{
 "generated_at": "2026-11-03T15:37:44+00:00",
 "build": "final",
 "pending": [],
 "warnings": [],
 "sources": [
  {
   "source": "repository: config.json",
   "sha256": "26357b8d133591f224f3db49e7ce69e411a823a3283a9c83655ab44597441ebf"
  },
  {
   "source": "repository: data/selected_tasks.json",
   "sha256": "6a81c8df0fab30dc5e60daa71ff763542ef11413556cda74de125c6c9419d495"
  },
  {
   "source": "repository: docs/status.md",
   "sha256": "9e38c71d83fca6138b7c490980f6e86d96505ca101be9c1e2a7a34e700479489"
  },
  {
   "source": "repository: preregistration.md",
   "sha256": "69aa23fe5d13f16259963c8ca99933ed6701e2e851bdf3f0ea8eb023c29fee68"
  },
  {
   "source": "repository: analysis/codex_failure_review/diagnosis.csv",
   "sha256": "4cb8e9ba04468ef9181a7106f4cce1898a32bfcf9cd40fe01ce372994e142939"
  },
  {
   "source": "repository: analysis/human_taxonomy.json",
   "sha256": "59dfc13a6fbb44a35aadc5dd30d838957db621b5f75804412916cae90d7ab49b"
  },
  {
   "source": "repository: analysis/requirement_source.json",
   "sha256": "db627c8c8be8c66ab6e2aed7972ef486b2457c1818bcc52d2876b94560ac9d52"
  },
  {
   "source": "repository: docs/expert_gold_fail_audit.md",
   "sha256": "696bca5da0f0646213ccc572feed8218bb649c1ffa65d18526d60d020a61b396"
  },
  {
   "source": "repository: site/img/crops.json (site/render_crops.py)",
   "sha256": "fc1e83de8615e5382f996fc4e02822e80e4f9ebcff0c2f115911dcd35bb23f9f"
  },
  {
   "source": "formal results summary, authoritative grading pass per run (local run store, not in repo); see report §7.2",
   "sha256": "aaf16feaeb4b4c809debcbdab979063d7806ef961c35d65f640f545c0139c045"
  },
  {
   "source": "human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "sha256": "02182bf8b7f0b645fb7962d14004f430439d58c21831048b0372a64f84f5a3e6"
  },
  {
   "source": "author adjudication worksheet (local run store, not in repo); see report §7.7",
   "sha256": "52ab3b31e55db51e08a8ecc219d93ed2a7e3842844c3943364303b434db7410d"
  },
  {
   "source": "failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7",
   "sha256": "a0a7fdaf157e0ec31e538c3562acd31fa44b926fc6b8ef714a81667f06292162"
  },
  {
   "source": "formal blind-label reveal (local run store, not in repo); see report §7.6",
   "sha256": "3a73176d32a9ac399f4b3d0e721929f89244c17b9e02824b7fdc6101b7e6797a"
  },
  {
   "source": "citation check of the AI draft (local run store, not in repo); see report §7.7",
   "sha256": "2a6cfe2c6789f25809ddc4781057871e8876b38c8d256b61c8b30c1d3510df1b"
  },
  {
   "source": "formal judge pass registry (local run store, not in repo); see report §1",
   "sha256": "307006217cd57e5c2fd16000269396c2998d54361154dcc32409df8a1a2a01be"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261102T133050Z",
   "sha256": "b95df770cddc74475db103b6028619291b567eb10551a4da56656f5efad666f7"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261102T185621Z",
   "sha256": "d32737c951dd25aa4e162f1828e040b6cda5f86e90cf8fa7bcd475932262470f"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261102T220501Z",
   "sha256": "c8e438fdac1363dbb6aac3221b360de6e7ee9e90704bc51ff7fc36a3ca26fcf1"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261102T224529Z",
   "sha256": "dc2b08d4ad7c41d5ad0a733410f47a08b8d2514c7565b942d64e98c4ae912886"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261102T232302Z",
   "sha256": "ebf590dea899078c48d670d89b81f330c58bed93087a4b4ddbc4726655f3127e"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261102T232303Z",
   "sha256": "b80e8b960504da9cc1665069c9db4e676aa6bc8873f4aaf3764d61ea79dac76f"
  },
  {
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · batch adjudication_20261103T023411Z",
   "sha256": "8f8cca6f861e51fc3e9fd9616213f08fff6259f01d233362fe8a20eadd9fea15"
  },
  {
   "source": "human visual-layer records (local run store, not in repo); see report §7.3 · batch visual_20261102T133023Z",
   "sha256": "1dd063aed4ad9c3c373f2d7e9a6090dc9bccd52096d9e2cf2db8dd5002d7af43"
  },
  {
   "source": "repository: site/img/opus_A11-C15.png",
   "sha256": "50a4bfb64798b5ed76845f3bf781f0956a52cf52ebe6f4ae29016448f1640d4e"
  },
  {
   "source": "repository: site/img/opus_A354-C356.png",
   "sha256": "a6c8f2d83abcc02c5d01281530eebc50eef9c52e6da979f0564eab955aa73d2c"
  },
  {
   "source": "repository: site/img/gpt_A13-C16.png",
   "sha256": "79ee66032edda4b4f1df80383f87a277d0486a173c6e59a4144ffc3fc1e4d77d"
  },
  {
   "source": "repository: site/img/gpt_A317-C319.png",
   "sha256": "720c461970b6381a0da9a8d81bf53ec69193bee55932ecc7cd6d56bbdb5e4cb7"
  },
  {
   "source": "repository: site/img/b_opus_A19-F35.png",
   "sha256": "304dcba73a7ad016042110cc385febc2e877f32c1019ee3c6176255ffcf55a5f"
  },
  {
   "source": "repository: site/img/b_gpt_A18-E32.png",
   "sha256": "9a9a70661334016645e28365097c8a3ebce81639d9ed61d41a6bcae9581f0969"
  },
  {
   "source": "repository: site/img/c_opus_summary.png",
   "sha256": "af5cc23eaa971682b65493a55d3df6c8b560436df9e01d239eb3c56fb1c48428"
  },
  {
   "source": "repository: site/img/c_gpt_summary.png",
   "sha256": "571e1be90d87f5697718aee23c0c4bcbc29a4d4c9cda28285f6d4ecc4591cf6b"
  }
 ],
 "metrics": {
  "tasks": {
   "value": 10,
   "source": "repository: data/selected_tasks.json",
   "verification": "fixed before generation"
  },
  "task_groups": {
   "value": {
    "order_clerks": 5,
    "accounting_and_it": 5,
    "occupations": 3
   },
   "source": "repository: data/selected_tasks.json (group, occupation)",
   "verification": "fixed before generation"
  },
  "models": {
   "value": 2,
   "source": "formal results summary, authoritative grading pass per run (local run store, not in repo); see report §7.2",
   "verification": "fixed before generation"
  },
  "runs_total": {
   "value": 20,
   "source": "formal results summary, authoritative grading pass per run (local run store, not in repo); see report §7.2",
   "verification": "one run per task and model"
  },
  "cost_formal_total_usd": {
   "value": "30.058664",
   "source": "paid-workload ledgers (local run store, not in repo); see report §8.1 · the four formal ledgers",
   "verification": "ledger settlement (harness accounting from provider-reported usage, not an invoice)"
  },
  "cost_formal_generation_per_model_usd": {
   "value": {
    "opus": "14.900901",
    "gpt": "3.078310"
   },
   "source": "paid-workload ledgers (local run store, not in repo); see report §8.1 · settlement entries of the three formal generation ledgers (shared units 1-2, Opus worker, GPT worker), by the unit's model",
   "verification": "ledger settlements; the per-model sums equal the three ledgers' settled totals; generation only, judge excluded"
  },
  "cost_all_workloads_usd": {
   "value": {
    "total": "88.329792",
    "per_ledger": {
     "Dev generation": "22.255219",
     "Validation generation": "17.488478",
     "Compatibility": "2.043342",
     "Judge Dev": "11.636271",
     "Judge Validation": "4.847818",
     "Formal generation, shared ledger (units 1-2)": "1.795593",
     "Formal generation, Opus worker": "13.407464",
     "Formal generation, GPT worker": "2.776154",
     "Formal judge": "12.079453"
    }
   },
   "source": "paid-workload ledgers (local run store, not in repo); see report §8.1 · harness ledger_admin status on every paid-workload ledger",
   "verification": "ledger settlements after manual resolutions, inferred portions included; equals report §8.1 recorded subtotal"
