",
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449",
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ],
          "artifact": "runtime/fyrr-release-full/quality.jsonl"
        },
        "recall_at_50": {
          "status": "measured",
          "value": 0.9884633569739952,
          "unit": "fraction",
          "n": 235,
          "attempted": 235,
          "expected": 235,
          "missing": 0,
          "ci95": [
            0.9756973995271868,
            0.9981087470449173
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175",
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299",
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340",
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377",
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410",
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428",
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449",
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ],
          "artifact": "runtime/fyrr-release-full/quality.jsonl"
        },
        "ndcg_at_10": {
          "status": "measured",
          "value": 0.8764508986376564,
          "unit": "fraction",
          "n": 235,
          "attempted": 235,
          "expected": 235,
          "missing": 0,
          "ci95": [
            0.8444762114305789,
            0.9063002496400836
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175",
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299",
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340",
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377",
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410",
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428",
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449",
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ],
          "artifact": "runtime/fyrr-release-full/quality.jsonl"
        },
        "first_ask_c1_failure_rate": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/latency.jsonl"
        },
        "first_ask_c1_p50_ms": {
          "status": "measured",
          "value": 263.17194756120443,
          "ci95": [
            261.51396706700325,
            265.08648600429296
          ],
          "unit": "ms",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "sample_ids": [
            "first_ask-c1-qst_0002",
            "first_ask-c1-qst_0004",
            "first_ask-c1-qst_0008",
            "first_ask-c1-qst_0009",
            "first_ask-c1-qst_0010",
            "first_ask-c1-qst_0011",
            "first_ask-c1-qst_0012",
            "first_ask-c1-qst_0013",
            "first_ask-c1-qst_0015",
            "first_ask-c1-qst_0016",
            "first_ask-c1-qst_0022",
            "first_ask-c1-qst_0026",
            "first_ask-c1-qst_0028",
            "first_ask-c1-qst_0030",
            "first_ask-c1-qst_0032",
            "first_ask-c1-qst_0033",
            "first_ask-c1-qst_0035",
            "first_ask-c1-qst_0040",
            "first_ask-c1-qst_0041",
            "first_ask-c1-qst_0043",
            "first_ask-c1-qst_0044",
            "first_ask-c1-qst_0046",
            "first_ask-c1-qst_0048",
            "first_ask-c1-qst_0049",
            "first_ask-c1-qst_0050",
            "first_ask-c1-qst_0051",
            "first_ask-c1-qst_0052",
            "first_ask-c1-qst_0053",
            "first_ask-c1-qst_0055",
            "first_ask-c1-qst_0059",
            "first_ask-c1-qst_0063",
            "first_ask-c1-qst_0064",
            "first_ask-c1-qst_0065",
            "first_ask-c1-qst_0066",
            "first_ask-c1-qst_0067",
            "first_ask-c1-qst_0069",
            "first_ask-c1-qst_0071",
            "first_ask-c1-qst_0074",
            "first_ask-c1-qst_0077",
            "first_ask-c1-qst_0079",
            "first_ask-c1-qst_0082",
            "first_ask-c1-qst_0083",
            "first_ask-c1-qst_0085",
            "first_ask-c1-qst_0086",
            "first_ask-c1-qst_0088",
            "first_ask-c1-qst_0089",
            "first_ask-c1-qst_0092",
            "first_ask-c1-qst_0096",
            "first_ask-c1-qst_0098",
            "first_ask-c1-qst_0100",
            "first_ask-c1-qst_0102",
            "first_ask-c1-qst_0103",
            "first_ask-c1-qst_0104",
            "first_ask-c1-qst_0105",
            "first_ask-c1-qst_0106",
            "first_ask-c1-qst_0111",
            "first_ask-c1-qst_0117",
            "first_ask-c1-qst_0122",
            "first_ask-c1-qst_0124",
            "first_ask-c1-qst_0125",
            "first_ask-c1-qst_0127",
            "first_ask-c1-qst_0129",
            "first_ask-c1-qst_0130",
            "first_ask-c1-qst_0132",
            "first_ask-c1-qst_0134",
            "first_ask-c1-qst_0138",
            "first_ask-c1-qst_0141",
            "first_ask-c1-qst_0145",
            "first_ask-c1-qst_0147",
            "first_ask-c1-qst_0148",
            "first_ask-c1-qst_0152",
            "first_ask-c1-qst_0153",
            "first_ask-c1-qst_0154",
            "first_ask-c1-qst_0155",
            "first_ask-c1-qst_0157",
            "first_ask-c1-qst_0158",
            "first_ask-c1-qst_0161",
            "first_ask-c1-qst_0162",
            "first_ask-c1-qst_0163",
            "first_ask-c1-qst_0165",
            "first_ask-c1-qst_0166",
            "first_ask-c1-qst_0167",
            "first_ask-c1-qst_0168",
            "first_ask-c1-qst_0169",
            "first_ask-c1-qst_0172",
            "first_ask-c1-qst_0173",
            "first_ask-c1-qst_0174",
            "first_ask-c1-qst_0175",
            "first_ask-c1-qst_0176",
            "first_ask-c1-qst_0177",
            "first_ask-c1-qst_0180",
            "first_ask-c1-qst_0181",
            "first_ask-c1-qst_0182",
            "first_ask-c1-qst_0186",
            "first_ask-c1-qst_0187",
            "first_ask-c1-qst_0189",
            "first_ask-c1-qst_0192",
            "first_ask-c1-qst_0195",
            "first_ask-c1-qst_0198",
            "first_ask-c1-qst_0199",
            "first_ask-c1-qst_0200",
            "first_ask-c1-qst_0209",
            "first_ask-c1-qst_0212",
            "first_ask-c1-qst_0213",
            "first_ask-c1-qst_0215",
            "first_ask-c1-qst_0216",
            "first_ask-c1-qst_0217",
            "first_ask-c1-qst_0218",
            "first_ask-c1-qst_0221",
            "first_ask-c1-qst_0223",
            "first_ask-c1-qst_0224",
            "first_ask-c1-qst_0226",
            "first_ask-c1-qst_0228",
            "first_ask-c1-qst_0229",
            "first_ask-c1-qst_0230",
            "first_ask-c1-qst_0231",
            "first_ask-c1-qst_0235",
            "first_ask-c1-qst_0238",
            "first_ask-c1-qst_0239",
            "first_ask-c1-qst_0241",
            "first_ask-c1-qst_0245",
            "first_ask-c1-qst_0247",
            "first_ask-c1-qst_0250",
            "first_ask-c1-qst_0252",
            "first_ask-c1-qst_0253",
            "first_ask-c1-qst_0254",
            "first_ask-c1-qst_0255",
            "first_ask-c1-qst_0256",
            "first_ask-c1-qst_0257",
            "first_ask-c1-qst_0260",
            "first_ask-c1-qst_0262",
            "first_ask-c1-qst_0265",
            "first_ask-c1-qst_0270",
            "first_ask-c1-qst_0271",
            "first_ask-c1-qst_0274",
            "first_ask-c1-qst_0275",
            "first_ask-c1-qst_0277",
            "first_ask-c1-qst_0280",
            "first_ask-c1-qst_0281",
            "first_ask-c1-qst_0283",
            "first_ask-c1-qst_0284",
            "first_ask-c1-qst_0285",
            "first_ask-c1-qst_0286",
            "first_ask-c1-qst_0289",
            "first_ask-c1-qst_0290",
            "first_ask-c1-qst_0291",
            "first_ask-c1-qst_0293",
            "first_ask-c1-qst_0295",
            "first_ask-c1-qst_0296",
            "first_ask-c1-qst_0299",
            "first_ask-c1-qst_0301",
            "first_ask-c1-qst_0302",
            "first_ask-c1-qst_0304",
            "first_ask-c1-qst_0305",
            "first_ask-c1-qst_0306",
            "first_ask-c1-qst_0307",
            "first_ask-c1-qst_0309",
            "first_ask-c1-qst_0314",
            "first_ask-c1-qst_0315",
            "first_ask-c1-qst_0317",
            "first_ask-c1-qst_0322",
            "first_ask-c1-qst_0323",
            "first_ask-c1-qst_0324",
            "first_ask-c1-qst_0331",
            "first_ask-c1-qst_0332",
            "first_ask-c1-qst_0333",
            "first_ask-c1-qst_0334",
            "first_ask-c1-qst_0336",
            "first_ask-c1-qst_0339",
            "first_ask-c1-qst_0340",
            "first_ask-c1-qst_0341",
            "first_ask-c1-qst_0342",
            "first_ask-c1-qst_0343",
            "first_ask-c1-qst_0345",
            "first_ask-c1-qst_0346",
            "first_ask-c1-qst_0347",
            "first_ask-c1-qst_0349",
            "first_ask-c1-qst_0350",
            "first_ask-c1-qst_0354",
            "first_ask-c1-qst_0355",
            "first_ask-c1-qst_0356",
            "first_ask-c1-qst_0360",
            "first_ask-c1-qst_0362",
            "first_ask-c1-qst_0363",
            "first_ask-c1-qst_0365",
            "first_ask-c1-qst_0367",
            "first_ask-c1-qst_0369",
            "first_ask-c1-qst_0375",
            "first_ask-c1-qst_0376",
            "first_ask-c1-qst_0377",
            "first_ask-c1-qst_0381",
            "first_ask-c1-qst_0386",
            "first_ask-c1-qst_0389",
            "first_ask-c1-qst_0390",
            "first_ask-c1-qst_0391",
            "first_ask-c1-qst_0392",
            "first_ask-c1-qst_0393",
            "first_ask-c1-qst_0394",
            "first_ask-c1-qst_0395",
            "first_ask-c1-qst_0396",
            "first_ask-c1-qst_0399",
            "first_ask-c1-qst_0402",
            "first_ask-c1-qst_0404",
            "first_ask-c1-qst_0407",
            "first_ask-c1-qst_0410",
            "first_ask-c1-qst_0411",
            "first_ask-c1-qst_0412",
            "first_ask-c1-qst_0414",
            "first_ask-c1-qst_0417",
            "first_ask-c1-qst_0419",
            "first_ask-c1-qst_0420",
            "first_ask-c1-qst_0422",
            "first_ask-c1-qst_0423",
            "first_ask-c1-qst_0426",
            "first_ask-c1-qst_0428",
            "first_ask-c1-qst_0434",
            "first_ask-c1-qst_0435",
            "first_ask-c1-qst_0436",
            "first_ask-c1-qst_0438",
            "first_ask-c1-qst_0439",
            "first_ask-c1-qst_0442",
            "first_ask-c1-qst_0443",
            "first_ask-c1-qst_0445",
            "first_ask-c1-qst_0447",
            "first_ask-c1-qst_0449",
            "first_ask-c1-qst_0451",
            "first_ask-c1-qst_0452",
            "first_ask-c1-qst_0455",
            "first_ask-c1-qst_0460",
            "first_ask-c1-qst_0461",
            "first_ask-c1-qst_0463",
            "first_ask-c1-qst_0464",
            "first_ask-c1-qst_0465",
            "first_ask-c1-qst_0467",
            "first_ask-c1-qst_0470",
            "first_ask-c1-qst_0472",
            "first_ask-c1-qst_0476",
            "first_ask-c1-qst_0477",
            "first_ask-c1-qst_0478",
            "first_ask-c1-qst_0479",
            "first_ask-c1-qst_0484",
            "first_ask-c1-qst_0486",
            "first_ask-c1-qst_0487",
            "first_ask-c1-qst_0488",
            "first_ask-c1-qst_0492",
            "first_ask-c1-qst_0495",
            "first_ask-c1-qst_0496",
            "first_ask-c1-qst_0497",
            "first_ask-c1-qst_0499",
            "first_ask-c1-qst_0500"
          ],
          "observed_valid": 250,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 3.4068827408859
        },
        "first_ask_c1_p95_ms": {
          "status": "measured",
          "value": 514.7065628436387,
          "ci95": [
            327.6849381509235,
            624.1304608993232
          ],
          "unit": "ms",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "sample_ids": [
            "first_ask-c1-qst_0002",
            "first_ask-c1-qst_0004",
            "first_ask-c1-qst_0008",
            "first_ask-c1-qst_0009",
            "first_ask-c1-qst_0010",
            "first_ask-c1-qst_0011",
            "first_ask-c1-qst_0012",
            "first_ask-c1-qst_0013",
            "first_ask-c1-qst_0015",
            "first_ask-c1-qst_0016",
            "first_ask-c1-qst_0022",
            "first_ask-c1-qst_0026",
            "first_ask-c1-qst_0028",
            "first_ask-c1-qst_0030",
            "first_ask-c1-qst_0032",
            "first_ask-c1-qst_0033",
            "first_ask-c1-qst_0035",
            "first_ask-c1-qst_0040",
            "first_ask-c1-qst_0041",
            "first_ask-c1-qst_0043",
            "first_ask-c1-qst_0044",
            "first_ask-c1-qst_0046",
            "first_ask-c1-qst_0048",
            "first_ask-c1-qst_0049",
            "first_ask-c1-qst_0050",
            "first_ask-c1-qst_0051",
            "first_ask-c1-qst_0052",
            "first_ask-c1-qst_0053",
            "first_ask-c1-qst_0055",
            "first_ask-c1-qst_0059",
            "first_ask-c1-qst_0063",
            "first_ask-c1-qst_0064",
            "first_ask-c1-qst_0065",
            "first_ask-c1-qst_0066",
            "first_ask-c1-qst_0067",
            "first_ask-c1-qst_0069",
            "first_ask-c1-qst_0071",
            "first_ask-c1-qst_0074",
            "first_ask-c1-qst_0077",
            "first_ask-c1-qst_0079",
            "first_ask-c1-qst_0082",
            "first_ask-c1-qst_0083",
            "first_ask-c1-qst_0085",
            "first_ask-c1-qst_0086",
            "first_ask-c1-qst_0088",
            "first_ask-c1-qst_0089",
            "first_ask-c1-qst_0092",
            "first_ask-c1-qst_0096",
            "first_ask-c1-qst_0098",
            "first_ask-c1-qst_0100",
            "first_ask-c1-qst_0102",
            "first_ask-c1-qst_0103",
            "first_ask-c1-qst_0104",
            "first_ask-c1-qst_0105",
            "first_ask-c1-qst_0106",
            "first_ask-c1-qst_0111",
            "first_ask-c1-qst_0117",
            "first_ask-c1-qst_0122",
            "first_ask-c1-qst_0124",
            "first_ask-c1-qst_0125",
            "first_ask-c1-qst_0127",
            "first_ask-c1-qst_0129",
            "first_ask-c1-qst_0130",
            "first_ask-c1-qst_0132",
            "first_ask-c1-qst_0134",
            "first_ask-c1-qst_0138",
            "first_ask-c1-qst_0141",
            "first_ask-c1-qst_0145",
            "first_ask-c1-qst_0147",
            "first_ask-c1-qst_0148",
            "first_ask-c1-qst_0152",
            "first_ask-c1-qst_0153",
            "first_ask-c1-qst_0154",
            "first_ask-c1-qst_0155",
            "first_ask-c1-qst_0157",
            "first_ask-c1-qst_0158",
            "first_ask-c1-qst_0161",
            "first_ask-c1-qst_0162",
            "first_ask-c1-qst_0163",
            "first_ask-c1-qst_0165",
            "first_ask-c1-qst_0166",
            "first_ask-c1-qst_0167",
            "first_ask-c1-qst_0168",
            "first_ask-c1-qst_0169",
            "first_ask-c1-qst_0172",
            "first_ask-c1-qst_0173",
            "first_ask-c1-qst_0174",
            "first_ask-c1-qst_0175",
            "first_ask-c1-qst_0176",
            "first_ask-c1-qst_0177",
            "first_ask-c1-qst_0180",
            "first_ask-c1-qst_0181",
            "first_ask-c1-qst_0182",
            "first_ask-c1-qst_0186",
            "first_ask-c1-qst_0187",
            "first_ask-c1-qst_0189",
            "first_ask-c1-qst_0192",
            "first_ask-c1-qst_0195",
            "first_ask-c1-qst_0198",
            "first_ask-c1-qst_0199",
            "first_ask-c1-qst_0200",
            "first_ask-c1-qst_0209",
            "first_ask-c1-qst_0212",
            "first_ask-c1-qst_0213",
            "first_ask-c1-qst_0215",
            "first_ask-c1-qst_0216",
            "first_ask-c1-qst_0217",
            "first_ask-c1-qst_0218",
            "first_ask-c1-qst_0221",
            "first_ask-c1-qst_0223",
            "first_ask-c1-qst_0224",
            "first_ask-c1-qst_0226",
            "first_ask-c1-qst_0228",
            "first_ask-c1-qst_0229",
            "first_ask-c1-qst_0230",
            "first_ask-c1-qst_0231",
            "first_ask-c1-qst_0235",
            "first_ask-c1-qst_0238",
            "first_ask-c1-qst_0239",
            "first_ask-c1-qst_0241",
            "first_ask-c1-qst_0245",
            "first_ask-c1-qst_0247",
            "first_ask-c1-qst_0250",
            "first_ask-c1-qst_0252",
            "first_ask-c1-qst_0253",
            "first_ask-c1-qst_0254",
            "first_ask-c1-qst_0255",
            "first_ask-c1-qst_0256",
            "first_ask-c1-qst_0257",
            "first_ask-c1-qst_0260",
            "first_ask-c1-qst_0262",
            "first_ask-c1-qst_0265",
            "first_ask-c1-qst_0270",
            "first_ask-c1-qst_0271",
            "first_ask-c1-qst_0274",
            "first_ask-c1-qst_0275",
            "first_ask-c1-qst_0277",
            "first_ask-c1-qst_0280",
            "first_ask-c1-qst_0281",
            "first_ask-c1-qst_0283",
            "first_ask-c1-qst_0284",
            "first_ask-c1-qst_0285",
            "first_ask-c1-qst_0286",
            "first_ask-c1-qst_0289",
            "first_ask-c1-qst_0290",
            "first_ask-c1-qst_0291",
            "first_ask-c1-qst_0293",
            "first_ask-c1-qst_0295",
            "first_ask-c1-qst_0296",
            "first_ask-c1-qst_0299",
            "first_ask-c1-qst_0301",
            "first_ask-c1-qst_0302",
            "first_ask-c1-qst_0304",
            "first_ask-c1-qst_0305",
            "first_ask-c1-qst_0306",
            "first_ask-c1-qst_0307",
            "first_ask-c1-qst_0309",
            "first_ask-c1-qst_0314",
            "first_ask-c1-qst_0315",
            "first_ask-c1-qst_0317",
            "first_ask-c1-qst_0322",
            "first_ask-c1-qst_0323",
            "first_ask-c1-qst_0324",
            "first_ask-c1-qst_0331",
            "first_ask-c1-qst_0332",
            "first_ask-c1-qst_0333",
            "first_ask-c1-qst_0334",
            "first_ask-c1-qst_0336",
            "first_ask-c1-qst_0339",
            "first_ask-c1-qst_0340",
            "first_ask-c1-qst_0341",
            "first_ask-c1-qst_0342",
            "first_ask-c1-qst_0343",
            "first_ask-c1-qst_0345",
            "first_ask-c1-qst_0346",
            "first_ask-c1-qst_0347",
            "first_ask-c1-qst_0349",
            "first_ask-c1-qst_0350",
            "first_ask-c1-qst_0354",
            "first_ask-c1-qst_0355",
            "first_ask-c1-qst_0356",
            "first_ask-c1-qst_0360",
            "first_ask-c1-qst_0362",
            "first_ask-c1-qst_0363",
            "first_ask-c1-qst_0365",
            "first_ask-c1-qst_0367",
            "first_ask-c1-qst_0369",
            "first_ask-c1-qst_0375",
            "first_ask-c1-qst_0376",
            "first_ask-c1-qst_0377",
            "first_ask-c1-qst_0381",
            "first_ask-c1-qst_0386",
            "first_ask-c1-qst_0389",
            "first_ask-c1-qst_0390",
            "first_ask-c1-qst_0391",
            "first_ask-c1-qst_0392",
            "first_ask-c1-qst_0393",
            "first_ask-c1-qst_0394",
            "first_ask-c1-qst_0395",
            "first_ask-c1-qst_0396",
            "first_ask-c1-qst_0399",
            "first_ask-c1-qst_0402",
            "first_ask-c1-qst_0404",
            "first_ask-c1-qst_0407",
            "first_ask-c1-qst_0410",
            "first_ask-c1-qst_0411",
            "first_ask-c1-qst_0412",
            "first_ask-c1-qst_0414",
            "first_ask-c1-qst_0417",
            "first_ask-c1-qst_0419",
            "first_ask-c1-qst_0420",
            "first_ask-c1-qst_0422",
            "first_ask-c1-qst_0423",
            "first_ask-c1-qst_0426",
            "first_ask-c1-qst_0428",
            "first_ask-c1-qst_0434",
            "first_ask-c1-qst_0435",
            "first_ask-c1-qst_0436",
            "first_ask-c1-qst_0438",
            "first_ask-c1-qst_0439",
            "first_ask-c1-qst_0442",
            "first_ask-c1-qst_0443",
            "first_ask-c1-qst_0445",
            "first_ask-c1-qst_0447",
            "first_ask-c1-qst_0449",
            "first_ask-c1-qst_0451",
            "first_ask-c1-qst_0452",
            "first_ask-c1-qst_0455",
            "first_ask-c1-qst_0460",
            "first_ask-c1-qst_0461",
            "first_ask-c1-qst_0463",
            "first_ask-c1-qst_0464",
            "first_ask-c1-qst_0465",
            "first_ask-c1-qst_0467",
            "first_ask-c1-qst_0470",
            "first_ask-c1-qst_0472",
            "first_ask-c1-qst_0476",
            "first_ask-c1-qst_0477",
            "first_ask-c1-qst_0478",
            "first_ask-c1-qst_0479",
            "first_ask-c1-qst_0484",
            "first_ask-c1-qst_0486",
            "first_ask-c1-qst_0487",
            "first_ask-c1-qst_0488",
            "first_ask-c1-qst_0492",
            "first_ask-c1-qst_0495",
            "first_ask-c1-qst_0496",
            "first_ask-c1-qst_0497",
            "first_ask-c1-qst_0499",
            "first_ask-c1-qst_0500"
          ],
          "observed_valid": 250,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 3.4068827408859
        },
        "first_ask_c1_embedding_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "first_ask_c1_embedding_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "first_ask_c1_engine_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "first_ask_c1_engine_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c1_failure_rate": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/latency.jsonl"
        },
        "warm_c1_p50_ms": {
          "status": "measured",
          "value": 15.031314571388066,
          "ci95": [
            14.847045997157693,
            15.525329858064651
          ],
          "unit": "ms",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "sample_ids": [
            "repeat-c1-qst_0002",
            "repeat-c1-qst_0004",
            "repeat-c1-qst_0008",
            "repeat-c1-qst_0009",
            "repeat-c1-qst_0010",
            "repeat-c1-qst_0011",
            "repeat-c1-qst_0012",
            "repeat-c1-qst_0013",
            "repeat-c1-qst_0015",
            "repeat-c1-qst_0016",
            "repeat-c1-qst_0022",
            "repeat-c1-qst_0026",
            "repeat-c1-qst_0028",
            "repeat-c1-qst_0030",
            "repeat-c1-qst_0032",
            "repeat-c1-qst_0033",
            "repeat-c1-qst_0035",
            "repeat-c1-qst_0040",
            "repeat-c1-qst_0041",
            "repeat-c1-qst_0043",
            "repeat-c1-qst_0044",
            "repeat-c1-qst_0046",
            "repeat-c1-qst_0048",
            "repeat-c1-qst_0049",
            "repeat-c1-qst_0050",
            "repeat-c1-qst_0051",
            "repeat-c1-qst_0052",
            "repeat-c1-qst_0053",
            "repeat-c1-qst_0055",
            "repeat-c1-qst_0059",
            "repeat-c1-qst_0063",
            "repeat-c1-qst_0064",
            "repeat-c1-qst_0065",
            "repeat-c1-qst_0066",
            "repeat-c1-qst_0067",
            "repeat-c1-qst_0069",
            "repeat-c1-qst_0071",
            "repeat-c1-qst_0074",
            "repeat-c1-qst_0077",
            "repeat-c1-qst_0079",
            "repeat-c1-qst_0082",
            "repeat-c1-qst_0083",
            "repeat-c1-qst_0085",
            "repeat-c1-qst_0086",
            "repeat-c1-qst_0088",
            "repeat-c1-qst_0089",
            "repeat-c1-qst_0092",
            "repeat-c1-qst_0096",
            "repeat-c1-qst_0098",
            "repeat-c1-qst_0100",
            "repeat-c1-qst_0102",
            "repeat-c1-qst_0103",
            "repeat-c1-qst_0104",
            "repeat-c1-qst_0105",
            "repeat-c1-qst_0106",
            "repeat-c1-qst_0111",
            "repeat-c1-qst_0117",
            "repeat-c1-qst_0122",
            "repeat-c1-qst_0124",
            "repeat-c1-qst_0125",
            "repeat-c1-qst_0127",
            "repeat-c1-qst_0129",
            "repeat-c1-qst_0130",
            "repeat-c1-qst_0132",
            "repeat-c1-qst_0134",
            "repeat-c1-qst_0138",
            "repeat-c1-qst_0141",
            "repeat-c1-qst_0145",
            "repeat-c1-qst_0147",
            "repeat-c1-qst_0148",
            "repeat-c1-qst_0152",
            "repeat-c1-qst_0153",
            "repeat-c1-qst_0154",
            "repeat-c1-qst_0155",
            "repeat-c1-qst_0157",
            "repeat-c1-qst_0158",
            "repeat-c1-qst_0161",
            "repeat-c1-qst_0162",
            "repeat-c1-qst_0163",
            "repeat-c1-qst_0165",
            "repeat-c1-qst_0166",
            "repeat-c1-qst_0167",
            "repeat-c1-qst_0168",
            "repeat-c1-qst_0169",
            "repeat-c1-qst_0172",
            "repeat-c1-qst_0173",
            "repeat-c1-qst_0174",
            "repeat-c1-qst_0175",
            "repeat-c1-qst_0176",
            "repeat-c1-qst_0177",
            "repeat-c1-qst_0180",
            "repeat-c1-qst_0181",
            "repeat-c1-qst_0182",
            "repeat-c1-qst_0186",
            "repeat-c1-qst_0187",
            "repeat-c1-qst_0189",
            "repeat-c1-qst_0192",
            "repeat-c1-qst_0195",
            "repeat-c1-qst_0198",
            "repeat-c1-qst_0199",
            "repeat-c1-qst_0200",
            "repeat-c1-qst_0209",
            "repeat-c1-qst_0212",
            "repeat-c1-qst_0213",
            "repeat-c1-qst_0215",
            "repeat-c1-qst_0216",
            "repeat-c1-qst_0217",
            "repeat-c1-qst_0218",
            "repeat-c1-qst_0221",
            "repeat-c1-qst_0223",
            "repeat-c1-qst_0224",
            "repeat-c1-qst_0226",
            "repeat-c1-qst_0228",
            "repeat-c1-qst_0229",
            "repeat-c1-qst_0230",
            "repeat-c1-qst_0231",
            "repeat-c1-qst_0235",
            "repeat-c1-qst_0238",
            "repeat-c1-qst_0239",
            "repeat-c1-qst_0241",
            "repeat-c1-qst_0245",
            "repeat-c1-qst_0247",
            "repeat-c1-qst_0250",
            "repeat-c1-qst_0252",
            "repeat-c1-qst_0253",
            "repeat-c1-qst_0254",
            "repeat-c1-qst_0255",
            "repeat-c1-qst_0256",
            "repeat-c1-qst_0257",
            "repeat-c1-qst_0260",
            "repeat-c1-qst_0262",
            "repeat-c1-qst_0265",
            "repeat-c1-qst_0270",
            "repeat-c1-qst_0271",
            "repeat-c1-qst_0274",
            "repeat-c1-qst_0275",
            "repeat-c1-qst_0277",
            "repeat-c1-qst_0280",
            "repeat-c1-qst_0281",
            "repeat-c1-qst_0283",
            "repeat-c1-qst_0284",
            "repeat-c1-qst_0285",
            "repeat-c1-qst_0286",
            "repeat-c1-qst_0289",
            "repeat-c1-qst_0290",
            "repeat-c1-qst_0291",
            "repeat-c1-qst_0293",
            "repeat-c1-qst_0295",
            "repeat-c1-qst_0296",
            "repeat-c1-qst_0299",
            "repeat-c1-qst_0301",
            "repeat-c1-qst_0302",
            "repeat-c1-qst_0304",
            "repeat-c1-qst_0305",
            "repeat-c1-qst_0306",
            "repeat-c1-qst_0307",
            "repeat-c1-qst_0309",
            "repeat-c1-qst_0314",
            "repeat-c1-qst_0315",
            "repeat-c1-qst_0317",
            "repeat-c1-qst_0322",
            "repeat-c1-qst_0323",
            "repeat-c1-qst_0324",
            "repeat-c1-qst_0331",
            "repeat-c1-qst_0332",
            "repeat-c1-qst_0333",
            "repeat-c1-qst_0334",
            "repeat-c1-qst_0336",
            "repeat-c1-qst_0339",
            "repeat-c1-qst_0340",
            "repeat-c1-qst_0341",
            "repeat-c1-qst_0342",
            "repeat-c1-qst_0343",
            "repeat-c1-qst_0345",
            "repeat-c1-qst_0346",
            "repeat-c1-qst_0347",
            "repeat-c1-qst_0349",
            "repeat-c1-qst_0350",
            "repeat-c1-qst_0354",
            "repeat-c1-qst_0355",
            "repeat-c1-qst_0356",
            "repeat-c1-qst_0360",
            "repeat-c1-qst_0362",
            "repeat-c1-qst_0363",
            "repeat-c1-qst_0365",
            "repeat-c1-qst_0367",
            "repeat-c1-qst_0369",
            "repeat-c1-qst_0375",
            "repeat-c1-qst_0376",
            "repeat-c1-qst_0377",
            "repeat-c1-qst_0381",
            "repeat-c1-qst_0386",
            "repeat-c1-qst_0389",
            "repeat-c1-qst_0390",
            "repeat-c1-qst_0391",
            "repeat-c1-qst_0392",
            "repeat-c1-qst_0393",
            "repeat-c1-qst_0394",
            "repeat-c1-qst_0395",
            "repeat-c1-qst_0396",
            "repeat-c1-qst_0399",
            "repeat-c1-qst_0402",
            "repeat-c1-qst_0404",
            "repeat-c1-qst_0407",
            "repeat-c1-qst_0410",
            "repeat-c1-qst_0411",
            "repeat-c1-qst_0412",
            "repeat-c1-qst_0414",
            "repeat-c1-qst_0417",
            "repeat-c1-qst_0419",
            "repeat-c1-qst_0420",
            "repeat-c1-qst_0422",
            "repeat-c1-qst_0423",
            "repeat-c1-qst_0426",
            "repeat-c1-qst_0428",
            "repeat-c1-qst_0434",
            "repeat-c1-qst_0435",
            "repeat-c1-qst_0436",
            "repeat-c1-qst_0438",
            "repeat-c1-qst_0439",
            "repeat-c1-qst_0442",
            "repeat-c1-qst_0443",
            "repeat-c1-qst_0445",
            "repeat-c1-qst_0447",
            "repeat-c1-qst_0449",
            "repeat-c1-qst_0451",
            "repeat-c1-qst_0452",
            "repeat-c1-qst_0455",
            "repeat-c1-qst_0460",
            "repeat-c1-qst_0461",
            "repeat-c1-qst_0463",
            "repeat-c1-qst_0464",
            "repeat-c1-qst_0465",
            "repeat-c1-qst_0467",
            "repeat-c1-qst_0470",
            "repeat-c1-qst_0472",
            "repeat-c1-qst_0476",
            "repeat-c1-qst_0477",
            "repeat-c1-qst_0478",
            "repeat-c1-qst_0479",
            "repeat-c1-qst_0484",
            "repeat-c1-qst_0486",
            "repeat-c1-qst_0487",
            "repeat-c1-qst_0488",
            "repeat-c1-qst_0492",
            "repeat-c1-qst_0495",
            "repeat-c1-qst_0496",
            "repeat-c1-qst_0497",
            "repeat-c1-qst_0499",
            "repeat-c1-qst_0500"
          ],
          "observed_valid": 250,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 64.5565053854925
        },
        "warm_c1_p95_ms": {
          "status": "measured",
          "value": 18.434702663216736,
          "ci95": [
            17.842773906886578,
            19.29764007218182
          ],
          "unit": "ms",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "sample_ids": [
            "repeat-c1-qst_0002",
            "repeat-c1-qst_0004",
            "repeat-c1-qst_0008",
            "repeat-c1-qst_0009",
            "repeat-c1-qst_0010",
            "repeat-c1-qst_0011",
            "repeat-c1-qst_0012",
            "repeat-c1-qst_0013",
            "repeat-c1-qst_0015",
            "repeat-c1-qst_0016",
            "repeat-c1-qst_0022",
            "repeat-c1-qst_0026",
            "repeat-c1-qst_0028",
            "repeat-c1-qst_0030",
            "repeat-c1-qst_0032",
            "repeat-c1-qst_0033",
            "repeat-c1-qst_0035",
            "repeat-c1-qst_0040",
            "repeat-c1-qst_0041",
            "repeat-c1-qst_0043",
            "repeat-c1-qst_0044",
            "repeat-c1-qst_0046",
            "repeat-c1-qst_0048",
            "repeat-c1-qst_0049",
            "repeat-c1-qst_0050",
            "repeat-c1-qst_0051",
            "repeat-c1-qst_0052",
            "repeat-c1-qst_0053",
            "repeat-c1-qst_0055",
            "repeat-c1-qst_0059",
            "repeat-c1-qst_0063",
            "repeat-c1-qst_0064",
            "repeat-c1-qst_0065",
            "repeat-c1-qst_0066",
            "repeat-c1-qst_0067",
            "repeat-c1-qst_0069",
            "repeat-c1-qst_0071",
            "repeat-c1-qst_0074",
            "repeat-c1-qst_0077",
            "repeat-c1-qst_0079",
            "repeat-c1-qst_0082",
            "repeat-c1-qst_0083",
            "repeat-c1-qst_0085",
            "repeat-c1-qst_0086",
            "repeat-c1-qst_0088",
            "repeat-c1-qst_0089",
            "repeat-c1-qst_0092",
            "repeat-c1-qst_0096",
            "repeat-c1-qst_0098",
            "repeat-c1-qst_0100",
            "repeat-c1-qst_0102",
            "repeat-c1-qst_0103",
            "repeat-c1-qst_0104",
            "repeat-c1-qst_0105",
            "repeat-c1-qst_0106",
            "repeat-c1-qst_0111",
            "repeat-c1-qst_0117",
            "repeat-c1-qst_0122",
            "repeat-c1-qst_0124",
            "repeat-c1-qst_0125",
            "repeat-c1-qst_0127",
            "repeat-c1-qst_0129",
            "repeat-c1-qst_0130",
            "repeat-c1-qst_0132",
            "repeat-c1-qst_0134",
            "repeat-c1-qst_0138",
            "repeat-c1-qst_0141",
            "repeat-c1-qst_0145",
            "repeat-c1-qst_0147",
            "repeat-c1-qst_0148",
            "repeat-c1-qst_0152",
            "repeat-c1-qst_0153",
            "repeat-c1-qst_0154",
            "repeat-c1-qst_0155",
            "repeat-c1-qst_0157",
            "repeat-c1-qst_0158",
            "repeat-c1-qst_0161",
            "repeat-c1-qst_0162",
            "repeat-c1-qst_0163",
            "repeat-c1-qst_0165",
            "repeat-c1-qst_0166",
            "repeat-c1-qst_0167",
            "repeat-c1-qst_0168",
            "repeat-c1-qst_0169",
            "repeat-c1-qst_0172",
            "repeat-c1-qst_0173",
            "repeat-c1-qst_0174",
            "repeat-c1-qst_0175",
            "repeat-c1-qst_0176",
            "repeat-c1-qst_0177",
            "repeat-c1-qst_0180",
            "repeat-c1-qst_0181",
            "repeat-c1-qst_0182",
            "repeat-c1-qst_0186",
            "repeat-c1-qst_0187",
            "repeat-c1-qst_0189",
            "repeat-c1-qst_0192",
            "repeat-c1-qst_0195",
            "repeat-c1-qst_0198",
            "repeat-c1-qst_0199",
            "repeat-c1-qst_0200",
            "repeat-c1-qst_0209",
            "repeat-c1-qst_0212",
            "repeat-c1-qst_0213",
            "repeat-c1-qst_0215",
            "repeat-c1-qst_0216",
            "repeat-c1-qst_0217",
            "repeat-c1-qst_0218",
            "repeat-c1-qst_0221",
            "repeat-c1-qst_0223",
            "repeat-c1-qst_0224",
            "repeat-c1-qst_0226",
            "repeat-c1-qst_0228",
            "repeat-c1-qst_0229",
            "repeat-c1-qst_0230",
            "repeat-c1-qst_0231",
            "repeat-c1-qst_0235",
            "repeat-c1-qst_0238",
            "repeat-c1-qst_0239",
            "repeat-c1-qst_0241",
            "repeat-c1-qst_0245",
            "repeat-c1-qst_0247",
            "repeat-c1-qst_0250",
            "repeat-c1-qst_0252",
            "repeat-c1-qst_0253",
            "repeat-c1-qst_0254",
            "repeat-c1-qst_0255",
            "repeat-c1-qst_0256",
            "repeat-c1-qst_0257",
            "repeat-c1-qst_0260",
            "repeat-c1-qst_0262",
            "repeat-c1-qst_0265",
            "repeat-c1-qst_0270",
            "repeat-c1-qst_0271",
            "repeat-c1-qst_0274",
            "repeat-c1-qst_0275",
            "repeat-c1-qst_0277",
            "repeat-c1-qst_0280",
            "repeat-c1-qst_0281",
            "repeat-c1-qst_0283",
            "repeat-c1-qst_0284",
            "repeat-c1-qst_0285",
            "repeat-c1-qst_0286",
            "repeat-c1-qst_0289",
            "repeat-c1-qst_0290",
            "repeat-c1-qst_0291",
            "repeat-c1-qst_0293",
            "repeat-c1-qst_0295",
            "repeat-c1-qst_0296",
            "repeat-c1-qst_0299",
            "repeat-c1-qst_0301",
            "repeat-c1-qst_0302",
            "repeat-c1-qst_0304",
            "repeat-c1-qst_0305",
            "repeat-c1-qst_0306",
            "repeat-c1-qst_0307",
            "repeat-c1-qst_0309",
            "repeat-c1-qst_0314",
            "repeat-c1-qst_0315",
            "repeat-c1-qst_0317",
            "repeat-c1-qst_0322",
            "repeat-c1-qst_0323",
            "repeat-c1-qst_0324",
            "repeat-c1-qst_0331",
            "repeat-c1-qst_0332",
            "repeat-c1-qst_0333",
            "repeat-c1-qst_0334",
            "repeat-c1-qst_0336",
            "repeat-c1-qst_0339",
            "repeat-c1-qst_0340",
            "repeat-c1-qst_0341",
            "repeat-c1-qst_0342",
            "repeat-c1-qst_0343",
            "repeat-c1-qst_0345",
            "repeat-c1-qst_0346",
            "repeat-c1-qst_0347",
            "repeat-c1-qst_0349",
            "repeat-c1-qst_0350",
            "repeat-c1-qst_0354",
            "repeat-c1-qst_0355",
            "repeat-c1-qst_0356",
            "repeat-c1-qst_0360",
            "repeat-c1-qst_0362",
            "repeat-c1-qst_0363",
            "repeat-c1-qst_0365",
            "repeat-c1-qst_0367",
            "repeat-c1-qst_0369",
            "repeat-c1-qst_0375",
            "repeat-c1-qst_0376",
            "repeat-c1-qst_0377",
            "repeat-c1-qst_0381",
            "repeat-c1-qst_0386",
            "repeat-c1-qst_0389",
            "repeat-c1-qst_0390",
            "repeat-c1-qst_0391",
            "repeat-c1-qst_0392",
            "repeat-c1-qst_0393",
            "repeat-c1-qst_0394",
            "repeat-c1-qst_0395",
            "repeat-c1-qst_0396",
            "repeat-c1-qst_0399",
            "repeat-c1-qst_0402",
            "repeat-c1-qst_0404",
            "repeat-c1-qst_0407",
            "repeat-c1-qst_0410",
            "repeat-c1-qst_0411",
            "repeat-c1-qst_0412",
            "repeat-c1-qst_0414",
            "repeat-c1-qst_0417",
            "repeat-c1-qst_0419",
            "repeat-c1-qst_0420",
            "repeat-c1-qst_0422",
            "repeat-c1-qst_0423",
            "repeat-c1-qst_0426",
            "repeat-c1-qst_0428",
            "repeat-c1-qst_0434",
            "repeat-c1-qst_0435",
            "repeat-c1-qst_0436",
            "repeat-c1-qst_0438",
            "repeat-c1-qst_0439",
            "repeat-c1-qst_0442",
            "repeat-c1-qst_0443",
            "repeat-c1-qst_0445",
            "repeat-c1-qst_0447",
            "repeat-c1-qst_0449",
            "repeat-c1-qst_0451",
            "repeat-c1-qst_0452",
            "repeat-c1-qst_0455",
            "repeat-c1-qst_0460",
            "repeat-c1-qst_0461",
            "repeat-c1-qst_0463",
            "repeat-c1-qst_0464",
            "repeat-c1-qst_0465",
            "repeat-c1-qst_0467",
            "repeat-c1-qst_0470",
            "repeat-c1-qst_0472",
            "repeat-c1-qst_0476",
            "repeat-c1-qst_0477",
            "repeat-c1-qst_0478",
            "repeat-c1-qst_0479",
            "repeat-c1-qst_0484",
            "repeat-c1-qst_0486",
            "repeat-c1-qst_0487",
            "repeat-c1-qst_0488",
            "repeat-c1-qst_0492",
            "repeat-c1-qst_0495",
            "repeat-c1-qst_0496",
            "repeat-c1-qst_0497",
            "repeat-c1-qst_0499",
            "repeat-c1-qst_0500"
          ],
          "observed_valid": 250,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 64.5565053854925
        },
        "warm_c1_embedding_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c1_embedding_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c1_engine_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c1_engine_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "voyage_probe_c1_p50_ms": {
          "status": "partial",
          "value": 296.0006578359753,
          "ci95": [
            261.1328649800271,
            764.1305821016431
          ],
          "unit": "ms",
          "n": 10,
          "expected": 11,
          "attempted": 11,
          "missing": 1,
          "artifact": "runtime/fyrr-release-full/voyage-probes.jsonl"
        },
        "voyage_probe_c1_p95_ms": {
          "status": "partial",
          "value": 4330.121272394893,
          "ci95": [
            437.2746180742979,
            7063.185462960973
          ],
          "unit": "ms",
          "n": 10,
          "expected": 11,
          "attempted": 11,
          "missing": 1,
          "artifact": "runtime/fyrr-release-full/voyage-probes.jsonl"
        },
        "first_ask_c8_failure_rate": {
          "status": "measured",
          "value": 0.984,
          "unit": "fraction",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/latency.jsonl"
        },
        "first_ask_c8_p50_ms": {
          "status": "partial",
          "value": 273.98472593631595,
          "ci95": [
            262.7075160853565,
            717.626933939755
          ],
          "unit": "ms",
          "n": 4,
          "expected": 250,
          "attempted": 250,
          "missing": 246,
          "sample_ids": [
            "first_ask-c8-qst_0198",
            "first_ask-c8-qst_0304",
            "first_ask-c8-qst_0399",
            "first_ask-c8-qst_0411"
          ],
          "observed_valid": 4,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 5.469415946088047
        },
        "first_ask_c8_p95_ms": {
          "status": "partial",
          "value": 652.0346460980361,
          "ci95": [
            267.6244368776679,
            717.626933939755
          ],
          "unit": "ms",
          "n": 4,
          "expected": 250,
          "attempted": 250,
          "missing": 246,
          "sample_ids": [
            "first_ask-c8-qst_0198",
            "first_ask-c8-qst_0304",
            "first_ask-c8-qst_0399",
            "first_ask-c8-qst_0411"
          ],
          "observed_valid": 4,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 5.469415946088047
        },
        "first_ask_c8_embedding_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "first_ask_c8_embedding_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "first_ask_c8_engine_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "first_ask_c8_engine_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c8_failure_rate": {
          "status": "measured",
          "value": 0.976,
          "unit": "fraction",
          "n": 250,
          "expected": 250,
          "attempted": 250,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/latency.jsonl"
        },
        "warm_c8_p50_ms": {
          "status": "partial",
          "value": 54.60896447766572,
          "ci95": [
            29.752586036920547,
            495.68285746499896
          ],
          "unit": "ms",
          "n": 6,
          "expected": 250,
          "attempted": 250,
          "missing": 244,
          "sample_ids": [
            "repeat-c8-qst_0198",
            "repeat-c8-qst_0293",
            "repeat-c8-qst_0304",
            "repeat-c8-qst_0399",
            "repeat-c8-qst_0411",
            "repeat-c8-qst_0492"
          ],
          "observed_valid": 6,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 8.012929276062259
        },
        "warm_c8_p95_ms": {
          "status": "partial",
          "value": 601.6876532230526,
          "ci95": [
            54.76847727550194,
            707.6924489811063
          ],
          "unit": "ms",
          "n": 6,
          "expected": 250,
          "attempted": 250,
          "missing": 244,
          "sample_ids": [
            "repeat-c8-qst_0198",
            "repeat-c8-qst_0293",
            "repeat-c8-qst_0304",
            "repeat-c8-qst_0399",
            "repeat-c8-qst_0411",
            "repeat-c8-qst_0492"
          ],
          "observed_valid": 6,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately",
          "throughput_success_per_second": 8.012929276062259
        },
        "warm_c8_embedding_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c8_embedding_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c8_engine_p50_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "warm_c8_engine_p95_ms": {
          "status": "n/e",
          "value": null,
          "ci95": null,
          "unit": "ms",
          "n": 0,
          "expected": 250,
          "attempted": 250,
          "missing": 250,
          "sample_ids": [],
          "observed_valid": 0,
          "artifact": "runtime/fyrr-release-full/latency.jsonl",
          "method": "One request per question, valid responses only; percentile bootstrap 10000 seed1729; failures reported separately"
        },
        "voyage_probe_c8_p50_ms": {
          "status": "measured",
          "value": 277.132909046486,
          "ci95": [
            260.14595991000533,
            755.3482928778976
          ],
          "unit": "ms",
          "n": 11,
          "expected": 11,
          "attempted": 11,
          "missing": 0,
          "artifact": "runtime/fyrr-release-full/voyage-probes.jsonl"
        },
        "voyage_probe_c8_p95_ms": {
          "status": "measured",
          "value": 809.4893966335803,
          "ci95": [
            745.2247784240171,
            855.1151161082089
          ],
          "unit": "ms",
          "n": 11,
          "expected": 11,
          "attempted": 11,
          "missing": 0,
          "artifact": "runtime/fyrr-release-full/voyage-probes.jsonl"
        },
        "ingest_submitted_docs": {
          "status": "measured",
          "value": 20000,
          "unit": "documents",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_accepted_docs": {
          "status": "measured",
          "value": 19999,
          "unit": "documents",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_rejected_docs": {
          "status": "measured",
          "value": 1,
          "unit": "documents",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_searchable_docs": {
          "status": "measured",
          "value": 19999,
          "unit": "documents",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_pending_docs": {
          "status": "measured",
          "value": 0,
          "unit": "documents",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_coverage_fraction": {
          "status": "measured",
          "value": 0.99995,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "searchable_coverage_fraction": {
          "status": "measured",
          "value": 0.99995,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_wall_s": {
          "status": "measured",
          "value": 670.0962180199567,
          "unit": "seconds",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_docs_per_min": {
          "status": "measured",
          "value": 1790.6980635492312,
          "unit": "docs/min",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "embedding_cost_usd": {
          "status": "n/e",
          "value": null,
          "unit": "USD",
          "n": 0,
          "expected": 1,
          "attempted": 0,
          "missing": 1,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "embedding_request_count": {
          "status": "measured",
          "value": 630,
          "unit": "requests",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json",
          "scope": "Writer sync only: requests counts document-embedding POST attempts, provider_http_attempts also includes non-embedding probes. Reader readiness/search probe and setup usage are not exposed here."
