{
  "artifact_version": 1,
  "experiment_id": "batch-alignment-guard",
  "scope": "historical-working-tree-experiment",
  "recorded_on": "2026-09-21",
  "runner": "prod-llm",
  "snapshot": "2026-09-02",
  "k": 12,
  "fixtures": 10,
  "original": {
    "source": {
      "path": "backend/evals/results/47edb50-prod-llm-2026-09-02.json",
      "sha256": "728c62a3f532978221f30b6c1ba70d22818c7a55e749fd97fc6760f2634826be",
      "recorded_git_sha": "47edb50",
      "recorded_created_at": "2026-09-02T07:14:29+00:00"
    },
    "execution_mode": "unknown",
    "summary": {
      "recall_at_k_mean": 0.2207,
      "recall_at_k_min": 0.0,
      "raw_recall_at_k_mean": 0.2026,
      "raw_recall_at_k_min": 0.0,
      "recall_at_retrieval_mean": 0.3211,
      "recall_at_retrieval_min": 0.1176,
      "need_to_know_recall_mean": 0.1739,
      "need_to_know_recall_min": 0.0,
      "followup_recall_mean": 0.1111,
      "followup_recall_min": 0.0,
      "never_rate_mean": 0.2694,
      "never_rate_min": 0.0,
      "event_delivery_mean": 0.25,
      "event_delivery_min": 0.0,
      "major_delivery_mean": 0.22,
      "major_delivery_min": 0.0,
      "false_major_rate_mean": null,
      "false_major_rate_min": null,
      "judge_precision_mean": 0.4561,
      "judge_precision_min": 0.0,
      "judge_recall_mean": 0.35,
      "judge_recall_min": 0.0,
      "needle_recall_mean": 0.6,
      "needle_recall_min": 0.0,
      "lookalike_rate_mean": 0.15,
      "lookalike_rate_min": 0.0,
      "feed_size_k_mean": 11.7,
      "feed_size_k_min": 9,
      "distinct_sources_mean": 6.7,
      "distinct_sources_min": 5,
      "latency_s_mean": 2.581,
      "latency_s_min": 0.39,
      "calls_total": 44,
      "calls_max_per_persona": 7,
      "cost_usd_total": 0.26526,
      "cache_misses_total": 0,
      "personas": 10
    }
  },
  "guard": {
    "source": {
      "path": ".context/batch-alignment-fix/after-fix.json",
      "sha256": "268c898b8e1cd5a9367713e181ff7ed613a999d739f86b28d758eb7e7ae072fa",
      "recorded_git_sha": "3b11a3c",
      "recorded_created_at": "2026-09-21T18:18:03+00:00"
    },
    "execution_mode": "offline-replay-reported",
    "revision_is_guard_commit": false,
    "implementation_sha256": null,
    "summary": {
      "recall_at_k_mean": 0.1866,
      "recall_at_k_min": 0.0,
      "raw_recall_at_k_mean": 0.1719,
      "raw_recall_at_k_min": 0.0,
      "recall_at_retrieval_mean": 0.3211,
      "recall_at_retrieval_min": 0.1176,
      "need_to_know_recall_mean": 0.1656,
      "need_to_know_recall_min": 0.0,
      "followup_recall_mean": 0.0952,
      "followup_recall_min": 0.0,
      "never_rate_mean": 0.425,
      "never_rate_min": 0.0,
      "event_delivery_mean": 0.1,
      "event_delivery_min": 0.0,
      "major_delivery_mean": 0.16,
      "major_delivery_min": 0.0,
      "false_major_rate_mean": null,
      "false_major_rate_min": null,
      "event_slot_contamination_mean": null,
      "event_slot_contamination_min": null,
      "judge_precision_mean": null,
      "judge_precision_min": null,
      "judge_recall_mean": 0.0,
      "judge_recall_min": 0.0,
      "needle_recall_mean": 0.45,
      "needle_recall_min": 0.0,
      "lookalike_rate_mean": 0.25,
      "lookalike_rate_min": 0.0,
      "feed_size_k_mean": 11.3,
      "feed_size_k_min": 5,
      "distinct_sources_mean": 5.9,
      "distinct_sources_min": 2,
      "latency_s_mean": 4.954,
      "latency_s_min": 4.52,
      "calls_total": 84,
      "calls_max_per_persona": 9,
      "cost_usd_total": 0.30196,
      "cache_misses_total": 0,
      "personas": 10,
      "reader_pipeline_calls_max_per_persona": 9
    }
  },
  "input_provenance": {
    "equivalence": "reported-by-historical-note",
    "finding": {
      "path": ".context/batch-alignment-fix/FINDING.md",
      "sha256": "4363ac4b24ce566fd6761f3509b0b8bfca6fe8d63db1b3619140d23c39ae3266"
    },
    "reference_revision": "b95a9a63562a6c7b7c36550f1ca7f90ce5e4ceaa",
    "reference_files": [
      {
        "path": "backend/evals/snapshots/2026-09-02.json.gz",
        "sha256": "d556f5244080c1917442f189022de32306d88113ca8c86009c064ba47e590b78"
      },
      {
        "path": "backend/evals/labels/2026-09-02/aisha.jsonl",
        "sha256": "7ec7bc18087396bd01cefab0f331f79d52023335fdaad8aff272dd96ffb05edc"
      },
      {
        "path": "backend/evals/labels/2026-09-02/cache_keys.json",
        "sha256": "2a769832096bf126cf278351e7911ef7444ac51d380d4824a813b64df8c4e77d"
      },
      {
        "path": "backend/evals/labels/2026-09-02/cold.jsonl",
        "sha256": "edb61217f40812403a376eb3926dfbb8dbdd92ad68f8eccaa4e62fc27a3c7e7b"
      },
      {
        "path": "backend/evals/labels/2026-09-02/dilshod.jsonl",
        "sha256": "ab013ee58167282b7ec88881c8c1a6662b4cfea6865a677e869cfea69bd34e3f"
      },
      {
        "path": "backend/evals/labels/2026-09-02/events.json",
        "sha256": "73cf182db49e4da037ce74907b8916674314620b46799f8b277f9786c2e3a9a6"