  },
  "adjudication_layer": {
   "value": {
    "items": 183,
    "records": 197,
    "by_class": {
     "grader_false_fail": 70,
     "grader_false_pass": 1,
     "verified_pass": 21,
     "confirmed_fail": 33,
     "verified_fail": 6,
     "leave": 42,
     "confirmed_pass": 10
    },
    "last_recorded_at": "2026-11-03T02:34:11+00:00"
   },
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 · latest record per item key",
   "verification": "human decisions; 'leave' marks an unresolved item-level coverage gap, not a verdict; later records supersede earlier ones"
  },
  "worksheet_breakdown": {
   "value": {
    "rows": 170,
    "by_class": {
     "verified_pass": 21,
     "confirmed_fail": 33,
     "leave": 42,
     "grader_false_fail": 67,
     "verified_fail": 6,
     "grader_false_pass": 1
    },
    "confirmed_failing": 40
   },
   "source": "author adjudication worksheet (local run store, not in repo); see report §7.7 joined with the latest adjudication records",
   "verification": "human decisions on the reviewed items (kept separate from the adjudication layer); confirmed failure = confirmed_fail + verified_fail + grader_false_pass"
  },
  "leave_comparison": {
   "value": {
    "leave_items": 42,
    "with_confirmed_counterpart_on_other_candidate": [],
    "by_criterion": {
     "opus|7d7fc9a7|C2": 6,
     "opus|7d7fc9a7|C3": 8,
     "gpt|2ea2e5b5|C3": 28
    },
    "all_in_criteria_failing_on_a_confirmed_item": true
   },
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 joined with human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "computed"
  },
  "review_scope": {
   "value": {
    "reviewed": 170,
    "critical_members": 145,
    "draft_suspected_model_failures": 24,
    "false_pass_from_blind_audit": 1,
    "reviewed_nonpass": 169,
    "draft_class_of_verified_pass_and_unresolved": {
     "verified_pass": {
      "grader_read_failure": 21
     },
     "leave": {
      "grader_read_failure": 42
     }
    },
    "opus_7d7fc9a7_C1_verified_pass_items": [
     13,
     14,
     33,
     34
    ],
    "not_reviewed": {
     "total": 96,
     "critical": 0,
     "fail": 60,
     "fail_draft_classes": {
      "rubric_ambiguous": 31,
      "grader_false_fail": 29
     },
     "cannot_assess": 10,
     "pending_visual": 24,
     "invalid_judge_output": 2
    }
   },
   "source": "author adjudication worksheet (local run store, not in repo); see report §7.7; failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7; human adjudication records (local run store, not in repo); see report §7.3 and §7.7",
   "verification": "computed from the reviewed item keys, the graded verdicts and the AI-draft classes"
  },
  "runs": {
   "value": [
    {
     "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
     "task": "Weekly sales analysis",
     "occupation": "Order Clerks",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "Mae4tizWVH2BiinB5zKQV2",
     "pass": "pass-20261102T115725665994Z-0e9c",
     "P": 58,
     "E": 60,
     "W": 65,
     "P_over_E": 0.9666666666666667,
     "E_over_W": 0.9230769230769231,
     "critical": {
      "pure_automated": "pass",
      "as_graded": "pass",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
     "task": "Weekly sales analysis",
     "occupation": "Order Clerks",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "d5sypgfTdCCwfsFV9jdzLd",
     "pass": "pass-20261102T115729995774Z-4fc2",
     "P": 42,
     "E": 58,
     "W": 65,
     "P_over_E": 0.7241379310344828,
     "E_over_W": 0.8923076923076924,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "f841ddcf-2a28-4f6d-bac3-61b607219d3e",
     "task": "June shipment recap",
     "occupation": "Order Clerks",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "hdDnxUR6sUhQD6PHLnhKSm",
     "pass": "pass-20261102T115733218914Z-a5e4",
     "P": 82,
     "E": 83,
     "W": 88,
     "P_over_E": 0.9879518072289156,
     "E_over_W": 0.9431818181818182,
     "critical": {
      "pure_automated": "pass",
      "as_graded": "pass",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "f841ddcf-2a28-4f6d-bac3-61b607219d3e",
     "task": "June shipment recap",
     "occupation": "Order Clerks",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "P8gA2U2jR7VzNWKqqdJ5SZ",
     "pass": "pass-20261102T115847754554Z-0d8f",
     "P": 61,
     "E": 81,
     "W": 88,
     "P_over_E": 0.7530864197530864,
     "E_over_W": 0.9204545454545454,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "47ef842d-8eac-4b90-bda8-dd934c228c96",
     "task": "Weeks of supply",
     "occupation": "Order Clerks",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "LVyjBBcU4iFqDKviwY2YJz",
     "pass": "pass-20261102T120000409606Z-f47b",
     "P": 82,
     "E": 92,
     "W": 99,
     "P_over_E": 0.8913043478260869,
     "E_over_W": 0.9292929292929293,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "47ef842d-8eac-4b90-bda8-dd934c228c96",
     "task": "Weeks of supply",
     "occupation": "Order Clerks",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "dPR5yvN9wEfcMyQiAC2njB",
     "pass": "pass-20261102T120015898362Z-9b7d",
     "P": 76,
     "E": 92,
     "W": 99,
     "P_over_E": 0.8260869565217391,
     "E_over_W": 0.9292929292929293,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "1137e2bb-bdf9-4876-b572-f29b7de5e595",
     "task": "PO entry audit",
     "occupation": "Order Clerks",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "iS5pLfPvTxhxxXScg7HXVw",
     "pass": "pass-20261102T120033449260Z-cd91",
     "P": 60,
     "E": 79,
     "W": 80,
     "P_over_E": 0.759493670886076,
     "E_over_W": 0.9875,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "1137e2bb-bdf9-4876-b572-f29b7de5e595",
     "task": "PO entry audit",
     "occupation": "Order Clerks",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "mE6xrcKcirXrQgh4ugBfbE",
     "pass": "pass-20261102T120138543561Z-511d",
     "P": 75,
     "E": 77,
     "W": 80,
     "P_over_E": 0.974025974025974,
     "E_over_W": 0.9625,
     "critical": {
      "pure_automated": "not_confirmed",
      "as_graded": "not_confirmed",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "c3525d4d-2012-45df-853e-2d2a0e902991",
     "task": "Holiday floor-stand budget",
     "occupation": "Order Clerks",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "fAFHskFwVPcLzn89SuYaxv",
     "pass": "pass-20261102T120246281223Z-e733",
     "P": 52,
     "E": 75,
     "W": 86,
     "P_over_E": 0.6933333333333334,
     "E_over_W": 0.872093023255814,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "c3525d4d-2012-45df-853e-2d2a0e902991",
     "task": "Holiday floor-stand budget",
     "occupation": "Order Clerks",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "SJbeNCc6EYF5dHerim5UD6",
     "pass": "pass-20261102T120327674401Z-ee13",
     "P": 55,
     "E": 75,
     "W": 86,
     "P_over_E": 0.7333333333333333,
     "E_over_W": 0.872093023255814,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "7b08cd4d-df60-41ae-9102-8aaa49306ba2",
     "task": "Tour profit and loss",
     "occupation": "Accountants and Auditors",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "KU4bWmuqsWUUbkTXEu9Fpj",
     "pass": "pass-20261102T120408339836Z-770e",
     "P": 62,
     "E": 83,
     "W": 89,
     "P_over_E": 0.7469879518072289,