        },
        "ingest_peak_rss_bytes": {
          "status": "measured",
          "value": 523472896,
          "unit": "bytes",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_stage_writer_sync_s": {
          "status": "measured",
          "value": 663.6318440919276,
          "unit": "seconds",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_stage_writer_commit_s": {
          "status": "measured",
          "value": 4.14176432415843,
          "unit": "seconds",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_stage_reader_check_s": {
          "status": "measured",
          "value": 0.2651936048641801,
          "unit": "seconds",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_stage_mcp_inventory_readiness_s": {
          "status": "measured",
          "value": 0.29550478304736316,
          "unit": "seconds",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "ingest_stage_late_document_probe_s": {
          "status": "measured",
          "value": 1.7604167989920825,
          "unit": "seconds",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/ingest.json"
        },
        "service_count": {
          "status": "measured",
          "value": 1,
          "unit": "services",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/footprint.json"
        },
        "idle_rss_bytes": {
          "status": "measured",
          "value": 96595968,
          "unit": "bytes",
          "n": 238,
          "expected": 238,
          "attempted": 238,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/footprint.json"
        },
        "load_rss_peak_bytes": {
          "status": "measured",
          "value": 171745280,
          "unit": "bytes",
          "n": 329,
          "expected": 329,
          "attempted": 329,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/footprint.json"
        },
        "index_and_data_bytes": {
          "status": "measured",
          "value": 439840768,
          "unit": "bytes",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "runtime/fyrr-release-full/footprint.json",
          "scope": "Owned index/stored application data, persistent document-vector cache and serving mirror; excludes the shared frozen input folder, binaries/images, static model/package caches and application logs. Collector retains component breakdown."
        },
        "monthly_cost_usd": {
          "status": "measured",
          "value": 42.29758737476408,
          "unit": "USD/month",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "${BENCH_ROOT}/results/metadata/fyrr-production-cost-model.json",
          "basis": "Documented production deployment estimate; not a measured invoice"
        }
      },
      "footprint": {
        "idle": {
          "rss_median_bytes": 96595968,
          "n": 238
        },
        "services": 1,
        "containers": 0,
        "processes": 1,
        "external_services": [
          "Voyage embeddings API"
        ],
        "notes": [
          "Batch writer stopped; Python driver excluded; additive CPU observer outside search timing",
          "Per-block fresh process/cgroup samples; idle median of block medians, peak=max of block peaks"
        ],
        "under_load": {
          "rss_peak_bytes": 171745280,
          "n": 329
        },
        "disk": {
          "index_and_data": {
            "logical_bytes": 140408223,
            "allocated_bytes": 141062144
          },
          "reader_mirror": {
            "logical_bytes": 134164385,
            "allocated_bytes": 134791168
          },
          "writer_vector_cache": {
            "logical_bytes": 83655222,
            "allocated_bytes": 163987456
          }
        },
        "cpu": {
          "status": "measured",
          "n": 24,
          "sample_interval_seconds": 0.25,
          "seconds": 6.1710512798745185,
          "cpu_seconds_observed": 1.08,
          "mean_logical_cores": 0.17501069931507046,
          "p95_logical_cores": 0.9981057313787681,
          "max_logical_cores": 1.6611557245370543,
          "median_logical_cores": 0,
          "artifact": "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c8.jsonl",
          "scope": "Owned native process tree user+system /proc counters; driver excluded",
          "limitations": [
            "Logical core demand on this host is not a cloud per-core performance guarantee",
            "Subinterval peaks and CPU after an exiting process's last sample are unobserved",
            "New process CPU before its first snapshot is excluded from rates and retained separately in raw samples"
          ]
        },
        "minimum_deployment_cost_estimate": {
          "status": "estimate",
          "total_usd": 24.232135890465372,
          "known_subtotal_usd": 24.232135890465372,
          "assumptions": {
            "provider": "AWS",
            "region": "eu-central-1",
            "currency": "USD",
            "hours_per_month": 730,
            "days_per_month": 30,
            "corpus_documents": 20000,
            "queries_per_month": 100000,
            "daily_changed_fraction": 0.01,
            "replicas": 1,
            "memory_headroom_fraction": 0.25,
            "os_reserved_gib": 1,
            "cpu_requirement": "Smallest passing pinned N in2,4,8; zeroerrors, p95firstask<=1.5x own unpinned. Light passesC1; team passesbothC1/C8.",
            "instance_choice": "Cheapest eligible x86 EBS-only Linux on-demand shape fitting measured passing N and peak RSS+25%+1GiB OS; burstable only with idle CPU and CPU-seconds/query reference-traffic qualification, otherwise m/c non-burstable",
            "fyrr_reader": "Always-on, sized separately from measured reader memory/CPU",
            "fyrr_writer": "Scheduled writer sync+commit only; one-time full job separate from recurring daily incremental projection",
            "incremental_projection": "Daily job runtime = measured full sync+commit seconds * changed fraction; fixed scan/commit costs not separately measured, so projection has that caveat",
            "vm_boot_seconds_assumed": 60,
            "vm_minimum_billed_seconds": 60,
            "writer_billing": "max(60,ceil(projected sync+commit seconds +60 bootseconds)) per daily job; no continuously running writer; persistent root/cache EBS remains billed all month",
            "competitor_compute": "Measured full default/configured stack always-on in this reference deployment",
            "root_volume_base_gib": 20,
            "block_storage_headroom_fraction": 0.25,
            "s3_gets_per_query_assumed": 2,
            "s3_puts_per_month_assumed": 1000,
            "public_ipv4_usd_hour": 0.005,
            "public_ipv4_source": "https://aws.amazon.com/vpc/pricing/",
            "billing_source": "https://aws.amazon.com/ec2/pricing/on-demand/",
            "embedding_workload": "Official-tokenizer canonical text counts; common normalized input assumption, not product-billed usage",
            "ingest_llm": "Observed product operational document enrichment usage and priced projection where available",
            "discounts": "None; no credits/free grants/reservations/spot used for any product",
            "excluded": [
              "HA replicas",
              "cross-region traffic",
              "taxes",
              "support",
              "operator labor",
              "backups/snapshots",
              "load balancer",
              "Any assumption of cloud capacity/SLA equivalence"
            ],
            "million_document_projection": "Same reference query volume; scale stored bytes, changed-doc volume and token-based usage linearly. Compute capacity at1M is not measured; any fixed-instance line must be explicitly conditional and never a measured SLA.",
            "cpu_statistic": "Pinned capacity pass/fail determines core count; CPU counters retained as diagnostics only",
            "memory_requirement": "Measured peak RSS during steady query load*1.25 plus1GiB explicit OS allowance; writer peakRSS during ingest",
            "selection_approval": "3528fe7d-8770-4b6a-8800-784df515d564",
            "required_caveats": [
              "Measurements on a20-core host",
              "Ingest can be slower on the selected smaller instance; displayed ingest times are benchmark-host measurements",
              "Cloud capacity/SLA has not been validated",
              "Burstable qualification requires measured idle CPU plus CPU-seconds/query at the same reference traffic; products without that evidence use non-burstable minimum shapes"
            ],
            "instance_family_approval": "d55f078a-9dfa-4865-8c1c-31cbb17a8fe4",
            "basis": "minimum-measured",
            "load_scenario": "light",
            "load_concurrency": 1,
            "instance_name": "t3a.small",
            "instance_ram_gib": 2,
            "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks"
          },
          "components": {
            "writer_hours_month": 0.5583333333333333,
            "compute_usd_month": 15.816575,
            "storage_usd_month": 4.001618671798706,
            "public_ipv4_usd_month": 3.6527916666666664,
            "object_requests_usd_month": 0.09140000000000001,
            "embedding_usd_per_document": 0.000074149092,
            "llm_enrichment_usd_per_document": 0,
            "changed_document_embeddings_usd_month": 0.444894552,
            "query_embeddings_usd_per_1000": 0.00224856,
            "query_embeddings_usd_month": 0.224856,
            "enrichment_usd_month": 0
          },
          "sizing": {
            "reader": {
              "name": "t3a.small",
              "vcpu": 2,
              "ram_gib": 2,
              "usd_hour": 0.0216,
              "rate_code": "774XP29ZBNQKJHC3.JRTCKXETXF.6YS6EN2CT7",
              "family": "General purpose",
              "burstable": true,
              "effective_usd_hour": 0.0216,
              "required_ram_gib": 1.2501869201660156,
              "required_vcpu": 2,
              "resources": {
                "peak_rss_bytes": 214908928,
                "p95_cores": 0.586274446513124,
                "peak_cores": 0.6254239913304944,
                "mean_cores": 0.3813949999048495,
                "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                "artifacts": [
                  "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
                ],
                "selected_cpus": 2,
                "cpu_seconds_per_query": 0.12400000000000001,
                "reference_mean_cores": 0.008061417586796186,
                "idle_mean_cores": 0.0033430005396120153,
                "burstable_qualified": true,
                "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
              }
            },
            "writer": {
              "name": "c5a.large",
              "vcpu": 2,
              "ram_gib": 4,
              "usd_hour": 0.087,
              "rate_code": "BBAVEJUPG9KCAM7G.JRTCKXETXF.6YS6EN2CT7",
              "family": "Compute optimized",
              "burstable": false,
              "effective_usd_hour": 0.087,
              "required_ram_gib": 1.6094026565551758,
              "required_vcpu": 2,
              "resources": {
                "peak_rss_bytes": 523472896,
                "p95_cores": 0.23272410605561036,
                "peak_cores": 17.298930892056188,
                "mean_cores": 0.09457849652583226,
                "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                "artifacts": [
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.jsonl"
                ],
                "selected_cpus": 2,
                "selection_basis": "Assumed2vCPU scheduled writer, measured writer peakRSS; pinned ingest capacity not tested. Host ingest time can be slower on this VM."
              }
            },
            "capacity_tests": [
              {
                "cpus": 2,
                "concurrency": 1,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 252.71095510106534,
                "p95_ms": 367.9228202207014,
                "threshold_p95_ms": 772.0598442654581,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 214908928,
                "artifact": "runtime/fyrr-capacity/n2/c1"
              },
              {
                "cpus": 2,
                "concurrency": 8,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 263.433939544484,
                "p95_ms": 309.47251981124276,
                "threshold_p95_ms": 431.47896916489117,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 222191616,
                "artifact": "runtime/fyrr-capacity/n2/c8"
              },
              {
                "cpus": 4,
                "concurrency": 1,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 253.72744188643992,
                "p95_ms": 588.8207679148762,
                "threshold_p95_ms": 772.0598442654581,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 219512832,
                "artifact": "runtime/fyrr-capacity/n4/c1"
              },
              {
                "cpus": 4,
                "concurrency": 8,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 258.01787059754133,
                "p95_ms": 337.4477885430678,
                "threshold_p95_ms": 431.47896916489117,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 221077504,
                "artifact": "runtime/fyrr-capacity/n4/c8"
              },
              {
                "cpus": 8,
                "concurrency": 1,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 270.42482141405344,
                "p95_ms": 569.3619827739875,
                "threshold_p95_ms": 772.0598442654581,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 229953536,
                "artifact": "runtime/fyrr-capacity/n8/c1"
              },
              {
                "cpus": 8,
                "concurrency": 8,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 252.25088198203593,
                "p95_ms": 289.69891923479736,
                "threshold_p95_ms": 431.47896916489117,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 239128576,
                "artifact": "runtime/fyrr-capacity/n8/c8"
              }
            ],
            "selected_cpus": 2,
            "ram_headroom_fraction": 0.25,
            "measured_peak_rss_bytes": 214908928
          },
          "inputs": {
            "system": "fyrr",
            "documents": 20000,
            "reader": {
              "peak_rss_bytes": 214908928,
              "p95_cores": 0.586274446513124,
              "peak_cores": 0.6254239913304944,
              "mean_cores": 0.3813949999048495,
              "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
              "artifacts": [
                "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
              ],
              "selected_cpus": 2,
              "cpu_seconds_per_query": 0.12400000000000001,
              "reference_mean_cores": 0.008061417586796186,
              "idle_mean_cores": 0.0033430005396120153,
              "burstable_qualified": true,
              "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
            },
            "scenario_resources": {
              "light": {
                "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks",
                "capacity_tests": [
                  {
                    "cpus": 2,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.71095510106534,
                    "p95_ms": 367.9228202207014,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 214908928,
                    "artifact": "runtime/fyrr-capacity/n2/c1"
                  },
                  {
                    "cpus": 2,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 263.433939544484,
                    "p95_ms": 309.47251981124276,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 222191616,
                    "artifact": "runtime/fyrr-capacity/n2/c8"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 253.72744188643992,
                    "p95_ms": 588.8207679148762,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 219512832,
                    "artifact": "runtime/fyrr-capacity/n4/c1"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 258.01787059754133,
                    "p95_ms": 337.4477885430678,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 221077504,
                    "artifact": "runtime/fyrr-capacity/n4/c8"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 270.42482141405344,
                    "p95_ms": 569.3619827739875,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 229953536,
                    "artifact": "runtime/fyrr-capacity/n8/c1"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.25088198203593,
                    "p95_ms": 289.69891923479736,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 239128576,
                    "artifact": "runtime/fyrr-capacity/n8/c8"
                  }
                ],
                "reader": {
                  "peak_rss_bytes": 214908928,
                  "p95_cores": 0.586274446513124,
                  "peak_cores": 0.6254239913304944,
                  "mean_cores": 0.3813949999048495,
                  "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                  "artifacts": [
                    "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
                  ],
                  "selected_cpus": 2,
                  "cpu_seconds_per_query": 0.12400000000000001,
                  "reference_mean_cores": 0.008061417586796186,
                  "idle_mean_cores": 0.0033430005396120153,
                  "burstable_qualified": true,
                  "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                },
                "notes": [
                  "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS"
                ]
              },
              "team": {
                "configuration_note": "C8, reader.concurrency=8; measured sized run and pinned capacity test, same index, no re-ingest",
                "capacity_tests": [
                  {
                    "cpus": 2,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.71095510106534,
                    "p95_ms": 367.9228202207014,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 214908928,
                    "artifact": "runtime/fyrr-capacity/n2/c1"
                  },
                  {
                    "cpus": 2,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 263.433939544484,
                    "p95_ms": 309.47251981124276,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 222191616,
                    "artifact": "runtime/fyrr-capacity/n2/c8"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 253.72744188643992,
                    "p95_ms": 588.8207679148762,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 219512832,
                    "artifact": "runtime/fyrr-capacity/n4/c1"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 258.01787059754133,
                    "p95_ms": 337.4477885430678,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 221077504,
                    "artifact": "runtime/fyrr-capacity/n4/c8"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 270.42482141405344,
                    "p95_ms": 569.3619827739875,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 229953536,
                    "artifact": "runtime/fyrr-capacity/n8/c1"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.25088198203593,
                    "p95_ms": 289.69891923479736,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 239128576,
                    "artifact": "runtime/fyrr-capacity/n8/c8"
                  }
                ],
                "reader": {
                  "peak_rss_bytes": 277237760,
                  "p95_cores": 13.693632726827053,
                  "peak_cores": 14.074346598170697,
                  "mean_cores": 4.098601693166352,
                  "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                  "artifacts": [
                    "${BENCH_ROOT}/runtime/fyrr-release-concurrency8/cpu-c8.jsonl"
                  ],
                  "selected_cpus": 2,
                  "cpu_seconds_per_query": 0.07986,
                  "cpu_work_scope": "250first-ask+250repeat; CPU counter diagnostic only",
                  "reference_mean_cores": 0.006381813325000143,
                  "idle_mean_cores": 0.0033430005396120153,
                  "burstable_qualified": true,
                  "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                },
                "notes": [
                  "C8 reader.concurrency=8, same index; CPU/RSS from full250 first-ask plus250 repeat sized-run samples; no sizing from default C8 refusals"
                ]
              }
            },
            "writer": {
              "peak_rss_bytes": 523472896,
              "p95_cores": 0.23272410605561036,
              "peak_cores": 17.298930892056188,
              "mean_cores": 0.09457849652583226,
              "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
              "artifacts": [
                "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.jsonl"
              ],
              "selected_cpus": 2,
              "selection_basis": "Assumed2vCPU scheduled writer, measured writer peakRSS; pinned ingest capacity not tested. Host ingest time can be slower on this VM."
            },
            "index_data_bytes": 141062144,
            "reader_cache_bytes": 134791168,
            "reader_software_bytes": 147272320,
            "writer_cache_bytes": 163987456,
            "writer_software_bytes": 147272320,
            "sync_commit_seconds": 667.7736084160861,
            "llm_enrichment_usd_per_document": 0,
            "notes": [
              "No operational LLM enrichment in the measured local-files ingest/search pipeline.",
              "Light and team scenarios use their own measured CPU/RSS. Team explicitly sets reader.concurrency=8; default C8 refusals are never used to size team compute.",
              "Writer CPU observer includes the short excluded preflight interval; measured writer RSS includes the full sync, commit and visibility check.",
              "Object-store size is measured using the native local object-store implementation; reference deployment prices those bytes on S3 Standard.",
              "Native binary size is included in both local role volumes; persistent vector cache and serving mirror are retained separately.",
              "Ingest accepted 19999 of 20000 documents; reference API costs use the common complete-corpus canonical token estimate.",
              "Writer core count is an explicit2vCPU assumption; query capacity tests cover the reader, not writer ingestion throughput.",
              "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS"
            ],
            "artifacts": [
              "${BENCH_ROOT}/runtime/fyrr-release-full/footprint.json",
              "${BENCH_ROOT}/runtime/fyrr-release-full/ingest.json",
              "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c1.json",
              "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c8.json",
              "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.json"
            ],
            "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks",
            "capacity_tests": [
              {
                "cpus": 2,
                "concurrency": 1,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 252.71095510106534,
                "p95_ms": 367.9228202207014,
                "threshold_p95_ms": 772.0598442654581,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 214908928,
                "artifact": "runtime/fyrr-capacity/n2/c1"
              },
              {
                "cpus": 2,
                "concurrency": 8,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 263.433939544484,
                "p95_ms": 309.47251981124276,
                "threshold_p95_ms": 431.47896916489117,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 222191616,
                "artifact": "runtime/fyrr-capacity/n2/c8"
              },
              {
                "cpus": 4,
                "concurrency": 1,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 253.72744188643992,
                "p95_ms": 588.8207679148762,
                "threshold_p95_ms": 772.0598442654581,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 219512832,
                "artifact": "runtime/fyrr-capacity/n4/c1"
              },
              {
                "cpus": 4,
                "concurrency": 8,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 258.01787059754133,
                "p95_ms": 337.4477885430678,
                "threshold_p95_ms": 431.47896916489117,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 221077504,
                "artifact": "runtime/fyrr-capacity/n4/c8"
              },
              {
                "cpus": 8,
                "concurrency": 1,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 270.42482141405344,
                "p95_ms": 569.3619827739875,
                "threshold_p95_ms": 772.0598442654581,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 229953536,
                "artifact": "runtime/fyrr-capacity/n8/c1"
              },
              {
                "cpus": 8,
                "concurrency": 8,
                "expected": 250,
                "valid": 250,
                "errors": 0,
                "p50_ms": 252.25088198203593,
                "p95_ms": 289.69891923479736,
                "threshold_p95_ms": 431.47896916489117,
                "pass": true,
                "timing_valid": true,
                "measured_peak_rss_bytes": 239128576,
                "artifact": "runtime/fyrr-capacity/n8/c8"
              }
            ]
          },
          "one_time_full_ingest": {
            "compute_usd": 0.017593333333333332,
            "normalized_embedding_usd": 1.4829818399999999,
            "llm_enrichment_usd": 0,
            "total_usd": 1.5005751733333332
          },
          "sources": [
            {
              "label": "aws-ec2-eu-central-1.json",
              "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
              "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
              "publication_date": "2026-11-14T01:29:26Z"
            },
            {
              "label": "aws-ebs.json",
              "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
              "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
              "publication_date": "2026-11-14T01:29:26Z"
            },
            {
              "label": "aws-s3-eu-central-1.json",
              "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
              "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
              "publication_date": "2026-11-16T21:17:45Z"
            },
            {
              "label": "Voyage list pricing",
              "url": "https://docs.voyageai.com/docs/pricing"
            },
            {
              "label": "Official tokenizer",
              "url": "https://docs.voyageai.com/docs/tokenization"
            },
            {
              "label": "Public IPv4",
              "url": "https://aws.amazon.com/vpc/pricing/"
            },
            {
              "label": "EC2 billing granularity",
              "url": "https://aws.amazon.com/ec2/pricing/on-demand/"
            }
          ],
          "missing": [],
          "notes": [
            "No operational LLM enrichment in the measured local-files ingest/search pipeline.",
            "Light and team scenarios use their own measured CPU/RSS. Team explicitly sets reader.concurrency=8; default C8 refusals are never used to size team compute.",
            "Writer CPU observer includes the short excluded preflight interval; measured writer RSS includes the full sync, commit and visibility check.",
            "Object-store size is measured using the native local object-store implementation; reference deployment prices those bytes on S3 Standard.",
            "Native binary size is included in both local role volumes; persistent vector cache and serving mirror are retained separately.",
            "Ingest accepted 19999 of 20000 documents; reference API costs use the common complete-corpus canonical token estimate.",
            "Writer core count is an explicit2vCPU assumption; query capacity tests cover the reader, not writer ingestion throughput.",
            "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS",
            "Daily incremental writer duration is a linear changed-document projection from measured full sync+commit plus assumed boot time; fixed scan/commit costs are not separately measured.",
            "Measurements on a20-core host",
            "Ingest can be slower on the selected smaller instance; displayed ingest times are benchmark-host measurements",
            "Cloud capacity/SLA has not been validated",
            "Burstable qualification requires measured idle CPU plus CPU-seconds/query at the same reference traffic; products without that evidence use non-burstable minimum shapes",
            "List-price deployment projection; provider-billed spend remains separate and can be n/e.",
            "Normalized embedding inputs use canonical text counts; native chunking, title prefixes, deduplication, retries and rejections can change actual product usage.",
            "CPU core demand is measured on the benchmark host; selected cloud-instance latency/capacity has not been benchmarked.",
            "Measured minimum quote uses a qualified burstable reader only where idle+query CPU evidence supports its baseline. Roles without that evidence stay on non-burstable m/c shapes. Persistent writer EBS stays billed while its VM is stopped."
          ],
          "first_month_total_usd": 25.732711063798707,
          "projections": [
            {
              "documents": 1000000,
              "status": "projection",
              "comparable": false,
              "total_usd": 48.32338385660197,
              "components": {
                "writer_hours_month": 3.283333333333333,
                "compute_usd_month": 16.05365,
                "storage_usd_month": 5.777733589935304,
                "public_ipv4_usd_month": 3.6664166666666667,
                "object_requests_usd_month": 0.35600000000000004,
                "embedding_usd_per_document": 0.000074149092,
                "llm_enrichment_usd_per_document": 0,
                "changed_document_embeddings_usd_month": 22.244727599999997,
                "query_embeddings_usd_per_1000": 0.00224856,
                "query_embeddings_usd_month": 0.224856,
                "enrichment_usd_month": 0
              },
              "assumptions": "Measured20k compute shapes held fixed; index/data and per-document API usage and writer time scale linearly; fixed software/model overhead stays fixed. Writer vector cache also scales linearly; software stays fixed.",
              "capacity_validation": "n/e; no1M memory/CPU/latency measurement or cloud SLA claim"
            }
          ],
          "scenarios": {
            "light": {
              "status": "estimate",
              "total_usd": 24.232135890465372,
              "known_subtotal_usd": 24.232135890465372,
              "assumptions": {
                "provider": "AWS",
                "region": "eu-central-1",
                "currency": "USD",
                "hours_per_month": 730,
                "days_per_month": 30,
                "corpus_documents": 20000,
                "queries_per_month": 100000,
                "daily_changed_fraction": 0.01,
                "replicas": 1,
                "memory_headroom_fraction": 0.25,
                "os_reserved_gib": 1,
                "cpu_requirement": "Smallest passing pinned N in2,4,8; zeroerrors, p95firstask<=1.5x own unpinned. Light passesC1; team passesbothC1/C8.",
                "instance_choice": "Cheapest eligible x86 EBS-only Linux on-demand shape fitting measured passing N and peak RSS+25%+1GiB OS; burstable only with idle CPU and CPU-seconds/query reference-traffic qualification, otherwise m/c non-burstable",
                "fyrr_reader": "Always-on, sized separately from measured reader memory/CPU",
                "fyrr_writer": "Scheduled writer sync+commit only; one-time full job separate from recurring daily incremental projection",
                "incremental_projection": "Daily job runtime = measured full sync+commit seconds * changed fraction; fixed scan/commit costs not separately measured, so projection has that caveat",
                "vm_boot_seconds_assumed": 60,
                "vm_minimum_billed_seconds": 60,
                "writer_billing": "max(60,ceil(projected sync+commit seconds +60 bootseconds)) per daily job; no continuously running writer; persistent root/cache EBS remains billed all month",
                "competitor_compute": "Measured full default/configured stack always-on in this reference deployment",
                "root_volume_base_gib": 20,
                "block_storage_headroom_fraction": 0.25,
                "s3_gets_per_query_assumed": 2,
                "s3_puts_per_month_assumed": 1000,
                "public_ipv4_usd_hour": 0.005,
                "public_ipv4_source": "https://aws.amazon.com/vpc/pricing/",
                "billing_source": "https://aws.amazon.com/ec2/pricing/on-demand/",
                "embedding_workload": "Official-tokenizer canonical text counts; common normalized input assumption, not product-billed usage",
                "ingest_llm": "Observed product operational document enrichment usage and priced projection where available",
                "discounts": "None; no credits/free grants/reservations/spot used for any product",
                "excluded": [
                  "HA replicas",
                  "cross-region traffic",
                  "taxes",
                  "support",
                  "operator labor",
                  "backups/snapshots",
                  "load balancer",
                  "Any assumption of cloud capacity/SLA equivalence"
                ],
                "million_document_projection": "Same reference query volume; scale stored bytes, changed-doc volume and token-based usage linearly. Compute capacity at1M is not measured; any fixed-instance line must be explicitly conditional and never a measured SLA.",
                "cpu_statistic": "Pinned capacity pass/fail determines core count; CPU counters retained as diagnostics only",
                "memory_requirement": "Measured peak RSS during steady query load*1.25 plus1GiB explicit OS allowance; writer peakRSS during ingest",
                "selection_approval": "3528fe7d-8770-4b6a-8800-784df515d564",
                "required_caveats": [
                  "Measurements on a20-core host",
                  "Ingest can be slower on the selected smaller instance; displayed ingest times are benchmark-host measurements",
                  "Cloud capacity/SLA has not been validated",
                  "Burstable qualification requires measured idle CPU plus CPU-seconds/query at the same reference traffic; products without that evidence use non-burstable minimum shapes"
                ],
                "instance_family_approval": "d55f078a-9dfa-4865-8c1c-31cbb17a8fe4",
                "basis": "minimum-measured",
                "load_scenario": "light",
                "load_concurrency": 1,
                "instance_name": "t3a.small",
                "instance_ram_gib": 2,
                "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks"
              },
              "components": {
                "writer_hours_month": 0.5583333333333333,
                "compute_usd_month": 15.816575,
                "storage_usd_month": 4.001618671798706,
                "public_ipv4_usd_month": 3.6527916666666664,
                "object_requests_usd_month": 0.09140000000000001,
                "embedding_usd_per_document": 0.000074149092,
                "llm_enrichment_usd_per_document": 0,
                "changed_document_embeddings_usd_month": 0.444894552,
                "query_embeddings_usd_per_1000": 0.00224856,
                "query_embeddings_usd_month": 0.224856,
                "enrichment_usd_month": 0
              },
              "sizing": {
                "reader": {
                  "name": "t3a.small",
                  "vcpu": 2,
                  "ram_gib": 2,
                  "usd_hour": 0.0216,
                  "rate_code": "774XP29ZBNQKJHC3.JRTCKXETXF.6YS6EN2CT7",
                  "family": "General purpose",
                  "burstable": true,
                  "effective_usd_hour": 0.0216,
                  "required_ram_gib": 1.2501869201660156,
                  "required_vcpu": 2,
                  "resources": {
                    "peak_rss_bytes": 214908928,
                    "p95_cores": 0.586274446513124,
                    "peak_cores": 0.6254239913304944,
                    "mean_cores": 0.3813949999048495,
                    "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                    "artifacts": [
                      "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
                    ],
                    "selected_cpus": 2,
                    "cpu_seconds_per_query": 0.12400000000000001,
                    "reference_mean_cores": 0.008061417586796186,
                    "idle_mean_cores": 0.0033430005396120153,
                    "burstable_qualified": true,
                    "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                  }
                },
                "writer": {
                  "name": "c5a.large",
                  "vcpu": 2,
                  "ram_gib": 4,
                  "usd_hour": 0.087,
                  "rate_code": "BBAVEJUPG9KCAM7G.JRTCKXETXF.6YS6EN2CT7",
                  "family": "Compute optimized",
                  "burstable": false,
                  "effective_usd_hour": 0.087,
                  "required_ram_gib": 1.6094026565551758,
                  "required_vcpu": 2,
                  "resources": {
                    "peak_rss_bytes": 523472896,
                    "p95_cores": 0.23272410605561036,
                    "peak_cores": 17.298930892056188,
                    "mean_cores": 0.09457849652583226,
                    "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                    "artifacts": [
                      "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.jsonl"
                    ],
                    "selected_cpus": 2,
                    "selection_basis": "Assumed2vCPU scheduled writer, measured writer peakRSS; pinned ingest capacity not tested. Host ingest time can be slower on this VM."