      },
      {
        "path": "backend/evals/labels/2026-09-02/farrukh.jsonl",
        "sha256": "c849d50c38e903c9832598d2b0852121b7f9174926869d0b22aabbe4584f38ea"
      },
      {
        "path": "backend/evals/labels/2026-09-02/lena.jsonl",
        "sha256": "50dadbee8893a41fee8572f1eeba3bd710198dad1e018a6e5916663c1673c0e6"
      },
      {
        "path": "backend/evals/labels/2026-09-02/maya.jsonl",
        "sha256": "f9fece8302d3dd193b4115bbb0ffb51c2642d53c33434e1963236fb46c160759"
      },
      {
        "path": "backend/evals/labels/2026-09-02/needles.json",
        "sha256": "fb321e8cd1918c69fcc0258738c8cbb985cc9c715cb600bfe72eb9d4ddda09bd"
      },
      {
        "path": "backend/evals/labels/2026-09-02/priya.jsonl",
        "sha256": "340041f14fa0eb44ee7ebf8b296df6f0c7be283262d086d563aee0decdb5e9a4"
      },
      {
        "path": "backend/evals/labels/2026-09-02/ray.jsonl",
        "sha256": "9cf72b556ee5c4dd55383b4474cd30d8181074e5f6e7b2300d02ad7ccbcd6c53"
      },
      {
        "path": "backend/evals/labels/2026-09-02/tom.jsonl",
        "sha256": "a2a4454e949c587da1975c72f00b8717003cbeb4c853895f2e5fa545983b41a2"
      },
      {
        "path": "backend/evals/labels/2026-09-02/wei.jsonl",
        "sha256": "f694e808570d485c76cbd28d9b0363cf99adad4d3c2b4c03df53d0610a9c17f9"
      }
    ],
    "recorded_cache_keys": {
      "original_count": 89,
      "guard_count": 30,
      "original_sha256": "f1a27a7767b4bce238eef53d7007b2f3aedd6cffcfc65b51827a31cbc4e64486",
      "guard_sha256": "94ddf2688e49f30a301e3c756a330988a6ba203af5fa12762407fc324dacc4af",
      "hash_encoding": "sorted keys joined by newline, no trailing newline"
    },
    "needles": true,
    "quiet_corpus": false
  },
  "observations": {
    "guard_fires": 63,
    "largest_reported_mismatch": {
      "returned": 201,
      "sent": 40
    }
  },
  "historical_test_result": {
    "failed": 6,
    "passed": 1969,
    "skipped": 185,
    "subtests_passed": 233,
    "baseline_re_recorded": false,
    "failures": [
      {
        "metric": "never_rate_mean",
        "snapshot": "2026-08-31"
      },
      {
        "metric": "never_rate_mean",
        "snapshot": "2026-08-31-quiet"
      },
      {
        "metric": "lookalike_rate_mean",
        "snapshot": "2026-08-31"
      },
      {
        "metric": "lookalike_rate_mean",
        "snapshot": "2026-08-31-quiet"
      },
      {
        "metric": "lookalike_rate_mean",
        "snapshot": "2026-09-02"
      },
      {
        "metric": "needle_recall_mean",
        "snapshot": "2026-09-02"
      }
    ]
  },
  "notes": [
    {
      "severity": "caution",
      "message": "This is a summary of the September 21, 2026 experiment, not a current CI result, a production release, or a Lab parser execution. Its baseline was not re-recorded.",
      "source": ".context/batch-alignment-fix/FINDING.md"
    },
    {
      "severity": "warning",
      "message": "The guard scorecard records 3b11a3c, the base checkout revision. It does not identify a commit containing the tested working-tree guard. The exact guard implementation bytes were not preserved with this summary.",
      "source": ".context/batch-alignment-fix/after-fix.json:git_sha"
    },
    {
      "severity": "caution",
      "message": "The historical note reports the same snapshot, ten fixtures, k=12 and offline replay with no cache misses on both sides. The original scorecard itself does not record execution mode; zero misses alone do not prove offline replay.",
      "source": ".context/batch-alignment-fix/FINDING.md; backend/evals/results/47edb50-prod-llm-2026-09-02.json"
    },
    {
      "severity": "info",
      "message": "Source hashes identify the original note and scorecard bytes used to make this summary. Reference-file hashes identify the committed corpus and labels at reference_revision; they were not captured during the historical execution.",
      "source": "input_provenance.reference_files"
    },
    {
      "severity": "caution",
      "message": "Recorded cache-key sets differ between the original scorecard and the guard scorecard. Their hashes preserve that difference instead of asserting byte-for-byte identical execution inputs.",
      "source": "input_provenance.recorded_cache_keys"
    },
    {
      "severity": "caution",
      "message": "Labels are provisional model and agent labels. These historical quality measurements are not human-reviewed ground truth.",
      "source": "web/public/artifacts/prod-llm__2026-09-02__47edb50.json:provenance.labels"
    },
    {
      "severity": "info",
      "message": "201 verdicts for 40 articles is the largest mismatch listed in the September 21 finding. It is not a claim about the largest recorded response across all Lab cases. Guard-fire and test counts are transcribed from that finding, not recomputed from raw logs.",
      "source": ".context/batch-alignment-fix/FINDING.md"
    }
  ]
}