     "E_over_W": 0.9325842696629213,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "7b08cd4d-df60-41ae-9102-8aaa49306ba2",
     "task": "Tour profit and loss",
     "occupation": "Accountants and Auditors",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "hX4B29t5iE2v3rFKMtqWqz",
     "pass": "pass-20261102T120617329284Z-5da8",
     "P": 62,
     "E": 82,
     "W": 89,
     "P_over_E": 0.7560975609756098,
     "E_over_W": 0.9213483146067416,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "7d7fc9a7-21a7-4b83-906f-416dea5ad04f",
     "task": "Prepaid amortization",
     "occupation": "Accountants and Auditors",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "czgd84YmwwTZzDpVKD5vhD",
     "pass": "pass-20261102T120837825114Z-bd90",
     "P": 36,
     "E": 39,
     "W": 95,
     "P_over_E": 0.9230769230769231,
     "E_over_W": 0.4105263157894737,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "not_confirmed"
     }
    },
    {
     "task_id": "7d7fc9a7-21a7-4b83-906f-416dea5ad04f",
     "task": "Prepaid amortization",
     "occupation": "Accountants and Auditors",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "FDVJTwSvqDJFQj6JyTpkHY",
     "pass": "pass-20261102T121133296465Z-6b7c",
     "P": 66,
     "E": 84,
     "W": 95,
     "P_over_E": 0.7857142857142857,
     "E_over_W": 0.8842105263157894,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "not_confirmed"
     }
    },
    {
     "task_id": "2ea2e5b5-257f-42e6-a7dc-93763f28b19d",
     "task": "Work-time study deck",
     "occupation": "Computer and Information Systems Managers",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "SaDUDAyEP45E3gNRVZWjgp",
     "pass": "pass-20261102T121440292802Z-cc25",
     "P": 78,
     "E": 85,
     "W": 85,
     "P_over_E": 0.9176470588235294,
     "E_over_W": 1.0,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "2ea2e5b5-257f-42e6-a7dc-93763f28b19d",
     "task": "Work-time study deck",
     "occupation": "Computer and Information Systems Managers",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "ehpELnbiUfAp5dFTeDNwt9",
     "pass": "pass-20261102T121505538896Z-e39c",
     "P": 28,
     "E": 30,
     "W": 85,
     "P_over_E": 0.9333333333333333,
     "E_over_W": 0.35294117647058826,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "c357f0e2-963d-4eb7-a6fa-3078fe55b3ba",
     "task": "UAT test plan",
     "occupation": "Computer and Information Systems Managers",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "AbudpM6dPVjwu2NSnwrZ6D",
     "pass": "pass-20261102T121530287546Z-1e75",
     "P": 92,
     "E": 95,
     "W": 101,
     "P_over_E": 0.968421052631579,
     "E_over_W": 0.9405940594059405,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "c357f0e2-963d-4eb7-a6fa-3078fe55b3ba",
     "task": "UAT test plan",
     "occupation": "Computer and Information Systems Managers",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "4FAJ3AmGzujC6FZ67Krt9B",
     "pass": "pass-20261102T122438506232Z-7829",
     "P": 89,
     "E": 96,
     "W": 101,
     "P_over_E": 0.9270833333333334,
     "E_over_W": 0.9504950495049505,
     "critical": {
      "pure_automated": "fail",
      "as_graded": "fail",
      "adjudicated": "fail",
      "sensitivity": "fail"
     }
    },
    {
     "task_id": "a45bc83b-22f9-4def-8d89-9c5661b2b86f",
     "task": "GCP migration design",
     "occupation": "Computer and Information Systems Managers",
     "model": "Claude Opus 5.5",
     "cand": "opus",
     "run_uuid": "BVeKQk8ZWp6j2uxp9hsuGf",
     "pass": "pass-20261102T124516281243Z-a9a6",
     "P": 66,
     "E": 73,
     "W": 87,
     "P_over_E": 0.9041095890410958,
     "E_over_W": 0.8390804597701149,
     "critical": {
      "pure_automated": "not_confirmed",
      "as_graded": "pass",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    },
    {
     "task_id": "a45bc83b-22f9-4def-8d89-9c5661b2b86f",
     "task": "GCP migration design",
     "occupation": "Computer and Information Systems Managers",
     "model": "GPT-6.1-sol",
     "cand": "gpt",
     "run_uuid": "NHSUKRMDDbYoJoGyWnpjcN",
     "pass": "pass-20261102T124517129297Z-bcc4",
     "P": 68,
     "E": 74,
     "W": 87,
     "P_over_E": 0.918918918918919,
     "E_over_W": 0.8505747126436781,
     "critical": {
      "pure_automated": "not_confirmed",
      "as_graded": "pass",
      "adjudicated": "pass",
      "sensitivity": "pass"
     }
    }
   ],
   "source": "formal results summary, authoritative grading pass per run (local run store, not in repo); see report §7.2 + human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "item scores as graded (frozen); critical status on four bases computed with the frozen scoring on copies"
  },
  "critical_counts": {
   "value": {
    "opus|pure_automated|pass": 24,
    "opus|pure_automated|fail": 7,
    "opus|pure_automated|not_confirmed": 4,
    "opus|as_graded|pass": 25,
    "opus|as_graded|fail": 7,
    "opus|as_graded|not_confirmed": 3,
    "opus|adjudicated|pass": 31,
    "opus|adjudicated|fail": 4,
    "opus|adjudicated|not_confirmed": 0,
    "opus|sensitivity|pass": 31,
    "opus|sensitivity|fail": 2,
    "opus|sensitivity|not_confirmed": 2,
    "gpt|pure_automated|pass": 15,
    "gpt|pure_automated|fail": 14,
    "gpt|pure_automated|not_confirmed": 6,
    "gpt|as_graded|pass": 16,
    "gpt|as_graded|fail": 14,
    "gpt|as_graded|not_confirmed": 5,
    "gpt|adjudicated|pass": 26,
    "gpt|adjudicated|fail": 9,
    "gpt|adjudicated|not_confirmed": 0,
    "gpt|sensitivity|pass": 26,
    "gpt|sensitivity|fail": 7,
    "gpt|sensitivity|not_confirmed": 2
   },
   "source": "human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "automated only = frozen verdicts; as graded = + preregistered visual layer; adjudicated = + human records (final); sensitivity = adjudicated with R-2.7-21 not binding (both models)"
  },
  "critical_criteria": {
   "value": {
    "per_model": 35,
    "per_task_min": 3,
    "per_task_max": 5,
    "approved": "2026-10-30",
    "revised_one": "2026-11-01",
    "revised_definition_same_status": true
   },
   "source": "repository: preregistration.md (closing list, item 1) and grading task definitions via human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "counted from the frozen task definitions; approval date as recorded in the preregistration"
  },
  "rule_dependence": {
   "value": {
    "opus": 2,
    "gpt": 2
   },
   "source": "human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "adjudicated fails minus sensitivity fails per model (R-2.7-21)"
  },
  "judge_coverage": {
   "value": {
    "with_verdict": 330,
    "items": 332,
    "invalid_judge_output": 2
   },
   "source": "formal results summary, authoritative grading pass per run (local run store, not in repo); see report §7.2",
   "verification": "automated count"
  },
  "agreement": {
   "value": {
    "deterministic": [
     8,
     10
    ],
    "judge": [
     9,
     10
    ],
    "all": [
     17,
     20
    ]
   },
   "source": "formal blind-label reveal (local run store, not in repo); see report §7.6",
   "verification": "human blind labels vs frozen graders on the sampled formal items"
  },
  "false_pass_search": {
   "value": {
    "deterministic": [
     0,
     8
    ],
    "judge": [
     1,
     10
    ],
    "overall": [
     1,
     18
    ]
   },