                  }
                },
                "capacity_tests": [
                  {
                    "cpus": 2,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.71095510106534,
                    "p95_ms": 367.9228202207014,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 214908928,
                    "artifact": "runtime/fyrr-capacity/n2/c1"
                  },
                  {
                    "cpus": 2,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 263.433939544484,
                    "p95_ms": 309.47251981124276,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 222191616,
                    "artifact": "runtime/fyrr-capacity/n2/c8"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 253.72744188643992,
                    "p95_ms": 588.8207679148762,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 219512832,
                    "artifact": "runtime/fyrr-capacity/n4/c1"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 258.01787059754133,
                    "p95_ms": 337.4477885430678,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 221077504,
                    "artifact": "runtime/fyrr-capacity/n4/c8"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 270.42482141405344,
                    "p95_ms": 569.3619827739875,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 229953536,
                    "artifact": "runtime/fyrr-capacity/n8/c1"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.25088198203593,
                    "p95_ms": 289.69891923479736,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 239128576,
                    "artifact": "runtime/fyrr-capacity/n8/c8"
                  }
                ],
                "selected_cpus": 2,
                "ram_headroom_fraction": 0.25,
                "measured_peak_rss_bytes": 214908928
              },
              "inputs": {
                "system": "fyrr",
                "documents": 20000,
                "reader": {
                  "peak_rss_bytes": 214908928,
                  "p95_cores": 0.586274446513124,
                  "peak_cores": 0.6254239913304944,
                  "mean_cores": 0.3813949999048495,
                  "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                  "artifacts": [
                    "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
                  ],
                  "selected_cpus": 2,
                  "cpu_seconds_per_query": 0.12400000000000001,
                  "reference_mean_cores": 0.008061417586796186,
                  "idle_mean_cores": 0.0033430005396120153,
                  "burstable_qualified": true,
                  "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                },
                "scenario_resources": {
                  "light": {
                    "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks",
                    "capacity_tests": [
                      {
                        "cpus": 2,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.71095510106534,
                        "p95_ms": 367.9228202207014,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 214908928,
                        "artifact": "runtime/fyrr-capacity/n2/c1"
                      },
                      {
                        "cpus": 2,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 263.433939544484,
                        "p95_ms": 309.47251981124276,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 222191616,
                        "artifact": "runtime/fyrr-capacity/n2/c8"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 253.72744188643992,
                        "p95_ms": 588.8207679148762,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 219512832,
                        "artifact": "runtime/fyrr-capacity/n4/c1"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 258.01787059754133,
                        "p95_ms": 337.4477885430678,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 221077504,
                        "artifact": "runtime/fyrr-capacity/n4/c8"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 270.42482141405344,
                        "p95_ms": 569.3619827739875,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 229953536,
                        "artifact": "runtime/fyrr-capacity/n8/c1"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.25088198203593,
                        "p95_ms": 289.69891923479736,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 239128576,
                        "artifact": "runtime/fyrr-capacity/n8/c8"
                      }
                    ],
                    "reader": {
                      "peak_rss_bytes": 214908928,
                      "p95_cores": 0.586274446513124,
                      "peak_cores": 0.6254239913304944,
                      "mean_cores": 0.3813949999048495,
                      "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                      "artifacts": [
                        "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
                      ],
                      "selected_cpus": 2,
                      "cpu_seconds_per_query": 0.12400000000000001,
                      "reference_mean_cores": 0.008061417586796186,
                      "idle_mean_cores": 0.0033430005396120153,
                      "burstable_qualified": true,
                      "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                    },
                    "notes": [
                      "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS"
                    ]
                  },
                  "team": {
                    "configuration_note": "C8, reader.concurrency=8; measured sized run and pinned capacity test, same index, no re-ingest",
                    "capacity_tests": [
                      {
                        "cpus": 2,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.71095510106534,
                        "p95_ms": 367.9228202207014,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 214908928,
                        "artifact": "runtime/fyrr-capacity/n2/c1"
                      },
                      {
                        "cpus": 2,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 263.433939544484,
                        "p95_ms": 309.47251981124276,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 222191616,
                        "artifact": "runtime/fyrr-capacity/n2/c8"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 253.72744188643992,
                        "p95_ms": 588.8207679148762,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 219512832,
                        "artifact": "runtime/fyrr-capacity/n4/c1"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 258.01787059754133,
                        "p95_ms": 337.4477885430678,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 221077504,
                        "artifact": "runtime/fyrr-capacity/n4/c8"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 270.42482141405344,
                        "p95_ms": 569.3619827739875,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 229953536,
                        "artifact": "runtime/fyrr-capacity/n8/c1"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.25088198203593,
                        "p95_ms": 289.69891923479736,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 239128576,
                        "artifact": "runtime/fyrr-capacity/n8/c8"
                      }
                    ],
                    "reader": {
                      "peak_rss_bytes": 277237760,
                      "p95_cores": 13.693632726827053,
                      "peak_cores": 14.074346598170697,
                      "mean_cores": 4.098601693166352,
                      "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                      "artifacts": [
                        "${BENCH_ROOT}/runtime/fyrr-release-concurrency8/cpu-c8.jsonl"
                      ],
                      "selected_cpus": 2,
                      "cpu_seconds_per_query": 0.07986,
                      "cpu_work_scope": "250first-ask+250repeat; CPU counter diagnostic only",
                      "reference_mean_cores": 0.006381813325000143,
                      "idle_mean_cores": 0.0033430005396120153,
                      "burstable_qualified": true,
                      "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                    },
                    "notes": [
                      "C8 reader.concurrency=8, same index; CPU/RSS from full250 first-ask plus250 repeat sized-run samples; no sizing from default C8 refusals"
                    ]
                  }
                },
                "writer": {
                  "peak_rss_bytes": 523472896,
                  "p95_cores": 0.23272410605561036,
                  "peak_cores": 17.298930892056188,
                  "mean_cores": 0.09457849652583226,
                  "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                  "artifacts": [
                    "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.jsonl"
                  ],
                  "selected_cpus": 2,
                  "selection_basis": "Assumed2vCPU scheduled writer, measured writer peakRSS; pinned ingest capacity not tested. Host ingest time can be slower on this VM."
                },
                "index_data_bytes": 141062144,
                "reader_cache_bytes": 134791168,
                "reader_software_bytes": 147272320,
                "writer_cache_bytes": 163987456,
                "writer_software_bytes": 147272320,
                "sync_commit_seconds": 667.7736084160861,
                "llm_enrichment_usd_per_document": 0,
                "notes": [
                  "No operational LLM enrichment in the measured local-files ingest/search pipeline.",
                  "Light and team scenarios use their own measured CPU/RSS. Team explicitly sets reader.concurrency=8; default C8 refusals are never used to size team compute.",
                  "Writer CPU observer includes the short excluded preflight interval; measured writer RSS includes the full sync, commit and visibility check.",
                  "Object-store size is measured using the native local object-store implementation; reference deployment prices those bytes on S3 Standard.",
                  "Native binary size is included in both local role volumes; persistent vector cache and serving mirror are retained separately.",
                  "Ingest accepted 19999 of 20000 documents; reference API costs use the common complete-corpus canonical token estimate.",
                  "Writer core count is an explicit2vCPU assumption; query capacity tests cover the reader, not writer ingestion throughput.",
                  "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS"
                ],
                "artifacts": [
                  "${BENCH_ROOT}/runtime/fyrr-release-full/footprint.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/ingest.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c1.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c8.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.json"
                ],
                "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks",
                "capacity_tests": [
                  {
                    "cpus": 2,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.71095510106534,
                    "p95_ms": 367.9228202207014,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 214908928,
                    "artifact": "runtime/fyrr-capacity/n2/c1"
                  },
                  {
                    "cpus": 2,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 263.433939544484,
                    "p95_ms": 309.47251981124276,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 222191616,
                    "artifact": "runtime/fyrr-capacity/n2/c8"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 253.72744188643992,
                    "p95_ms": 588.8207679148762,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 219512832,
                    "artifact": "runtime/fyrr-capacity/n4/c1"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 258.01787059754133,
                    "p95_ms": 337.4477885430678,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 221077504,
                    "artifact": "runtime/fyrr-capacity/n4/c8"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 270.42482141405344,
                    "p95_ms": 569.3619827739875,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 229953536,
                    "artifact": "runtime/fyrr-capacity/n8/c1"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.25088198203593,
                    "p95_ms": 289.69891923479736,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 239128576,
                    "artifact": "runtime/fyrr-capacity/n8/c8"
                  }
                ]
              },
              "one_time_full_ingest": {
                "compute_usd": 0.017593333333333332,
                "normalized_embedding_usd": 1.4829818399999999,
                "llm_enrichment_usd": 0,
                "total_usd": 1.5005751733333332
              },
              "sources": [
                {
                  "label": "aws-ec2-eu-central-1.json",
                  "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
                  "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
                  "publication_date": "2026-11-14T01:29:26Z"
                },
                {
                  "label": "aws-ebs.json",
                  "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
                  "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
                  "publication_date": "2026-11-14T01:29:26Z"
                },
                {
                  "label": "aws-s3-eu-central-1.json",
                  "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                  "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                  "publication_date": "2026-11-16T21:17:45Z"
                },
                {
                  "label": "Voyage list pricing",
                  "url": "https://docs.voyageai.com/docs/pricing"
                },
                {
                  "label": "Official tokenizer",
                  "url": "https://docs.voyageai.com/docs/tokenization"
                },
                {
                  "label": "Public IPv4",
                  "url": "https://aws.amazon.com/vpc/pricing/"
                },
                {
                  "label": "EC2 billing granularity",
                  "url": "https://aws.amazon.com/ec2/pricing/on-demand/"
                }
              ],
              "missing": [],
              "notes": [
                "No operational LLM enrichment in the measured local-files ingest/search pipeline.",
                "Light and team scenarios use their own measured CPU/RSS. Team explicitly sets reader.concurrency=8; default C8 refusals are never used to size team compute.",
                "Writer CPU observer includes the short excluded preflight interval; measured writer RSS includes the full sync, commit and visibility check.",
                "Object-store size is measured using the native local object-store implementation; reference deployment prices those bytes on S3 Standard.",
                "Native binary size is included in both local role volumes; persistent vector cache and serving mirror are retained separately.",
                "Ingest accepted 19999 of 20000 documents; reference API costs use the common complete-corpus canonical token estimate.",
                "Writer core count is an explicit2vCPU assumption; query capacity tests cover the reader, not writer ingestion throughput.",
                "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS",
                "Daily incremental writer duration is a linear changed-document projection from measured full sync+commit plus assumed boot time; fixed scan/commit costs are not separately measured.",
                "Measurements on a20-core host",
                "Ingest can be slower on the selected smaller instance; displayed ingest times are benchmark-host measurements",
                "Cloud capacity/SLA has not been validated",
                "Burstable qualification requires measured idle CPU plus CPU-seconds/query at the same reference traffic; products without that evidence use non-burstable minimum shapes",
                "List-price deployment projection; provider-billed spend remains separate and can be n/e.",
                "Normalized embedding inputs use canonical text counts; native chunking, title prefixes, deduplication, retries and rejections can change actual product usage.",
                "CPU core demand is measured on the benchmark host; selected cloud-instance latency/capacity has not been benchmarked.",
                "Measured minimum quote uses a qualified burstable reader only where idle+query CPU evidence supports its baseline. Roles without that evidence stay on non-burstable m/c shapes. Persistent writer EBS stays billed while its VM is stopped."
              ],
              "first_month_total_usd": 25.732711063798707,
              "projections": [
                {
                  "documents": 1000000,
                  "status": "projection",
                  "comparable": false,
                  "total_usd": 48.32338385660197,
                  "components": {
                    "writer_hours_month": 3.283333333333333,
                    "compute_usd_month": 16.05365,
                    "storage_usd_month": 5.777733589935304,
                    "public_ipv4_usd_month": 3.6664166666666667,
                    "object_requests_usd_month": 0.35600000000000004,
                    "embedding_usd_per_document": 0.000074149092,
                    "llm_enrichment_usd_per_document": 0,
                    "changed_document_embeddings_usd_month": 22.244727599999997,
                    "query_embeddings_usd_per_1000": 0.00224856,
                    "query_embeddings_usd_month": 0.224856,
                    "enrichment_usd_month": 0
                  },
                  "assumptions": "Measured20k compute shapes held fixed; index/data and per-document API usage and writer time scale linearly; fixed software/model overhead stays fixed. Writer vector cache also scales linearly; software stays fixed.",
                  "capacity_validation": "n/e; no1M memory/CPU/latency measurement or cloud SLA claim"
                }
              ]
            },
            "team": {
              "status": "estimate",
              "total_usd": 24.232135890465372,
              "known_subtotal_usd": 24.232135890465372,
              "assumptions": {
                "provider": "AWS",
                "region": "eu-central-1",
                "currency": "USD",
                "hours_per_month": 730,
                "days_per_month": 30,
                "corpus_documents": 20000,
                "queries_per_month": 100000,
                "daily_changed_fraction": 0.01,
                "replicas": 1,
                "memory_headroom_fraction": 0.25,
                "os_reserved_gib": 1,
                "cpu_requirement": "Smallest passing pinned N in2,4,8; zeroerrors, p95firstask<=1.5x own unpinned. Light passesC1; team passesbothC1/C8.",
                "instance_choice": "Cheapest eligible x86 EBS-only Linux on-demand shape fitting measured passing N and peak RSS+25%+1GiB OS; burstable only with idle CPU and CPU-seconds/query reference-traffic qualification, otherwise m/c non-burstable",
                "fyrr_reader": "Always-on, sized separately from measured reader memory/CPU",
                "fyrr_writer": "Scheduled writer sync+commit only; one-time full job separate from recurring daily incremental projection",
                "incremental_projection": "Daily job runtime = measured full sync+commit seconds * changed fraction; fixed scan/commit costs not separately measured, so projection has that caveat",
                "vm_boot_seconds_assumed": 60,
                "vm_minimum_billed_seconds": 60,
                "writer_billing": "max(60,ceil(projected sync+commit seconds +60 bootseconds)) per daily job; no continuously running writer; persistent root/cache EBS remains billed all month",
                "competitor_compute": "Measured full default/configured stack always-on in this reference deployment",
                "root_volume_base_gib": 20,
                "block_storage_headroom_fraction": 0.25,
                "s3_gets_per_query_assumed": 2,
                "s3_puts_per_month_assumed": 1000,
                "public_ipv4_usd_hour": 0.005,
                "public_ipv4_source": "https://aws.amazon.com/vpc/pricing/",
                "billing_source": "https://aws.amazon.com/ec2/pricing/on-demand/",
                "embedding_workload": "Official-tokenizer canonical text counts; common normalized input assumption, not product-billed usage",
                "ingest_llm": "Observed product operational document enrichment usage and priced projection where available",
                "discounts": "None; no credits/free grants/reservations/spot used for any product",
                "excluded": [
                  "HA replicas",
                  "cross-region traffic",
                  "taxes",
                  "support",
                  "operator labor",
                  "backups/snapshots",
                  "load balancer",
                  "Any assumption of cloud capacity/SLA equivalence"
                ],
                "million_document_projection": "Same reference query volume; scale stored bytes, changed-doc volume and token-based usage linearly. Compute capacity at1M is not measured; any fixed-instance line must be explicitly conditional and never a measured SLA.",
                "cpu_statistic": "Pinned capacity pass/fail determines core count; CPU counters retained as diagnostics only",
                "memory_requirement": "Measured peak RSS during steady query load*1.25 plus1GiB explicit OS allowance; writer peakRSS during ingest",
                "selection_approval": "3528fe7d-8770-4b6a-8800-784df515d564",
                "required_caveats": [
                  "Measurements on a20-core host",
                  "Ingest can be slower on the selected smaller instance; displayed ingest times are benchmark-host measurements",
                  "Cloud capacity/SLA has not been validated",
                  "Burstable qualification requires measured idle CPU plus CPU-seconds/query at the same reference traffic; products without that evidence use non-burstable minimum shapes"
                ],
                "instance_family_approval": "d55f078a-9dfa-4865-8c1c-31cbb17a8fe4",
                "basis": "minimum-measured",
                "load_scenario": "team",
                "load_concurrency": 8,
                "instance_name": "t3a.small",
                "instance_ram_gib": 2,
                "configuration_note": "C8, reader.concurrency=8; measured sized run and pinned capacity test, same index, no re-ingest"
              },
              "components": {
                "writer_hours_month": 0.5583333333333333,
                "compute_usd_month": 15.816575,
                "storage_usd_month": 4.001618671798706,
                "public_ipv4_usd_month": 3.6527916666666664,
                "object_requests_usd_month": 0.09140000000000001,
                "embedding_usd_per_document": 0.000074149092,
                "llm_enrichment_usd_per_document": 0,
                "changed_document_embeddings_usd_month": 0.444894552,
                "query_embeddings_usd_per_1000": 0.00224856,
                "query_embeddings_usd_month": 0.224856,
                "enrichment_usd_month": 0
              },
              "sizing": {
                "reader": {
                  "name": "t3a.small",
                  "vcpu": 2,
                  "ram_gib": 2,
                  "usd_hour": 0.0216,
                  "rate_code": "774XP29ZBNQKJHC3.JRTCKXETXF.6YS6EN2CT7",
                  "family": "General purpose",
                  "burstable": true,
                  "effective_usd_hour": 0.0216,
                  "required_ram_gib": 1.3227472305297852,
                  "required_vcpu": 2,
                  "resources": {
                    "peak_rss_bytes": 277237760,
                    "p95_cores": 13.693632726827053,
                    "peak_cores": 14.074346598170697,
                    "mean_cores": 4.098601693166352,
                    "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                    "artifacts": [
                      "${BENCH_ROOT}/runtime/fyrr-release-concurrency8/cpu-c8.jsonl"
                    ],
                    "selected_cpus": 2,
                    "cpu_seconds_per_query": 0.07986,
                    "cpu_work_scope": "250first-ask+250repeat; CPU counter diagnostic only",
                    "reference_mean_cores": 0.006381813325000143,
                    "idle_mean_cores": 0.0033430005396120153,
                    "burstable_qualified": true,
                    "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                  }
                },
                "writer": {
                  "name": "c5a.large",
                  "vcpu": 2,
                  "ram_gib": 4,
                  "usd_hour": 0.087,
                  "rate_code": "BBAVEJUPG9KCAM7G.JRTCKXETXF.6YS6EN2CT7",
                  "family": "Compute optimized",
                  "burstable": false,
                  "effective_usd_hour": 0.087,
                  "required_ram_gib": 1.6094026565551758,
                  "required_vcpu": 2,
                  "resources": {
                    "peak_rss_bytes": 523472896,
                    "p95_cores": 0.23272410605561036,
                    "peak_cores": 17.298930892056188,
                    "mean_cores": 0.09457849652583226,
                    "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                    "artifacts": [
                      "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.jsonl"
                    ],
                    "selected_cpus": 2,
                    "selection_basis": "Assumed2vCPU scheduled writer, measured writer peakRSS; pinned ingest capacity not tested. Host ingest time can be slower on this VM."
                  }
                },
                "capacity_tests": [
                  {
                    "cpus": 2,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.71095510106534,
                    "p95_ms": 367.9228202207014,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 214908928,
                    "artifact": "runtime/fyrr-capacity/n2/c1"
                  },
                  {
                    "cpus": 2,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 263.433939544484,
                    "p95_ms": 309.47251981124276,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 222191616,
                    "artifact": "runtime/fyrr-capacity/n2/c8"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 253.72744188643992,
                    "p95_ms": 588.8207679148762,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 219512832,
                    "artifact": "runtime/fyrr-capacity/n4/c1"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 258.01787059754133,
                    "p95_ms": 337.4477885430678,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 221077504,
                    "artifact": "runtime/fyrr-capacity/n4/c8"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 270.42482141405344,
                    "p95_ms": 569.3619827739875,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 229953536,
                    "artifact": "runtime/fyrr-capacity/n8/c1"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.25088198203593,
                    "p95_ms": 289.69891923479736,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 239128576,
                    "artifact": "runtime/fyrr-capacity/n8/c8"
                  }
                ],
                "selected_cpus": 2,
                "ram_headroom_fraction": 0.25,
                "measured_peak_rss_bytes": 277237760
              },
              "inputs": {
                "system": "fyrr",
                "documents": 20000,
                "reader": {
                  "peak_rss_bytes": 277237760,
                  "p95_cores": 13.693632726827053,
                  "peak_cores": 14.074346598170697,
                  "mean_cores": 4.098601693166352,
                  "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                  "artifacts": [
                    "${BENCH_ROOT}/runtime/fyrr-release-concurrency8/cpu-c8.jsonl"
                  ],
                  "selected_cpus": 2,
                  "cpu_seconds_per_query": 0.07986,
                  "cpu_work_scope": "250first-ask+250repeat; CPU counter diagnostic only",
                  "reference_mean_cores": 0.006381813325000143,
                  "idle_mean_cores": 0.0033430005396120153,
                  "burstable_qualified": true,
                  "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                },
                "scenario_resources": {
                  "light": {
                    "configuration_note": "C1 release default headline; pinned capacity test reader.concurrency=8 for all blocks",
                    "capacity_tests": [
                      {
                        "cpus": 2,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.71095510106534,
                        "p95_ms": 367.9228202207014,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 214908928,
                        "artifact": "runtime/fyrr-capacity/n2/c1"
                      },
                      {
                        "cpus": 2,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 263.433939544484,
                        "p95_ms": 309.47251981124276,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 222191616,
                        "artifact": "runtime/fyrr-capacity/n2/c8"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 253.72744188643992,
                        "p95_ms": 588.8207679148762,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 219512832,
                        "artifact": "runtime/fyrr-capacity/n4/c1"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 258.01787059754133,
                        "p95_ms": 337.4477885430678,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 221077504,
                        "artifact": "runtime/fyrr-capacity/n4/c8"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 270.42482141405344,
                        "p95_ms": 569.3619827739875,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 229953536,
                        "artifact": "runtime/fyrr-capacity/n8/c1"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.25088198203593,
                        "p95_ms": 289.69891923479736,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 239128576,
                        "artifact": "runtime/fyrr-capacity/n8/c8"
                      }
                    ],
                    "reader": {
                      "peak_rss_bytes": 214908928,
                      "p95_cores": 0.586274446513124,
                      "peak_cores": 0.6254239913304944,
                      "mean_cores": 0.3813949999048495,
                      "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                      "artifacts": [
                        "runtime/fyrr-release-full/cpu-query/c1/raw.jsonl"
                      ],
                      "selected_cpus": 2,
                      "cpu_seconds_per_query": 0.12400000000000001,
                      "reference_mean_cores": 0.008061417586796186,
                      "idle_mean_cores": 0.0033430005396120153,
                      "burstable_qualified": true,
                      "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                    },
                    "notes": [
                      "C1 default reader.concurrency=2; CPU common30 first-ask only; full primary query RSS"
                    ]
                  },
                  "team": {
                    "configuration_note": "C8, reader.concurrency=8; measured sized run and pinned capacity test, same index, no re-ingest",
                    "capacity_tests": [
                      {
                        "cpus": 2,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.71095510106534,
                        "p95_ms": 367.9228202207014,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 214908928,
                        "artifact": "runtime/fyrr-capacity/n2/c1"
                      },
                      {
                        "cpus": 2,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 263.433939544484,
                        "p95_ms": 309.47251981124276,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 222191616,
                        "artifact": "runtime/fyrr-capacity/n2/c8"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 253.72744188643992,
                        "p95_ms": 588.8207679148762,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 219512832,
                        "artifact": "runtime/fyrr-capacity/n4/c1"
                      },
                      {
                        "cpus": 4,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 258.01787059754133,
                        "p95_ms": 337.4477885430678,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 221077504,
                        "artifact": "runtime/fyrr-capacity/n4/c8"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 1,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 270.42482141405344,
                        "p95_ms": 569.3619827739875,
                        "threshold_p95_ms": 772.0598442654581,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 229953536,
                        "artifact": "runtime/fyrr-capacity/n8/c1"
                      },
                      {
                        "cpus": 8,
                        "concurrency": 8,
                        "expected": 250,
                        "valid": 250,
                        "errors": 0,
                        "p50_ms": 252.25088198203593,
                        "p95_ms": 289.69891923479736,
                        "threshold_p95_ms": 431.47896916489117,
                        "pass": true,
                        "timing_valid": true,
                        "measured_peak_rss_bytes": 239128576,
                        "artifact": "runtime/fyrr-capacity/n8/c8"
                      }
                    ],
                    "reader": {
                      "peak_rss_bytes": 277237760,
                      "p95_cores": 13.693632726827053,
                      "peak_cores": 14.074346598170697,
                      "mean_cores": 4.098601693166352,
                      "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                      "artifacts": [
                        "${BENCH_ROOT}/runtime/fyrr-release-concurrency8/cpu-c8.jsonl"
                      ],
                      "selected_cpus": 2,
                      "cpu_seconds_per_query": 0.07986,
                      "cpu_work_scope": "250first-ask+250repeat; CPU counter diagnostic only",
                      "reference_mean_cores": 0.006381813325000143,
                      "idle_mean_cores": 0.0033430005396120153,
                      "burstable_qualified": true,
                      "qualification_scope": "Measured native idle + CPU/query at100k/month; cloud per-core equivalence unvalidated"
                    },
                    "notes": [
                      "C8 reader.concurrency=8, same index; CPU/RSS from full250 first-ask plus250 repeat sized-run samples; no sizing from default C8 refusals"
                    ]
                  }
                },
                "writer": {
                  "peak_rss_bytes": 523472896,
                  "p95_cores": 0.23272410605561036,
                  "peak_cores": 17.298930892056188,
                  "mean_cores": 0.09457849652583226,
                  "aggregation": "Maximum of measured query-block statistics; writer uses ingest observer",
                  "artifacts": [
                    "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.jsonl"
                  ],
                  "selected_cpus": 2,
                  "selection_basis": "Assumed2vCPU scheduled writer, measured writer peakRSS; pinned ingest capacity not tested. Host ingest time can be slower on this VM."
                },
                "index_data_bytes": 141062144,
                "reader_cache_bytes": 134791168,
                "reader_software_bytes": 147272320,
                "writer_cache_bytes": 163987456,
                "writer_software_bytes": 147272320,
                "sync_commit_seconds": 667.7736084160861,
                "llm_enrichment_usd_per_document": 0,
                "notes": [
                  "No operational LLM enrichment in the measured local-files ingest/search pipeline.",
                  "Light and team scenarios use their own measured CPU/RSS. Team explicitly sets reader.concurrency=8; default C8 refusals are never used to size team compute.",
                  "Writer CPU observer includes the short excluded preflight interval; measured writer RSS includes the full sync, commit and visibility check.",
                  "Object-store size is measured using the native local object-store implementation; reference deployment prices those bytes on S3 Standard.",
                  "Native binary size is included in both local role volumes; persistent vector cache and serving mirror are retained separately.",
                  "Ingest accepted 19999 of 20000 documents; reference API costs use the common complete-corpus canonical token estimate.",
                  "Writer core count is an explicit2vCPU assumption; query capacity tests cover the reader, not writer ingestion throughput.",
                  "C8 reader.concurrency=8, same index; CPU/RSS from full250 first-ask plus250 repeat sized-run samples; no sizing from default C8 refusals"
                ],
                "artifacts": [
                  "${BENCH_ROOT}/runtime/fyrr-release-full/footprint.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/ingest.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c1.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-c8.json",
                  "${BENCH_ROOT}/runtime/fyrr-release-full/cpu-ingest.json"
                ],
                "configuration_note": "C8, reader.concurrency=8; measured sized run and pinned capacity test, same index, no re-ingest",
                "capacity_tests": [
                  {
                    "cpus": 2,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.71095510106534,
                    "p95_ms": 367.9228202207014,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 214908928,
                    "artifact": "runtime/fyrr-capacity/n2/c1"
                  },
                  {
                    "cpus": 2,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 263.433939544484,
                    "p95_ms": 309.47251981124276,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 222191616,
                    "artifact": "runtime/fyrr-capacity/n2/c8"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 253.72744188643992,
                    "p95_ms": 588.8207679148762,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 219512832,
                    "artifact": "runtime/fyrr-capacity/n4/c1"
                  },
                  {
                    "cpus": 4,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 258.01787059754133,
                    "p95_ms": 337.4477885430678,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 221077504,
                    "artifact": "runtime/fyrr-capacity/n4/c8"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 1,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 270.42482141405344,
                    "p95_ms": 569.3619827739875,
                    "threshold_p95_ms": 772.0598442654581,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 229953536,
                    "artifact": "runtime/fyrr-capacity/n8/c1"
                  },
                  {
                    "cpus": 8,
                    "concurrency": 8,
                    "expected": 250,
                    "valid": 250,
                    "errors": 0,
                    "p50_ms": 252.25088198203593,
                    "p95_ms": 289.69891923479736,
                    "threshold_p95_ms": 431.47896916489117,
                    "pass": true,
                    "timing_valid": true,
                    "measured_peak_rss_bytes": 239128576,
                    "artifact": "runtime/fyrr-capacity/n8/c8"
                  }
                ]
              },
              "one_time_full_ingest": {
                "compute_usd": 0.017593333333333332,
                "normalized_embedding_usd": 1.4829818399999999,
                "llm_enrichment_usd": 0,
                "total_usd": 1.5005751733333332
              },
              "sources": [
                {
                  "label": "aws-ec2-eu-central-1.json",
                  "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
                  "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
                  "publication_date": "2026-11-14T01:29:26Z"
                },
                {
                  "label": "aws-ebs.json",
                  "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
                  "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
                  "publication_date": "2026-11-14T01:29:26Z"
                },
                {
                  "label": "aws-s3-eu-central-1.json",
                  "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                  "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                  "publication_date": "2026-11-16T21:17:45Z"
                },
                {
                  "label": "Voyage list pricing",
                  "url": "https://docs.voyageai.com/docs/pricing"
                },
                {
                  "label": "Official tokenizer",
                  "url": "https://docs.voyageai.com/docs/tokenization"
                },
                {
                  "label": "Public IPv4",
                  "url": "https://aws.amazon.com/vpc/pricing/"
                },
                {
                  "label": "EC2 billing granularity",
                  "url": "https://aws.amazon.com/ec2/pricing/on-demand/"
                }
              ],
              "missing": [],
              "notes": [
                "No operational LLM enrichment in the measured local-files ingest/search pipeline.",
                "Light and team scenarios use their own measured CPU/RSS. Team explicitly sets reader.concurrency=8; default C8 refusals are never used to size team compute.",
                "Writer CPU observer includes the short excluded preflight interval; measured writer RSS includes the full sync, commit and visibility check.",
                "Object-store size is measured using the native local object-store implementation; reference deployment prices those bytes on S3 Standard.",
                "Native binary size is included in both local role volumes; persistent vector cache and serving mirror are retained separately.",
                "Ingest accepted 19999 of 20000 documents; reference API costs use the common complete-corpus canonical token estimate.",
                "Writer core count is an explicit2vCPU assumption; query capacity tests cover the reader, not writer ingestion throughput.",
                "C8 reader.concurrency=8, same index; CPU/RSS from full250 first-ask plus250 repeat sized-run samples; no sizing from default C8 refusals",
                "Daily incremental writer duration is a linear changed-document projection from measured full sync+commit plus assumed boot time; fixed scan/commit costs are not separately measured.",
                "Measurements on a20-core host",
                "Ingest can be slower on the selected smaller instance; displayed ingest times are benchmark-host measurements",
                "Cloud capacity/SLA has not been validated",
                "Burstable qualification requires measured idle CPU plus CPU-seconds/query at the same reference traffic; products without that evidence use non-burstable minimum shapes",
                "List-price deployment projection; provider-billed spend remains separate and can be n/e.",
                "Normalized embedding inputs use canonical text counts; native chunking, title prefixes, deduplication, retries and rejections can change actual product usage.",
                "CPU core demand is measured on the benchmark host; selected cloud-instance latency/capacity has not been benchmarked.",
                "Measured minimum quote uses a qualified burstable reader only where idle+query CPU evidence supports its baseline. Roles without that evidence stay on non-burstable m/c shapes. Persistent writer EBS stays billed while its VM is stopped."
              ],
              "first_month_total_usd": 25.732711063798707,
              "projections": [
                {
                  "documents": 1000000,
                  "status": "projection",
                  "comparable": false,
                  "total_usd": 48.32338385660197,
                  "components": {
                    "writer_hours_month": 3.283333333333333,
                    "compute_usd_month": 16.05365,
                    "storage_usd_month": 5.777733589935304,
                    "public_ipv4_usd_month": 3.6664166666666667,
                    "object_requests_usd_month": 0.35600000000000004,
                    "embedding_usd_per_document": 0.000074149092,
                    "llm_enrichment_usd_per_document": 0,
                    "changed_document_embeddings_usd_month": 22.244727599999997,
                    "query_embeddings_usd_per_1000": 0.00224856,
                    "query_embeddings_usd_month": 0.224856,
                    "enrichment_usd_month": 0
                  },
                  "assumptions": "Measured20k compute shapes held fixed; index/data and per-document API usage and writer time scale linearly; fixed software/model overhead stays fixed. Writer vector cache also scales linearly; software stays fixed.",
                  "capacity_validation": "n/e; no1M memory/CPU/latency measurement or cloud SLA claim"
                }
              ]
            }
          },
          "basis": "minimum-measured"
        },
        "monthly_cost_estimate": {
          "basis": "documented-production",
          "model_version": 2,
          "profile": "fyrr documented minimal production: HTTP reader, scheduled writer, S3",
          "status": "estimate",
          "total_usd": 42.29758737476408,
          "known_subtotal_usd": 42.29758737476408,
          "assumptions": {
            "provider": "AWS",
            "region": "eu-central-1",
            "currency": "USD",
            "price_date": "2026-11-17",
            "documents": 20000,
            "queries_per_month": 100000,
            "daily_changed_fraction": 0.01,
            "days_per_month": 30,
            "hours_per_month": 730,
            "eks_version": "1.36, current standard support where EKS is used",
            "client_response_bytes_assumed": 131072,
            "provider_response_bytes_per_query_assumed": 32768,
            "provider_bytes_per_changed_doc_assumed": 65536,
            "existing_idp_dns": "Existing organization IdP and DNS assumed; incremental fees excluded consistently",
            "promotions": "No free credits/grants, reservations or spot discounts applied",
            "cloud_validated": false,
            "existing_scheduler": "Existing job scheduler assumed; compute duty-billed but writer persistent EBS charged all month",
            "instance_name": "t3a.medium",
            "instance_ram_gib": 4,
            "reader_required_ram_gib": 3.7351112365722656,
            "reader_memory_max_bytes": 2415919104,
            "reader_memory_basis": "Maximum of observed native RSS plus25% and configured cache/preparation ceiling plus measured idle native overhead; then128MiB HTTP auth cache,256MiB TLS proxy,1GiB OS",
            "reader_memory_configuration_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/tuning.md#L91",
            "reader_idle_measured_cores": 0.0033430005396120153,
            "reader_cpu_seconds_per_query": 0.12400000000000001,
            "http_cpu_seconds_per_query_assumed": 0.01,
            "proxy_idle_cores_assumed": 0.025,
            "reference_mean_cores": 0.03344193509060136,
            "burstable_baseline_cores": 0.4,
            "burstable_surplus_forecast": 0,
            "writer_instance": "c5a.large",
            "writer_hours_month": 12.558333333333334,
            "configuration_note": "Documented HTTP reader with TLS proxy; existing IdP. Writer runs scheduled jobs, not continuously. Reader uses release DEFAULT memory_max_bytes=2304MiB; reader uses qualified burstable capacity; writer conservatively fixed performance 2 vCPU / 4 GiB."
          },
          "components": {
            "compute_usd_month": 32.628575,
            "storage_usd_month": 4.187837343597412,
            "network_usd_month": 4.811424479166666,
            "control_plane_usd_month": 0,
            "security_usd_month": 0,
            "observability_usd_month": 0,
            "api_usd_month": 0.669750552,
            "embedding_usd_per_document": 0.000074149092,
            "llm_enrichment_usd_per_document": 0,
            "query_embeddings_usd_per_1000": 0.00224856
          },
          "line_items": [
            {
              "key": "reader_compute",
              "label": "Always-on reader VM",
              "kind": "compute",
              "count": 1,
              "quantity": 730,
              "unit": "Hrs",
              "unit_price_usd": 0.0432,
              "usd_month": 31.536,
              "required": true,
              "quantity_basis": "Measured idle CPU + measured first-ask CPU/query at reference traffic; HTTP/proxy allowances explicit",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "label": "aws-ec2-eu-central-1.json",
                "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
                "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
                "publication_date": "2026-11-14T01:29:26Z"
              },
              "rate_code": "FCTDXPA9FM2E8Y98.JRTCKXETXF.6YS6EN2CT7"
            },
            {
              "key": "writer_compute",
              "label": "Scheduled sync/commit and maintenance allowance",
              "kind": "compute",
              "count": 1,
              "quantity": 12.558333333333334,
              "unit": "Hrs",
              "unit_price_usd": 0.087,
              "usd_month": 1.0925749999999999,
              "required": true,
              "quantity_basis": "Daily 1% runtime linear from full sync+commit +60s boot;4 maintenance jobs/day each assumed 300s + 60s boot, not measured cloud runtime",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md#scheduling",
              "price_source": {
                "label": "aws-ec2-eu-central-1.json",
                "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
                "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
                "publication_date": "2026-11-14T01:29:26Z"
              },
              "rate_code": "BBAVEJUPG9KCAM7G.JRTCKXETXF.6YS6EN2CT7"
            },
            {
              "key": "reader_root",
              "label": "Reader root/mirror volume",
              "kind": "storage",
              "count": 1,
              "quantity": 21,
              "unit": "GB-Mo",
              "unit_price_usd": 0.0952,
              "usd_month": 1.9992,
              "required": true,
              "quantity_basis": "20GiB base plus measured/projected mirror and fixed software allowance",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "label": "aws-ebs.json",
                "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
                "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
                "publication_date": "2026-11-14T01:29:26Z"
              },
              "rate_code": "43AUDFP7M3XV6YHQ.JRTCKXETXF.6YS6EN2CT7"
            },
            {
              "key": "writer_root",
              "label": "Persistent writer source/cache volume",
              "kind": "storage",
              "count": 1,
              "quantity": 21,
              "unit": "GB-Mo",
              "unit_price_usd": 0.0952,
              "usd_month": 1.9992,
              "required": true,
              "quantity_basis": "Provisioned writer volume remains billed while VM is stopped; cache/source data retained between jobs",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "label": "aws-ebs.json",
                "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
                "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
                "publication_date": "2026-11-14T01:29:26Z"
              },
              "rate_code": "43AUDFP7M3XV6YHQ.JRTCKXETXF.6YS6EN2CT7"
            },
            {
              "key": "ipv4",
              "label": "Reader/writer public IPv4",
              "kind": "network",
              "count": 1,
              "quantity": 742.5583333333333,
              "unit": "Hrs",
              "unit_price_usd": 0.005,
              "usd_month": 3.7127916666666665,
              "required": true,
              "quantity_basis": "reference forecast",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "url": "https://aws.amazon.com/vpc/pricing/",
                "verified_date": "2026-11-17"
              },
              "rate_code": null
            },
            {
              "key": "index_s3",
              "label": "Native namespace on S3 Standard",
              "kind": "storage",
              "count": 1,
              "quantity": 0.26274871826171875,
              "unit": "GiB-month",
              "unit_price_usd": 0.0245,
              "usd_month": 0.006437343597412109,
              "required": true,
              "quantity_basis": "Measured native local-store namespace bytes plus one full previous-generation storage allowance; remote S3 deployment not measured",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "label": "aws-s3-eu-central-1.json",
                "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                "publication_date": "2026-11-16T21:17:45Z"
              },
              "rate_code": null
            },
            {
              "key": "s3_reads",
              "label": "S3 GET/HEAD forecast",
              "kind": "storage",
              "count": 1,
              "quantity": 300000,
              "unit": "Requests",
              "unit_price_usd": 4.3e-7,
              "usd_month": 0.129,
              "required": true,
              "quantity_basis": "200k query admission/read checks plus 30s background checks and publication allowance",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "label": "aws-s3-eu-central-1.json",
                "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                "publication_date": "2026-11-16T21:17:45Z"
              },
              "rate_code": null
            },
            {
              "key": "s3_writes",
              "label": "S3 PUT/LIST forecast",
              "kind": "storage",
              "count": 1,
              "quantity": 10000,
              "unit": "Requests",
              "unit_price_usd": 0.0000054,
              "usd_month": 0.054,
              "required": true,
              "quantity_basis": "Explicit publication/maintenance object-operation allowance, not native observed S3 usage",
              "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
              "price_source": {
                "label": "aws-s3-eu-central-1.json",
                "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                "publication_date": "2026-11-16T21:17:45Z"
              },
              "rate_code": null
            },
            {
              "key": "internet",
              "label": "Client internet data-transfer forecast",
              "kind": "network",
              "count": 1,
              "quantity": 12.20703125,
              "unit": "GB",
              "unit_price_usd": 0.09,
              "usd_month": 1.0986328125,
              "required": true,
              "quantity_basis": "128KiB response per query; beyond account-level free transfer allocation",
              "deployment_source": "https://aws.amazon.com/ec2/pricing/on-demand/",
              "price_source": {
                "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AWSDataTransfer/current/eu-central-1/index.json",
                "bytes": 1522473,
                "sha256": "3e30dbdd219997662c3435bb80a6cb060627cffbc6c8ed5c02584d1960374230",
                "publication_date": "2026-11-17T00:49:47Z",
                "retrieved_at": "2026-11-17T08:24:28.810960+00:00",
                "version": "20261117004947",
                "version_url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AWSDataTransfer/20261117004947/eu-central-1/index.json"
              },
              "rate_code": "F62Y6TKR8NC6Q2ZB.JRTCKXETXF.Q3Z75P77EN"
            },
            {
              "key": "document_embeddings",
              "label": "Changed-document embedding estimate",
              "kind": "api",
              "count": 1,
              "quantity": 6000,
              "unit": "documents",
              "unit_price_usd": 0.000074149092,
              "usd_month": 0.444894552,
              "required": true,
              "quantity_basis": "Normalized canonical text estimate; native usage can differ",
              "price_source": {
                "url": "https://docs.voyageai.com/docs/pricing"
              }
            },
            {
              "key": "query_embeddings",
              "label": "First-ask query embedding estimate",
              "kind": "api",
              "count": 1,
              "quantity": 100,
              "unit": "1000queries",
              "unit_price_usd": 0.00224856,
              "usd_month": 0.224856,
              "required": true,
              "quantity_basis": "Normalized canonical text estimate; native usage can differ",
              "price_source": {
                "url": "https://docs.voyageai.com/docs/pricing"
              }
            },
            {
              "key": "llm_enrichment",
              "label": "Operational document enrichment estimate",
              "kind": "api",
              "count": 1,
              "quantity": 6000,
              "unit": "documents",
              "unit_price_usd": 0,
              "usd_month": 0,
              "required": true,
              "quantity_basis": "No operational LLM enrichment step in the measured ingest path; zero is a pipeline property, not an LLM unit price",
              "price_source": {
                "url": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md"
              }
            }
          ],
          "component_counts": {
            "always_on_reader_instances": 1,
            "tls_proxy_processes": 1,
            "scheduled_writer_role": 1,
            "s3_namespaces": 1,
            "external_embedding_apis": 1
          },
          "sources": [
            {
              "label": "fyrr deployment and scheduling",
              "url": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md"
            },
            {
              "label": "HTTP auth/cache settings",
              "url": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/reader-writer.md"
            },
            {
              "label": "Official normalized Voyage token estimate",
              "url": "https://docs.voyageai.com/docs/tokenization"
            }
          ],
          "missing": [],
          "notes": [
            "Vendor-documented production profile, not measured in the cloud; HA/SLA are not equivalent across these profiles.",
            "Production profile differs from the single-host benchmark topology; host latency/ingest measurements are not claims about these cloud deployments.",
            "List-price forecast under explicit traffic/storage assumptions, not actual provider billing.",
            "Internet transfer is priced beyond account free tiers; actual account-level allowances are not inferred.",
            "Production reader retains release DEFAULT memory_max_bytes=2304MiB. The VM fits that cache/preparation ceiling plus measured idle native overhead, HTTP auth/TLS allowances and1GiB OS. This is a deployment estimate, not cloud capacity validation.",
            "Reader CPU/RSS measured with native stdio/local-store path; HTTP authentication, TLS proxy and remote S3 overhead are forecast allowances, not cloud measurements.",
            "Writer c5a.large is a conservative fixed performance job shape; incremental duration and maintenance allowance are assumptions. Cloud ingestion can be slower than host measurements.",
            "Bursty reader eligibility uses measured idle CPU plus measured CPU/query, then adds explicit HTTP/proxy CPU. Expected average is below the bundled baseline; surplus credits are not assumed from account balances."
          ],
          "one_time_full_corpus_embedding_estimate_usd": 1.4829818399999999,
          "cloud_capacity_validation": "n/e",
          "projections": [
            {
              "basis": "documented-production",
              "model_version": 2,
              "profile": "fyrr documented minimal production: HTTP reader, scheduled writer, S3",
              "status": "projection",
              "total_usd": 102.05801825903727,
              "known_subtotal_usd": 102.05801825903727,
              "assumptions": {
                "provider": "AWS",
                "region": "eu-central-1",
                "currency": "USD",
                "price_date": "2026-11-17",
                "documents": 1000000,
                "queries_per_month": 100000,
                "daily_changed_fraction": 0.01,
                "days_per_month": 30,
                "hours_per_month": 730,
                "eks_version": "1.36, current standard support where EKS is used",
                "client_response_bytes_assumed": 131072,
                "provider_response_bytes_per_query_assumed": 32768,
                "provider_bytes_per_changed_doc_assumed": 65536,
                "existing_idp_dns": "Existing organization IdP and DNS assumed; incremental fees excluded consistently",
                "promotions": "No free credits/grants, reservations or spot discounts applied",
                "cloud_validated": false,
                "existing_scheduler": "Existing job scheduler assumed; compute duty-billed but writer persistent EBS charged all month",
                "instance_name": "t3a.large",
                "instance_ram_gib": 8,
                "reader_required_ram_gib": 7.485111236572266,
                "reader_memory_max_bytes": 6442450944,
                "reader_memory_basis": "Maximum of observed native RSS plus25% and configured cache/preparation ceiling plus measured idle native overhead; then128MiB HTTP auth cache,256MiB TLS proxy,1GiB OS",
                "reader_memory_configuration_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/tuning.md#L91",
                "reader_idle_measured_cores": 0.0033430005396120153,
                "reader_cpu_seconds_per_query": 6.200000000000001,
                "http_cpu_seconds_per_query_assumed": 0.01,
                "proxy_idle_cores_assumed": 0.025,
                "reference_mean_cores": 0.2646443704026258,
                "burstable_baseline_cores": 0.6,
                "burstable_surplus_forecast": 0,
                "writer_instance": "c5a.large",
                "writer_hours_month": 15.283333333333333,
                "configuration_note": "Documented HTTP reader with TLS proxy; existing IdP. Writer runs scheduled jobs, not continuously. Projected reader memory_max_bytes=6GiB explicitly, unvalidated; reader uses qualified burstable capacity; writer conservatively fixed performance 2 vCPU / 4 GiB.",
                "reader_projection": "Scale active-index RAM above measured idle linearly; retain observed request-workspace overhead; scale CPU work linearly and allow 6 GiB native cache/preparation budget; no 1M capacity or latency measurement"
              },
              "components": {
                "compute_usd_month": 64.40165,
                "storage_usd_month": 10.361735179870607,
                "network_usd_month": 4.825049479166667,
                "control_plane_usd_month": 0,
                "security_usd_month": 0,
                "observability_usd_month": 0,
                "api_usd_month": 22.469583599999996,
                "embedding_usd_per_document": 0.000074149092,
                "llm_enrichment_usd_per_document": 0,
                "query_embeddings_usd_per_1000": 0.00224856
              },
              "line_items": [
                {
                  "key": "reader_compute",
                  "label": "Always-on reader VM",
                  "kind": "compute",
                  "count": 1,
                  "quantity": 730,
                  "unit": "Hrs",
                  "unit_price_usd": 0.0864,
                  "usd_month": 63.072,
                  "required": true,
                  "quantity_basis": "Measured idle CPU + measured first-ask CPU/query at reference traffic; HTTP/proxy allowances explicit",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "label": "aws-ec2-eu-central-1.json",
                    "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
                    "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
                    "publication_date": "2026-11-14T01:29:26Z"
                  },
                  "rate_code": "W2TKRCZGWBBYBYMF.JRTCKXETXF.6YS6EN2CT7"
                },
                {
                  "key": "writer_compute",
                  "label": "Scheduled sync/commit and maintenance allowance",
                  "kind": "compute",
                  "count": 1,
                  "quantity": 15.283333333333333,
                  "unit": "Hrs",
                  "unit_price_usd": 0.087,
                  "usd_month": 1.32965,
                  "required": true,
                  "quantity_basis": "Daily 1% runtime linear from full sync+commit +60s boot;4 maintenance jobs/day each assumed 300s + 60s boot, not measured cloud runtime",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md#scheduling",
                  "price_source": {
                    "label": "aws-ec2-eu-central-1.json",
                    "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/EU%20%28Frankfurt%29/Linux/index.json",
                    "sha256": "947eb1357a3b6c1ca317a97cd7eb5e2f9b655ce0cc7e10ade3d25e9f8d7a09bd",
                    "publication_date": "2026-11-14T01:29:26Z"
                  },
                  "rate_code": "BBAVEJUPG9KCAM7G.JRTCKXETXF.6YS6EN2CT7"
                },
                {
                  "key": "reader_root",
                  "label": "Reader root/mirror volume",
                  "kind": "storage",
                  "count": 1,
                  "quantity": 29,
                  "unit": "GB-Mo",
                  "unit_price_usd": 0.0952,
                  "usd_month": 2.7608,
                  "required": true,
                  "quantity_basis": "20GiB base plus measured/projected mirror and fixed software allowance",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "label": "aws-ebs.json",
                    "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
                    "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
                    "publication_date": "2026-11-14T01:29:26Z"
                  },
                  "rate_code": "43AUDFP7M3XV6YHQ.JRTCKXETXF.6YS6EN2CT7"
                },
                {
                  "key": "writer_root",
                  "label": "Persistent writer source/cache volume",
                  "kind": "storage",
                  "count": 1,
                  "quantity": 44,
                  "unit": "GB-Mo",
                  "unit_price_usd": 0.0952,
                  "usd_month": 4.1888000000000005,
                  "required": true,
                  "quantity_basis": "Provisioned writer volume remains billed while VM is stopped; cache/source data retained between jobs",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "label": "aws-ebs.json",
                    "url": "https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ebs.json",
                    "sha256": "f50008b5fd99600fd181127ebf033edd2f98b9722dacf63bccb39747fa75ceec",
                    "publication_date": "2026-11-14T01:29:26Z"
                  },
                  "rate_code": "43AUDFP7M3XV6YHQ.JRTCKXETXF.6YS6EN2CT7"
                },
                {
                  "key": "ipv4",
                  "label": "Reader/writer public IPv4",
                  "kind": "network",
                  "count": 1,
                  "quantity": 745.2833333333333,
                  "unit": "Hrs",
                  "unit_price_usd": 0.005,
                  "usd_month": 3.7264166666666667,
                  "required": true,
                  "quantity_basis": "reference forecast",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "url": "https://aws.amazon.com/vpc/pricing/",
                    "verified_date": "2026-11-17"
                  },
                  "rate_code": null
                },
                {
                  "key": "index_s3",
                  "label": "Native namespace on S3 Standard",
                  "kind": "storage",
                  "count": 1,
                  "quantity": 13.137435913085938,
                  "unit": "GiB-month",
                  "unit_price_usd": 0.0245,
                  "usd_month": 0.3218671798706055,
                  "required": true,
                  "quantity_basis": "Measured native local-store namespace bytes plus one full previous-generation storage allowance; remote S3 deployment not measured",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "label": "aws-s3-eu-central-1.json",
                    "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                    "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                    "publication_date": "2026-11-16T21:17:45Z"
                  },
                  "rate_code": null
                },
                {
                  "key": "s3_reads",
                  "label": "S3 GET/HEAD forecast",
                  "kind": "storage",
                  "count": 1,
                  "quantity": 907600,
                  "unit": "Requests",
                  "unit_price_usd": 4.3e-7,
                  "usd_month": 0.390268,
                  "required": true,
                  "quantity_basis": "200k query admission/read checks plus 30s background checks and publication allowance",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "label": "aws-s3-eu-central-1.json",
                    "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                    "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                    "publication_date": "2026-11-16T21:17:45Z"
                  },
                  "rate_code": null
                },
                {
                  "key": "s3_writes",
                  "label": "S3 PUT/LIST forecast",
                  "kind": "storage",
                  "count": 1,
                  "quantity": 500000,
                  "unit": "Requests",
                  "unit_price_usd": 0.0000054,
                  "usd_month": 2.7,
                  "required": true,
                  "quantity_basis": "Explicit publication/maintenance object-operation allowance, not native observed S3 usage",
                  "deployment_source": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md",
                  "price_source": {
                    "label": "aws-s3-eu-central-1.json",
                    "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/eu-central-1/index.json",
                    "sha256": "3ce274a4cfcad418bdae1089f092486b2ce33f59978ec874122f5749144e3453",
                    "publication_date": "2026-11-16T21:17:45Z"
                  },
                  "rate_code": null
                },
                {
                  "key": "internet",
                  "label": "Client internet data-transfer forecast",
                  "kind": "network",
                  "count": 1,
                  "quantity": 12.20703125,
                  "unit": "GB",
                  "unit_price_usd": 0.09,
                  "usd_month": 1.0986328125,
                  "required": true,
                  "quantity_basis": "128KiB response per query; beyond account-level free transfer allocation",
                  "deployment_source": "https://aws.amazon.com/ec2/pricing/on-demand/",
                  "price_source": {
                    "url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AWSDataTransfer/current/eu-central-1/index.json",
                    "bytes": 1522473,
                    "sha256": "3e30dbdd219997662c3435bb80a6cb060627cffbc6c8ed5c02584d1960374230",
                    "publication_date": "2026-11-17T00:49:47Z",
                    "retrieved_at": "2026-11-17T08:24:28.810960+00:00",
                    "version": "20261117004947",
                    "version_url": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AWSDataTransfer/20261117004947/eu-central-1/index.json"
                  },
                  "rate_code": "F62Y6TKR8NC6Q2ZB.JRTCKXETXF.Q3Z75P77EN"
                },
                {
                  "key": "document_embeddings",
                  "label": "Changed-document embedding estimate",
                  "kind": "api",
                  "count": 1,
                  "quantity": 300000,
                  "unit": "documents",
                  "unit_price_usd": 0.000074149092,
                  "usd_month": 22.244727599999997,
                  "required": true,
                  "quantity_basis": "Normalized canonical text estimate; native usage can differ",
                  "price_source": {
                    "url": "https://docs.voyageai.com/docs/pricing"
                  }
                },
                {
                  "key": "query_embeddings",
                  "label": "First-ask query embedding estimate",
                  "kind": "api",
                  "count": 1,
                  "quantity": 100,
                  "unit": "1000queries",
                  "unit_price_usd": 0.00224856,
                  "usd_month": 0.224856,
                  "required": true,
                  "quantity_basis": "Normalized canonical text estimate; native usage can differ",
                  "price_source": {
                    "url": "https://docs.voyageai.com/docs/pricing"
                  }
                },
                {
                  "key": "llm_enrichment",
                  "label": "Operational document enrichment estimate",
                  "kind": "api",
                  "count": 1,
                  "quantity": 300000,
                  "unit": "documents",
                  "unit_price_usd": 0,
                  "usd_month": 0,
                  "required": true,
                  "quantity_basis": "No operational LLM enrichment step in the measured ingest path; zero is a pipeline property, not an LLM unit price",
                  "price_source": {
                    "url": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md"
                  }
                }
              ],
              "component_counts": {
                "always_on_reader_instances": 1,
                "tls_proxy_processes": 1,
                "scheduled_writer_role": 1,
                "s3_namespaces": 1,
                "external_embedding_apis": 1
              },
              "sources": [
                {
                  "label": "fyrr deployment and scheduling",
                  "url": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/deployment.md"
                },
                {
                  "label": "HTTP auth/cache settings",
                  "url": "https://github.com/fyrr-ai/fyrr/blob/38e7df7198c52f340da1737dca6af665acc69b96/docs/reader-writer.md"
                },
                {
                  "label": "Official normalized Voyage token estimate",
                  "url": "https://docs.voyageai.com/docs/tokenization"
                }
              ],
              "missing": [],
              "notes": [
                "Vendor-documented production profile, not measured in the cloud; HA/SLA are not equivalent across these profiles.",
                "Production profile differs from the single-host benchmark topology; host latency/ingest measurements are not claims about these cloud deployments.",
                "List-price forecast under explicit traffic/storage assumptions, not actual provider billing.",
                "Internet transfer is priced beyond account free tiers; actual account-level allowances are not inferred.",
                "The 1M projection increases native memory_max_bytes to 6 GiB and the VM to 8 GiB to accommodate projected active-index memory; neither working-set scaling nor this configuration has been measured.",
                "Reader CPU/RSS measured with native stdio/local-store path; HTTP authentication, TLS proxy and remote S3 overhead are forecast allowances, not cloud measurements.",
                "Writer c5a.large is a conservative fixed performance job shape; incremental duration and maintenance allowance are assumptions. Cloud ingestion can be slower than host measurements.",
                "Bursty reader eligibility uses measured idle CPU plus measured CPU/query, then adds explicit HTTP/proxy CPU. Expected average is below the bundled baseline; surplus credits are not assumed from account balances.",
                "One-million-document profile/work forecast; native reader memory/CPU scaling and every cloud capacity/SLA remain unvalidated"
              ],
              "one_time_full_corpus_embedding_estimate_usd": 74.149092,
              "cloud_capacity_validation": "n/e",
              "comparable": false
            }
          ]
        }
      },
      "ingest_coverage": {
        "submitted": 20000,
        "accepted": 19999,
        "rejected": 1,
        "rejected_documents": [
          {
            "doc_id": "dsid_fd0cc1a9394840a6a0642be4ee470a18",
            "reason": "document_invalid",
            "native_status": "quarantined",
            "native_root_digest": "sha256:f97d75222bf1ce0c877afe15ffe64f9a2c192eee0ff39102b657f70323f12e9a"
          }
        ],
        "searchable": 19999,
        "pending": 0,
        "pending_documents": []
      },
      "ingest_completion": {
        "status": "verified_complete",
        "verified_searchable": true,
        "verified_inventory": true,
        "all_accepted_searchable": true,
        "searchable": 19999,
        "pending": 0,
        "canonical_timing_valid": true,
        "artifact": "runtime/fyrr-release-full/ingest.json",
        "resume_evidence": null
      },
      "notes": [
        "Ingest coverage 19,999/20,000; 1 rejected document(s), with native reasons and IDs in ingest.json. No gold questions excluded."
      ],
      "publication_ready_primary": false,
      "artifacts": {
        "run_dir": "runtime/fyrr-release-full",
        "adapter_files_sha256": {
          "adapters/fyrr.py": "4abcffc292f1b38e55c08ff665a48728f1597996377417f93366db62ab38ea4b"
        },
        "protocol_amendment": null,
        "ingest_status": "ok"
      },
      "source_slices": {
        "confluence": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.860204081632653,
              "unit": "fraction",
              "n": 49,
              "attempted": 49,
              "expected": 49,
              "missing": 0,
              "ci95": [
                0.7929109977324262,
                0.9206391723356009
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0011",
                "qst_0013",
                "qst_0016",
                "qst_0022",
                "qst_0035",
                "qst_0040",
                "qst_0102",
                "qst_0104",
                "qst_0117",
                "qst_0174",
                "qst_0216",
                "qst_0221",
                "qst_0252",
                "qst_0270",
                "qst_0341",
                "qst_0342",
                "qst_0345",
                "qst_0354",
                "qst_0355",
                "qst_0356",
                "qst_0362",
                "qst_0363",
                "qst_0365",
                "qst_0367",
                "qst_0369",
                "qst_0375",
                "qst_0376",
                "qst_0377",
                "qst_0381",
                "qst_0386",
                "qst_0390",
                "qst_0393",
                "qst_0394",
                "qst_0402",
                "qst_0404",
                "qst_0412",
                "qst_0417",
                "qst_0419",
                "qst_0420",
                "qst_0422",
                "qst_0426",
                "qst_0428",
                "qst_0434",
                "qst_0435",
                "qst_0438",
                "qst_0439",
                "qst_0442",
                "qst_0445",
                "qst_0447"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8492411381376049,
              "unit": "fraction",
              "n": 49,
              "attempted": 49,
              "expected": 49,
              "missing": 0,
              "ci95": [
                0.7825905270670565,
                0.9089144065768502
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0011",
                "qst_0013",
                "qst_0016",
                "qst_0022",
                "qst_0035",
                "qst_0040",
                "qst_0102",
                "qst_0104",
                "qst_0117",
                "qst_0174",
                "qst_0216",
                "qst_0221",
                "qst_0252",
                "qst_0270",
                "qst_0341",
                "qst_0342",
                "qst_0345",
                "qst_0354",
                "qst_0355",
                "qst_0356",
                "qst_0362",
                "qst_0363",
                "qst_0365",
                "qst_0367",
                "qst_0369",
                "qst_0375",
                "qst_0376",
                "qst_0377",
                "qst_0381",
                "qst_0386",
                "qst_0390",
                "qst_0393",
                "qst_0394",
                "qst_0402",
                "qst_0404",
                "qst_0412",
                "qst_0417",
                "qst_0419",
                "qst_0420",
                "qst_0422",
                "qst_0426",
                "qst_0428",
                "qst_0434",
                "qst_0435",
                "qst_0438",
                "qst_0439",
                "qst_0442",
                "qst_0445",
                "qst_0447"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 0.9895691609977324,
              "unit": "fraction",
              "n": 49,
              "attempted": 49,
              "expected": 49,
              "missing": 0,
              "ci95": [
                0.9709750566893424,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0011",
                "qst_0013",
                "qst_0016",
                "qst_0022",
                "qst_0035",
                "qst_0040",
                "qst_0102",
                "qst_0104",
                "qst_0117",
                "qst_0174",
                "qst_0216",
                "qst_0221",
                "qst_0252",
                "qst_0270",
                "qst_0341",
                "qst_0342",
                "qst_0345",
                "qst_0354",
                "qst_0355",
                "qst_0356",
                "qst_0362",
                "qst_0363",
                "qst_0365",
                "qst_0367",
                "qst_0369",
                "qst_0375",
                "qst_0376",
                "qst_0377",
                "qst_0381",
                "qst_0386",
                "qst_0390",
                "qst_0393",
                "qst_0394",
                "qst_0402",
                "qst_0404",
                "qst_0412",
                "qst_0417",
                "qst_0419",
                "qst_0420",
                "qst_0422",
                "qst_0426",
                "qst_0428",
                "qst_0434",
                "qst_0435",
                "qst_0438",
                "qst_0439",
                "qst_0442",
                "qst_0445",
                "qst_0447"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "fireflies": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.9090909090909091,
              "unit": "fraction",
              "n": 11,
              "attempted": 11,
              "expected": 11,
              "missing": 0,
              "ci95": [
                0.7272727272727273,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0004",
                "qst_0033",
                "qst_0041",
                "qst_0082",
                "qst_0247",
                "qst_0260",
                "qst_0333",
                "qst_0417",
                "qst_0443",
                "qst_0460",
                "qst_0470"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8046982307621477,
              "unit": "fraction",
              "n": 11,
              "attempted": 11,
              "expected": 11,
              "missing": 0,
              "ci95": [
                0.5867036845192354,
                0.9781056697676876
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0004",
                "qst_0033",
                "qst_0041",
                "qst_0082",
                "qst_0247",
                "qst_0260",
                "qst_0333",
                "qst_0417",
                "qst_0443",
                "qst_0460",
                "qst_0470"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 1,
              "unit": "fraction",
              "n": 11,
              "attempted": 11,
              "expected": 11,
              "missing": 0,
              "ci95": [
                1,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0004",
                "qst_0033",
                "qst_0041",
                "qst_0082",
                "qst_0247",
                "qst_0260",
                "qst_0333",
                "qst_0417",
                "qst_0443",
                "qst_0460",
                "qst_0470"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "github": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.9241013071895425,
              "unit": "fraction",
              "n": 34,
              "attempted": 34,
              "expected": 34,
              "missing": 0,
              "ci95": [
                0.8449346405228758,
                0.9852941176470589
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0002",
                "qst_0010",
                "qst_0052",
                "qst_0074",
                "qst_0077",
                "qst_0089",
                "qst_0103",
                "qst_0125",
                "qst_0127",
                "qst_0147",
                "qst_0148",
                "qst_0161",
                "qst_0167",
                "qst_0168",
                "qst_0180",
                "qst_0189",
                "qst_0192",
                "qst_0230",
                "qst_0290",
                "qst_0322",
                "qst_0339",
                "qst_0343",
                "qst_0346",
                "qst_0347",
                "qst_0349",
                "qst_0350",
                "qst_0354",
                "qst_0362",
                "qst_0363",
                "qst_0365",
                "qst_0369",
                "qst_0375",
                "qst_0436",
                "qst_0467"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8996671858824913,
              "unit": "fraction",
              "n": 34,
              "attempted": 34,
              "expected": 34,
              "missing": 0,
              "ci95": [
                0.8264662297893071,
                0.9605308376394527
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0002",
                "qst_0010",
                "qst_0052",
                "qst_0074",
                "qst_0077",
                "qst_0089",
                "qst_0103",
                "qst_0125",
                "qst_0127",
                "qst_0147",
                "qst_0148",
                "qst_0161",
                "qst_0167",
                "qst_0168",
                "qst_0180",
                "qst_0189",
                "qst_0192",
                "qst_0230",
                "qst_0290",
                "qst_0322",
                "qst_0339",
                "qst_0343",
                "qst_0346",
                "qst_0347",
                "qst_0349",
                "qst_0350",
                "qst_0354",
                "qst_0362",
                "qst_0363",
                "qst_0365",
                "qst_0369",
                "qst_0375",
                "qst_0436",
                "qst_0467"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 0.9794117647058823,
              "unit": "fraction",
              "n": 34,
              "attempted": 34,
              "expected": 34,
              "missing": 0,
              "ci95": [
                0.9382352941176471,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0002",
                "qst_0010",
                "qst_0052",
                "qst_0074",
                "qst_0077",
                "qst_0089",
                "qst_0103",
                "qst_0125",
                "qst_0127",
                "qst_0147",
                "qst_0148",
                "qst_0161",
                "qst_0167",
                "qst_0168",
                "qst_0180",
                "qst_0189",
                "qst_0192",
                "qst_0230",
                "qst_0290",
                "qst_0322",
                "qst_0339",
                "qst_0343",
                "qst_0346",
                "qst_0347",
                "qst_0349",
                "qst_0350",
                "qst_0354",
                "qst_0362",
                "qst_0363",
                "qst_0365",
                "qst_0369",
                "qst_0375",
                "qst_0436",
                "qst_0467"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "gmail": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.9105902777777778,
              "unit": "fraction",
              "n": 32,
              "attempted": 32,
              "expected": 32,
              "missing": 0,
              "ci95": [
                0.8263888888888888,
                0.9756944444444444
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0009",
                "qst_0026",
                "qst_0044",
                "qst_0046",
                "qst_0049",
                "qst_0055",
                "qst_0098",
                "qst_0145",
                "qst_0165",
                "qst_0166",
                "qst_0177",
                "qst_0231",
                "qst_0235",
                "qst_0250",
                "qst_0255",
                "qst_0257",
                "qst_0281",
                "qst_0284",
                "qst_0299",
                "qst_0304",
                "qst_0314",
                "qst_0324",
                "qst_0336",
                "qst_0342",
                "qst_0345",
                "qst_0346",
                "qst_0355",
                "qst_0356",
                "qst_0362",
                "qst_0367",
                "qst_0414",
                "qst_0443"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8346031513094934,
              "unit": "fraction",
              "n": 32,
              "attempted": 32,
              "expected": 32,
              "missing": 0,
              "ci95": [
                0.7399100024106432,
                0.9169594599549985
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0009",
                "qst_0026",
                "qst_0044",
                "qst_0046",
                "qst_0049",
                "qst_0055",
                "qst_0098",
                "qst_0145",
                "qst_0165",
                "qst_0166",
                "qst_0177",
                "qst_0231",
                "qst_0235",
                "qst_0250",
                "qst_0255",
                "qst_0257",
                "qst_0281",
                "qst_0284",
                "qst_0299",
                "qst_0304",
                "qst_0314",
                "qst_0324",
                "qst_0336",
                "qst_0342",
                "qst_0345",
                "qst_0346",
                "qst_0355",
                "qst_0356",
                "qst_0362",
                "qst_0367",
                "qst_0414",
                "qst_0443"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 0.984375,
              "unit": "fraction",
              "n": 32,
              "attempted": 32,
              "expected": 32,
              "missing": 0,
              "ci95": [
                0.953125,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0009",
                "qst_0026",
                "qst_0044",
                "qst_0046",
                "qst_0049",
                "qst_0055",
                "qst_0098",
                "qst_0145",
                "qst_0165",
                "qst_0166",
                "qst_0177",
                "qst_0231",
                "qst_0235",
                "qst_0250",
                "qst_0255",
                "qst_0257",
                "qst_0281",
                "qst_0284",
                "qst_0299",
                "qst_0304",
                "qst_0314",
                "qst_0324",
                "qst_0336",
                "qst_0342",
                "qst_0345",
                "qst_0346",
                "qst_0355",
                "qst_0356",
                "qst_0362",
                "qst_0367",
                "qst_0414",
                "qst_0443"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "google_drive": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.9234126984126984,
              "unit": "fraction",
              "n": 35,
              "attempted": 35,
              "expected": 35,
              "missing": 0,
              "ci95": [
                0.844047619047619,
                0.9841269841269841
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0008",
                "qst_0050",
                "qst_0069",
                "qst_0079",
                "qst_0085",
                "qst_0105",
                "qst_0122",
                "qst_0134",
                "qst_0141",
                "qst_0155",
                "qst_0169",
                "qst_0186",
                "qst_0195",
                "qst_0199",
                "qst_0200",
                "qst_0212",
                "qst_0238",
                "qst_0280",
                "qst_0283",
                "qst_0291",
                "qst_0350",
                "qst_0354",
                "qst_0365",
                "qst_0369",
                "qst_0377",
                "qst_0391",
                "qst_0411",
                "qst_0412",
                "qst_0419",
                "qst_0420",
                "qst_0428",
                "qst_0434",
                "qst_0438",
                "qst_0452",
                "qst_0461"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8109134892436375,
              "unit": "fraction",
              "n": 35,
              "attempted": 35,
              "expected": 35,
              "missing": 0,
              "ci95": [
                0.7175870064222313,
                0.8943299618401187
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0008",
                "qst_0050",
                "qst_0069",
                "qst_0079",
                "qst_0085",
                "qst_0105",
                "qst_0122",
                "qst_0134",
                "qst_0141",
                "qst_0155",
                "qst_0169",
                "qst_0186",
                "qst_0195",
                "qst_0199",
                "qst_0200",
                "qst_0212",
                "qst_0238",
                "qst_0280",
                "qst_0283",
                "qst_0291",
                "qst_0350",
                "qst_0354",
                "qst_0365",
                "qst_0369",
                "qst_0377",
                "qst_0391",
                "qst_0411",
                "qst_0412",
                "qst_0419",
                "qst_0420",
                "qst_0428",
                "qst_0434",
                "qst_0438",
                "qst_0452",
                "qst_0461"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 0.9968253968253967,
              "unit": "fraction",
              "n": 35,
              "attempted": 35,
              "expected": 35,
              "missing": 0,
              "ci95": [
                0.9904761904761904,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0008",
                "qst_0050",
                "qst_0069",
                "qst_0079",
                "qst_0085",
                "qst_0105",
                "qst_0122",
                "qst_0134",
                "qst_0141",
                "qst_0155",
                "qst_0169",
                "qst_0186",
                "qst_0195",
                "qst_0199",
                "qst_0200",
                "qst_0212",
                "qst_0238",
                "qst_0280",
                "qst_0283",
                "qst_0291",
                "qst_0350",
                "qst_0354",
                "qst_0365",
                "qst_0369",
                "qst_0377",
                "qst_0391",
                "qst_0411",
                "qst_0412",
                "qst_0419",
                "qst_0420",
                "qst_0428",
                "qst_0434",
                "qst_0438",
                "qst_0452",
                "qst_0461"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "hubspot": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 1,
              "unit": "fraction",
              "n": 15,
              "attempted": 15,
              "expected": 15,
              "missing": 0,
              "ci95": [
                1,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0015",
                "qst_0063",
                "qst_0066",
                "qst_0124",
                "qst_0130",
                "qst_0138",
                "qst_0209",
                "qst_0213",
                "qst_0239",
                "qst_0245",
                "qst_0253",
                "qst_0254",
                "qst_0274",
                "qst_0277",
                "qst_0286"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.9128357376810872,
              "unit": "fraction",
              "n": 15,
              "attempted": 15,
              "expected": 15,
              "missing": 0,
              "ci95": [
                0.8123212623289701,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0015",
                "qst_0063",
                "qst_0066",
                "qst_0124",
                "qst_0130",
                "qst_0138",
                "qst_0209",
                "qst_0213",
                "qst_0239",
                "qst_0245",
                "qst_0253",
                "qst_0254",
                "qst_0274",
                "qst_0277",
                "qst_0286"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 1,
              "unit": "fraction",
              "n": 15,
              "attempted": 15,
              "expected": 15,
              "missing": 0,
              "ci95": [
                1,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0015",
                "qst_0063",
                "qst_0066",
                "qst_0124",
                "qst_0130",
                "qst_0138",
                "qst_0209",
                "qst_0213",
                "qst_0239",
                "qst_0245",
                "qst_0253",
                "qst_0254",
                "qst_0274",
                "qst_0277",
                "qst_0286"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "jira": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.895352564102564,
              "unit": "fraction",
              "n": 52,
              "attempted": 52,
              "expected": 52,
              "missing": 0,
              "ci95": [
                0.8241439636752136,
                0.9553966346153846
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0012",
                "qst_0059",
                "qst_0065",
                "qst_0083",
                "qst_0086",
                "qst_0096",
                "qst_0129",
                "qst_0132",
                "qst_0157",
                "qst_0181",
                "qst_0198",
                "qst_0215",
                "qst_0217",
                "qst_0223",
                "qst_0228",
                "qst_0241",
                "qst_0256",
                "qst_0262",
                "qst_0265",
                "qst_0271",
                "qst_0289",
                "qst_0293",
                "qst_0331",
                "qst_0341",
                "qst_0343",
                "qst_0346",
                "qst_0347",
                "qst_0349",
                "qst_0350",
                "qst_0354",
                "qst_0356",
                "qst_0362",
                "qst_0363",
                "qst_0367",
                "qst_0369",
                "qst_0375",
                "qst_0376",
                "qst_0377",
                "qst_0389",
                "qst_0390",
                "qst_0392",
                "qst_0393",
                "qst_0395",
                "qst_0396",
                "qst_0402",
                "qst_0407",
                "qst_0411",
                "qst_0414",
                "qst_0423",
                "qst_0436",
                "qst_0438",
                "qst_0449"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8790205851695014,
              "unit": "fraction",
              "n": 52,
              "attempted": 52,
              "expected": 52,
              "missing": 0,
              "ci95": [
                0.8098518565959419,
                0.9385181253652212
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0012",
                "qst_0059",
                "qst_0065",
                "qst_0083",
                "qst_0086",
                "qst_0096",
                "qst_0129",
                "qst_0132",
                "qst_0157",
                "qst_0181",
                "qst_0198",
                "qst_0215",
                "qst_0217",
                "qst_0223",
                "qst_0228",
                "qst_0241",
                "qst_0256",
                "qst_0262",
                "qst_0265",
                "qst_0271",
                "qst_0289",
                "qst_0293",
                "qst_0331",
                "qst_0341",
                "qst_0343",
                "qst_0346",
                "qst_0347",
                "qst_0349",
                "qst_0350",
                "qst_0354",
                "qst_0356",
                "qst_0362",
                "qst_0363",
                "qst_0367",
                "qst_0369",
                "qst_0375",
                "qst_0376",
                "qst_0377",
                "qst_0389",
                "qst_0390",
                "qst_0392",
                "qst_0393",
                "qst_0395",
                "qst_0396",
                "qst_0402",
                "qst_0407",
                "qst_0411",
                "qst_0414",
                "qst_0423",
                "qst_0436",
                "qst_0438",
                "qst_0449"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 0.9747863247863248,
              "unit": "fraction",
              "n": 52,
              "attempted": 52,
              "expected": 52,
              "missing": 0,
              "ci95": [
                0.9382478632478632,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0012",
                "qst_0059",
                "qst_0065",
                "qst_0083",
                "qst_0086",
                "qst_0096",
                "qst_0129",
                "qst_0132",
                "qst_0157",
                "qst_0181",
                "qst_0198",
                "qst_0215",
                "qst_0217",
                "qst_0223",
                "qst_0228",
                "qst_0241",
                "qst_0256",
                "qst_0262",
                "qst_0265",
                "qst_0271",
                "qst_0289",
                "qst_0293",
                "qst_0331",
                "qst_0341",
                "qst_0343",
                "qst_0346",
                "qst_0347",
                "qst_0349",
                "qst_0350",
                "qst_0354",
                "qst_0356",
                "qst_0362",
                "qst_0363",
                "qst_0367",
                "qst_0369",
                "qst_0375",
                "qst_0376",
                "qst_0377",
                "qst_0389",
                "qst_0390",
                "qst_0392",
                "qst_0393",
                "qst_0395",
                "qst_0396",
                "qst_0402",
                "qst_0407",
                "qst_0411",
                "qst_0414",
                "qst_0423",
                "qst_0436",
                "qst_0438",
                "qst_0449"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "linear": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.9615384615384616,
              "unit": "fraction",
              "n": 26,
              "attempted": 26,
              "expected": 26,
              "missing": 0,
              "ci95": [
                0.9102564102564104,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0030",
                "qst_0032",
                "qst_0048",
                "qst_0067",
                "qst_0071",
                "qst_0106",
                "qst_0111",
                "qst_0158",
                "qst_0162",
                "qst_0163",
                "qst_0175",
                "qst_0182",
                "qst_0187",
                "qst_0218",
                "qst_0275",
                "qst_0285",
                "qst_0301",
                "qst_0315",
                "qst_0317",
                "qst_0323",
                "qst_0360",
                "qst_0365",
                "qst_0377",
                "qst_0422",
                "qst_0423",
                "qst_0426"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8771948757934364,
              "unit": "fraction",
              "n": 26,
              "attempted": 26,
              "expected": 26,
              "missing": 0,
              "ci95": [
                0.7831074171780352,
                0.9561816037659683
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0030",
                "qst_0032",
                "qst_0048",
                "qst_0067",
                "qst_0071",
                "qst_0106",
                "qst_0111",
                "qst_0158",
                "qst_0162",
                "qst_0163",
                "qst_0175",
                "qst_0182",
                "qst_0187",
                "qst_0218",
                "qst_0275",
                "qst_0285",
                "qst_0301",
                "qst_0315",
                "qst_0317",
                "qst_0323",
                "qst_0360",
                "qst_0365",
                "qst_0377",
                "qst_0422",
                "qst_0423",
                "qst_0426"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 1,
              "unit": "fraction",
              "n": 26,
              "attempted": 26,
              "expected": 26,
              "missing": 0,
              "ci95": [
                1,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0030",
                "qst_0032",
                "qst_0048",
                "qst_0067",
                "qst_0071",
                "qst_0106",
                "qst_0111",
                "qst_0158",
                "qst_0162",
                "qst_0163",
                "qst_0175",
                "qst_0182",
                "qst_0187",
                "qst_0218",
                "qst_0275",
                "qst_0285",
                "qst_0301",
                "qst_0315",
                "qst_0317",
                "qst_0323",
                "qst_0360",
                "qst_0365",
                "qst_0377",
                "qst_0422",
                "qst_0423",
                "qst_0426"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        },
        "slack": {
          "metrics": {
            "recall_at_10": {
              "status": "measured",
              "value": 0.8848484848484848,
              "unit": "fraction",
              "n": 44,
              "attempted": 44,
              "expected": 44,
              "missing": 0,
              "ci95": [
                0.8,
                0.9558080808080809
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0028",
                "qst_0043",
                "qst_0051",
                "qst_0053",
                "qst_0064",
                "qst_0088",
                "qst_0092",
                "qst_0100",
                "qst_0152",
                "qst_0153",
                "qst_0154",
                "qst_0172",
                "qst_0173",
                "qst_0176",
                "qst_0224",
                "qst_0226",
                "qst_0229",
                "qst_0295",
                "qst_0296",
                "qst_0302",
                "qst_0305",
                "qst_0306",
                "qst_0307",
                "qst_0309",
                "qst_0332",
                "qst_0334",
                "qst_0340",
                "qst_0343",
                "qst_0355",
                "qst_0360",
                "qst_0362",
                "qst_0365",
                "qst_0367",
                "qst_0377",
                "qst_0399",
                "qst_0410",
                "qst_0436",
                "qst_0438",
                "qst_0443",
                "qst_0451",
                "qst_0455",
                "qst_0463",
                "qst_0464",
                "qst_0465"
              ]
            },
            "ndcg_at_10": {
              "status": "measured",
              "value": 0.8264482916877589,
              "unit": "fraction",
              "n": 44,
              "attempted": 44,
              "expected": 44,
              "missing": 0,
              "ci95": [
                0.7378993463937408,
                0.9065722758796401
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0028",
                "qst_0043",
                "qst_0051",
                "qst_0053",
                "qst_0064",
                "qst_0088",
                "qst_0092",
                "qst_0100",
                "qst_0152",
                "qst_0153",
                "qst_0154",
                "qst_0172",
                "qst_0173",
                "qst_0176",
                "qst_0224",
                "qst_0226",
                "qst_0229",
                "qst_0295",
                "qst_0296",
                "qst_0302",
                "qst_0305",
                "qst_0306",
                "qst_0307",
                "qst_0309",
                "qst_0332",
                "qst_0334",
                "qst_0340",
                "qst_0343",
                "qst_0355",
                "qst_0360",
                "qst_0362",
                "qst_0365",
                "qst_0367",
                "qst_0377",
                "qst_0399",
                "qst_0410",
                "qst_0436",
                "qst_0438",
                "qst_0443",
                "qst_0451",
                "qst_0455",
                "qst_0463",
                "qst_0464",
                "qst_0465"
              ]
            },
            "recall_at_50": {
              "status": "measured",
              "value": 0.9588383838383838,
              "unit": "fraction",
              "n": 44,
              "attempted": 44,
              "expected": 44,
              "missing": 0,
              "ci95": [
                0.8977209595959595,
                1
              ],
              "method": "question-bootstrap-10000-seed1729",
              "sample_ids": [
                "qst_0028",
                "qst_0043",
                "qst_0051",
                "qst_0053",
                "qst_0064",
                "qst_0088",
                "qst_0092",
                "qst_0100",
                "qst_0152",
                "qst_0153",
                "qst_0154",
                "qst_0172",
                "qst_0173",
                "qst_0176",
                "qst_0224",
                "qst_0226",
                "qst_0229",
                "qst_0295",
                "qst_0296",
                "qst_0302",
                "qst_0305",
                "qst_0306",
                "qst_0307",
                "qst_0309",
                "qst_0332",
                "qst_0334",
                "qst_0340",
                "qst_0343",
                "qst_0355",
                "qst_0360",
                "qst_0362",
                "qst_0365",
                "qst_0367",
                "qst_0377",
                "qst_0399",
                "qst_0410",
                "qst_0436",
                "qst_0438",
                "qst_0443",
                "qst_0451",
                "qst_0455",
                "qst_0463",
                "qst_0464",
                "qst_0465"
              ]
            }
          },
          "membership": "Official question source_types; multi-source questions appear in each matching slice"
        }
      }
    }
  ],
  "details": [
    {
      "id": "basic",
      "label": "basic",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9772727272727273,