   "source": "formal blind-label reveal (local run store, not in repo); see report §7.6 · packets of the formal-v4 checkpoint sample",
   "verification": "human blind labels on the sampled items originally graded pass, per layer; counts only"
  },
  "judge_validation_agreement": {
   "value": [
    3,
    3
   ],
   "source": "repository: docs/status.md, 'Judge Validation complete; judge-human agreement' (report §4.2, §4.4)",
   "verification": "human blind labels vs the frozen judge on the held-out Validation judge-only items; the debug-set item excluded"
  },
  "judge_status": {
   "value": {
    "provisional": false,
    "condition_agreement_recorded": true,
    "condition_recheck_v1_9_recorded": true,
    "recheck_outcome": "BLOCKED (B1-B3, N1), closed by the author as known limitations (stop / pause control)"
   },
   "source": "repository: preregistration.md 1.25; docs/status.md 'Judge Validation complete; judge-human agreement' and 'Codex recheck 2 of V1.9 closed as known limitations'",
   "verification": "preregistration 1.25: formal judge verdicts are provisional until both the Judge Validation agreement and the independent recheck of V1.9 exist; the label is lifted when both are recorded (site spec v3.1)"
  },
  "triage": {
   "value": {
    "reviewed": 239,
    "nonpass_total": 265,
    "pending_visual": 24,
    "invalid_judge": 2,
    "classes": {
     "model_error": 33,
     "grader_false_fail": 92,
     "rubric_ambiguous": 35,
     "model_format_deviation": 2,
     "grader_read_failure": 77
    }
   },
   "source": "repository: analysis/codex_failure_review/diagnosis.csv; failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7",
   "verification": "AI-assisted draft (OpenAI Codex static review), not human-adjudicated"
  },
  "d08_items": {
   "value": 10,
   "source": "repository: analysis/codex_failure_review/analysis.md, D08 affected keys",
   "verification": "AI-draft defect diagnosis; author-verified (report §7.6); the items are adjudicated grader false fails"
  },
  "citations": {
   "value": {
    "checked": 2341,
    "by_status": {
     "verified": 1567,
     "not_checkable": 774
    },
    "rows": 239
   },
   "source": "citation check of the AI draft (local run store, not in repo); see report §7.7",
   "verification": "mechanical checks of the draft's citations; 'verified' = the check passed, not that a value or conclusion is confirmed"
  },
  "blind_review": {
   "value": {
    "representatives": 24,
    "satisfied": 22,
    "rubric_ambiguous": 2,
    "author_verified": 5,
    "author_verified_ids": [
     "D07",
     "D08",
     "D11",
     "D18",
     "D23"
    ],
    "others": 19
   },
   "source": "defect-representative review records (local run store, not in repo); see report §7.6",
   "verification": "second-model blind review; author verification recorded per representative"
  },
  "visual_verdicts": {
   "value": 12,
   "source": "human visual-layer records (local run store, not in repo); see report §7.3",
   "verification": "human-authoritative (preregistered)"
  },
  "human_review_time": {
   "value": "about 3 hours",
   "source": "author estimate; report §8.1",
   "verification": "author-reported, not measured"
  },
  "requirement_source": {
   "value": {
    "status": "approved",
    "rule": {
     "prompt": "The prompt itself asks for the missing element (a level, measure, tab, section or statement), so a reader of the prompt alone would expect it.",
     "rubric": "The prompt leaves the element open: the rubric fixes its exact wording, form or detail, or adds it beyond the prompt. Where the deliverable covers what the prompt names in another form, the item is rubric.",
     "rule": "The item fails only under a rule this evaluation added before the runs (the pre-registered additions rule)."
    },
    "per_model": {
     "opus": {
      "prompt": 0,
      "rubric": 8,
      "rule": 2
     },
     "gpt": {
      "prompt": 12,
      "rubric": 16,
      "rule": 2
     }
    },
    "prompt_tasks": {
     "opus": 0,
     "gpt": 4
    },
    "debatable": [
     "gpt|47ef842d|28",
     "gpt|47ef842d|35",
     "gpt|47ef842d|42",
     "gpt|47ef842d|49",
     "gpt|47ef842d|56",
     "gpt|47ef842d|57",
     "gpt|7b08cd4d|16",
     "gpt|7b08cd4d|17",
     "gpt|7d7fc9a7|3",
     "gpt|7d7fc9a7|4",
     "gpt|c357f0e2|50",
     "gpt|1137e2bb|23"
    ],
    "items": {
     "gpt|47ef842d|28": "prompt",
     "gpt|47ef842d|35": "prompt",
     "gpt|47ef842d|42": "prompt",
     "gpt|47ef842d|49": "prompt",
     "gpt|47ef842d|56": "prompt",
     "gpt|47ef842d|57": "prompt",
     "gpt|7b08cd4d|16": "rubric",
     "gpt|7b08cd4d|17": "rubric",
     "gpt|b5d2e6f1|16": "prompt",
     "gpt|b5d2e6f1|25": "prompt",
     "gpt|b5d2e6f1|26": "prompt",
     "gpt|2ea2e5b5|27": "rubric",
     "opus|2ea2e5b5|27": "rubric",
     "gpt|2ea2e5b5|38": "rubric",
     "opus|2ea2e5b5|38": "rubric",
     "gpt|47ef842d|13": "rubric",
     "opus|47ef842d|13": "rubric",
     "gpt|47ef842d|17": "rubric",
     "gpt|7d7fc9a7|3": "prompt",
     "gpt|7d7fc9a7|4": "prompt",
     "gpt|c357f0e2|12": "rubric",
     "gpt|a45bc83b|60": "rubric",
     "opus|a45bc83b|64": "rubric",
     "opus|a45bc83b|68": "rubric",
     "gpt|c357f0e2|66": "rubric",
     "gpt|7d7fc9a7|31": "rubric",
     "gpt|7d7fc9a7|32": "rubric",
     "gpt|7d7fc9a7|49": "rubric",
     "gpt|7d7fc9a7|50": "rubric",
     "gpt|c357f0e2|15": "rubric",
     "opus|c357f0e2|34": "rubric",
     "gpt|c357f0e2|50": "rubric",
     "opus|1137e2bb|22": "rubric",
     "gpt|1137e2bb|23": "prompt",
     "gpt|1137e2bb|21": "rubric",
     "opus|1137e2bb|21": "rubric",
     "gpt|7d7fc9a7|21": "rule",
     "opus|7d7fc9a7|21": "rule",
     "gpt|7d7fc9a7|39": "rule",
     "opus|7d7fc9a7|39": "rule"
    }
   },
   "source": "repository: analysis/requirement_source.json; task prompts of the formal task inputs (local run store, not in repo)",
   "verification": "tagged after the runs from the task prompts (Claude's draft); status and the author's decision on each judgment call as recorded in the file; every prompt quote checked verbatim against the task prompt; display only, no verdict changes"
  },
  "categories": {
   "value": {
    "rows": [
     {
      "id": "A",
      "types": [
       "M1",
       "ADDS",
       "R03"
      ],
      "per_model": {
       "opus": {
        "items": 3,
        "tasks": 2,
        "critical": 2,
        "prompt": 0,
        "severity": "medium"
       },
       "gpt": {
        "items": 14,
        "tasks": 5,
        "critical": 7,
        "prompt": 9,
        "severity": "high"
       }
      },
      "highest_severity": "high",
      "seen_here": {
       "opus": 3,
       "gpt": 14
      },
      "inside_critical_requirements": {
       "opus": 2,
       "gpt": 7
      },
      "total_items": 17
     },
     {
      "id": "B",
      "types": [
       "M2",
       "M6"
      ],
      "per_model": {
       "opus": {
        "items": 4,
        "tasks": 3,
        "critical": 3,
        "prompt": 0,
        "severity": "low"
       },
       "gpt": {
        "items": 7,
        "tasks": 4,
        "critical": 4,
        "prompt": 2,
        "severity": "low"
       }
      },
      "highest_severity": "low",
      "seen_here": {
       "opus": 4,
       "gpt": 7
      },
      "inside_critical_requirements": {
       "opus": 3,
       "gpt": 4
      },
      "total_items": 11
     },
     {
      "id": "C",
      "types": [
       "M3",
       "M5",
       "M7"
      ],
      "per_model": {
       "opus": {
        "items": 3,
        "tasks": 2,
        "critical": 0,
        "prompt": 0,
        "severity": "medium"
       },
       "gpt": {
        "items": 5,
        "tasks": 3,
        "critical": 3,
        "prompt": 1,
        "severity": "medium"
       }
      },
      "highest_severity": "medium",
      "seen_here": {
       "opus": 3,
       "gpt": 5
      },
      "inside_critical_requirements": {
       "opus": 0,
       "gpt": 3
      },
      "total_items": 8
     },
     {
      "id": "D",
      "types": [
       "M4"
      ],
      "per_model": {
       "opus": {
        "items": 0,
        "tasks": 0,
        "critical": 0,
        "prompt": 0,
        "severity": null
       },
       "gpt": {
        "items": 4,
        "tasks": 1,
        "critical": 0,
        "prompt": 0,
        "severity": "low"
       }
      },
      "highest_severity": "low",
      "seen_here": {
       "opus": 0,
       "gpt": 4
      },
      "inside_critical_requirements": {
       "opus": 0,
       "gpt": 0
      },
      "total_items": 4
     }
    ],
    "totals": {
     "opus": {
      "items": 10,
      "critical": 5,
      "prompt": 0
     },
     "gpt": {
      "items": 30,
      "critical": 14,
      "prompt": 12
     }
    },
    "scale_max": 14,
    "types_per_category": {
     "A": 3,
     "B": 2,
     "C": 3,
     "D": 1
    }
   },
   "source": "repository: analysis/human_taxonomy.json grouped by site/content.json categories; critical membership from human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "display grouping of the existing failure types; no item changes type and no decision changes; counts are rubric items"
  },
  "taxonomy_types": {
   "value": [
    {
     "id": "M1",
     "name": "Required reporting levels or measures are omitted",
     "definition": "The workbook has the underlying inputs or higher-level totals, but omits a requested measure or aggregation level from the delivered report. Likely mechanism (inference): presentation simplification was not checked against a field-by-field and level-by-level requirement list. This is not evidence of an arithmetic propagation error.",
     "severity": "high",
     "per_model": {
      "opus": {
       "items": 0,
       "tasks": 0
      },
      "gpt": {
       "items": 11,
       "tasks": 3
      }
     },
     "critical_rubric_items": 5,
     "flagged_by": {
      "deterministic": 10,
      "judge": 1
     },
     "item_keys": [
      "gpt|47ef842d|28",
      "gpt|47ef842d|35",
      "gpt|47ef842d|42",
      "gpt|47ef842d|49",
      "gpt|47ef842d|56",
      "gpt|47ef842d|57",
      "gpt|7b08cd4d|16",
      "gpt|7b08cd4d|17",
      "gpt|b5d2e6f1|16",
      "gpt|b5d2e6f1|25",
      "gpt|b5d2e6f1|26"
     ],
     "note": "7b08cd4d items 16 / 17 follow the expert reference's country-level approach (the rubric is stricter)"
    },
    {
     "id": "M2",
     "name": "Semantic substitutes do not meet explicit naming or display requirements",
     "definition": "The intended information is often present, but exact required identifiers, field labels, sheet names or decimal presentation are replaced with a different form. Likely mechanism (inference): readability and inherited source labels were prioritized without a final literal-contract check. These are predominantly presentation mismatches, not missing transactions.",
     "severity": "low",
     "per_model": {
      "opus": {
       "items": 3,
       "tasks": 2
      },
      "gpt": {
       "items": 7,
       "tasks": 4
      }
     },
     "critical_rubric_items": 7,
     "flagged_by": {
      "deterministic": 9,
      "judge": 1
     },
     "item_keys": [
      "gpt|2ea2e5b5|27",
      "opus|2ea2e5b5|27",
      "gpt|2ea2e5b5|38",
      "opus|2ea2e5b5|38",
      "gpt|47ef842d|13",
      "opus|47ef842d|13",
      "gpt|47ef842d|17",
      "gpt|7d7fc9a7|3",
      "gpt|7d7fc9a7|4",
      "gpt|c357f0e2|12"
     ],
     "note": null
    },
    {
     "id": "M3",
     "name": "An operational control or concrete validation step is left unspecified",
     "definition": "The artifact discusses a goal or supplies functional content but omits the named operational mechanism, access decision or performance validation procedure. Likely mechanism (inference): broad design coverage was accepted without checking that each required operational step had an executable or explicit counterpart.",
     "severity": "medium",
     "per_model": {
      "opus": {
       "items": 2,
       "tasks": 1
      },
      "gpt": {
       "items": 2,
       "tasks": 2
      }
     },
     "critical_rubric_items": 0,
     "flagged_by": {
      "judge": 4
     },
     "item_keys": [
      "gpt|a45bc83b|60",
      "opus|a45bc83b|64",
      "opus|a45bc83b|68",
      "gpt|c357f0e2|66"
     ],
     "note": null
    },
    {
     "id": "M4",
     "name": "A disclosed rounding adjustment conflicts with a source-only accounting rule",
     "definition": "A synthetic fractional-cent line aligns rounded invoice calculations with the full-precision account balance, creating negative amortization and a rising balance unsupported by a source invoice adjustment. Likely mechanism (inference): the model reconciled two rounding conventions by adding an explicit precision-adjustment line, without testing that line against the rubric's no-negative/no-unfunded-increase constraints.",
     "severity": "low",
     "per_model": {
      "opus": {
       "items": 0,
       "tasks": 0
      },
      "gpt": {
       "items": 4,
       "tasks": 1
      }
     },
     "critical_rubric_items": 0,
     "flagged_by": {
      "judge": 2,
      "deterministic": 2
     },
     "item_keys": [
      "gpt|7d7fc9a7|31",
      "gpt|7d7fc9a7|32",
      "gpt|7d7fc9a7|49",
      "gpt|7d7fc9a7|50"
     ],
     "note": "the four items describe one rounding strategy (a labelled fractional-cent line on each schedule), not four independent mistakes"
    },
    {
     "id": "M5",
     "name": "A nearby test scenario omits the exact required condition or assertion",
     "definition": "A test addresses the general feature but fails to instantiate the required missing-input condition, specified fields or both-organization mutation. Likely mechanism (inference): related scenarios were treated as interchangeable during coverage review; generic references to configured fields replaced concrete test data.",
     "severity": "medium",
     "per_model": {
      "opus": {
       "items": 1,
       "tasks": 1
      },
      "gpt": {
       "items": 2,
       "tasks": 1
      }
     },
     "critical_rubric_items": 2,
     "flagged_by": {
      "judge": 3
     },
     "item_keys": [
      "gpt|c357f0e2|15",
      "opus|c357f0e2|34",
      "gpt|c357f0e2|50"
     ],
     "note": null
    },
    {
     "id": "M6",
     "name": "A grouped report lacks the requested prioritization interaction",
     "definition": "The summary is useful for drilldown but its interleaved group structure does not offer an easy sort of SKU totals by error count. Likely mechanism (inference): the author supplied a rank column and outline groups but did not validate the requested sorting workflow.",
     "severity": "low",
     "per_model": {