          "unit": "fraction",
          "n": 88,
          "attempted": 88,
          "expected": 88,
          "missing": 0,
          "ci95": [
            0.9431818181818182,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "completeness",
      "label": "completeness",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.7402777777777778,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.5433333333333333,
            0.9125
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "conflicting_info",
      "label": "conflicting info",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.75,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.5,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "constrained",
      "label": "constrained",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9666666666666667,
          "unit": "fraction",
          "n": 15,
          "attempted": 15,
          "expected": 15,
          "missing": 0,
          "ci95": [
            0.9,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "high_level",
      "label": "high level",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "info_not_found",
      "label": "info not found",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "intra_document_reasoning",
      "label": "intra document reasoning",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 20,
          "attempted": 20,
          "expected": 20,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "miscellaneous",
      "label": "miscellaneous",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "project_related",
      "label": "project related",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8173611111111111,
          "unit": "fraction",
          "n": 20,
          "attempted": 20,
          "expected": 20,
          "missing": 0,
          "ci95": [
            0.7229166666666667,
            0.9069444444444444
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "semantic",
      "label": "semantic",
      "kind": "category",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9838709677419355,
          "unit": "fraction",
          "n": 62,
          "attempted": 62,
          "expected": 62,
          "missing": 0,
          "ci95": [
            0.9516129032258065,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0002",
      "label": "What is the name of the new metric added so SRE can track when server-side streaming sessions get finalized due to hitting the time limit?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0002"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0004",
      "label": "In the meeting about onboarding a SaaS product to Google Cloud Marketplace, what did the GCP team recommend for handling delays where a new subscription entitlement is not immediately available during the customer onboarding flow?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0004"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0008",
      "label": "In the internal shiproom runner notes, what is the target time-to-recover threshold for a verified rollback in staging?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0008"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0009",
      "label": "In the EdgePath evaluation email thread, what alternative Year 1 pricing package did Redwood propose instead of matching the competitor's 50 percent first-year discount and migration credit?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0009"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0010",
      "label": "How does the new alerting approach group model-serving requests by the stages they pass through to reduce noisy alerts and speed up root cause analysis?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0010"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0011",
      "label": "How long is the validity period for the telemetry driven runbook author certification from the operator training bootcamp?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0011"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0012",
      "label": "What was identified as the root cause of the intermittent VPN connectivity blackholes and duplicate IP warnings for users on the corporate Wi-Fi in the SF2 building behind a specific edge switch?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0012"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0013",
      "label": "What is the recommended two-stage process for rotating the signing credentials used by GPU image build and promotion pipelines, including how to cut over safely and roll back if something goes wrong?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0013"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0015",
      "label": "What was the high-percentile latency concern reported after a smoke test of about 50 concurrent streaming chat sessions for Satellite Grove?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0015"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0016",
      "label": "What is the company policy for how long contractor access should last by default before it expires, according to the access and permissions playbook?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0016"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0022",
      "label": "What is the recommended liability cap language in the procurement legal triage SOP for vendor contracts (including the key exceptions and how the cap amount is determined)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0022"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0026",
      "label": "For a customer upgrading a private deployment in their own cloud environment, what restore time did support say is typical if the upgrade fails and a rollback is needed (assuming snapshots were taken)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0026"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0028",
      "label": "What steps and commands were recommended for creating and pushing a signed annotated Git tag and then making a draft GitHub release for the v1.14.0 release?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0028"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0030",
      "label": "In the Console feature that compares canary or A-B cohorts and links metric anomalies to request traces, what customer tiers are allowed into the beta via a feature flag gate?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0030"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0032",
      "label": "What performance goal was set for reducing GPU out of memory incidents in the stress tests for the windowed KV cache proposal?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0032"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0033",
      "label": "In the security review meeting about private cloud upgrades for a financial services customer, what was the stated minimum audit log retention period they wanted?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0033"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0035",
      "label": "For regulated customer private VPC deployments, what is the tenant-facing monthly availability target for the inference endpoint?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0035"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0040",
      "label": "What are the retention time windows for these lightweight kernel state snapshots (in-memory cache, indexed store, and any cold archive) used for post-incident diagnostics?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0040"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0041",
      "label": "In the security review call about buying an LLM inference service through AWS Marketplace, what was said about whether cross-region failover is enabled by default for the hosted endpoint?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0041"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0043",
      "label": "What caused the brief p99 latency jump on the hosted text generation endpoint in us-west-2, and what mitigation was applied to bring it back down?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0043"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0044",
      "label": "What support response time commitments did Acme approve for being shown in their AWS and Azure marketplace listing for incident priority levels?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0044"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0046",
      "label": "In the Medicore procurement email thread about data use and telemetry, what turnaround timeline did Redwood propose for delivering the operational runbook?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0046"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0048",
      "label": "What are the initial weighting percentages for the signals used in the Model Resilience Propensity Index, before the 4-week recalibration?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0048"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0049",
      "label": "In the email thread about BrightPath considering a marketplace pilot versus a direct contract for an AI inference vendor, what lead time was given for getting marketplace entitlement after onboarding?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0049"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0050",
      "label": "What is the proposed HTTP header scheme for attributing API traffic to quickstart templates, including how the header value is formatted and the opt-in header?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0050"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0051",
      "label": "In the cost-aware model routing proposal, what telemetry fields were suggested to emit so the team can analyze routing decisions and tradeoffs between spend and quality?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0051"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0052",
      "label": "What is the sampling policy for request routes of different telemetry priority levels in the new path-based observability change?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0052"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0053",
      "label": "What metric label allowlist and caps did the SRE team decide for v1 shadow traffic metrics to prevent high-cardinality issues (including limits for route and experiment labels)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0053"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0055",
      "label": "What audit log retention period did Redwood commit to for pilot customers during the 2027 API model migration, in response to the Acme Health security questions?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0055"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0059",
      "label": "In the incident where a CI runner service token was suspected of being used from two different datacenter IP ranges minutes apart, what time window did the team decide to audit for artifact uploads and repository pushes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0059"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0063",
      "label": "For the NorthPoint Signalworks dedicated pilot, what p95 latency target did they set for short chat interactions?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0063"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0064",
      "label": "What remediation timeline by severity did the security team propose for the upcoming external penetration test findings?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0064"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0065",
      "label": "During the November 2025 us-east incident where streaming chat responses were dropping, what was the internal start and mitigation time that support shared with the affected customer?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0065"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0066",
      "label": "What p95 latency goal did the founder set for the on-page product Q and A widget for the streetwear brand that expects big traffic spikes during limited edition drops?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0066"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0067",
      "label": "What did the internal usability test with six reviewers conclude about the table column compression feature, especially regarding its discoverability?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0067"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0069",
      "label": "In an ops day log about GPU rebalancing and brief west region network instability, what was the target slot capacity for the small hosted pool in the slot metrics dashboard?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0069"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0071",
      "label": "In the Private VPC console prototype for an interactive network path storyboard used to rehearse cluster upgrades, what is explicitly out of scope for the ticket?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0071"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0074",
      "label": "How does the new v1 chat compatibility layer handle invalid tool or function schemas, and what HTTP status does it return when validation fails?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0074"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0077",
      "label": "What is the three-step message sequence used for the new cross-language streaming startup handshake in the SDKs?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0077"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0079",
      "label": "In a morning account blitz note from May 2026, what was the proposed mitigation to address repeated inference timeouts for a healthcare customer with a mid-June renewal?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0079"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0082",
      "label": "During the year end handoff meeting, what was identified as the cause of the early December latency incident impacting some p95 requests?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0082"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0083",
      "label": "What was the root cause of the customer invoice overcharge where an organization-wide promotional credit was mistakenly redistributed as charges across multiple child accounts during a billing reconciliation run in a hybrid dedicated plus hosted setup?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0083"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0085",
      "label": "What are the three request quality and cost tiers proposed in the lightweight interaction tiering model for user facing features?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0085"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0086",
      "label": "Why would Apple Silicon MacBooks coming back from logic board repair get stuck during automated DEP or MDM setup at first boot, with the device checking in but the management profile and SSO certificate never installing?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0086"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0088",
      "label": "In the Slack chat about a desk plant rap contest, where was the remaining sticker stock said to be stored?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0088"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0089",
      "label": "What are the three runtime precision settings introduced for controlling fp32 vs lower precision execution, and what does each mode allow?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0089"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0092",
      "label": "In the engineering rollout chat about deprecating v2, what was the start time and region for the initial canary deployment and what traffic ramp was planned?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0092"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0096",
      "label": "When an employee is marked for immediate termination in Workday, why can a third party SaaS seat stay billable for around two weeks even after Okta deactivates the user?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0096"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0098",
      "label": "In the MedSecure logging and PII redaction POC email thread, what method did Redwood propose for pseudonymizing trace or request IDs so correlation works within a tenant but not across customers?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0098"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0100",
      "label": "What change was applied to reduce the latency spike caused by a KV cache and continuous batching regression on the us-west-2 inference production cluster?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0100"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0102",
      "label": "What are the default significance threshold and minimum sample size guidelines for deciding whether a model regression is actionable in the causal regression investigation handbook?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0102"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0103",
      "label": "How is the baseline token cost computed for the per-span cost delta metric in the new tracing observability change?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0103"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0104",
      "label": "What is the standard amount of time a new hire buddy is expected to spend per day during the first two weeks when a long-term contractor is converted to a full-time employee?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0104"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0105",
      "label": "What are the proposed internal time to mitigate targets by severity for incident response, regardless of customer deployment tier?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0105"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0106",
      "label": "What baseline metrics are listed for the Console activation health panel in the quarterly activation plan for customer-hosted Private VPC deployments?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0106"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0111",
      "label": "What are the acceptance thresholds for rerank performance in the mini-bundle unified inference API, including the required top-1 precision and the P95 latency limit?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0111"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0117",
      "label": "In the PRD about adding managed orchestration for multi-model inference pipelines, what are the rollout milestones and target quarters for alpha, beta, and general availability?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0117"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0122",
      "label": "Which GPU cluster or pool was flagged as having relatively low average usage but high churn, making its free capacity unreliable during a specific morning UTC window?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0122"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0124",
      "label": "What percent reduction in monthly model or token spend was the founder of a small developer tools micro-SaaS aiming for while keeping acceptable latency?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0124"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0125",
      "label": "What is the formula used by the new autoscaling approach that combines a high-percentile request rate with request variability to compute effective demand, including the default weights?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0125"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0127",
      "label": "What is the phased rollout plan for the change that smooths dedicated autoscaling signals and adds an advisory gateway-to-controller backpressure mechanism?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0127"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0129",
      "label": "What was the temporary workaround used to prevent the Windows 11 imaging task sequence from failing when a provisioning script breaks under certain non-default locale collation settings?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0129"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0130",
      "label": "What latency goals did the founder of Birch Branch Apps mention for short in-app chat responses during their early discovery call?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0130"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0132",
      "label": "In a case where server-sent event streaming responses sometimes lose the very last space or period only when traffic goes through an HTTP/2 edge proxy with stream preemption, what mitigation was proposed to stop the final bytes from being dropped?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0132"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0134",
      "label": "What was the suspected root cause behind the brief p99 latency spikes on the hosted chat inference route during small burst traffic?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0134"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0138",
      "label": "What was the CTOs non negotiable security requirement mentioned on the early 2026 demo call for the analytics SaaS customer moving from hosted API to a dedicated VPC deployment?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0138"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0141",
      "label": "What was the target goal for average router queue time for very short token sequences, compared to the baseline of 28 ms, in the weekly sprint anchor about reducing staging inference queuing latency?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0141"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0145",
      "label": "For the upcoming joint demo with a Japan-based logistics customer, what document did they prefer for trial procurement instead of a full purchase order?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0145"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0147",
      "label": "What is the planned cutoff date when the old model capability format may stop being accepted for new writes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0147"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0148",
      "label": "In the recent SDK update that added typed streaming errors and a reconnecting iterator API across Python, TypeScript, and Go, what package versions were released for each SDK?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0148"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0152",
      "label": "What caused the GPU training slowdowns in one us-east rack where same-rack pod latency jumped to tens of milliseconds and NCCL allreduce started timing out, and what network symptom on the top-of-rack switch pointed to it?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0152"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0153",
      "label": "In the engineering chat about a Go refactor PR, what did they change in the linter config to get CI passing when a gocritic rule complained about context parameter types in tests?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0153"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0154",
      "label": "What success targets were suggested for the four week micro credential pilot with SI partners (completion percentage and number of integrations started within about two months)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0154"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0155",
      "label": "What are the lifecycle states and key transitions for the edge adapter service that registers with a control plane and can fall back to hosted inference?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0155"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0157",
      "label": "What is the corrective journal number created to move a supplier credit amount from the DEV cost center to the OPS cost center after it was posted incorrectly during the March close?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0157"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0158",
      "label": "What are the three default preset options in the autoscaling policy builder for the dedicated capacity trial onboarding flow?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0158"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0161",
      "label": "In the OpenAI compatibility adapter update for Redwood deployments, what priority order does the normalizer use to resolve conflicting tool selection hints?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0161"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0162",
      "label": "What caused the brief period where a few tenants were not being throttled correctly after a shard reassignment happened at the same time as a runtime policy reload?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0162"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0163",
      "label": "What are the pass fail threshold percentages for the release guardrails for critical versus lower risk controls in the SSO and audit log compliance rollout plan?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0163"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0165",
      "label": "What was the agreed condition and timing for approving the additional senior infrastructure engineering hire under the hiring freeze plan?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0165"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0166",
      "label": "What procurement target date did ClearPath Health give for signing an agreement after the streaming model benchmark and security review?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0166"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0167",
      "label": "What configuration option was added to let repositories explicitly allow Docker-in-Docker during the staged rollout of the new GitHub composite action CI setup?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0167"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0168",
      "label": "What is the HTTP endpoint and method for the new service that returns recommended default profiles for batching, cache lifetimes, and quantization based on recent traffic telemetry?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0168"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0169",
      "label": "Who is responsible for signing off on latency regression in the weekly customer rollout tracker for Nimbus Health?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0169"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0172",
      "label": "In the eng-platform standup about staging 401 spikes on the generate endpoint, what header value caused the requests to miss both the primary and fallback route predicates and hit the fallthrough route?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0172"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0173",
      "label": "What maintenance window and expected downtime were planned for the rollout bringing up new dedicated GPU capacity nodes, including the canary and broader rollout dates?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0173"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0174",
      "label": "What are the required fields that each entry in the contract conformance results ledger must include?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0174"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0175",
      "label": "What keyboard-only interaction model is required for the compact dashboard widgets, including how focus enters the card and how users move and activate controls inside it?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0175"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0176",
      "label": "When does booking open for the new top end 80GB accelerator on dedicated clusters in the EU Central and India South locations?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0176"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0177",
      "label": "What were the final concession terms approved for the retail partner purchase through the major cloud providers marketplace, specifically the first year price reduction and the referral payout schedule after launch?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0177"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0180",
      "label": "In our rollout system, what is the name of the new mechanism that prevents a candidate release from getting full user traffic until a dry run with replayed requests and smoke checks has passed?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0180"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0181",
      "label": "During a large overnight upload run using short lived credentials refreshed by many transient workers, what client side scheduling change was recommended to stop periodic too many requests errors after a partial worker restart caused refresh bursts?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0181"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0182",
      "label": "During a failover between two active locations, what is the default time window where a short-lived resume credential is accepted on the destination before doing full authentication again?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0182"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0186",
      "label": "In the notes about keeping long, stop-and-go chat sessions cheap without replaying the whole history, what storage setup and time-to-live were proposed for keeping the compact per-session state for recent sessions versus longer retention?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0186"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0187",
      "label": "When a production model experiment shows a sustained increase in harmful output warnings for two evaluation windows, what automated actions should the platform take regarding traffic shifting and incident/oncall notifications?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0187"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0189",
      "label": "In our offline installer release process, how do we expose per-component supply chain provenance so an operator can see which build system and commit produced each piece and optionally enforce that missing or bad attestations fail the index step?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0189"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0192",
      "label": "In our GPU inference runtime, what change was introduced to cut the worst-case temporary device-memory spike when short and long requests are interleaved, especially for attention-heavy ops that used to grab per-op temp buffers?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0192"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0195",
      "label": "From the internal sandbox microbench in the EU West lab around May 2029, what was the estimated dollars per one million tokens for the biggest model on a single 40GB GPU when using the no-batching low-latency mode on the long output workload (about 300 tokens per response)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0195"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0198",
      "label": "In the EU West production setup where token streaming to browsers sometimes reconnects after a short handshake reset at the CDN layer, what change on the gateway side reduced cases where the last bit of generated text never arrives even though the backend says it finished?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0198"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0199",
      "label": "After a recent refactor to request grouping and the per request memoization layer, what is the specific risk if the cache key canonicalization now treats a trailing slash differently between callers, especially in a multi customer environment?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0199"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0200",
      "label": "For the hospital system that wants to run an interactive intake chatbot and auto-generate discharge writeups entirely inside its own locked-down data center with no patient data leaving the network, what end-to-end response time target did they set for producing about 200 tokens under peak load?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0200"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0209",
      "label": "For a small B2B SaaS embedding a live chat helper inside their product that needs partial replies immediately while it fetches recent user-event context, what first-response-time target and concurrent-load threshold were set as the pass criteria for their short trial?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0209"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0212",
      "label": "How should we set up low-noise oncall dashboards and tiered paging for a streaming text generation service so that responders can jump from a latency spike to a trace with hashed prompt references and cache hotness context and resolve issues within about 15 minutes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0212"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0213",
      "label": "For a content templating company planning tens of millions of very short generations during a Q3 ramp, what was the requested plan for a time limited evaluation that included a large free usage allowance?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0213"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0215",
      "label": "On managed iPhones on corporate Wi-Fi, why do long-lived server push text streams sometimes reconnect after the app is backgrounded and then end with a few seconds of silence and a cut-off final JSON fragment instead of a clean end marker?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0215"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0216",
      "label": "In our cross team outage process, what is the expected update rhythm for the shared incident channel from the first message through containment, including when the interval should be shortened?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0216"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0217",
      "label": "In a customer owned cloud network peered to our control plane, why would big model artifact downloads over long lived encrypted multiplexed streams start hanging after a transit vendor policy change, then immediately work again when the node packet size limit is lowered?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0217"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0218",
      "label": "In the private network hosted cluster console, what are the planned quick actions on the compact outage summary tile that on call engineers can use to stop a risky release or switch to a safer model option?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0218"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0221",
      "label": "When someone new starts at Redwood Inference, what is the target turnaround time for getting them fully set up with accounts and tools in North America versus Europe/Asia-Pacific?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0221"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0223",
      "label": "Why did our telemetry collectors in the dedicated production clusters stop being able to mint short lived credentials for a customer cloud identity when going through the AWS to GCP federation bridge, after an IAM policy tightening with conditional rules, causing missing metrics and traces for about a day in US East and EU West?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0223"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0224",
      "label": "In the public documentation for searching system activity records, what identifier should examples use instead of showing a persons email address to avoid exposing sensitive personal data?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0224"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0226",
      "label": "During the 10:00 UTC partial rollout that caused structured function-invocation outputs to sometimes drop or swap argument fields, what small hotfix was added in the batching/serving path to prevent argument-buffer interleaving?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0226"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0228",
      "label": "On Windows ARM tablets in our SF office, why does switching between the two WiFi bands make our always-on VPN still show connected but suddenly stop reaching private 10.x services until a reboot, with routes preferring the wrong network after the handoff?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0228"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0229",
      "label": "After a customer rotated credentials during a migration, their staging probes started alternating between auth failures complaining about a signature mismatch and throttling errors; what short lived workaround did support create so both legacy and new signing methods would be accepted during the rollout?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0229"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0230",
      "label": "For a chat style server sent events stream that must survive mobile disconnects without re running tool side effects, what mechanism was introduced to let a client reconnect from a safe checkpoint using a server issued marker plus an explicit receipt from the client, while keeping compatibility with existing OpenAI like streaming clients?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0230"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0231",
      "label": "For the healthcare client contract review about sending patient records through our connector into the vendor platform, what is the promised retention duration for ingestion related audit traces and the default way customers can obtain exports?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0231"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0235",
      "label": "In the email thread about an AI service pilot for a payments company, what were the four main categories of changes their procurement and counsel asked for in the first contract markup?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0235"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0238",
      "label": "How should I structure a small prompt experiment to reduce overconfident mistakes by mixing a couple of correct examples with one intentionally wrong example plus an immediate fix, while keeping the total examples very short and tracking things like made-up details and format compliance over a few hundred test prompts?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0238"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0239",
      "label": "For the fintech prospect doing merchant reconciliation and dispute handling, what are the specific compliance items their ops and security teams said must be in place before moving past the pilot phase?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0239"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0241",
      "label": "For a EU-hosted dedicated banking tenant facing a regulator check in late March 2026, how can we produce a tamper-evident 7-day audit activity export that includes per-event masking justification plus the GDPR lawful-ground and any consent time, with a signed checksum manifest and a trusted timestamp?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0241"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0245",
      "label": "Which fintech support-chat prospect dropped us after their lawyers rejected non-audited compliance proof and chose a rival that could deliver a private install with formal audit paperwork?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0245"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0247",
      "label": "In the partner sales planning call between a vendor and an ISV, what were the concrete due dates in late December for each side to send the pilot inputs and the vendor deliverables?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0247"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0250",
      "label": "In the dedicated GPU hosting plan with prepaid overage credits, what is the default time window after a demand spike during which any additional spikes are treated as part of the same billable overage event?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0250"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0252",
      "label": "In our incident response process, what is the rule for when a quick time-boxed after-action review is acceptable instead of writing the full formal analysis, based on impact duration and whether the issue hit one customer versus many?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0252"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0253",
      "label": "For the outsourced customer support company planning an isolated always-on deployment for AI agent guidance and end-of-chat ticket writeups, what peak simultaneous staff load and response-time target did they specify for real-time streaming suggestions?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0253"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0254",
      "label": "Need the account notes for a tiny two person dev shop building a browser based coding assistant that struggled with keeping API secrets safe in a serverless front end, is pushing hard for incremental responses to keep the UI snappy, and is asking about cheap usage plus some starter credit before converting from prototype to paid.",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0254"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0255",
      "label": "When is the 45 minute video call finally locked in for the vendor walkthrough that got moved because the client team was traveling to a Denver conference in early October 2026?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0255"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0256",
      "label": "During a gradual rollout of a new large model on a single production tenant in early March 2026, some sessions that were supposed to stay on a customer-specified version briefly got routed to a cheaper compressed build, causing worse answers without any server error spike. What was the immediate mitigation applied at the request routing layer to prevent this kind of misrouting when a version lock is provided?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0256"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0257",
      "label": "For the healthcare chat rollout that ramps from a few thousand to about fifteen thousand monthly active users by month three, what was the projected total monthly language model token consumption at steady state based on the stated session and prompt/response length assumptions?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0257"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0260",
      "label": "From a late July 2025 call with a vendor, what was the specific console tuning recommendation for the search route that included both a suggested batch size and a quantization setting?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0260"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0262",
      "label": "For the EU production rollout of a card-industry regulated bank, can you provide a tamper-evident proof package for a 3-day early-March 2026 activity export to their file-drop security tool that lets them verify no per-call audit records were lost or reordered (including a hardware-backed signed index and local validation steps)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0262"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0265",
      "label": "Why do some long range analytics charts in the US East production console briefly show zero or outdated values and the drill through to traces disappears, then fixes itself after about 10 to 30 minutes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0265"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0270",
      "label": "Which bank client got a temporary signoff to keep normal processing in Europe but let traffic switch to a US site only during disaster recovery drills or when the EU cluster is degraded, with payload logging turned off, and what were the start and end dates of that allowance?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0270"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0271",
      "label": "How should we handle a short term external engineering team working out of our London site so they can get into the lab and then send two sensitive prototype boards back to the US with proper custody tracking and customs clearance after the week ends?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0271"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0274",
      "label": "For the small legal document review firm that came in via a website contact form in late Feb 2026, what was their target p95 response time for an interactive single-user chat experience?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0274"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0275",
      "label": "Im looking for the doc about building a fast system to store and query model performance run histories and a simple UI to compare two runs plus an oncall workflow for investigating detected slowdowns, including how it groups repeated issues and how long detailed vs rolled up data is kept",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0275"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0277",
      "label": "For the mid sized software company building outbound sales chat reps that trigger actions in email, calendar, and a CRM, what latency targets did they set for live token streaming and for the full message to action acknowledgement flow during their planned short evaluation?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0277"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0280",
      "label": "In the rollout plan for the internal service that only hands out application secrets to verified runtime hosts using short lived machine health proofs and hardware backed key unwrapping, what is the default hot and cold retention period for the tamper resistant audit trail?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0280"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0281",
      "label": "In the GPU reservation proposal for a pharma company with nightly model scoring and some real time triage traffic, what was the estimated cache warmup cost expressed in token hours for the smallest configuration?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0281"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0283",
      "label": "In our long conversation serving experiments, what mitigation is suggested for the tendency to start making up references right after a retrieval step when older context has been aggressively shrunk to save memory?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0283"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0284",
      "label": "What phone number did the recruiter provide for the frontend engineering candidate so the interviewers could reach her quickly if she was running late?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0284"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0285",
      "label": "How does the proposed multi-region traffic controller proactively avoid tail-latency blowups by warming likely failover locations ahead of time and using a three-level degrade/shed mechanism before hard failures occur?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0285"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0286",
      "label": "For a small mobile game studio adding realtime global chat and match lobby messaging, what latency goal did they set for typical short messages so players dont notice added delay?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0286"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0289",
      "label": "What caused the March 2026 morning spike where a dedicated east coast deployment started cutting off live generated replies mid stream and briefly overcounted usage charges due to retry and double counting behavior?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0289"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0290",
      "label": "In the service that lets clients discover which language models are available, where is the advisory 0 to 100 score documented that helps automatic selection based on feature match between our capability names and an OpenAI style interface?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0290"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0291",
      "label": "For the shipping customer rollout checklist, what is the requirement for output stability across repeated cold and warm starts before approving deployment?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0291"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0293",
      "label": "How does the temporary cross team sandbox get automatically shut down, including the default lifetime and the advance warning sent to the owners before everything is archived?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0293"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0295",
      "label": "In the late March 2026 payments rollout, what was the prior deployment version the team reverted to after the early Europe canary started restart looping due to a shutdown hook issue?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0295"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0296",
      "label": "Who on the team committed to producing a short animated demo to embed in the upcoming renewal business review materials, and what was the demo supposed to showcase?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0296"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0299",
      "label": "For the short evaluation engagement with the retail client where we offered a temporary sandbox to run about a couple thousand acceptance examples and promised a signed deletion attestation afterward, what date did we say we needed to receive their materials in order to hit the staging deadline?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0299"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0301",
      "label": "During the 2025-02-10 capacity incident, what two thresholds should trigger noisy-tenant detection in the runbook update, and how long were SLA targets breached?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0301"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0302",
      "label": "For the 90-day AcmeSec eval, what do they need to send to get Splunk HEC audit log exports wired up, and which internal team members are responsible for wiring the export and sending the pricing quote?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0302"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0304",
      "label": "For the PayRight Redwood DPA negotiation, what incident notification timeframe did PayRight request at the start and what notification window did Redwood propose in the final tracked wording before sending the DocuSign link?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0304"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0305",
      "label": "When using the python async batching helper for embeddings then rerank, what singleton-wrapped rerank response shape can show up during auth refresh, and what redwood.utils normalizer did the sampleapp add to fix the output shape?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0305"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0306",
      "label": "During the us-west-2 incident with TLS handshake failures on edge.redwood.ai, what alert triggered the PagerDuty page and what two preventative follow-ups were planned before closing the thread?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0306"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0307",
      "label": "For the StrataCore joint benchmark on the S100 card, what model and metrics are they benchmarking, and what artifacts will be included in the post run internal results bundle for signoff?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0307"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0309",
      "label": "After the docs deploy failed due to mkdocs linkcheck reporting broken anchor fragments from an include refactor, what final deploy-bot status was reported once prod was updated?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0309"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0314",
      "label": "For the Jupiter Therapeutics rollout, if the target go-live is 2028-07-22 and the canary is planned as 7 days with a 10/90 traffic split, what date should the canary period start?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0314"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0315",
      "label": "What UI indicator did the team plan to show for tenants in escalation mode, and what default grace-window duration was chosen for the circuit-breaker?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0315"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0317",
      "label": "In the Adaptive Log Zipper rollout plan, what dependency blocks per-tenant admission signaling in the edge gateway and what fallback mode is planned until it is added?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0317"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0322",
      "label": "What config flag controls enabling the bandwidth-aware fanout planner for TP allreduce on heterogeneous link clusters, and what other new components are called out in the release notes as helping reduce comm and compute stalls?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0322"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0323",
      "label": "Which issue filed after two customer escalations describes transient admin privileges or permission denied after SSO due to SAML assertion ordering and session stitching, and what environment(s) was it reproduced on?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0323"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0324",
      "label": "For the Q3 KPI dashboard email thread, what was the quarterly uptime percentage and how far below the target was it, and which JIRA ticket was Jonas going to create to track the infra changes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0324"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0331",
      "label": "In the incident where Meridian Analytics saw pinned sessions revert to the fallback variant during a tiered canary in us-east, which model rollout was affected and what type of customer contract did they have?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0331"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0332",
      "label": "What are the acceptance criteria for the trace export UI changes that include the corr-id copy behavior, the consent modal requirements, and the keyboard help overlay, and what else did the PM ask to include before merging?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0332"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0333",
      "label": "In the runtime inference forensics walkthrough for Arcadia Retail, what is their p95 latency target for interactive queries, and by what day did they say they would upload the sample requests?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0333"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0334",
      "label": "Why were SSE streaming connections getting cleanly closed after about 60 seconds only when going through the edge proxy, and what heartbeat change was implemented to mitigate it?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0334"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0336",
      "label": "For the CartPilot Redwood trial POC, when is the kickoff sync scheduled and what peak request rate did they say they expect during bursts?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0336"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0339",
      "label": "For the Usage Insights reconciliation drift alert, what drift percentage and time window trigger warn vs page, and which severity level includes workspace_id in the alert labels?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0339"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0340",
      "label": "For the private customer collector debug bundle upload 403 escalation, what support ticket number is being updated and what request ID should be included if escalating to infra?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0340"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0341",
      "label": "For the Proxima Bank 429 spike after priority routing rollout, what caused the throttling and what temporary policy exception did we apply, and how do we verify it is not burning the enterprise route SLOs?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0341"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0342",
      "label": "Our Key Management Standard says we must not cache plaintext DEKs, but ADR-007 allows a short-lived in-memory DEK cache. What is the actual policy we should follow, and how should we describe DEK caching and KMS outage behavior in customer-facing rotation materials?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0342"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0343",
      "label": "What caused the savings $/day estimates in Optimization Suggestions beta to not match Arkadia Finance billing, and what UI or API changes should we make to prevent this confusion going forward?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0343"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0345",
      "label": "For the Trident Financial Private VPC deal, what is our approved stance on US-only data residency and 7-year audit log retention/export, and which evidence pack artifacts should we share (and how) to back it up?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0345"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0346",
      "label": "After the Northwind latency spike, what metric wording ended up approved for the public case study and what exactly can Sales share privately without triggering new legal approval?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0346"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0347",
      "label": "For an air-gapped on-prem restore into a brand new cluster, what do we need to install before running restore apply, and what will the installer check and report if those prerequisites are missing?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0347"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0349",
      "label": "What caused the EU-West activation funnel and onboarding email issues in late January, and which code/config changes fixed them?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0349"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0350",
      "label": "Why did Northpeak Search see their Dedicated Burst Usage billed higher than expected, and how do we prevent retry attempts from double-counting burst GPU-seconds now?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0350"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0354",
      "label": "How should we explain and remediate a case where a customers invoice token total is higher than their Usage API export because streaming retries were double-counted, and what approvals are required before issuing a credit or changing the ledger?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.375,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0354"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0355",
      "label": "What is the approved wording to use when a prospect demands EU-only data residency and no non-EU support access for the Hosted API, and who do we need to escalate to if they want that as a contractual commitment?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6666666666666666,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0355"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0356",
      "label": "For the Acme AI dedicated failover incident with intermittent 500s and fallback churn, what v1 incident type and primary owner should we set, and what time-to-mitigate and time-to-fix targets should CS use in customer updates?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.75,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0356"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0360",
      "label": "Customers behind strict corporate proxies are seeing 4xx failures when the quickstart templates add telemetry tags. What tagging approach and opt-out setting should we use, and how will that affect TTFS and activation KPI measurement?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0360"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0362",
      "label": "What is our v1 standard for streaming timeout enforcement and retry behavior, and do any of the Python, TypeScript, or Go SDKs still violate it based on the parity matrix and recent support tickets?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.4444444444444444,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0362"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0363",
      "label": "Northbridge Bank reported that request log TTL was not being enforced in us-east and that audit log exports were missing key admin event types; what caused these issues and what customer-facing compliance pack artifacts should we provide to show both retention enforcement and audit log evidence are now correct?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8333333333333334,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0363"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0365",
      "label": "Fast tier p99 is regressing in canary, what SLO targets and abort criteria apply and what rollback steps should we take? Which dashboards and reason codes should we use to confirm whether admission control or requeue behavior is causing it.",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6666666666666666,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0365"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0367",
      "label": "For an enterprise ticket like SUP-1842 where streaming disconnects happen in us-east, when should we escalate from a Support bridge to a formal incident and post a status page update, and what customer update cadence and credits wording should we use?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.7777777777777778,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0367"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0369",
      "label": "Why were some Hosted API overload errors missing Retry-After, and what changes and alerts ensure it stays consistent across gateway, routing, and streaming paths going forward?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0369"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0375",
      "label": "When a quantization profile is blocked due to a quality regression, how is that enforced differently from an incompatibility fallback, and what should oncall do to verify a customer route is back on baseline using logs, dashboards, and audit history?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0375"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0376",
      "label": "Northbridge Bank is seeing request logs older than their configured retention window in us-east after a maintenance window. What is the recommended remediation and what customer-safe evidence artifacts should we provide to prove the retention policy is configured correctly and enforcement has caught up?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0376"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0377",
      "label": "If the demo tenant starts returning 429s mid-demo and the Console dashboards are empty, what are the fastest recovery steps to keep the call moving and what fixes were put in place to prevent this from happening again?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8333333333333334,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0377"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0381",
      "label": "In the planned controlled failover game day on 2026-01-15 (Hosted API us-east → eu-west), what caused the routing automation to oscillate (flip/rollback), and what follow-up ticket and target ship date were created to fix the streaming disconnect/reconnect behavior clients saw during cross-region redirects?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0381"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0386",