      "opus": {
       "items": 1,
       "tasks": 1
      },
      "gpt": {
       "items": 0,
       "tasks": 0
      }
     },
     "critical_rubric_items": 0,
     "flagged_by": {
      "deterministic": 1
     },
     "item_keys": [
      "opus|1137e2bb|22"
     ],
     "note": null
    },
    {
     "id": "M7",
     "name": "A plain-language explanation of a check omits its decisive criterion",
     "definition": "The deliverable applies a rule correctly in the workbook but its narrative explanation of the check leaves out the condition that defines it, so a reader could not re-apply the check from the explanation alone.",
     "severity": "medium",
     "per_model": {
      "opus": {
       "items": 0,
       "tasks": 0
      },
      "gpt": {
       "items": 1,
       "tasks": 1
      }
     },
     "critical_rubric_items": 1,
     "flagged_by": {
      "human": 1
     },
     "item_keys": [
      "gpt|1137e2bb|23"
     ],
     "note": null
    },
    {
     "id": "R03",
     "name": "Missing separate invalid-case-pack indicator",
     "definition": "The workbook flags case-pack violations but carries no separate indicator for a missing or invalid case pack (rubric item 21, which goes beyond the prompt; A-21). All source case packs are positive, so no count is affected.",
     "severity": "low",
     "per_model": {
      "opus": {
       "items": 1,
       "tasks": 1
      },
      "gpt": {
       "items": 1,
       "tasks": 1
      }
     },
     "critical_rubric_items": 0,
     "flagged_by": {
      "deterministic": 2
     },
     "item_keys": [
      "gpt|1137e2bb|21",
      "opus|1137e2bb|21"
     ],
     "note": null
    },
    {
     "id": "ADDS",
     "name": "Per-line additions column absent (both models; depends on R-2.7-21)",
     "definition": "Neither prepaid schedule presents a per-line monthly additions column; the roll-forwards are arithmetically valid, and the items fail under the binding pre-run rule R-2.7-21 (additions are never inferred). Mutually exclusive with M1.",
     "severity": "medium",
     "per_model": {
      "opus": {
       "items": 2,
       "tasks": 1
      },
      "gpt": {
       "items": 2,
       "tasks": 1
      }
     },
     "critical_rubric_items": 4,
     "flagged_by": {
      "deterministic": 4
     },
     "item_keys": [
      "gpt|7d7fc9a7|21",
      "opus|7d7fc9a7|21",
      "gpt|7d7fc9a7|39",
      "opus|7d7fc9a7|39"
     ],
     "note": "kept out of M1 so that M1 counts only levels and measures omitted without a rule in play; the sensitivity of R-2.7-21 is reported with the critical status"
    }
   ],
   "source": "repository: analysis/human_taxonomy.json",
   "verification": "human-confirmed items only; severity is the author's rating"
  },
  "failed_critical_criteria": {
   "value": {
    "opus": 4,
    "gpt": 9
   },
   "source": "repository: analysis/human_taxonomy.json",
   "verification": "adjudicated"
  },
  "m2_leverage": {
   "value": {
    "m2_opus_sole": 2,
    "m2_gpt_with": 3,
    "gpt_failed": 9,
    "m2_gpt_m5": 1
   },
   "source": "repository: analysis/human_taxonomy.json (failed critical criteria by failure type)",
   "verification": "human-verified"
  },
  "m1": {
   "value": {
    "gpt_items": 11,
    "gpt_tasks": 3,
    "critical": 5,
    "opus_items": 0
   },
   "source": "repository: analysis/human_taxonomy.json",
   "verification": "human-verified"
  },
  "opus_levels_verified": {
   "value": {
    "b5d2e6f1|12": "confirmed_pass",
    "b5d2e6f1|16": "confirmed_pass",
    "b5d2e6f1|25": "confirmed_pass",
    "b5d2e6f1|26": "confirmed_pass",
    "7b08cd4d|16": "grader_false_fail",
    "7b08cd4d|17": "grader_false_fail",
    "47ef842d|28": "confirmed_pass",
    "47ef842d|35": "confirmed_pass",
    "47ef842d|42": "confirmed_pass",
    "47ef842d|49": "confirmed_pass",
    "47ef842d|56": "confirmed_pass",
    "47ef842d|57": "confirmed_pass"
   },
   "source": "human adjudication records (local run store, not in repo); see report §7.3 and §7.7 (confirmed_pass / grader_false_fail with the author's evidence)",
   "verification": "author-verified; backs 'Opus 5.5 includes them'"
  },
  "cards": {
   "value": {
    "A": {
     "gpt_items": [
      16,
      25,
      26
     ],
     "gpt_store_subtotal_rows": 0,
     "opus_store_total_rows": 39,
     "stores": 39,
     "grand_total_wtd_qty": {
      "opus": "55,380",
      "gpt": "55,380"
     },
     "gpt_grand_total_item17_frozen_verdict": "pass"
    },
    "B": {
     "item": 13,
     "task": "47ef842d-8eac-4b90-bda8-dd934c228c96",
     "both_confirmed": true,
     "required_upcs": 5,
     "opus_failed_critical": 4,
     "m2_sole_cause_opus": 2
    },
    "C": {
     "item": 23,
     "task": "1137e2bb-bdf9-4876-b572-f29b7de5e595",
     "gpt_decision": "grader_false_pass",
     "frozen_verdicts": {
      "opus": [
       "judge",
       "pass"
      ],
      "gpt": [
       "judge",
       "pass"
      ]
     },
     "decides_gpt_task_alone": true,
     "document_terms": {
      "opus": {
       "document": {
        "multiple": 3,
        "divisib": 1
       },
       "rendered": {
        "multiple": 3,
        "divisib": 1
       }
      },
      "gpt": {
       "document": {
        "multiple": 0,
        "divisib": 1
       },
       "rendered": {
        "multiple": 0,
        "divisib": 1
       }
      }
     }
    },
    "source_rows": 936
   },
   "source": "repository: analysis/human_taxonomy.json; human adjudication records (local run store, not in repo); see report §7.3 and §7.7; the candidates' deliverables and the task inputs (local run store, not in repo); see report §6.5; failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7 (rubric text, frozen verdicts)",
   "verification": "card items are human-confirmed failures; store and total rows read from the deliverables; card C term counts read from the Word documents and their renderings"
  },
  "crops": {
   "value": [
    {
     "card": "A",
     "cand": "opus",
     "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
     "run_uuid": "Mae4tizWVH2BiinB5zKQV2",
     "file": "Weekly Sales Data - Beauty Sales Analysis.xlsx",
     "sha256": "dcc1c9e4827e7a5480228bdcb93896af86eb24adafad8c32dbea1227020cc353",
     "sheet": "Sales by Store",
     "range": "A11:C15",
     "header_row": 4,
     "kind": "store_boundary",
     "mark": "box_row",
     "cells": {
      "header": [
       "Store",
       "Brand Name",
       "WTD Sales Quantity"
      ],
      "rows": [
       [
        11,
        "Store 1",
        "Men's Treatment",
        "242"
       ],
       [
        12,
        "Store 1",
        "Shiseido",
        "162"
       ],
       [
        13,
        "Store 1 Total",
        "",
        "1,420"
       ],
       [
        14,
        "Store 2",
        "Clinique",
        "181"
       ],
       [
        15,
        "Store 2",
        "Elizabeth Arden",
        "119"
       ]
      ]
     },
     "chart": null,
     "text": null,
     "page": null,
     "clip_pt": null,
     "png": "opus_A11-C15.png",
     "png_px": [
      861,
      331
     ]
    },
    {
     "card": "A",
     "cand": "opus",
     "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
     "run_uuid": "Mae4tizWVH2BiinB5zKQV2",
     "file": "Weekly Sales Data - Beauty Sales Analysis.xlsx",
     "sha256": "dcc1c9e4827e7a5480228bdcb93896af86eb24adafad8c32dbea1227020cc353",
     "sheet": "Sales by Store",
     "range": "A354:C356",
     "header_row": 4,
     "kind": "grand_total",
     "mark": null,
     "cells": {
      "header": [
       "Store",
       "Brand Name",
       "WTD Sales Quantity"
      ],
      "rows": [
       [
        354,
        "Store 39",
        "Shiseido",
        "162"
       ],
       [
        355,
        "Store 39 Total",
        "",
        "1,420"
       ],
       [
        356,
        "Grand Total",
        "",
        "55,380"
       ]
      ]
     },
     "chart": null,
     "text": null,
     "page": null,
     "clip_pt": null,
     "png": "opus_A354-C356.png",
     "png_px": [
      861,
      251
     ]
    },
    {
     "card": "A",
     "cand": "gpt",
     "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
     "run_uuid": "d5sypgfTdCCwfsFV9jdzLd",
     "file": "Weekly Sales Analysis.xlsx",
     "sha256": "3e021dd2f4644b85466f750a19416a5b24f593018a9d6981cbd31777994c6305",
     "sheet": "Sales by Store",
     "range": "A13:C16",
     "header_row": 6,
     "kind": "store_boundary",
     "mark": "line_before",
     "cells": {
      "header": [
       "Store",
       "Brand Name",
       "WTD Sales Quantity"
      ],
      "rows": [
       [
        13,
        "",
        "Men's Treatment",
        "242"
       ],
       [
        14,
        "",
        "Shiseido",
        "162"
       ],
       [
        15,
        "Store 2",
        "Clinique",
        "181"
       ],
       [
        16,
        "",
        "Elizabeth Arden",
        "119"
       ]
      ]
     },
     "chart": null,
     "text": null,
     "page": null,
     "clip_pt": null,
     "png": "gpt_A13-C16.png",
     "png_px": [
      903,
      361
     ]
    },
    {
     "card": "A",
     "cand": "gpt",
     "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
     "run_uuid": "d5sypgfTdCCwfsFV9jdzLd",
     "file": "Weekly Sales Analysis.xlsx",
     "sha256": "3e021dd2f4644b85466f750a19416a5b24f593018a9d6981cbd31777994c6305",
     "sheet": "Sales by Store",
     "range": "A317:C319",
     "header_row": 6,
     "kind": "grand_total",
     "mark": null,
     "cells": {
      "header": [
       "Store",
       "Brand Name",
       "WTD Sales Quantity"
      ],
      "rows": [
       [
        317,
        "",
        "Men's Treatment",
        "242"
       ],
       [
        318,
        "",
        "Shiseido",
        "162"
       ],
       [
        319,
        "Grand Total",
        "",
        "55,380"
       ]
      ]
     },
     "chart": null,
     "text": null,
     "page": null,
     "clip_pt": null,
     "png": "gpt_A317-C319.png",
     "png_px": [
      903,
      337
     ]
    },
    {
     "card": "B",
     "cand": "opus",
     "task_id": "47ef842d-8eac-4b90-bda8-dd934c228c96",
     "run_uuid": "LVyjBBcU4iFqDKviwY2YJz",
     "file": "Top5_UPC_Inventory_Position_Wk28_2025.xlsx",
     "sha256": "b8d9f3dd3dc2f415ed34606b0d7a3bb5777aa6383d644c9da16bed5bf5b40e36",
     "sheet": "Summary",
     "range": "A19:F35",
     "header_row": null,
     "kind": "chart",
     "mark": null,
     "cells": null,
     "chart": {
      "category_ref": "'Summary'!$C$48:$C$52",
      "categories": [
       "AC AMBER NOIR EDP SPRAY",
       "HONEY BLISS KISS EDT",
       "LVNC VANILLA REVERIE EDT",
       "HONEYCREST EDP SPRAY",
       "A DURAND CINNAMON SPARK ROLLERBALL"
      ],
      "chart_title": "% of Active Stores Out of Stock – Top 5 UPCs (Wk 28 2025)",
      "upc_cells": "B48:B52",
      "upcs": [
       "875218534223",
       "375301052429",
       "567219040266",
       "901153373247",
       "217313054556"
      ]
     },
     "text": null,
     "page": null,
     "clip_pt": null,
     "png": "b_opus_A19-F35.png",
     "png_px": [
      1395,
      663
     ]
    },
    {
     "card": "B",
     "cand": "gpt",
     "task_id": "47ef842d-8eac-4b90-bda8-dd934c228c96",
     "run_uuid": "dPR5yvN9wEfcMyQiAC2njB",
     "file": "Inventory_Health_Top_5_2025_W28.xlsx",
     "sha256": "acceda689a53b986277c5d2d5d40ec008fcfe234c3903922c4d9de5edadd607e",
     "sheet": "Summary",
     "range": "A18:E32",
     "header_row": null,
     "kind": "chart",
     "mark": null,
     "cells": null,
     "chart": {
      "category_ref": "Summary!$K$11:$K$15",
      "categories": [
       "AC Amber Noir",
       "Honey Bliss Kiss",
       "LVNC Vanilla Reverie",
       "Honeycrest",
       "A Durand Cinnamon Spark"
      ],
      "chart_title": null,
      "upc_cells": "B11:B15",
      "upcs": [
       "875218534223",
       "375301052429",
       "567219040266",
       "901153373247",
       "217313054556"
      ]
     },
     "text": null,
     "page": null,
     "clip_pt": null,
     "png": "b_gpt_A18-E32.png",
     "png_px": [
      1372,
      518
     ]
    },
    {
     "card": "C",
     "cand": "opus",
     "task_id": "1137e2bb-bdf9-4876-b572-f29b7de5e595",
     "run_uuid": "iS5pLfPvTxhxxXScg7HXVw",
     "file": "PO_Audit_Error_Summary.docx",
     "sha256": "f687de7ed5ef52ab69c590be76863b0903e24f37b9d3a72ec42252aefed05487",
     "sheet": null,
     "range": null,
     "header_row": null,
     "kind": "doc_paragraph",
     "mark": null,
     "cells": null,
     "chart": null,
     "text": {
      "heading": "1. Summary",
      "paragraph": "I audited all 41 PO lines in the export (10 POs, 16 SKUs, 4 ship-to locations) against two checks: Price Mismatch (entered unit price ≠ expected unit price) and Case Pack (a CASE-UOM line whose ordered units are not a whole multiple of the case pack).",
      "document_terms": {
       "multiple": 3,
       "divisib": 1
      },
      "rendered_terms": {
       "multiple": 3,
       "divisib": 1
      }
     },
     "page": 1,
     "clip_pt": [
      62.5,
      141.4,
      540.7,
      204.4
     ],
     "png": "c_opus_summary.png",
     "png_px": [
      1288,
      181
     ]
    },
    {
     "card": "C",
     "cand": "gpt",
     "task_id": "1137e2bb-bdf9-4876-b572-f29b7de5e595",
     "run_uuid": "mE6xrcKcirXrQgh4ugBfbE",
     "file": "Wholesale_Order_Audit_Summary.docx",
     "sha256": "b798ae94af50f517ad6f9a4f2e63e04b5d1766e1e3b7e81085c69c6bc2bf0359",
     "sheet": null,
     "range": null,
     "header_row": null,
     "kind": "doc_paragraph",
     "mark": null,
     "cells": null,
     "chart": null,
     "text": {
      "heading": "Method & workbook use",
      "paragraph": "Price errors compare entered vs. expected unit prices at 2-decimal precision, assuming expected prices are authoritative. Case-pack errors count invalid CASE lines once; EA lines are exempt. Total errors sum the two flags. The SKU Summary tab expands via + to PO totals; filter RawData for individual ship-to lines. This is a 41-line snapshot, not a measure of invoice losses, short shipments or assessed fines; no order dates or shipment status were supplied.",
      "document_terms": {
       "multiple": 0,
       "divisib": 1
      },
      "rendered_terms": {
       "multiple": 0,
       "divisib": 1
      }
     },
     "page": 1,
     "clip_pt": [
      46.9,
      634.7,
      564.5,
      724.4
     ],
     "png": "c_gpt_summary.png",
     "png_px": [
      1393,
      252
     ]
    }
   ],
   "source": "repository: site/img/crops.json, written by site/render_crops.py from the candidates' deliverables and the task inputs (local run store, not in repo); see report §6.5",