      "label": "In the Feb 2026 SEV-2 incident where the Dedicated autoscaler in us-west stopped scaling above minimum because its metrics query returned no data, what was the underlying cause and what hotfix restored scale-out?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0386"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0389",
      "label": "After the Console v4.3.2 rollout (March 2026), what caused saved dashboard filter presets to revert to the default right after you edit and save a linked sub-report in the widget editor modal, and what hotfix fixed it?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0389"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0390",
      "label": "During a March 2026 prod us-east-1 dedicated incident where a sudden burst of high-concurrency streaming traffic caused intermittent 502/504s due to proxy ephemeral port pressure, what mitigations were applied (replica changes and any kernel/sysctl tuning) and-per the ingress proxy connection-entropy playbook-what RCA artifacts should be captured while the issue is ongoing?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0390"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0391",
      "label": "In the hosted API incident in late April 2026 where streaming requests started returning 502/504 and non-streaming requests also saw a sharp tail-latency increase, what was the underlying trigger (involving an edge keepalive/connection-TTL change) and what immediate mitigations were used to restore service?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0391"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0392",
      "label": "During the us-east canary that normalized/rotated the session cookie domain for cross-subdomain routing, long-poll streaming clients started losing their pinned model version and getting routed to a fallback model. What was the root cause in the edge/session-manager flow, and what immediate mitigations were applied (platform-side and customer-side)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0392"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0393",
      "label": "After the March 9, 2026 embedding-catalog promotion that made a 1024-d embedding model the default, a dedicated Hosted API tenant (us-west with eu-west failover) saw /v1/rerank start timing out with 512-vs-1024 vector length mismatches when using the alias open-embed-default. What was the confirmed root cause, and what short-term mitigation and permanent fix did we deploy?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0393"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0394",
      "label": "For Incident Bot v2, when responders use break-glass during a Sev0/Sev1 incident to view tenant-scoped config diffs or run privileged telemetry queries, which audit events must be emitted and how long are those audit events retained (SIEM vs immutable archive)? Also, what baseline vs break-glass permission scopes does the bot use for these reads?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0394"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0395",
      "label": "What was the root cause and planned server-side fix for the March 2026 incident where function-call webhook callbacks were retried after a timeout and the JSON payload stayed the same, but the HMAC signature header changed between the first delivery and the retry due to gzip/encoding differences?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0395"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0396",
      "label": "After rotating a Hosted API key in the Console, why might the Usage → Ad-hoc Query still show request IDs tied to the old key and 'View trace' links that 404, and what immediate mitigation did Support/SRE apply before a proper fix shipped?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0396"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0399",
      "label": "In the mid-March 2026 eu-west canary rollout where sticky fallback kept pinning requests back onto the failing canary instead of the stable quantized variant, what configuration mistake caused the loop and what mitigations were applied to stop it?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0399"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0402",
      "label": "In a Redwood Private (VPC/on-prem) upgrade using the private installer, the UI has been stuck on \"Running migrations\" for over an hour and the migration pod logs repeatedly say it's waiting on a Postgres advisory lock (the Postgres database is externally managed, not the in-cluster StatefulSet). What's the underlying cause, what Support-guided action resolved it, and what specific logs/artifacts should be collected before attempting a retry or rollback?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0402"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0404",
      "label": "We saw quality regressions after enabling an INT8/mixed-precision variant and suspect KV-cache rounding on long-context requests. What request header can we set as a quick mitigation to bypass the KV cache, and in the dynamic quant handoff policy for the interactive-chat route what minimum KV-cache hit-rate and max sequence length are required before routing to a quant variant?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0404"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0407",
      "label": "In March 2026, what caused duplicate tool/webhook executions during *streaming* chat retries only when the idempotency key contained non-ASCII characters (e.g., emoji/diacritics) and the request went over HTTP/2 through the edge proxy-and what were the recommended workaround and the server-side fix (including what normalization we applied)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0407"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0410",
      "label": "In the EU hosted API, streaming generate requests over HTTP/2 intermittently returned 502 only when the client included a ~12KB base64-encoded user-metadata header. What was the root cause in our proxy chain, and what server-side mitigation did we deploy to stop the 502s (include the header and cutoff/behavior)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0410"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0411",
      "label": "On Streamly AI's dedicated pool dp-132-usw, what % of interactive burst credits should be reserved exclusively for priority=high routes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0411"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0412",
      "label": "For extraction, what price-normalization correctness is reported at temp=0.0 with strict JSON + canonicalization, and what earlier % is it compared to?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0412"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0414",
      "label": "INC-9821: was the degraded GPU node an OOM or intermittent driver/kernel launch stalls?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0414"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0417",
      "label": "What does Redwood Private on-prem recommend for signing backup manifests?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0417"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0419",
      "label": "What is the API endpoint to start a Unified Capacity Transition migration?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0419"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0420",
      "label": "For EXP-002 (eu-west→us-east), what egress cost rate and measurement basis does the cost penalty catalog use?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0420"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0422",
      "label": "How do audit log exports access records older than the retention cutoff (super-admin override vs legal hold)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0422"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0423",
      "label": "For the office keycard audit, how many months of access logs should we export?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0423"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0426",
      "label": "What are the automated rollback trigger thresholds (safety sentinel, composite quality, business KPI) in the Optimize cohort-driven gates PRD?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0426"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0428",
      "label": "What monthly token volume discount breakpoints apply for Hosted pricing in the enterprise playbook?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0428"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0434",
      "label": "In Q4 2025, which customer industry segment was featured in the most published customer stories (case studies + one-page success stories)?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0434"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0435",
      "label": "How many weekly status updates for Project Orion marked the project as 'At Risk'?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0435"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0436",
      "label": "Across Redwood's Go, Python, and TypeScript SDKs, which SDK has the highest number of customer-reported auth-related bug reports, and what are the corresponding support ticket IDs?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0436"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0438",
      "label": "Which Redwood Inference engineering team owned the most production incidents involving GPU capacity or quota exhaustion in H1 2025?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.4444444444444444,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0438"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0439",
      "label": "How many internal incident writeups at Redwood Inference describe activating automatic fallback (to a different model or region) as part of mitigation?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.4,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0439"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0442",
      "label": "What is Redwood Inference's end-to-end process for making a production change (definition, required approvals, pre-deploy verification, deploy execution, customer/internal communications, and rollback/post-change requirements) for the Hosted API and Console?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.75,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0442"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0443",
      "label": "List every internal communication thread (email, Slack, and meeting notes) about the Redwood Private upgrade 'rollback loop' bug (including references to RRB-17 or 'stuck rollback').",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0443"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0445",
      "label": "\"What is the complete required content of a production rollout plan for introducing a new LLM model version on Redwood Inference (hosted, dedicated, and private), including required metrics/thresholds, approvals, rollback/fallback rules, and customer-facing communications?\"",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.875,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0445"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0447",
      "label": "What is Redwood Inference's complete end-to-end process for shipping an emergency serving-runtime hotfix to production (Hosted + Dedicated), including required approvals, checklists, rollout/rollback steps, and mandatory customer communications?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8333333333333334,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0447"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0449",
      "label": "In March 2026, which deployment offering-Hosted API, Dedicated, or Private-had the most customer support escalations where the primary symptom was request timeouts?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0449"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0451",
      "label": "In the team chat about posting jokes when an AI makes things up in production, what tagging requirement did someone propose for the memes?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0451"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0452",
      "label": "Who owns the messy go-to-market misc assets dump with placeholder marketing files and micro-copy ideas created during a last-minute scramble before a partner call?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0452"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0455",
      "label": "Are sneakers acceptable for the first few weeks of the company summer softball league, or do we need cleats right away?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0455"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0460",
      "label": "In a first round technical interview for a senior sales engineer candidate, what were the agreed next steps after the call?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0460"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0461",
      "label": "What is the proposed weekly cleanup schedule and routine for tidying the engineering kitchen fridge so old opened items get discarded?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0461"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0463",
      "label": "What time are people planning to start the office foosball tournament lunch bracket on Friday?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0463"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0464",
      "label": "For the casual soccer game being organized on Friday after work, what was the agreed maximum number of players?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0464"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0465",
      "label": "When are signups due for the informal company softball team being organized for summer?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0465"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0467",
      "label": "Is there an internal hackathon demo project that uses a small web UI and a fake backend to visualize model serving behavior like batching and cache hit rates, and can be run locally or deployed for a quick demo?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0467"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0470",
      "label": "In a backend inference engineer interview about multi-tenant GPU scheduling, what strategy did the candidate propose to reduce cold misses for growing attention cache in long chat histories?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0470"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0472",
      "label": "What is the company's thesis about where competitive advantage shifts as LLM ecosystems expand?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0476",
      "label": "What is the stated differentiation around reliability and graceful degradation in Redwood's platform?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0477",
      "label": "What are the four main revenue streams in Redwood Inference's business model?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0478",
      "label": "What add-on categories are mentioned for Redwood's commercial offering?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0479",
      "label": "What triggers Redwood's expansion from PLG-led adoption to sales-assisted enterprise for Dedicated and Private?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0484",
      "label": "What is the complete mapping of `billing_tag` → Finance GL code (and cost center owner) used for chargebacks in the Eval Budget Ledger, and where is that mapping source-of-truth maintained?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0486",
      "label": "How do we compute and invoice the **per-request CO2e** for Hosted API usage (including the region-specific carbon intensity factors and any cache-hit adjustments), and where is that CO2e exposed in the customer Usage API/export schema?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0487",
      "label": "When a streaming resume attempt fails with HTTP 409 (checkpoint mismatch), what is the full JSON schema of the server's error payload (including any `resume_details` fields), and how is the `kv_crc` value inside v2 resume tokens computed?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0488",
      "label": "When we use the GPU firmware operator to do a BIOS/firmware rollout, what BIOS settings are included in Redwood's \"safe mode boot\" profile for A100/H100 nodes (e.g., Secure Boot, Above 4G decoding, Resizable BAR), and what are the default values for each?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0492",
      "label": "For the \"microburst insurance\" negotiated SLA bundles, what's the default burst surcharge charged when burst capacity units are consumed, and which GL account should that surcharge be posted to?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0495",
      "label": "For the Remote-first Pilot, what are the exact preferred vendors and model SKUs for the standing desk converter loaners, and which of those vendors can ship to Canada with delivery in under 48 hours?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0496",
      "label": "Does Incident Bot v2 support sending its incident context cards (deploy/config diff/saturation + pinned escalation context) to Microsoft Teams via an incoming webhook/Adaptive Cards, and if so what's the required payload schema and idempotency/retry behavior for updates?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0497",
      "label": "When the Security Risk Committee reviews a high-risk control exception (e.g., risk rating above 7), what quorum is required for the vote to be valid, and how are tie votes resolved?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0499",
      "label": "For the `profiling.signals` Kafka topic used by trace-orch and fidelity-proxy, what are the current production settings (partition count and message retention time), and where can I find the approval/change record for the most recent update?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0500",
      "label": "When a high-risk production change requires Change Advisory Board (CAB) approval, what quorum is required and what's the tie-break process if the CAB can't reach consensus-and who has final decision authority?",
      "kind": "question",
      "metric": "recall_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "basic",
      "label": "basic",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9121412716733218,
          "unit": "fraction",
          "n": 88,
          "attempted": 88,
          "expected": 88,
          "missing": 0,
          "ci95": [
            0.8637114156682625,
            0.9543744906585032
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "completeness",
      "label": "completeness",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.7842508359485264,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.6183152623949391,
            0.9252788326799577
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "conflicting_info",
      "label": "conflicting info",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6943738064224131,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.4184575696779568,
            0.9518324734889125
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "constrained",
      "label": "constrained",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9742098128510306,
          "unit": "fraction",
          "n": 15,
          "attempted": 15,
          "expected": 15,
          "missing": 0,
          "ci95": [
            0.9226294385530917,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "high_level",
      "label": "high level",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "info_not_found",
      "label": "info not found",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "intra_document_reasoning",
      "label": "intra document reasoning",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9815464876785729,
          "unit": "fraction",
          "n": 20,
          "attempted": 20,
          "expected": 20,
          "missing": 0,
          "ci95": [
            0.9446394630357187,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "miscellaneous",
      "label": "miscellaneous",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9430676558073394,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.8292029674220179,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "project_related",
      "label": "project related",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.831581723706968,
          "unit": "fraction",
          "n": 20,
          "attempted": 20,
          "expected": 20,
          "missing": 0,
          "ci95": [
            0.7593178514275063,
            0.8979972119022821
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "semantic",
      "label": "semantic",
      "kind": "category",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.816207981779643,
          "unit": "fraction",
          "n": 62,
          "attempted": 62,
          "expected": 62,
          "missing": 0,
          "ci95": [
            0.7461047831658707,
            0.8826757688839865
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0002",
      "label": "What is the name of the new metric added so SRE can track when server-side streaming sessions get finalized due to hitting the time limit?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0002"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0004",
      "label": "In the meeting about onboarding a SaaS product to Google Cloud Marketplace, what did the GCP team recommend for handling delays where a new subscription entitlement is not immediately available during the customer onboarding flow?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0004"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0008",
      "label": "In the internal shiproom runner notes, what is the target time-to-recover threshold for a verified rollback in staging?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0008"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0009",
      "label": "In the EdgePath evaluation email thread, what alternative Year 1 pricing package did Redwood propose instead of matching the competitor's 50 percent first-year discount and migration credit?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0009"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0010",
      "label": "How does the new alerting approach group model-serving requests by the stages they pass through to reduce noisy alerts and speed up root cause analysis?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0010"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0011",
      "label": "How long is the validity period for the telemetry driven runbook author certification from the operator training bootcamp?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0011"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0012",
      "label": "What was identified as the root cause of the intermittent VPN connectivity blackholes and duplicate IP warnings for users on the corporate Wi-Fi in the SF2 building behind a specific edge switch?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0012"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0013",
      "label": "What is the recommended two-stage process for rotating the signing credentials used by GPU image build and promotion pipelines, including how to cut over safely and roll back if something goes wrong?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0013"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0015",
      "label": "What was the high-percentile latency concern reported after a smoke test of about 50 concurrent streaming chat sessions for Satellite Grove?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0015"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0016",
      "label": "What is the company policy for how long contractor access should last by default before it expires, according to the access and permissions playbook?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3333333333333333,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0016"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0022",
      "label": "What is the recommended liability cap language in the procurement legal triage SOP for vendor contracts (including the key exceptions and how the cap amount is determined)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3562071871080222,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0022"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0026",
      "label": "For a customer upgrading a private deployment in their own cloud environment, what restore time did support say is typical if the upgrade fails and a rollback is needed (assuming snapshots were taken)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0026"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0028",
      "label": "What steps and commands were recommended for creating and pushing a signed annotated Git tag and then making a draft GitHub release for the v1.14.0 release?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0028"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0030",
      "label": "In the Console feature that compares canary or A-B cohorts and links metric anomalies to request traces, what customer tiers are allowed into the beta via a feature flag gate?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0030"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0032",
      "label": "What performance goal was set for reducing GPU out of memory incidents in the stress tests for the windowed KV cache proposal?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0032"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0033",
      "label": "In the security review meeting about private cloud upgrades for a financial services customer, what was the stated minimum audit log retention period they wanted?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0033"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0035",
      "label": "For regulated customer private VPC deployments, what is the tenant-facing monthly availability target for the inference endpoint?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0035"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0040",
      "label": "What are the retention time windows for these lightweight kernel state snapshots (in-memory cache, indexed store, and any cold archive) used for post-incident diagnostics?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0040"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0041",
      "label": "In the security review call about buying an LLM inference service through AWS Marketplace, what was said about whether cross-region failover is enabled by default for the hosted endpoint?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0041"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0043",
      "label": "What caused the brief p99 latency jump on the hosted text generation endpoint in us-west-2, and what mitigation was applied to bring it back down?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0043"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0044",
      "label": "What support response time commitments did Acme approve for being shown in their AWS and Azure marketplace listing for incident priority levels?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0044"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0046",
      "label": "In the Medicore procurement email thread about data use and telemetry, what turnaround timeline did Redwood propose for delivering the operational runbook?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0046"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0048",
      "label": "What are the initial weighting percentages for the signals used in the Model Resilience Propensity Index, before the 4-week recalibration?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0048"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0049",
      "label": "In the email thread about BrightPath considering a marketplace pilot versus a direct contract for an AI inference vendor, what lead time was given for getting marketplace entitlement after onboarding?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0049"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0050",
      "label": "What is the proposed HTTP header scheme for attributing API traffic to quickstart templates, including how the header value is formatted and the opt-in header?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.43067655807339306,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0050"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0051",
      "label": "In the cost-aware model routing proposal, what telemetry fields were suggested to emit so the team can analyze routing decisions and tradeoffs between spend and quality?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0051"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0052",
      "label": "What is the sampling policy for request routes of different telemetry priority levels in the new path-based observability change?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0052"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0053",
      "label": "What metric label allowlist and caps did the SRE team decide for v1 shadow traffic metrics to prevent high-cardinality issues (including limits for route and experiment labels)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0053"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0055",
      "label": "What audit log retention period did Redwood commit to for pilot customers during the 2027 API model migration, in response to the Acme Health security questions?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0055"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0059",
      "label": "In the incident where a CI runner service token was suspected of being used from two different datacenter IP ranges minutes apart, what time window did the team decide to audit for artifact uploads and repository pushes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0059"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0063",
      "label": "For the NorthPoint Signalworks dedicated pilot, what p95 latency target did they set for short chat interactions?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0063"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0064",
      "label": "What remediation timeline by severity did the security team propose for the upcoming external penetration test findings?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0064"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0065",
      "label": "During the November 2025 us-east incident where streaming chat responses were dropping, what was the internal start and mitigation time that support shared with the affected customer?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0065"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0066",
      "label": "What p95 latency goal did the founder set for the on-page product Q and A widget for the streetwear brand that expects big traffic spikes during limited edition drops?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0066"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0067",
      "label": "What did the internal usability test with six reviewers conclude about the table column compression feature, especially regarding its discoverability?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0067"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0069",
      "label": "In an ops day log about GPU rebalancing and brief west region network instability, what was the target slot capacity for the small hosted pool in the slot metrics dashboard?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0069"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0071",
      "label": "In the Private VPC console prototype for an interactive network path storyboard used to rehearse cluster upgrades, what is explicitly out of scope for the ticket?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0071"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0074",
      "label": "How does the new v1 chat compatibility layer handle invalid tool or function schemas, and what HTTP status does it return when validation fails?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.43067655807339306,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0074"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0077",
      "label": "What is the three-step message sequence used for the new cross-language streaming startup handshake in the SDKs?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0077"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0079",
      "label": "In a morning account blitz note from May 2026, what was the proposed mitigation to address repeated inference timeouts for a healthcare customer with a mid-June renewal?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0079"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0082",
      "label": "During the year end handoff meeting, what was identified as the cause of the early December latency incident impacting some p95 requests?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0082"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0083",
      "label": "What was the root cause of the customer invoice overcharge where an organization-wide promotional credit was mistakenly redistributed as charges across multiple child accounts during a billing reconciliation run in a hybrid dedicated plus hosted setup?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0083"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0085",
      "label": "What are the three request quality and cost tiers proposed in the lightweight interaction tiering model for user facing features?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0085"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0086",
      "label": "Why would Apple Silicon MacBooks coming back from logic board repair get stuck during automated DEP or MDM setup at first boot, with the device checking in but the management profile and SSO certificate never installing?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0086"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0088",
      "label": "In the Slack chat about a desk plant rap contest, where was the remaining sticker stock said to be stored?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0088"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0089",
      "label": "What are the three runtime precision settings introduced for controlling fp32 vs lower precision execution, and what does each mode allow?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0089"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0092",
      "label": "In the engineering rollout chat about deprecating v2, what was the start time and region for the initial canary deployment and what traffic ramp was planned?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0092"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0096",
      "label": "When an employee is marked for immediate termination in Workday, why can a third party SaaS seat stay billable for around two weeks even after Okta deactivates the user?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0096"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0098",
      "label": "In the MedSecure logging and PII redaction POC email thread, what method did Redwood propose for pseudonymizing trace or request IDs so correlation works within a tenant but not across customers?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0098"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0100",
      "label": "What change was applied to reduce the latency spike caused by a KV cache and continuous batching regression on the us-west-2 inference production cluster?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0100"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0102",
      "label": "What are the default significance threshold and minimum sample size guidelines for deciding whether a model regression is actionable in the causal regression investigation handbook?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0102"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0103",
      "label": "How is the baseline token cost computed for the per-span cost delta metric in the new tracing observability change?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0103"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0104",
      "label": "What is the standard amount of time a new hire buddy is expected to spend per day during the first two weeks when a long-term contractor is converted to a full-time employee?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0104"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0105",
      "label": "What are the proposed internal time to mitigate targets by severity for incident response, regardless of customer deployment tier?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0105"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0106",
      "label": "What baseline metrics are listed for the Console activation health panel in the quarterly activation plan for customer-hosted Private VPC deployments?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0106"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0111",
      "label": "What are the acceptance thresholds for rerank performance in the mini-bundle unified inference API, including the required top-1 precision and the P95 latency limit?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0111"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0117",
      "label": "In the PRD about adding managed orchestration for multi-model inference pipelines, what are the rollout milestones and target quarters for alpha, beta, and general availability?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0117"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0122",
      "label": "Which GPU cluster or pool was flagged as having relatively low average usage but high churn, making its free capacity unreliable during a specific morning UTC window?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0122"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0124",
      "label": "What percent reduction in monthly model or token spend was the founder of a small developer tools micro-SaaS aiming for while keeping acceptable latency?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0124"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0125",
      "label": "What is the formula used by the new autoscaling approach that combines a high-percentile request rate with request variability to compute effective demand, including the default weights?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0125"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0127",
      "label": "What is the phased rollout plan for the change that smooths dedicated autoscaling signals and adds an advisory gateway-to-controller backpressure mechanism?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0127"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0129",
      "label": "What was the temporary workaround used to prevent the Windows 11 imaging task sequence from failing when a provisioning script breaks under certain non-default locale collation settings?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0129"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0130",
      "label": "What latency goals did the founder of Birch Branch Apps mention for short in-app chat responses during their early discovery call?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0130"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0132",
      "label": "In a case where server-sent event streaming responses sometimes lose the very last space or period only when traffic goes through an HTTP/2 edge proxy with stream preemption, what mitigation was proposed to stop the final bytes from being dropped?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0132"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0134",
      "label": "What was the suspected root cause behind the brief p99 latency spikes on the hosted chat inference route during small burst traffic?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0134"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0138",
      "label": "What was the CTOs non negotiable security requirement mentioned on the early 2026 demo call for the analytics SaaS customer moving from hosted API to a dedicated VPC deployment?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0138"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0141",
      "label": "What was the target goal for average router queue time for very short token sequences, compared to the baseline of 28 ms, in the weekly sprint anchor about reducing staging inference queuing latency?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0141"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0145",
      "label": "For the upcoming joint demo with a Japan-based logistics customer, what document did they prefer for trial procurement instead of a full purchase order?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0145"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0147",
      "label": "What is the planned cutoff date when the old model capability format may stop being accepted for new writes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0147"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0148",
      "label": "In the recent SDK update that added typed streaming errors and a reconnecting iterator API across Python, TypeScript, and Go, what package versions were released for each SDK?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0148"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0152",
      "label": "What caused the GPU training slowdowns in one us-east rack where same-rack pod latency jumped to tens of milliseconds and NCCL allreduce started timing out, and what network symptom on the top-of-rack switch pointed to it?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0152"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0153",
      "label": "In the engineering chat about a Go refactor PR, what did they change in the linter config to get CI passing when a gocritic rule complained about context parameter types in tests?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0153"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0154",
      "label": "What success targets were suggested for the four week micro credential pilot with SI partners (completion percentage and number of integrations started within about two months)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0154"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0155",
      "label": "What are the lifecycle states and key transitions for the edge adapter service that registers with a control plane and can fall back to hosted inference?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0155"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0157",
      "label": "What is the corrective journal number created to move a supplier credit amount from the DEV cost center to the OPS cost center after it was posted incorrectly during the March close?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0157"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0158",
      "label": "What are the three default preset options in the autoscaling policy builder for the dedicated capacity trial onboarding flow?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0158"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0161",
      "label": "In the OpenAI compatibility adapter update for Redwood deployments, what priority order does the normalizer use to resolve conflicting tool selection hints?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0161"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0162",
      "label": "What caused the brief period where a few tenants were not being throttled correctly after a shard reassignment happened at the same time as a runtime policy reload?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0162"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0163",
      "label": "What are the pass fail threshold percentages for the release guardrails for critical versus lower risk controls in the SSO and audit log compliance rollout plan?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0163"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0165",
      "label": "What was the agreed condition and timing for approving the additional senior infrastructure engineering hire under the hiring freeze plan?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0165"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0166",
      "label": "What procurement target date did ClearPath Health give for signing an agreement after the streaming model benchmark and security review?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0166"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0167",
      "label": "What configuration option was added to let repositories explicitly allow Docker-in-Docker during the staged rollout of the new GitHub composite action CI setup?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0167"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0168",
      "label": "What is the HTTP endpoint and method for the new service that returns recommended default profiles for batching, cache lifetimes, and quantization based on recent traffic telemetry?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0168"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0169",
      "label": "Who is responsible for signing off on latency regression in the weekly customer rollout tracker for Nimbus Health?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0169"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0172",
      "label": "In the eng-platform standup about staging 401 spikes on the generate endpoint, what header value caused the requests to miss both the primary and fallback route predicates and hit the fallthrough route?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0172"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0173",
      "label": "What maintenance window and expected downtime were planned for the rollout bringing up new dedicated GPU capacity nodes, including the canary and broader rollout dates?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3010299956639812,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0173"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0174",
      "label": "What are the required fields that each entry in the contract conformance results ledger must include?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0174"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0175",
      "label": "What keyboard-only interaction model is required for the compact dashboard widgets, including how focus enters the card and how users move and activate controls inside it?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0175"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0176",
      "label": "When does booking open for the new top end 80GB accelerator on dedicated clusters in the EU Central and India South locations?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0176"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0177",
      "label": "What were the final concession terms approved for the retail partner purchase through the major cloud providers marketplace, specifically the first year price reduction and the referral payout schedule after launch?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0177"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0180",
      "label": "In our rollout system, what is the name of the new mechanism that prevents a candidate release from getting full user traffic until a dry run with replayed requests and smoke checks has passed?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0180"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0181",
      "label": "During a large overnight upload run using short lived credentials refreshed by many transient workers, what client side scheduling change was recommended to stop periodic too many requests errors after a partial worker restart caused refresh bursts?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0181"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0182",
      "label": "During a failover between two active locations, what is the default time window where a short-lived resume credential is accepted on the destination before doing full authentication again?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0182"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0186",
      "label": "In the notes about keeping long, stop-and-go chat sessions cheap without replaying the whole history, what storage setup and time-to-live were proposed for keeping the compact per-session state for recent sessions versus longer retention?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3562071871080222,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0186"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0187",
      "label": "When a production model experiment shows a sustained increase in harmful output warnings for two evaluation windows, what automated actions should the platform take regarding traffic shifting and incident/oncall notifications?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.31546487678572877,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0187"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0189",
      "label": "In our offline installer release process, how do we expose per-component supply chain provenance so an operator can see which build system and commit produced each piece and optionally enforce that missing or bad attestations fail the index step?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0189"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0192",
      "label": "In our GPU inference runtime, what change was introduced to cut the worst-case temporary device-memory spike when short and long requests are interleaved, especially for attention-heavy ops that used to grab per-op temp buffers?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0192"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0195",
      "label": "From the internal sandbox microbench in the EU West lab around May 2029, what was the estimated dollars per one million tokens for the biggest model on a single 40GB GPU when using the no-batching low-latency mode on the long output workload (about 300 tokens per response)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0195"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0198",
      "label": "In the EU West production setup where token streaming to browsers sometimes reconnects after a short handshake reset at the CDN layer, what change on the gateway side reduced cases where the last bit of generated text never arrives even though the backend says it finished?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0198"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0199",
      "label": "After a recent refactor to request grouping and the per request memoization layer, what is the specific risk if the cache key canonicalization now treats a trailing slash differently between callers, especially in a multi customer environment?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0199"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0200",
      "label": "For the hospital system that wants to run an interactive intake chatbot and auto-generate discharge writeups entirely inside its own locked-down data center with no patient data leaving the network, what end-to-end response time target did they set for producing about 200 tokens under peak load?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0200"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0209",
      "label": "For a small B2B SaaS embedding a live chat helper inside their product that needs partial replies immediately while it fetches recent user-event context, what first-response-time target and concurrent-load threshold were set as the pass criteria for their short trial?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.43067655807339306,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0209"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0212",
      "label": "How should we set up low-noise oncall dashboards and tiered paging for a streaming text generation service so that responders can jump from a latency spike to a trace with hashed prompt references and cache hotness context and resolve issues within about 15 minutes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.43067655807339306,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0212"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0213",
      "label": "For a content templating company planning tens of millions of very short generations during a Q3 ramp, what was the requested plan for a time limited evaluation that included a large free usage allowance?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0213"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0215",
      "label": "On managed iPhones on corporate Wi-Fi, why do long-lived server push text streams sometimes reconnect after the app is backgrounded and then end with a few seconds of silence and a cut-off final JSON fragment instead of a clean end marker?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0215"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0216",
      "label": "In our cross team outage process, what is the expected update rhythm for the shared incident channel from the first message through containment, including when the interval should be shortened?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0216"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0217",
      "label": "In a customer owned cloud network peered to our control plane, why would big model artifact downloads over long lived encrypted multiplexed streams start hanging after a transit vendor policy change, then immediately work again when the node packet size limit is lowered?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0217"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0218",
      "label": "In the private network hosted cluster console, what are the planned quick actions on the compact outage summary tile that on call engineers can use to stop a risky release or switch to a safer model option?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0218"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0221",
      "label": "When someone new starts at Redwood Inference, what is the target turnaround time for getting them fully set up with accounts and tools in North America versus Europe/Asia-Pacific?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0221"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0223",
      "label": "Why did our telemetry collectors in the dedicated production clusters stop being able to mint short lived credentials for a customer cloud identity when going through the AWS to GCP federation bridge, after an IAM policy tightening with conditional rules, causing missing metrics and traces for about a day in US East and EU West?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0223"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0224",
      "label": "In the public documentation for searching system activity records, what identifier should examples use instead of showing a persons email address to avoid exposing sensitive personal data?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0224"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0226",
      "label": "During the 10:00 UTC partial rollout that caused structured function-invocation outputs to sometimes drop or swap argument fields, what small hotfix was added in the batching/serving path to prevent argument-buffer interleaving?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0226"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0228",
      "label": "On Windows ARM tablets in our SF office, why does switching between the two WiFi bands make our always-on VPN still show connected but suddenly stop reaching private 10.x services until a reboot, with routes preferring the wrong network after the handoff?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0228"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0229",
      "label": "After a customer rotated credentials during a migration, their staging probes started alternating between auth failures complaining about a signature mismatch and throttling errors; what short lived workaround did support create so both legacy and new signing methods would be accepted during the rollout?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0229"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0230",
      "label": "For a chat style server sent events stream that must survive mobile disconnects without re running tool side effects, what mechanism was introduced to let a client reconnect from a safe checkpoint using a server issued marker plus an explicit receipt from the client, while keeping compatibility with existing OpenAI like streaming clients?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0230"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0231",
      "label": "For the healthcare client contract review about sending patient records through our connector into the vendor platform, what is the promised retention duration for ingestion related audit traces and the default way customers can obtain exports?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.2890648263178879,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0231"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0235",
      "label": "In the email thread about an AI service pilot for a payments company, what were the four main categories of changes their procurement and counsel asked for in the first contract markup?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0235"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0238",
      "label": "How should I structure a small prompt experiment to reduce overconfident mistakes by mixing a couple of correct examples with one intentionally wrong example plus an immediate fix, while keeping the total examples very short and tracking things like made-up details and format compliance over a few hundred test prompts?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0238"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0239",
      "label": "For the fintech prospect doing merchant reconciliation and dispute handling, what are the specific compliance items their ops and security teams said must be in place before moving past the pilot phase?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0239"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0241",
      "label": "For a EU-hosted dedicated banking tenant facing a regulator check in late March 2026, how can we produce a tamper-evident 7-day audit activity export that includes per-event masking justification plus the GDPR lawful-ground and any consent time, with a signed checksum manifest and a trusted timestamp?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0241"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0245",
      "label": "Which fintech support-chat prospect dropped us after their lawyers rejected non-audited compliance proof and chose a rival that could deliver a private install with formal audit paperwork?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0245"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0247",
      "label": "In the partner sales planning call between a vendor and an ISV, what were the concrete due dates in late December for each side to send the pilot inputs and the vendor deliverables?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3010299956639812,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0247"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0250",
      "label": "In the dedicated GPU hosting plan with prepaid overage credits, what is the default time window after a demand spike during which any additional spikes are treated as part of the same billable overage event?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0250"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0252",
      "label": "In our incident response process, what is the rule for when a quick time-boxed after-action review is acceptable instead of writing the full formal analysis, based on impact duration and whether the issue hit one customer versus many?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0252"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0253",
      "label": "For the outsourced customer support company planning an isolated always-on deployment for AI agent guidance and end-of-chat ticket writeups, what peak simultaneous staff load and response-time target did they specify for real-time streaming suggestions?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0253"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0254",
      "label": "Need the account notes for a tiny two person dev shop building a browser based coding assistant that struggled with keeping API secrets safe in a serverless front end, is pushing hard for incremental responses to keep the UI snappy, and is asking about cheap usage plus some starter credit before converting from prototype to paid.",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0254"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0255",
      "label": "When is the 45 minute video call finally locked in for the vendor walkthrough that got moved because the client team was traveling to a Denver conference in early October 2026?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3562071871080222,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0255"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0256",
      "label": "During a gradual rollout of a new large model on a single production tenant in early March 2026, some sessions that were supposed to stay on a customer-specified version briefly got routed to a cheaper compressed build, causing worse answers without any server error spike. What was the immediate mitigation applied at the request routing layer to prevent this kind of misrouting when a version lock is provided?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0256"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0257",
      "label": "For the healthcare chat rollout that ramps from a few thousand to about fifteen thousand monthly active users by month three, what was the projected total monthly language model token consumption at steady state based on the stated session and prompt/response length assumptions?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0257"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0260",
      "label": "From a late July 2025 call with a vendor, what was the specific console tuning recommendation for the search route that included both a suggested batch size and a quantization setting?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0260"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0262",
      "label": "For the EU production rollout of a card-industry regulated bank, can you provide a tamper-evident proof package for a 3-day early-March 2026 activity export to their file-drop security tool that lets them verify no per-call audit records were lost or reordered (including a hardware-backed signed index and local validation steps)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0262"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0265",
      "label": "Why do some long range analytics charts in the US East production console briefly show zero or outdated values and the drill through to traces disappears, then fixes itself after about 10 to 30 minutes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0265"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0270",
      "label": "Which bank client got a temporary signoff to keep normal processing in Europe but let traffic switch to a US site only during disaster recovery drills or when the EU cluster is degraded, with payload logging turned off, and what were the start and end dates of that allowance?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0270"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0271",
      "label": "How should we handle a short term external engineering team working out of our London site so they can get into the lab and then send two sensitive prototype boards back to the US with proper custody tracking and customs clearance after the week ends?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0271"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0274",
      "label": "For the small legal document review firm that came in via a website contact form in late Feb 2026, what was their target p95 response time for an interactive single-user chat experience?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0274"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0275",
      "label": "Im looking for the doc about building a fast system to store and query model performance run histories and a simple UI to compare two runs plus an oncall workflow for investigating detected slowdowns, including how it groups repeated issues and how long detailed vs rolled up data is kept",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0275"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0277",
      "label": "For the mid sized software company building outbound sales chat reps that trigger actions in email, calendar, and a CRM, what latency targets did they set for live token streaming and for the full message to action acknowledgement flow during their planned short evaluation?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0277"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0280",
      "label": "In the rollout plan for the internal service that only hands out application secrets to verified runtime hosts using short lived machine health proofs and hardware backed key unwrapping, what is the default hot and cold retention period for the tamper resistant audit trail?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0280"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0281",
      "label": "In the GPU reservation proposal for a pharma company with nightly model scoring and some real time triage traffic, what was the estimated cache warmup cost expressed in token hours for the smallest configuration?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0281"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0283",
      "label": "In our long conversation serving experiments, what mitigation is suggested for the tendency to start making up references right after a retrieval step when older context has been aggressively shrunk to save memory?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0283"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0284",
      "label": "What phone number did the recruiter provide for the frontend engineering candidate so the interviewers could reach her quickly if she was running late?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0284"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0285",
      "label": "How does the proposed multi-region traffic controller proactively avoid tail-latency blowups by warming likely failover locations ahead of time and using a three-level degrade/shed mechanism before hard failures occur?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0285"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0286",
      "label": "For a small mobile game studio adding realtime global chat and match lobby messaging, what latency goal did they set for typical short messages so players dont notice added delay?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0286"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0289",
      "label": "What caused the March 2026 morning spike where a dedicated east coast deployment started cutting off live generated replies mid stream and briefly overcounted usage charges due to retry and double counting behavior?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0289"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0290",
      "label": "In the service that lets clients discover which language models are available, where is the advisory 0 to 100 score documented that helps automatic selection based on feature match between our capability names and an OpenAI style interface?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0290"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0291",
      "label": "For the shipping customer rollout checklist, what is the requirement for output stability across repeated cold and warm starts before approving deployment?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.2890648263178879,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0291"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0293",
      "label": "How does the temporary cross team sandbox get automatically shut down, including the default lifetime and the advance warning sent to the owners before everything is archived?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0293"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0295",
      "label": "In the late March 2026 payments rollout, what was the prior deployment version the team reverted to after the early Europe canary started restart looping due to a shutdown hook issue?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0295"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0296",
      "label": "Who on the team committed to producing a short animated demo to embed in the upcoming renewal business review materials, and what was the demo supposed to showcase?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.2890648263178879,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0296"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0299",
      "label": "For the short evaluation engagement with the retail client where we offered a temporary sandbox to run about a couple thousand acceptance examples and promised a signed deletion attestation afterward, what date did we say we needed to receive their materials in order to hit the staging deadline?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0299"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0301",
      "label": "During the 2025-02-10 capacity incident, what two thresholds should trigger noisy-tenant detection in the runbook update, and how long were SLA targets breached?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0301"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0302",
      "label": "For the 90-day AcmeSec eval, what do they need to send to get Splunk HEC audit log exports wired up, and which internal team members are responsible for wiring the export and sending the pricing quote?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0302"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0304",
      "label": "For the PayRight Redwood DPA negotiation, what incident notification timeframe did PayRight request at the start and what notification window did Redwood propose in the final tracked wording before sending the DocuSign link?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0304"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0305",
      "label": "When using the python async batching helper for embeddings then rerank, what singleton-wrapped rerank response shape can show up during auth refresh, and what redwood.utils normalizer did the sampleapp add to fix the output shape?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0305"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0306",
      "label": "During the us-west-2 incident with TLS handshake failures on edge.redwood.ai, what alert triggered the PagerDuty page and what two preventative follow-ups were planned before closing the thread?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0306"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0307",
      "label": "For the StrataCore joint benchmark on the S100 card, what model and metrics are they benchmarking, and what artifacts will be included in the post run internal results bundle for signoff?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0307"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0309",
      "label": "After the docs deploy failed due to mkdocs linkcheck reporting broken anchor fragments from an include refactor, what final deploy-bot status was reported once prod was updated?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0309"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0314",
      "label": "For the Jupiter Therapeutics rollout, if the target go-live is 2028-07-22 and the canary is planned as 7 days with a 10/90 traffic split, what date should the canary period start?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0314"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0315",
      "label": "What UI indicator did the team plan to show for tenants in escalation mode, and what default grace-window duration was chosen for the circuit-breaker?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0315"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0317",
      "label": "In the Adaptive Log Zipper rollout plan, what dependency blocks per-tenant admission signaling in the edge gateway and what fallback mode is planned until it is added?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0317"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0322",
      "label": "What config flag controls enabling the bandwidth-aware fanout planner for TP allreduce on heterogeneous link clusters, and what other new components are called out in the release notes as helping reduce comm and compute stalls?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0322"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0323",
      "label": "Which issue filed after two customer escalations describes transient admin privileges or permission denied after SSO due to SAML assertion ordering and session stitching, and what environment(s) was it reproduced on?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0323"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0324",
      "label": "For the Q3 KPI dashboard email thread, what was the quarterly uptime percentage and how far below the target was it, and which JIRA ticket was Jonas going to create to track the infra changes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0324"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0331",
      "label": "In the incident where Meridian Analytics saw pinned sessions revert to the fallback variant during a tiered canary in us-east, which model rollout was affected and what type of customer contract did they have?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0331"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0332",
      "label": "What are the acceptance criteria for the trace export UI changes that include the corr-id copy behavior, the consent modal requirements, and the keyboard help overlay, and what else did the PM ask to include before merging?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0332"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0333",
      "label": "In the runtime inference forensics walkthrough for Arcadia Retail, what is their p95 latency target for interactive queries, and by what day did they say they would upload the sample requests?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0333"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0334",
      "label": "Why were SSE streaming connections getting cleanly closed after about 60 seconds only when going through the edge proxy, and what heartbeat change was implemented to mitigate it?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0334"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0336",
      "label": "For the CartPilot Redwood trial POC, when is the kickoff sync scheduled and what peak request rate did they say they expect during bursts?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0336"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0339",
      "label": "For the Usage Insights reconciliation drift alert, what drift percentage and time window trigger warn vs page, and which severity level includes workspace_id in the alert labels?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0339"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0340",
      "label": "For the private customer collector debug bundle upload 403 escalation, what support ticket number is being updated and what request ID should be included if escalating to infra?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6309297535714575,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0340"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0341",
      "label": "For the Proxima Bank 429 spike after priority routing rollout, what caused the throttling and what temporary policy exception did we apply, and how do we verify it is not burning the enterprise route SLOs?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6131471927654584,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0341"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0342",
      "label": "Our Key Management Standard says we must not cache plaintext DEKs, but ADR-007 allows a short-lived in-memory DEK cache. What is the actual policy we should follow, and how should we describe DEK caching and KMS outage behavior in customer-facing rotation materials?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9550236580992476,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0342"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0343",
      "label": "What caused the savings $/day estimates in Optimization Suggestions beta to not match Arkadia Finance billing, and what UI or API changes should we make to prevent this confusion going forward?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9674679834891693,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0343"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0345",
      "label": "For the Trident Financial Private VPC deal, what is our approved stance on US-only data residency and 7-year audit log retention/export, and which evidence pack artifacts should we share (and how) to back it up?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6131471927654584,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0345"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0346",
      "label": "After the Northwind latency spike, what metric wording ended up approved for the public case study and what exactly can Sales share privately without triggering new legal approval?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9066276098484507,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0346"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0347",
      "label": "For an air-gapped on-prem restore into a brand new cluster, what do we need to install before running restore apply, and what will the installer check and report if those prerequisites are missing?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0347"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0349",
      "label": "What caused the EU-West activation funnel and onboarding email issues in late January, and which code/config changes fixed them?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9469024295259745,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0349"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0350",
      "label": "Why did Northpeak Search see their Dedicated Burst Usage billed higher than expected, and how do we prevent retry attempts from double-counting burst GPU-seconds now?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9325210919548239,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0350"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0354",
      "label": "How should we explain and remediate a case where a customers invoice token total is higher than their Usage API export because streaming retries were double-counted, and what approvals are required before issuing a credit or changing the ledger?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.48867512060993706,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0354"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0355",
      "label": "What is the approved wording to use when a prospect demands EU-only data residency and no non-EU support access for the Hosted API, and who do we need to escalate to if they want that as a contractual commitment?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.7653606369886217,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0355"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0356",
      "label": "For the Acme AI dedicated failover incident with intermittent 500s and fallback churn, what v1 incident type and primary owner should we set, and what time-to-mitigate and time-to-fix targets should CS use in customer updates?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.7598336331031966,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0356"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0360",
      "label": "Customers behind strict corporate proxies are seeing 4xx failures when the quickstart templates add telemetry tags. What tagging approach and opt-out setting should we use, and how will that affect TTFS and activation KPI measurement?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0360"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0362",
      "label": "What is our v1 standard for streaming timeout enforcement and retry behavior, and do any of the Python, TypeScript, or Go SDKs still violate it based on the parity matrix and recent support tickets?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.4776418949509485,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0362"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0363",
      "label": "Northbridge Bank reported that request log TTL was not being enforced in us-east and that audit log exports were missing key admin event types; what caused these issues and what customer-facing compliance pack artifacts should we provide to show both retention enforcement and audit log evidence are now correct?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8829374068620366,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0363"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0365",
      "label": "Fast tier p99 is regressing in canary, what SLO targets and abort criteria apply and what rollback steps should we take? Which dashboards and reason codes should we use to confirm whether admission control or requeue behavior is causing it.",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.7767471075223498,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0365"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0367",
      "label": "For an enterprise ticket like SUP-1842 where streaming disconnects happen in us-east, when should we escalate from a Support bridge to a formal incident and post a status page update, and what customer update cadence and credits wording should we use?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.822534695579038,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0367"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0369",
      "label": "Why were some Hosted API overload errors missing Retry-After, and what changes and alerts ensure it stays consistent across gateway, routing, and streaming paths going forward?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9445735153413953,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0369"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0375",
      "label": "When a quantization profile is blocked due to a quality regression, how is that enforced differently from an incompatibility fallback, and what should oncall do to verify a customer route is back on baseline using logs, dashboards, and audit history?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9828920819566879,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0375"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0376",
      "label": "Northbridge Bank is seeing request logs older than their configured retention window in us-east after a maintenance window. What is the recommended remediation and what customer-safe evidence artifacts should we provide to prove the retention policy is configured correctly and enforcement has caught up?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9469024295259745,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0376"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0377",
      "label": "If the demo tenant starts returning 429s mid-demo and the Console dashboards are empty, what are the fastest recovery steps to keep the call moving and what fixes were put in place to prevent this from happening again?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8486987932505934,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0377"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0381",
      "label": "In the planned controlled failover game day on 2026-01-15 (Hosted API us-east → eu-west), what caused the routing automation to oscillate (flip/rollback), and what follow-up ticket and target ship date were created to fix the streaming disconnect/reconnect behavior clients saw during cross-region redirects?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0381"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0386",
      "label": "In the Feb 2026 SEV-2 incident where the Dedicated autoscaler in us-west stopped scaling above minimum because its metrics query returned no data, what was the underlying cause and what hotfix restored scale-out?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0386"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0389",
      "label": "After the Console v4.3.2 rollout (March 2026), what caused saved dashboard filter presets to revert to the default right after you edit and save a linked sub-report in the widget editor modal, and what hotfix fixed it?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0389"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0390",
      "label": "During a March 2026 prod us-east-1 dedicated incident where a sudden burst of high-concurrency streaming traffic caused intermittent 502/504s due to proxy ephemeral port pressure, what mitigations were applied (replica changes and any kernel/sysctl tuning) and-per the ingress proxy connection-entropy playbook-what RCA artifacts should be captured while the issue is ongoing?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0390"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0391",
      "label": "In the hosted API incident in late April 2026 where streaming requests started returning 502/504 and non-streaming requests also saw a sharp tail-latency increase, what was the underlying trigger (involving an edge keepalive/connection-TTL change) and what immediate mitigations were used to restore service?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0391"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0392",
      "label": "During the us-east canary that normalized/rotated the session cookie domain for cross-subdomain routing, long-poll streaming clients started losing their pinned model version and getting routed to a fallback model. What was the root cause in the edge/session-manager flow, and what immediate mitigations were applied (platform-side and customer-side)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0392"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0393",
      "label": "After the March 9, 2026 embedding-catalog promotion that made a 1024-d embedding model the default, a dedicated Hosted API tenant (us-west with eu-west failover) saw /v1/rerank start timing out with 512-vs-1024 vector length mismatches when using the alias open-embed-default. What was the confirmed root cause, and what short-term mitigation and permanent fix did we deploy?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0393"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0394",
      "label": "For Incident Bot v2, when responders use break-glass during a Sev0/Sev1 incident to view tenant-scoped config diffs or run privileged telemetry queries, which audit events must be emitted and how long are those audit events retained (SIEM vs immutable archive)? Also, what baseline vs break-glass permission scopes does the bot use for these reads?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0394"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0395",
      "label": "What was the root cause and planned server-side fix for the March 2026 incident where function-call webhook callbacks were retried after a timeout and the JSON payload stayed the same, but the HMAC signature header changed between the first delivery and the retry due to gzip/encoding differences?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0395"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0396",
      "label": "After rotating a Hosted API key in the Console, why might the Usage → Ad-hoc Query still show request IDs tied to the old key and 'View trace' links that 404, and what immediate mitigation did Support/SRE apply before a proper fix shipped?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0396"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0399",
      "label": "In the mid-March 2026 eu-west canary rollout where sticky fallback kept pinning requests back onto the failing canary instead of the stable quantized variant, what configuration mistake caused the loop and what mitigations were applied to stop it?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0399"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0402",
      "label": "In a Redwood Private (VPC/on-prem) upgrade using the private installer, the UI has been stuck on \"Running migrations\" for over an hour and the migration pod logs repeatedly say it's waiting on a Postgres advisory lock (the Postgres database is externally managed, not the in-cluster StatefulSet). What's the underlying cause, what Support-guided action resolved it, and what specific logs/artifacts should be collected before attempting a retry or rollback?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0402"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0404",
      "label": "We saw quality regressions after enabling an INT8/mixed-precision variant and suspect KV-cache rounding on long-context requests. What request header can we set as a quick mitigation to bypass the KV cache, and in the dynamic quant handoff policy for the interactive-chat route what minimum KV-cache hit-rate and max sequence length are required before routing to a quant variant?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6131471927654584,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0404"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0407",
      "label": "In March 2026, what caused duplicate tool/webhook executions during *streaming* chat retries only when the idempotency key contained non-ASCII characters (e.g., emoji/diacritics) and the request went over HTTP/2 through the edge proxy-and what were the recommended workaround and the server-side fix (including what normalization we applied)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0407"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0410",
      "label": "In the EU hosted API, streaming generate requests over HTTP/2 intermittently returned 502 only when the client included a ~12KB base64-encoded user-metadata header. What was the root cause in our proxy chain, and what server-side mitigation did we deploy to stop the 502s (include the header and cutoff/behavior)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0410"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0411",
      "label": "On Streamly AI's dedicated pool dp-132-usw, what % of interactive burst credits should be reserved exclusively for priority=high routes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0411"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0412",
      "label": "For extraction, what price-normalization correctness is reported at temp=0.0 with strict JSON + canonicalization, and what earlier % is it compared to?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0412"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0414",
      "label": "INC-9821: was the degraded GPU node an OOM or intermittent driver/kernel launch stalls?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0414"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0417",
      "label": "What does Redwood Private on-prem recommend for signing backup manifests?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9197207891481876,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0417"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0419",
      "label": "What is the API endpoint to start a Unified Capacity Transition migration?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0419"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0420",
      "label": "For EXP-002 (eu-west→us-east), what egress cost rate and measurement basis does the cost penalty catalog use?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9197207891481876,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0420"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0422",
      "label": "How do audit log exports access records older than the retention cutoff (super-admin override vs legal hold)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0422"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0423",
      "label": "For the office keycard audit, how many months of access logs should we export?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.18457569677956817,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0423"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0426",
      "label": "What are the automated rollback trigger thresholds (safety sentinel, composite quality, business KPI) in the Optimize cohort-driven gates PRD?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9197207891481876,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0426"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0428",
      "label": "What monthly token volume discount breakpoints apply for Hosted pricing in the enterprise playbook?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0428"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0434",
      "label": "In Q4 2025, which customer industry segment was featured in the most published customer stories (case studies + one-page success stories)?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1.0000000000000002,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0434"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0435",
      "label": "How many weekly status updates for Project Orion marked the project as 'At Risk'?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1.0000000000000002,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0435"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0436",
      "label": "Across Redwood's Go, Python, and TypeScript SDKs, which SDK has the highest number of customer-reported auth-related bug reports, and what are the corresponding support ticket IDs?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.22009176629808017,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0436"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0438",
      "label": "Which Redwood Inference engineering team owned the most production incidents involving GPU capacity or quota exhaustion in H1 2025?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5716213154129579,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0438"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0439",
      "label": "How many internal incident writeups at Redwood Inference describe activating automatic fallback (to a different model or region) as part of mitigation?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5087403079104241,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0439"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0442",
      "label": "What is Redwood Inference's end-to-end process for making a production change (definition, required approvals, pre-deploy verification, deploy execution, customer/internal communications, and rollback/post-change requirements) for the Hosted API and Console?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8141829732638892,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0442"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0443",
      "label": "List every internal communication thread (email, Slack, and meeting notes) about the Redwood Private upgrade 'rollback loop' bug (including references to RRB-17 or 'stuck rollback').",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0443"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0445",
      "label": "\"What is the complete required content of a production rollout plan for introducing a new LLM model version on Redwood Inference (hosted, dedicated, and private), including required metrics/thresholds, approvals, rollback/fallback rules, and customer-facing communications?\"",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9156857657441143,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0445"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0447",
      "label": "What is Redwood Inference's complete end-to-end process for shipping an emergency serving-runtime hotfix to production (Hosted + Dedicated), including required approvals, checklists, rollout/rollback steps, and mandatory customer communications?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8626199664389337,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0447"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0449",
      "label": "In March 2026, which deployment offering-Hosted API, Dedicated, or Private-had the most customer support escalations where the primary symptom was request timeouts?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9495662644168646,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0449"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0451",
      "label": "In the team chat about posting jokes when an AI makes things up in production, what tagging requirement did someone propose for the memes?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.43067655807339306,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0451"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0452",
      "label": "Who owns the messy go-to-market misc assets dump with placeholder marketing files and micro-copy ideas created during a last-minute scramble before a partner call?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0452"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0455",
      "label": "Are sneakers acceptable for the first few weeks of the company summer softball league, or do we need cleats right away?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0455"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0460",
      "label": "In a first round technical interview for a senior sales engineer candidate, what were the agreed next steps after the call?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0460"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0461",
      "label": "What is the proposed weekly cleanup schedule and routine for tidying the engineering kitchen fridge so old opened items get discarded?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0461"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0463",
      "label": "What time are people planning to start the office foosball tournament lunch bracket on Friday?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0463"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0464",
      "label": "For the casual soccer game being organized on Friday after work, what was the agreed maximum number of players?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0464"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0465",
      "label": "When are signups due for the informal company softball team being organized for summer?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0465"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0467",
      "label": "Is there an internal hackathon demo project that uses a small web UI and a fake backend to visualize model serving behavior like batching and cache hit rates, and can be run locally or deployed for a quick demo?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0467"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0470",
      "label": "In a backend inference engineer interview about multi-tenant GPU scheduling, what strategy did the candidate propose to reduce cold misses for growing attention cache in long chat histories?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0470"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0472",
      "label": "What is the company's thesis about where competitive advantage shifts as LLM ecosystems expand?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0476",
      "label": "What is the stated differentiation around reliability and graceful degradation in Redwood's platform?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0477",
      "label": "What are the four main revenue streams in Redwood Inference's business model?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0478",
      "label": "What add-on categories are mentioned for Redwood's commercial offering?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0479",
      "label": "What triggers Redwood's expansion from PLG-led adoption to sales-assisted enterprise for Dedicated and Private?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0484",
      "label": "What is the complete mapping of `billing_tag` → Finance GL code (and cost center owner) used for chargebacks in the Eval Budget Ledger, and where is that mapping source-of-truth maintained?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0486",
      "label": "How do we compute and invoice the **per-request CO2e** for Hosted API usage (including the region-specific carbon intensity factors and any cache-hit adjustments), and where is that CO2e exposed in the customer Usage API/export schema?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0487",
      "label": "When a streaming resume attempt fails with HTTP 409 (checkpoint mismatch), what is the full JSON schema of the server's error payload (including any `resume_details` fields), and how is the `kv_crc` value inside v2 resume tokens computed?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0488",
      "label": "When we use the GPU firmware operator to do a BIOS/firmware rollout, what BIOS settings are included in Redwood's \"safe mode boot\" profile for A100/H100 nodes (e.g., Secure Boot, Above 4G decoding, Resizable BAR), and what are the default values for each?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0492",
      "label": "For the \"microburst insurance\" negotiated SLA bundles, what's the default burst surcharge charged when burst capacity units are consumed, and which GL account should that surcharge be posted to?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0495",
      "label": "For the Remote-first Pilot, what are the exact preferred vendors and model SKUs for the standing desk converter loaners, and which of those vendors can ship to Canada with delivery in under 48 hours?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0496",
      "label": "Does Incident Bot v2 support sending its incident context cards (deploy/config diff/saturation + pinned escalation context) to Microsoft Teams via an incoming webhook/Adaptive Cards, and if so what's the required payload schema and idempotency/retry behavior for updates?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0497",
      "label": "When the Security Risk Committee reviews a high-risk control exception (e.g., risk rating above 7), what quorum is required for the vote to be valid, and how are tie votes resolved?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0499",
      "label": "For the `profiling.signals` Kafka topic used by trace-orch and fidelity-proxy, what are the current production settings (partition count and message retention time), and where can I find the approval/change record for the most recent update?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0500",
      "label": "When a high-risk production change requires Change Advisory Board (CAB) approval, what quorum is required and what's the tie-break process if the CAB can't reach consensus-and who has final decision authority?",
      "kind": "question",
      "metric": "ndcg_at_10",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "basic",
      "label": "basic",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.9886363636363636,
          "unit": "fraction",
          "n": 88,
          "attempted": 88,
          "expected": 88,
          "missing": 0,
          "ci95": [
            0.9659090909090909,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0002",
            "qst_0004",
            "qst_0008",
            "qst_0009",
            "qst_0010",
            "qst_0011",
            "qst_0012",
            "qst_0013",
            "qst_0015",
            "qst_0016",
            "qst_0022",
            "qst_0026",
            "qst_0028",
            "qst_0030",
            "qst_0032",
            "qst_0033",
            "qst_0035",
            "qst_0040",
            "qst_0041",
            "qst_0043",
            "qst_0044",
            "qst_0046",
            "qst_0048",
            "qst_0049",
            "qst_0050",
            "qst_0051",
            "qst_0052",
            "qst_0053",
            "qst_0055",
            "qst_0059",
            "qst_0063",
            "qst_0064",
            "qst_0065",
            "qst_0066",
            "qst_0067",
            "qst_0069",
            "qst_0071",
            "qst_0074",
            "qst_0077",
            "qst_0079",
            "qst_0082",
            "qst_0083",
            "qst_0085",
            "qst_0086",
            "qst_0088",
            "qst_0089",
            "qst_0092",
            "qst_0096",
            "qst_0098",
            "qst_0100",
            "qst_0102",
            "qst_0103",
            "qst_0104",
            "qst_0105",
            "qst_0106",
            "qst_0111",
            "qst_0117",
            "qst_0122",
            "qst_0124",
            "qst_0125",
            "qst_0127",
            "qst_0129",
            "qst_0130",
            "qst_0132",
            "qst_0134",
            "qst_0138",
            "qst_0141",
            "qst_0145",
            "qst_0147",
            "qst_0148",
            "qst_0152",
            "qst_0153",
            "qst_0154",
            "qst_0155",
            "qst_0157",
            "qst_0158",
            "qst_0161",
            "qst_0162",
            "qst_0163",
            "qst_0165",
            "qst_0166",
            "qst_0167",
            "qst_0168",
            "qst_0169",
            "qst_0172",
            "qst_0173",
            "qst_0174",
            "qst_0175"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "completeness",
      "label": "completeness",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8788888888888889,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.7277777777777777,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0434",
            "qst_0435",
            "qst_0436",
            "qst_0438",
            "qst_0439",
            "qst_0442",
            "qst_0443",
            "qst_0445",
            "qst_0447",
            "qst_0449"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "conflicting_info",
      "label": "conflicting info",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.95,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            0.85,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0411",
            "qst_0412",
            "qst_0414",
            "qst_0417",
            "qst_0419",
            "qst_0420",
            "qst_0422",
            "qst_0423",
            "qst_0426",
            "qst_0428"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "constrained",
      "label": "constrained",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 15,
          "attempted": 15,
          "expected": 15,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0381",
            "qst_0386",
            "qst_0389",
            "qst_0390",
            "qst_0391",
            "qst_0392",
            "qst_0393",
            "qst_0394",
            "qst_0395",
            "qst_0396",
            "qst_0399",
            "qst_0402",
            "qst_0404",
            "qst_0407",
            "qst_0410"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "high_level",
      "label": "high level",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "info_not_found",
      "label": "info not found",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "intra_document_reasoning",
      "label": "intra document reasoning",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 20,
          "attempted": 20,
          "expected": 20,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0301",
            "qst_0302",
            "qst_0304",
            "qst_0305",
            "qst_0306",
            "qst_0307",
            "qst_0309",
            "qst_0314",
            "qst_0315",
            "qst_0317",
            "qst_0322",
            "qst_0323",
            "qst_0324",
            "qst_0331",
            "qst_0332",
            "qst_0333",
            "qst_0334",
            "qst_0336",
            "qst_0339",
            "qst_0340"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "miscellaneous",
      "label": "miscellaneous",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 10,
          "attempted": 10,
          "expected": 10,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0451",
            "qst_0452",
            "qst_0455",
            "qst_0460",
            "qst_0461",
            "qst_0463",
            "qst_0464",
            "qst_0465",
            "qst_0467",
            "qst_0470"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "project_related",
      "label": "project related",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 20,
          "attempted": 20,
          "expected": 20,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0341",
            "qst_0342",
            "qst_0343",
            "qst_0345",
            "qst_0346",
            "qst_0347",
            "qst_0349",
            "qst_0350",
            "qst_0354",
            "qst_0355",
            "qst_0356",
            "qst_0360",
            "qst_0362",
            "qst_0363",
            "qst_0365",
            "qst_0367",
            "qst_0369",
            "qst_0375",
            "qst_0376",
            "qst_0377"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "semantic",
      "label": "semantic",
      "kind": "category",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 62,
          "attempted": 62,
          "expected": 62,
          "missing": 0,
          "ci95": [
            1,
            1
          ],
          "method": "question-bootstrap-10000-seed1729",
          "sample_ids": [
            "qst_0176",
            "qst_0177",
            "qst_0180",
            "qst_0181",
            "qst_0182",
            "qst_0186",
            "qst_0187",
            "qst_0189",
            "qst_0192",
            "qst_0195",
            "qst_0198",
            "qst_0199",
            "qst_0200",
            "qst_0209",
            "qst_0212",
            "qst_0213",
            "qst_0215",
            "qst_0216",
            "qst_0217",
            "qst_0218",
            "qst_0221",
            "qst_0223",
            "qst_0224",
            "qst_0226",
            "qst_0228",
            "qst_0229",
            "qst_0230",
            "qst_0231",
            "qst_0235",
            "qst_0238",
            "qst_0239",
            "qst_0241",
            "qst_0245",
            "qst_0247",
            "qst_0250",
            "qst_0252",
            "qst_0253",
            "qst_0254",
            "qst_0255",
            "qst_0256",
            "qst_0257",
            "qst_0260",
            "qst_0262",
            "qst_0265",
            "qst_0270",
            "qst_0271",
            "qst_0274",
            "qst_0275",
            "qst_0277",
            "qst_0280",
            "qst_0281",
            "qst_0283",
            "qst_0284",
            "qst_0285",
            "qst_0286",
            "qst_0289",
            "qst_0290",
            "qst_0291",
            "qst_0293",
            "qst_0295",
            "qst_0296",
            "qst_0299"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0002",
      "label": "What is the name of the new metric added so SRE can track when server-side streaming sessions get finalized due to hitting the time limit?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0002"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0004",
      "label": "In the meeting about onboarding a SaaS product to Google Cloud Marketplace, what did the GCP team recommend for handling delays where a new subscription entitlement is not immediately available during the customer onboarding flow?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0004"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0008",
      "label": "In the internal shiproom runner notes, what is the target time-to-recover threshold for a verified rollback in staging?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0008"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0009",
      "label": "In the EdgePath evaluation email thread, what alternative Year 1 pricing package did Redwood propose instead of matching the competitor's 50 percent first-year discount and migration credit?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0009"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0010",
      "label": "How does the new alerting approach group model-serving requests by the stages they pass through to reduce noisy alerts and speed up root cause analysis?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0010"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0011",
      "label": "How long is the validity period for the telemetry driven runbook author certification from the operator training bootcamp?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0011"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0012",
      "label": "What was identified as the root cause of the intermittent VPN connectivity blackholes and duplicate IP warnings for users on the corporate Wi-Fi in the SF2 building behind a specific edge switch?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0012"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0013",
      "label": "What is the recommended two-stage process for rotating the signing credentials used by GPU image build and promotion pipelines, including how to cut over safely and roll back if something goes wrong?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0013"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0015",
      "label": "What was the high-percentile latency concern reported after a smoke test of about 50 concurrent streaming chat sessions for Satellite Grove?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0015"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0016",
      "label": "What is the company policy for how long contractor access should last by default before it expires, according to the access and permissions playbook?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0016"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0022",
      "label": "What is the recommended liability cap language in the procurement legal triage SOP for vendor contracts (including the key exceptions and how the cap amount is determined)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0022"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0026",
      "label": "For a customer upgrading a private deployment in their own cloud environment, what restore time did support say is typical if the upgrade fails and a rollback is needed (assuming snapshots were taken)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0026"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0028",
      "label": "What steps and commands were recommended for creating and pushing a signed annotated Git tag and then making a draft GitHub release for the v1.14.0 release?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0028"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0030",
      "label": "In the Console feature that compares canary or A-B cohorts and links metric anomalies to request traces, what customer tiers are allowed into the beta via a feature flag gate?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0030"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0032",
      "label": "What performance goal was set for reducing GPU out of memory incidents in the stress tests for the windowed KV cache proposal?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0032"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0033",
      "label": "In the security review meeting about private cloud upgrades for a financial services customer, what was the stated minimum audit log retention period they wanted?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0033"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0035",
      "label": "For regulated customer private VPC deployments, what is the tenant-facing monthly availability target for the inference endpoint?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0035"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0040",
      "label": "What are the retention time windows for these lightweight kernel state snapshots (in-memory cache, indexed store, and any cold archive) used for post-incident diagnostics?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0040"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0041",
      "label": "In the security review call about buying an LLM inference service through AWS Marketplace, what was said about whether cross-region failover is enabled by default for the hosted endpoint?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0041"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0043",
      "label": "What caused the brief p99 latency jump on the hosted text generation endpoint in us-west-2, and what mitigation was applied to bring it back down?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0043"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0044",
      "label": "What support response time commitments did Acme approve for being shown in their AWS and Azure marketplace listing for incident priority levels?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0044"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0046",
      "label": "In the Medicore procurement email thread about data use and telemetry, what turnaround timeline did Redwood propose for delivering the operational runbook?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0046"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0048",
      "label": "What are the initial weighting percentages for the signals used in the Model Resilience Propensity Index, before the 4-week recalibration?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0048"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0049",
      "label": "In the email thread about BrightPath considering a marketplace pilot versus a direct contract for an AI inference vendor, what lead time was given for getting marketplace entitlement after onboarding?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0049"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0050",
      "label": "What is the proposed HTTP header scheme for attributing API traffic to quickstart templates, including how the header value is formatted and the opt-in header?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0050"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0051",
      "label": "In the cost-aware model routing proposal, what telemetry fields were suggested to emit so the team can analyze routing decisions and tradeoffs between spend and quality?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0051"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0052",
      "label": "What is the sampling policy for request routes of different telemetry priority levels in the new path-based observability change?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0052"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0053",
      "label": "What metric label allowlist and caps did the SRE team decide for v1 shadow traffic metrics to prevent high-cardinality issues (including limits for route and experiment labels)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0053"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0055",
      "label": "What audit log retention period did Redwood commit to for pilot customers during the 2027 API model migration, in response to the Acme Health security questions?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0055"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0059",
      "label": "In the incident where a CI runner service token was suspected of being used from two different datacenter IP ranges minutes apart, what time window did the team decide to audit for artifact uploads and repository pushes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0059"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0063",
      "label": "For the NorthPoint Signalworks dedicated pilot, what p95 latency target did they set for short chat interactions?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0063"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0064",
      "label": "What remediation timeline by severity did the security team propose for the upcoming external penetration test findings?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0064"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0065",
      "label": "During the November 2025 us-east incident where streaming chat responses were dropping, what was the internal start and mitigation time that support shared with the affected customer?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0065"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0066",
      "label": "What p95 latency goal did the founder set for the on-page product Q and A widget for the streetwear brand that expects big traffic spikes during limited edition drops?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0066"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0067",
      "label": "What did the internal usability test with six reviewers conclude about the table column compression feature, especially regarding its discoverability?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0067"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0069",
      "label": "In an ops day log about GPU rebalancing and brief west region network instability, what was the target slot capacity for the small hosted pool in the slot metrics dashboard?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0069"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0071",
      "label": "In the Private VPC console prototype for an interactive network path storyboard used to rehearse cluster upgrades, what is explicitly out of scope for the ticket?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0071"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0074",
      "label": "How does the new v1 chat compatibility layer handle invalid tool or function schemas, and what HTTP status does it return when validation fails?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0074"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0077",
      "label": "What is the three-step message sequence used for the new cross-language streaming startup handshake in the SDKs?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0077"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0079",
      "label": "In a morning account blitz note from May 2026, what was the proposed mitigation to address repeated inference timeouts for a healthcare customer with a mid-June renewal?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0079"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0082",
      "label": "During the year end handoff meeting, what was identified as the cause of the early December latency incident impacting some p95 requests?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0082"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0083",
      "label": "What was the root cause of the customer invoice overcharge where an organization-wide promotional credit was mistakenly redistributed as charges across multiple child accounts during a billing reconciliation run in a hybrid dedicated plus hosted setup?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0083"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0085",
      "label": "What are the three request quality and cost tiers proposed in the lightweight interaction tiering model for user facing features?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0085"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0086",
      "label": "Why would Apple Silicon MacBooks coming back from logic board repair get stuck during automated DEP or MDM setup at first boot, with the device checking in but the management profile and SSO certificate never installing?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0086"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0088",
      "label": "In the Slack chat about a desk plant rap contest, where was the remaining sticker stock said to be stored?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0088"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0089",
      "label": "What are the three runtime precision settings introduced for controlling fp32 vs lower precision execution, and what does each mode allow?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0089"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0092",
      "label": "In the engineering rollout chat about deprecating v2, what was the start time and region for the initial canary deployment and what traffic ramp was planned?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0092"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0096",
      "label": "When an employee is marked for immediate termination in Workday, why can a third party SaaS seat stay billable for around two weeks even after Okta deactivates the user?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0096"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0098",
      "label": "In the MedSecure logging and PII redaction POC email thread, what method did Redwood propose for pseudonymizing trace or request IDs so correlation works within a tenant but not across customers?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0098"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0100",
      "label": "What change was applied to reduce the latency spike caused by a KV cache and continuous batching regression on the us-west-2 inference production cluster?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0100"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0102",
      "label": "What are the default significance threshold and minimum sample size guidelines for deciding whether a model regression is actionable in the causal regression investigation handbook?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0102"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0103",
      "label": "How is the baseline token cost computed for the per-span cost delta metric in the new tracing observability change?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0103"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0104",
      "label": "What is the standard amount of time a new hire buddy is expected to spend per day during the first two weeks when a long-term contractor is converted to a full-time employee?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0104"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0105",
      "label": "What are the proposed internal time to mitigate targets by severity for incident response, regardless of customer deployment tier?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0105"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0106",
      "label": "What baseline metrics are listed for the Console activation health panel in the quarterly activation plan for customer-hosted Private VPC deployments?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0106"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0111",
      "label": "What are the acceptance thresholds for rerank performance in the mini-bundle unified inference API, including the required top-1 precision and the P95 latency limit?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0111"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0117",
      "label": "In the PRD about adding managed orchestration for multi-model inference pipelines, what are the rollout milestones and target quarters for alpha, beta, and general availability?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0117"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0122",
      "label": "Which GPU cluster or pool was flagged as having relatively low average usage but high churn, making its free capacity unreliable during a specific morning UTC window?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0122"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0124",
      "label": "What percent reduction in monthly model or token spend was the founder of a small developer tools micro-SaaS aiming for while keeping acceptable latency?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0124"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0125",
      "label": "What is the formula used by the new autoscaling approach that combines a high-percentile request rate with request variability to compute effective demand, including the default weights?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0125"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0127",
      "label": "What is the phased rollout plan for the change that smooths dedicated autoscaling signals and adds an advisory gateway-to-controller backpressure mechanism?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0127"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0129",
      "label": "What was the temporary workaround used to prevent the Windows 11 imaging task sequence from failing when a provisioning script breaks under certain non-default locale collation settings?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0129"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0130",
      "label": "What latency goals did the founder of Birch Branch Apps mention for short in-app chat responses during their early discovery call?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0130"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0132",
      "label": "In a case where server-sent event streaming responses sometimes lose the very last space or period only when traffic goes through an HTTP/2 edge proxy with stream preemption, what mitigation was proposed to stop the final bytes from being dropped?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0132"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0134",
      "label": "What was the suspected root cause behind the brief p99 latency spikes on the hosted chat inference route during small burst traffic?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0134"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0138",
      "label": "What was the CTOs non negotiable security requirement mentioned on the early 2026 demo call for the analytics SaaS customer moving from hosted API to a dedicated VPC deployment?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0138"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0141",
      "label": "What was the target goal for average router queue time for very short token sequences, compared to the baseline of 28 ms, in the weekly sprint anchor about reducing staging inference queuing latency?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0141"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0145",
      "label": "For the upcoming joint demo with a Japan-based logistics customer, what document did they prefer for trial procurement instead of a full purchase order?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0145"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0147",
      "label": "What is the planned cutoff date when the old model capability format may stop being accepted for new writes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0147"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0148",
      "label": "In the recent SDK update that added typed streaming errors and a reconnecting iterator API across Python, TypeScript, and Go, what package versions were released for each SDK?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0148"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0152",
      "label": "What caused the GPU training slowdowns in one us-east rack where same-rack pod latency jumped to tens of milliseconds and NCCL allreduce started timing out, and what network symptom on the top-of-rack switch pointed to it?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0152"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0153",
      "label": "In the engineering chat about a Go refactor PR, what did they change in the linter config to get CI passing when a gocritic rule complained about context parameter types in tests?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0153"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0154",
      "label": "What success targets were suggested for the four week micro credential pilot with SI partners (completion percentage and number of integrations started within about two months)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0154"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0155",
      "label": "What are the lifecycle states and key transitions for the edge adapter service that registers with a control plane and can fall back to hosted inference?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0155"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0157",
      "label": "What is the corrective journal number created to move a supplier credit amount from the DEV cost center to the OPS cost center after it was posted incorrectly during the March close?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0157"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0158",
      "label": "What are the three default preset options in the autoscaling policy builder for the dedicated capacity trial onboarding flow?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0158"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0161",
      "label": "In the OpenAI compatibility adapter update for Redwood deployments, what priority order does the normalizer use to resolve conflicting tool selection hints?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0161"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0162",
      "label": "What caused the brief period where a few tenants were not being throttled correctly after a shard reassignment happened at the same time as a runtime policy reload?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0162"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0163",
      "label": "What are the pass fail threshold percentages for the release guardrails for critical versus lower risk controls in the SSO and audit log compliance rollout plan?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0163"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0165",
      "label": "What was the agreed condition and timing for approving the additional senior infrastructure engineering hire under the hiring freeze plan?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0165"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0166",
      "label": "What procurement target date did ClearPath Health give for signing an agreement after the streaming model benchmark and security review?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0166"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0167",
      "label": "What configuration option was added to let repositories explicitly allow Docker-in-Docker during the staged rollout of the new GitHub composite action CI setup?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0167"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0168",
      "label": "What is the HTTP endpoint and method for the new service that returns recommended default profiles for batching, cache lifetimes, and quantization based on recent traffic telemetry?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0168"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0169",
      "label": "Who is responsible for signing off on latency regression in the weekly customer rollout tracker for Nimbus Health?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0169"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0172",
      "label": "In the eng-platform standup about staging 401 spikes on the generate endpoint, what header value caused the requests to miss both the primary and fallback route predicates and hit the fallthrough route?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0172"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0173",
      "label": "What maintenance window and expected downtime were planned for the rollout bringing up new dedicated GPU capacity nodes, including the canary and broader rollout dates?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0173"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0174",
      "label": "What are the required fields that each entry in the contract conformance results ledger must include?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0174"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0175",
      "label": "What keyboard-only interaction model is required for the compact dashboard widgets, including how focus enters the card and how users move and activate controls inside it?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0175"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0176",
      "label": "When does booking open for the new top end 80GB accelerator on dedicated clusters in the EU Central and India South locations?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0176"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0177",
      "label": "What were the final concession terms approved for the retail partner purchase through the major cloud providers marketplace, specifically the first year price reduction and the referral payout schedule after launch?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0177"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0180",
      "label": "In our rollout system, what is the name of the new mechanism that prevents a candidate release from getting full user traffic until a dry run with replayed requests and smoke checks has passed?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0180"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0181",
      "label": "During a large overnight upload run using short lived credentials refreshed by many transient workers, what client side scheduling change was recommended to stop periodic too many requests errors after a partial worker restart caused refresh bursts?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0181"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0182",
      "label": "During a failover between two active locations, what is the default time window where a short-lived resume credential is accepted on the destination before doing full authentication again?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0182"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0186",
      "label": "In the notes about keeping long, stop-and-go chat sessions cheap without replaying the whole history, what storage setup and time-to-live were proposed for keeping the compact per-session state for recent sessions versus longer retention?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0186"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0187",
      "label": "When a production model experiment shows a sustained increase in harmful output warnings for two evaluation windows, what automated actions should the platform take regarding traffic shifting and incident/oncall notifications?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0187"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0189",
      "label": "In our offline installer release process, how do we expose per-component supply chain provenance so an operator can see which build system and commit produced each piece and optionally enforce that missing or bad attestations fail the index step?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0189"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0192",
      "label": "In our GPU inference runtime, what change was introduced to cut the worst-case temporary device-memory spike when short and long requests are interleaved, especially for attention-heavy ops that used to grab per-op temp buffers?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0192"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0195",
      "label": "From the internal sandbox microbench in the EU West lab around May 2029, what was the estimated dollars per one million tokens for the biggest model on a single 40GB GPU when using the no-batching low-latency mode on the long output workload (about 300 tokens per response)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0195"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0198",
      "label": "In the EU West production setup where token streaming to browsers sometimes reconnects after a short handshake reset at the CDN layer, what change on the gateway side reduced cases where the last bit of generated text never arrives even though the backend says it finished?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0198"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0199",
      "label": "After a recent refactor to request grouping and the per request memoization layer, what is the specific risk if the cache key canonicalization now treats a trailing slash differently between callers, especially in a multi customer environment?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0199"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0200",
      "label": "For the hospital system that wants to run an interactive intake chatbot and auto-generate discharge writeups entirely inside its own locked-down data center with no patient data leaving the network, what end-to-end response time target did they set for producing about 200 tokens under peak load?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0200"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0209",
      "label": "For a small B2B SaaS embedding a live chat helper inside their product that needs partial replies immediately while it fetches recent user-event context, what first-response-time target and concurrent-load threshold were set as the pass criteria for their short trial?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0209"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0212",
      "label": "How should we set up low-noise oncall dashboards and tiered paging for a streaming text generation service so that responders can jump from a latency spike to a trace with hashed prompt references and cache hotness context and resolve issues within about 15 minutes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0212"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0213",
      "label": "For a content templating company planning tens of millions of very short generations during a Q3 ramp, what was the requested plan for a time limited evaluation that included a large free usage allowance?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0213"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0215",
      "label": "On managed iPhones on corporate Wi-Fi, why do long-lived server push text streams sometimes reconnect after the app is backgrounded and then end with a few seconds of silence and a cut-off final JSON fragment instead of a clean end marker?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0215"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0216",
      "label": "In our cross team outage process, what is the expected update rhythm for the shared incident channel from the first message through containment, including when the interval should be shortened?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0216"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0217",
      "label": "In a customer owned cloud network peered to our control plane, why would big model artifact downloads over long lived encrypted multiplexed streams start hanging after a transit vendor policy change, then immediately work again when the node packet size limit is lowered?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0217"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0218",
      "label": "In the private network hosted cluster console, what are the planned quick actions on the compact outage summary tile that on call engineers can use to stop a risky release or switch to a safer model option?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0218"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0221",
      "label": "When someone new starts at Redwood Inference, what is the target turnaround time for getting them fully set up with accounts and tools in North America versus Europe/Asia-Pacific?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0221"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0223",
      "label": "Why did our telemetry collectors in the dedicated production clusters stop being able to mint short lived credentials for a customer cloud identity when going through the AWS to GCP federation bridge, after an IAM policy tightening with conditional rules, causing missing metrics and traces for about a day in US East and EU West?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0223"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0224",
      "label": "In the public documentation for searching system activity records, what identifier should examples use instead of showing a persons email address to avoid exposing sensitive personal data?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0224"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0226",
      "label": "During the 10:00 UTC partial rollout that caused structured function-invocation outputs to sometimes drop or swap argument fields, what small hotfix was added in the batching/serving path to prevent argument-buffer interleaving?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0226"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0228",
      "label": "On Windows ARM tablets in our SF office, why does switching between the two WiFi bands make our always-on VPN still show connected but suddenly stop reaching private 10.x services until a reboot, with routes preferring the wrong network after the handoff?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0228"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0229",
      "label": "After a customer rotated credentials during a migration, their staging probes started alternating between auth failures complaining about a signature mismatch and throttling errors; what short lived workaround did support create so both legacy and new signing methods would be accepted during the rollout?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0229"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0230",
      "label": "For a chat style server sent events stream that must survive mobile disconnects without re running tool side effects, what mechanism was introduced to let a client reconnect from a safe checkpoint using a server issued marker plus an explicit receipt from the client, while keeping compatibility with existing OpenAI like streaming clients?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0230"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0231",
      "label": "For the healthcare client contract review about sending patient records through our connector into the vendor platform, what is the promised retention duration for ingestion related audit traces and the default way customers can obtain exports?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0231"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0235",
      "label": "In the email thread about an AI service pilot for a payments company, what were the four main categories of changes their procurement and counsel asked for in the first contract markup?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0235"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0238",
      "label": "How should I structure a small prompt experiment to reduce overconfident mistakes by mixing a couple of correct examples with one intentionally wrong example plus an immediate fix, while keeping the total examples very short and tracking things like made-up details and format compliance over a few hundred test prompts?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0238"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0239",
      "label": "For the fintech prospect doing merchant reconciliation and dispute handling, what are the specific compliance items their ops and security teams said must be in place before moving past the pilot phase?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0239"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0241",
      "label": "For a EU-hosted dedicated banking tenant facing a regulator check in late March 2026, how can we produce a tamper-evident 7-day audit activity export that includes per-event masking justification plus the GDPR lawful-ground and any consent time, with a signed checksum manifest and a trusted timestamp?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0241"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0245",
      "label": "Which fintech support-chat prospect dropped us after their lawyers rejected non-audited compliance proof and chose a rival that could deliver a private install with formal audit paperwork?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0245"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0247",
      "label": "In the partner sales planning call between a vendor and an ISV, what were the concrete due dates in late December for each side to send the pilot inputs and the vendor deliverables?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0247"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0250",
      "label": "In the dedicated GPU hosting plan with prepaid overage credits, what is the default time window after a demand spike during which any additional spikes are treated as part of the same billable overage event?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0250"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0252",
      "label": "In our incident response process, what is the rule for when a quick time-boxed after-action review is acceptable instead of writing the full formal analysis, based on impact duration and whether the issue hit one customer versus many?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0252"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0253",
      "label": "For the outsourced customer support company planning an isolated always-on deployment for AI agent guidance and end-of-chat ticket writeups, what peak simultaneous staff load and response-time target did they specify for real-time streaming suggestions?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0253"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0254",
      "label": "Need the account notes for a tiny two person dev shop building a browser based coding assistant that struggled with keeping API secrets safe in a serverless front end, is pushing hard for incremental responses to keep the UI snappy, and is asking about cheap usage plus some starter credit before converting from prototype to paid.",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0254"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0255",
      "label": "When is the 45 minute video call finally locked in for the vendor walkthrough that got moved because the client team was traveling to a Denver conference in early October 2026?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0255"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0256",
      "label": "During a gradual rollout of a new large model on a single production tenant in early March 2026, some sessions that were supposed to stay on a customer-specified version briefly got routed to a cheaper compressed build, causing worse answers without any server error spike. What was the immediate mitigation applied at the request routing layer to prevent this kind of misrouting when a version lock is provided?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0256"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0257",
      "label": "For the healthcare chat rollout that ramps from a few thousand to about fifteen thousand monthly active users by month three, what was the projected total monthly language model token consumption at steady state based on the stated session and prompt/response length assumptions?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0257"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0260",
      "label": "From a late July 2025 call with a vendor, what was the specific console tuning recommendation for the search route that included both a suggested batch size and a quantization setting?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0260"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0262",
      "label": "For the EU production rollout of a card-industry regulated bank, can you provide a tamper-evident proof package for a 3-day early-March 2026 activity export to their file-drop security tool that lets them verify no per-call audit records were lost or reordered (including a hardware-backed signed index and local validation steps)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0262"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0265",
      "label": "Why do some long range analytics charts in the US East production console briefly show zero or outdated values and the drill through to traces disappears, then fixes itself after about 10 to 30 minutes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0265"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0270",
      "label": "Which bank client got a temporary signoff to keep normal processing in Europe but let traffic switch to a US site only during disaster recovery drills or when the EU cluster is degraded, with payload logging turned off, and what were the start and end dates of that allowance?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0270"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0271",
      "label": "How should we handle a short term external engineering team working out of our London site so they can get into the lab and then send two sensitive prototype boards back to the US with proper custody tracking and customs clearance after the week ends?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0271"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0274",
      "label": "For the small legal document review firm that came in via a website contact form in late Feb 2026, what was their target p95 response time for an interactive single-user chat experience?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0274"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0275",
      "label": "Im looking for the doc about building a fast system to store and query model performance run histories and a simple UI to compare two runs plus an oncall workflow for investigating detected slowdowns, including how it groups repeated issues and how long detailed vs rolled up data is kept",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0275"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0277",
      "label": "For the mid sized software company building outbound sales chat reps that trigger actions in email, calendar, and a CRM, what latency targets did they set for live token streaming and for the full message to action acknowledgement flow during their planned short evaluation?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0277"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0280",
      "label": "In the rollout plan for the internal service that only hands out application secrets to verified runtime hosts using short lived machine health proofs and hardware backed key unwrapping, what is the default hot and cold retention period for the tamper resistant audit trail?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0280"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0281",
      "label": "In the GPU reservation proposal for a pharma company with nightly model scoring and some real time triage traffic, what was the estimated cache warmup cost expressed in token hours for the smallest configuration?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0281"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0283",
      "label": "In our long conversation serving experiments, what mitigation is suggested for the tendency to start making up references right after a retrieval step when older context has been aggressively shrunk to save memory?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0283"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0284",
      "label": "What phone number did the recruiter provide for the frontend engineering candidate so the interviewers could reach her quickly if she was running late?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0284"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0285",
      "label": "How does the proposed multi-region traffic controller proactively avoid tail-latency blowups by warming likely failover locations ahead of time and using a three-level degrade/shed mechanism before hard failures occur?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0285"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0286",
      "label": "For a small mobile game studio adding realtime global chat and match lobby messaging, what latency goal did they set for typical short messages so players dont notice added delay?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0286"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0289",
      "label": "What caused the March 2026 morning spike where a dedicated east coast deployment started cutting off live generated replies mid stream and briefly overcounted usage charges due to retry and double counting behavior?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0289"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0290",
      "label": "In the service that lets clients discover which language models are available, where is the advisory 0 to 100 score documented that helps automatic selection based on feature match between our capability names and an OpenAI style interface?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0290"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0291",
      "label": "For the shipping customer rollout checklist, what is the requirement for output stability across repeated cold and warm starts before approving deployment?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0291"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0293",
      "label": "How does the temporary cross team sandbox get automatically shut down, including the default lifetime and the advance warning sent to the owners before everything is archived?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0293"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0295",
      "label": "In the late March 2026 payments rollout, what was the prior deployment version the team reverted to after the early Europe canary started restart looping due to a shutdown hook issue?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0295"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0296",
      "label": "Who on the team committed to producing a short animated demo to embed in the upcoming renewal business review materials, and what was the demo supposed to showcase?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0296"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0299",
      "label": "For the short evaluation engagement with the retail client where we offered a temporary sandbox to run about a couple thousand acceptance examples and promised a signed deletion attestation afterward, what date did we say we needed to receive their materials in order to hit the staging deadline?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0299"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0301",
      "label": "During the 2025-02-10 capacity incident, what two thresholds should trigger noisy-tenant detection in the runbook update, and how long were SLA targets breached?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0301"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0302",
      "label": "For the 90-day AcmeSec eval, what do they need to send to get Splunk HEC audit log exports wired up, and which internal team members are responsible for wiring the export and sending the pricing quote?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0302"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0304",
      "label": "For the PayRight Redwood DPA negotiation, what incident notification timeframe did PayRight request at the start and what notification window did Redwood propose in the final tracked wording before sending the DocuSign link?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0304"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0305",
      "label": "When using the python async batching helper for embeddings then rerank, what singleton-wrapped rerank response shape can show up during auth refresh, and what redwood.utils normalizer did the sampleapp add to fix the output shape?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0305"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0306",
      "label": "During the us-west-2 incident with TLS handshake failures on edge.redwood.ai, what alert triggered the PagerDuty page and what two preventative follow-ups were planned before closing the thread?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0306"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0307",
      "label": "For the StrataCore joint benchmark on the S100 card, what model and metrics are they benchmarking, and what artifacts will be included in the post run internal results bundle for signoff?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0307"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0309",
      "label": "After the docs deploy failed due to mkdocs linkcheck reporting broken anchor fragments from an include refactor, what final deploy-bot status was reported once prod was updated?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0309"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0314",
      "label": "For the Jupiter Therapeutics rollout, if the target go-live is 2028-07-22 and the canary is planned as 7 days with a 10/90 traffic split, what date should the canary period start?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0314"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0315",
      "label": "What UI indicator did the team plan to show for tenants in escalation mode, and what default grace-window duration was chosen for the circuit-breaker?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0315"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0317",
      "label": "In the Adaptive Log Zipper rollout plan, what dependency blocks per-tenant admission signaling in the edge gateway and what fallback mode is planned until it is added?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0317"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0322",
      "label": "What config flag controls enabling the bandwidth-aware fanout planner for TP allreduce on heterogeneous link clusters, and what other new components are called out in the release notes as helping reduce comm and compute stalls?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0322"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0323",
      "label": "Which issue filed after two customer escalations describes transient admin privileges or permission denied after SSO due to SAML assertion ordering and session stitching, and what environment(s) was it reproduced on?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0323"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0324",
      "label": "For the Q3 KPI dashboard email thread, what was the quarterly uptime percentage and how far below the target was it, and which JIRA ticket was Jonas going to create to track the infra changes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0324"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0331",
      "label": "In the incident where Meridian Analytics saw pinned sessions revert to the fallback variant during a tiered canary in us-east, which model rollout was affected and what type of customer contract did they have?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0331"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0332",
      "label": "What are the acceptance criteria for the trace export UI changes that include the corr-id copy behavior, the consent modal requirements, and the keyboard help overlay, and what else did the PM ask to include before merging?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0332"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0333",
      "label": "In the runtime inference forensics walkthrough for Arcadia Retail, what is their p95 latency target for interactive queries, and by what day did they say they would upload the sample requests?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0333"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0334",
      "label": "Why were SSE streaming connections getting cleanly closed after about 60 seconds only when going through the edge proxy, and what heartbeat change was implemented to mitigate it?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0334"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0336",
      "label": "For the CartPilot Redwood trial POC, when is the kickoff sync scheduled and what peak request rate did they say they expect during bursts?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0336"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0339",
      "label": "For the Usage Insights reconciliation drift alert, what drift percentage and time window trigger warn vs page, and which severity level includes workspace_id in the alert labels?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0339"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0340",
      "label": "For the private customer collector debug bundle upload 403 escalation, what support ticket number is being updated and what request ID should be included if escalating to infra?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0340"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0341",
      "label": "For the Proxima Bank 429 spike after priority routing rollout, what caused the throttling and what temporary policy exception did we apply, and how do we verify it is not burning the enterprise route SLOs?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0341"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0342",
      "label": "Our Key Management Standard says we must not cache plaintext DEKs, but ADR-007 allows a short-lived in-memory DEK cache. What is the actual policy we should follow, and how should we describe DEK caching and KMS outage behavior in customer-facing rotation materials?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0342"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0343",
      "label": "What caused the savings $/day estimates in Optimization Suggestions beta to not match Arkadia Finance billing, and what UI or API changes should we make to prevent this confusion going forward?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0343"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0345",
      "label": "For the Trident Financial Private VPC deal, what is our approved stance on US-only data residency and 7-year audit log retention/export, and which evidence pack artifacts should we share (and how) to back it up?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0345"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0346",
      "label": "After the Northwind latency spike, what metric wording ended up approved for the public case study and what exactly can Sales share privately without triggering new legal approval?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0346"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0347",
      "label": "For an air-gapped on-prem restore into a brand new cluster, what do we need to install before running restore apply, and what will the installer check and report if those prerequisites are missing?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0347"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0349",
      "label": "What caused the EU-West activation funnel and onboarding email issues in late January, and which code/config changes fixed them?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0349"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0350",
      "label": "Why did Northpeak Search see their Dedicated Burst Usage billed higher than expected, and how do we prevent retry attempts from double-counting burst GPU-seconds now?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0350"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0354",
      "label": "How should we explain and remediate a case where a customers invoice token total is higher than their Usage API export because streaming retries were double-counted, and what approvals are required before issuing a credit or changing the ledger?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0354"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0355",
      "label": "What is the approved wording to use when a prospect demands EU-only data residency and no non-EU support access for the Hosted API, and who do we need to escalate to if they want that as a contractual commitment?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0355"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0356",
      "label": "For the Acme AI dedicated failover incident with intermittent 500s and fallback churn, what v1 incident type and primary owner should we set, and what time-to-mitigate and time-to-fix targets should CS use in customer updates?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0356"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0360",
      "label": "Customers behind strict corporate proxies are seeing 4xx failures when the quickstart templates add telemetry tags. What tagging approach and opt-out setting should we use, and how will that affect TTFS and activation KPI measurement?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0360"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0362",
      "label": "What is our v1 standard for streaming timeout enforcement and retry behavior, and do any of the Python, TypeScript, or Go SDKs still violate it based on the parity matrix and recent support tickets?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0362"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0363",
      "label": "Northbridge Bank reported that request log TTL was not being enforced in us-east and that audit log exports were missing key admin event types; what caused these issues and what customer-facing compliance pack artifacts should we provide to show both retention enforcement and audit log evidence are now correct?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0363"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0365",
      "label": "Fast tier p99 is regressing in canary, what SLO targets and abort criteria apply and what rollback steps should we take? Which dashboards and reason codes should we use to confirm whether admission control or requeue behavior is causing it.",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0365"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0367",
      "label": "For an enterprise ticket like SUP-1842 where streaming disconnects happen in us-east, when should we escalate from a Support bridge to a formal incident and post a status page update, and what customer update cadence and credits wording should we use?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0367"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0369",
      "label": "Why were some Hosted API overload errors missing Retry-After, and what changes and alerts ensure it stays consistent across gateway, routing, and streaming paths going forward?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0369"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0375",
      "label": "When a quantization profile is blocked due to a quality regression, how is that enforced differently from an incompatibility fallback, and what should oncall do to verify a customer route is back on baseline using logs, dashboards, and audit history?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0375"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0376",
      "label": "Northbridge Bank is seeing request logs older than their configured retention window in us-east after a maintenance window. What is the recommended remediation and what customer-safe evidence artifacts should we provide to prove the retention policy is configured correctly and enforcement has caught up?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0376"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0377",
      "label": "If the demo tenant starts returning 429s mid-demo and the Console dashboards are empty, what are the fastest recovery steps to keep the call moving and what fixes were put in place to prevent this from happening again?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0377"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0381",
      "label": "In the planned controlled failover game day on 2026-01-15 (Hosted API us-east → eu-west), what caused the routing automation to oscillate (flip/rollback), and what follow-up ticket and target ship date were created to fix the streaming disconnect/reconnect behavior clients saw during cross-region redirects?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0381"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0386",
      "label": "In the Feb 2026 SEV-2 incident where the Dedicated autoscaler in us-west stopped scaling above minimum because its metrics query returned no data, what was the underlying cause and what hotfix restored scale-out?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0386"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0389",
      "label": "After the Console v4.3.2 rollout (March 2026), what caused saved dashboard filter presets to revert to the default right after you edit and save a linked sub-report in the widget editor modal, and what hotfix fixed it?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0389"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0390",
      "label": "During a March 2026 prod us-east-1 dedicated incident where a sudden burst of high-concurrency streaming traffic caused intermittent 502/504s due to proxy ephemeral port pressure, what mitigations were applied (replica changes and any kernel/sysctl tuning) and-per the ingress proxy connection-entropy playbook-what RCA artifacts should be captured while the issue is ongoing?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0390"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0391",
      "label": "In the hosted API incident in late April 2026 where streaming requests started returning 502/504 and non-streaming requests also saw a sharp tail-latency increase, what was the underlying trigger (involving an edge keepalive/connection-TTL change) and what immediate mitigations were used to restore service?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0391"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0392",
      "label": "During the us-east canary that normalized/rotated the session cookie domain for cross-subdomain routing, long-poll streaming clients started losing their pinned model version and getting routed to a fallback model. What was the root cause in the edge/session-manager flow, and what immediate mitigations were applied (platform-side and customer-side)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0392"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0393",
      "label": "After the March 9, 2026 embedding-catalog promotion that made a 1024-d embedding model the default, a dedicated Hosted API tenant (us-west with eu-west failover) saw /v1/rerank start timing out with 512-vs-1024 vector length mismatches when using the alias open-embed-default. What was the confirmed root cause, and what short-term mitigation and permanent fix did we deploy?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0393"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0394",
      "label": "For Incident Bot v2, when responders use break-glass during a Sev0/Sev1 incident to view tenant-scoped config diffs or run privileged telemetry queries, which audit events must be emitted and how long are those audit events retained (SIEM vs immutable archive)? Also, what baseline vs break-glass permission scopes does the bot use for these reads?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0394"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0395",
      "label": "What was the root cause and planned server-side fix for the March 2026 incident where function-call webhook callbacks were retried after a timeout and the JSON payload stayed the same, but the HMAC signature header changed between the first delivery and the retry due to gzip/encoding differences?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0395"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0396",
      "label": "After rotating a Hosted API key in the Console, why might the Usage → Ad-hoc Query still show request IDs tied to the old key and 'View trace' links that 404, and what immediate mitigation did Support/SRE apply before a proper fix shipped?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0396"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0399",
      "label": "In the mid-March 2026 eu-west canary rollout where sticky fallback kept pinning requests back onto the failing canary instead of the stable quantized variant, what configuration mistake caused the loop and what mitigations were applied to stop it?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0399"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0402",
      "label": "In a Redwood Private (VPC/on-prem) upgrade using the private installer, the UI has been stuck on \"Running migrations\" for over an hour and the migration pod logs repeatedly say it's waiting on a Postgres advisory lock (the Postgres database is externally managed, not the in-cluster StatefulSet). What's the underlying cause, what Support-guided action resolved it, and what specific logs/artifacts should be collected before attempting a retry or rollback?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0402"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0404",
      "label": "We saw quality regressions after enabling an INT8/mixed-precision variant and suspect KV-cache rounding on long-context requests. What request header can we set as a quick mitigation to bypass the KV cache, and in the dynamic quant handoff policy for the interactive-chat route what minimum KV-cache hit-rate and max sequence length are required before routing to a quant variant?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0404"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0407",
      "label": "In March 2026, what caused duplicate tool/webhook executions during *streaming* chat retries only when the idempotency key contained non-ASCII characters (e.g., emoji/diacritics) and the request went over HTTP/2 through the edge proxy-and what were the recommended workaround and the server-side fix (including what normalization we applied)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0407"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0410",
      "label": "In the EU hosted API, streaming generate requests over HTTP/2 intermittently returned 502 only when the client included a ~12KB base64-encoded user-metadata header. What was the root cause in our proxy chain, and what server-side mitigation did we deploy to stop the 502s (include the header and cutoff/behavior)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0410"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0411",
      "label": "On Streamly AI's dedicated pool dp-132-usw, what % of interactive burst credits should be reserved exclusively for priority=high routes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0411"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0412",
      "label": "For extraction, what price-normalization correctness is reported at temp=0.0 with strict JSON + canonicalization, and what earlier % is it compared to?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0412"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0414",
      "label": "INC-9821: was the degraded GPU node an OOM or intermittent driver/kernel launch stalls?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.5,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0414"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0417",
      "label": "What does Redwood Private on-prem recommend for signing backup manifests?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0417"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0419",
      "label": "What is the API endpoint to start a Unified Capacity Transition migration?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0419"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0420",
      "label": "For EXP-002 (eu-west→us-east), what egress cost rate and measurement basis does the cost penalty catalog use?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0420"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0422",
      "label": "How do audit log exports access records older than the retention cutoff (super-admin override vs legal hold)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0422"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0423",
      "label": "For the office keycard audit, how many months of access logs should we export?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0423"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0426",
      "label": "What are the automated rollback trigger thresholds (safety sentinel, composite quality, business KPI) in the Optimize cohort-driven gates PRD?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0426"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0428",
      "label": "What monthly token volume discount breakpoints apply for Hosted pricing in the enterprise playbook?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0428"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0434",
      "label": "In Q4 2025, which customer industry segment was featured in the most published customer stories (case studies + one-page success stories)?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0434"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0435",
      "label": "How many weekly status updates for Project Orion marked the project as 'At Risk'?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0435"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0436",
      "label": "Across Redwood's Go, Python, and TypeScript SDKs, which SDK has the highest number of customer-reported auth-related bug reports, and what are the corresponding support ticket IDs?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.3,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0436"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0438",
      "label": "Which Redwood Inference engineering team owned the most production incidents involving GPU capacity or quota exhaustion in H1 2025?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.8888888888888888,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0438"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0439",
      "label": "How many internal incident writeups at Redwood Inference describe activating automatic fallback (to a different model or region) as part of mitigation?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 0.6,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0439"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0442",
      "label": "What is Redwood Inference's end-to-end process for making a production change (definition, required approvals, pre-deploy verification, deploy execution, customer/internal communications, and rollback/post-change requirements) for the Hosted API and Console?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0442"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0443",
      "label": "List every internal communication thread (email, Slack, and meeting notes) about the Redwood Private upgrade 'rollback loop' bug (including references to RRB-17 or 'stuck rollback').",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0443"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0445",
      "label": "\"What is the complete required content of a production rollout plan for introducing a new LLM model version on Redwood Inference (hosted, dedicated, and private), including required metrics/thresholds, approvals, rollback/fallback rules, and customer-facing communications?\"",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0445"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0447",
      "label": "What is Redwood Inference's complete end-to-end process for shipping an emergency serving-runtime hotfix to production (Hosted + Dedicated), including required approvals, checklists, rollout/rollback steps, and mandatory customer communications?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0447"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0449",
      "label": "In March 2026, which deployment offering-Hosted API, Dedicated, or Private-had the most customer support escalations where the primary symptom was request timeouts?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0449"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0451",
      "label": "In the team chat about posting jokes when an AI makes things up in production, what tagging requirement did someone propose for the memes?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0451"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0452",
      "label": "Who owns the messy go-to-market misc assets dump with placeholder marketing files and micro-copy ideas created during a last-minute scramble before a partner call?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0452"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0455",
      "label": "Are sneakers acceptable for the first few weeks of the company summer softball league, or do we need cleats right away?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0455"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0460",
      "label": "In a first round technical interview for a senior sales engineer candidate, what were the agreed next steps after the call?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0460"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0461",
      "label": "What is the proposed weekly cleanup schedule and routine for tidying the engineering kitchen fridge so old opened items get discarded?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0461"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0463",
      "label": "What time are people planning to start the office foosball tournament lunch bracket on Friday?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0463"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0464",
      "label": "For the casual soccer game being organized on Friday after work, what was the agreed maximum number of players?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0464"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0465",
      "label": "When are signups due for the informal company softball team being organized for summer?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0465"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0467",
      "label": "Is there an internal hackathon demo project that uses a small web UI and a fake backend to visualize model serving behavior like batching and cache hit rates, and can be run locally or deployed for a quick demo?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0467"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0470",
      "label": "In a backend inference engineer interview about multi-tenant GPU scheduling, what strategy did the candidate propose to reduce cold misses for growing attention cache in long chat histories?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "measured",
          "value": 1,
          "unit": "fraction",
          "n": 1,
          "expected": 1,
          "attempted": 1,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": [
            "qst_0470"
          ]
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0472",
      "label": "What is the company's thesis about where competitive advantage shifts as LLM ecosystems expand?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0476",
      "label": "What is the stated differentiation around reliability and graceful degradation in Redwood's platform?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0477",
      "label": "What are the four main revenue streams in Redwood Inference's business model?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0478",
      "label": "What add-on categories are mentioned for Redwood's commercial offering?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0479",
      "label": "What triggers Redwood's expansion from PLG-led adoption to sales-assisted enterprise for Dedicated and Private?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0484",
      "label": "What is the complete mapping of `billing_tag` → Finance GL code (and cost center owner) used for chargebacks in the Eval Budget Ledger, and where is that mapping source-of-truth maintained?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0486",
      "label": "How do we compute and invoice the **per-request CO2e** for Hosted API usage (including the region-specific carbon intensity factors and any cache-hit adjustments), and where is that CO2e exposed in the customer Usage API/export schema?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0487",
      "label": "When a streaming resume attempt fails with HTTP 409 (checkpoint mismatch), what is the full JSON schema of the server's error payload (including any `resume_details` fields), and how is the `kv_crc` value inside v2 resume tokens computed?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0488",
      "label": "When we use the GPU firmware operator to do a BIOS/firmware rollout, what BIOS settings are included in Redwood's \"safe mode boot\" profile for A100/H100 nodes (e.g., Secure Boot, Above 4G decoding, Resizable BAR), and what are the default values for each?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0492",
      "label": "For the \"microburst insurance\" negotiated SLA bundles, what's the default burst surcharge charged when burst capacity units are consumed, and which GL account should that surcharge be posted to?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0495",
      "label": "For the Remote-first Pilot, what are the exact preferred vendors and model SKUs for the standing desk converter loaners, and which of those vendors can ship to Canada with delivery in under 48 hours?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0496",
      "label": "Does Incident Bot v2 support sending its incident context cards (deploy/config diff/saturation + pinned escalation context) to Microsoft Teams via an incoming webhook/Adaptive Cards, and if so what's the required payload schema and idempotency/retry behavior for updates?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0497",
      "label": "When the Security Risk Committee reviews a high-risk control exception (e.g., risk rating above 7), what quorum is required for the vote to be valid, and how are tie votes resolved?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0499",
      "label": "For the `profiling.signals` Kafka topic used by trace-orch and fidelity-proxy, what are the current production settings (partition count and message retention time), and where can I find the approval/change record for the most recent update?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    },
    {
      "id": "qst_0500",
      "label": "When a high-risk production change requires Change Advisory Board (CAB) approval, what quorum is required and what's the tie-break process if the CAB can't reach consensus-and who has final decision authority?",
      "kind": "question",
      "metric": "recall_at_50",
      "results": {
        "fyrr": {
          "status": "n/e",
          "value": null,
          "unit": "fraction",
          "n": 0,
          "expected": 0,
          "attempted": 0,
          "missing": 0,
          "ci95": null,
          "artifact": "",
          "sample_ids": []
        }
      },
      "variant": "default"
    }
  ],
  "publication_ready_primary": false,
  "draft": false
}