   "verification": "LibreOffice in the frozen sandbox image sha256:18cde712cde1…, no network; display-only preparation of copies (card A: PivotTable relationship removed so the saved cells are shown; card C: Word documents unchanged); every shown cell token (card A), every referenced category label and no UPC (card B), or every word of the heading and paragraph (card C) checked in the rendering"
  },
  "run_configuration": {
   "value": {
    "models": {
     "opus": "anthropic/claude-opus-5-5",
     "gpt": "openai/gpt-6.1-sol"
    },
    "reasoning_effort": "high",
    "max_output_tokens": 32000,
    "max_retries": 3,
    "request_timeout_s": 2700,
    "attempt_timeout_s": 900,
    "provider_sdk_retries": 0,
    "time_limit_s": 7200,
    "working_limit_s": 7200,
    "message_limit": 200,
    "attempts": 1,
    "generation_start": "2026-11-02T09:57:12+00:00",
    "generation_end": "2026-11-02T11:04:32+00:00",
    "judge_pass_start": "2026-11-02T11:57:25+00:00",
    "judge_pass_end": "2026-11-02T12:55:28+00:00",
    "labels_saved": "2026-11-02T12:48:10+00:00",
    "freeze_generation_commit": "dda234f",
    "freeze_grading_commit": "cc684ef",
    "config_sha256": "77f765314253c2c3a2251443e952593e295652c5f947b1b2322ecab7be186e4b",
    "judge": {
     "model": "google/gemini-3.1-pro-preview",
     "temperature": 1.0,
     "max_tokens": 16384,
     "reasoning_effort": "high",
     "prompt_sha256": "1de955e607e16bd8d236469603bde099b3d309a885be170c21d7e759c57487b5"
    }
   },
   "source": "repository: config.json at freeze-v1-generation and freeze-v1-grading, harness/freeze/freeze-v1-generation.json; Inspect logs of the 20 formal runs (local run store, not in repo); see report §2.3; formal run inventories (local run store, not in repo); see report §7.1; formal judge pass registry (local run store, not in repo); see report §1; formal blind-label reveal (local run store, not in repo); see report §7.6",
   "verification": "frozen config cross-checked against the generate config and limits recorded in all 20 formal Inspect logs"
  },
  "run_limits_reached": {
   "value": {
    "runs": 20,
    "sample_limit_hits": [],
    "max_tokens_stops": 0,
    "model_requests": 409,
    "largest_output_tokens_one_request": 23598,
    "longest_run_s": 1039,
    "most_messages": 100,
    "request_retries": 0,
    "inventory_recovered_attempts": 0
   },
   "source": "Inspect logs of the 20 formal runs (local run store, not in repo); see report §2.3; formal run inventories (local run store, not in repo); see report §7.1",
   "verification": "read from every model event of the 20 formal samples; limits: 32000 output tokens per request, 7200 s per run, 200 messages per run"
  },
  "rubric_items": {
   "value": {
    "per_model": 580,
    "inside_critical_requirements": 320,
    "critical_requirements": 35,
    "critical_requirements_per_task": {
     "b5d2e6f1": 4,
     "f841ddcf": 5,
     "47ef842d": 4,
     "1137e2bb": 3,
     "c3525d4d": 3,
     "7b08cd4d": 3,
     "7d7fc9a7": 4,
     "2ea2e5b5": 3,
     "c357f0e2": 3,
     "a45bc83b": 3
    }
   },
   "source": "failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7; critical membership from human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "counted per model from the item bundle; the two models face the same rubric items"
  },
  "task_verdicts": {
   "value": {
    "rows": [
     {
      "task_id": "b5d2e6f1-62a2-433a-bcdd-95b260cdd860",
      "task": "Weekly sales analysis",
      "occupation": "Order Clerks",
      "opus": {
       "after_review": "pass",
       "before_review": "pass",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C3",
         "item": 16,
         "failure_type": "M1",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C3",
         "item": 25,
         "failure_type": "M1",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C3",
         "item": 26,
         "failure_type": "M1",
         "decision": "confirmed_fail"
        }
       ],
       "label": "missing store subtotals",
       "requirement_source": "prompt"
      }
     },
     {
      "task_id": "f841ddcf-2a28-4f6d-bac3-61b607219d3e",
      "task": "June shipment recap",
      "occupation": "Order Clerks",
      "opus": {
       "after_review": "pass",
       "before_review": "pass",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "pass",
       "before_review": "fail",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      }
     },
     {
      "task_id": "47ef842d-8eac-4b90-bda8-dd934c228c96",
      "task": "Weeks of supply",
      "occupation": "Order Clerks",
      "opus": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C4"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C4",
         "item": 13,
         "failure_type": "M2",
         "decision": "confirmed_fail"
        }
       ],
       "label": "exact label",
       "requirement_source": "rubric"
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C4"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C4",
         "item": 13,
         "failure_type": "M2",
         "decision": "confirmed_fail"
        }
       ],
       "label": "exact label",
       "requirement_source": "rubric"
      }
     },
     {
      "task_id": "1137e2bb-bdf9-4876-b572-f29b7de5e595",
      "task": "PO entry audit",
      "occupation": "Order Clerks",
      "opus": {
       "after_review": "pass",
       "before_review": "fail",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "not_confirmed",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C3",
         "item": 23,
         "failure_type": "M7",
         "decision": "grader_false_pass"
        }
       ],
       "label": "explanation omits the rule",
       "requirement_source": "prompt"
      }
     },
     {
      "task_id": "c3525d4d-2012-45df-853e-2d2a0e902991",
      "task": "Holiday floor-stand budget",
      "occupation": "Order Clerks",
      "opus": {
       "after_review": "pass",
       "before_review": "fail",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "pass",
       "before_review": "fail",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      }
     },
     {
      "task_id": "7b08cd4d-df60-41ae-9102-8aaa49306ba2",
      "task": "Tour profit and loss",
      "occupation": "Accountants and Auditors",
      "opus": {
       "after_review": "pass",
       "before_review": "fail",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C1"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C1",
         "item": 16,
         "failure_type": "M1",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C1",
         "item": 17,
         "failure_type": "M1",
         "decision": "confirmed_fail"
        }
       ],
       "label": "missing per-stop tax *",
       "requirement_source": "rubric"
      }
     },
     {
      "task_id": "7d7fc9a7-21a7-4b83-906f-416dea5ad04f",
      "task": "Prepaid amortization",
      "occupation": "Accountants and Auditors",
      "opus": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "not_confirmed",
       "failed_requirements": [
        "C2",
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C2",
         "item": 21,
         "failure_type": "ADDS",
         "decision": "verified_fail"
        },
        {
         "requirement": "C3",
         "item": 39,
         "failure_type": "ADDS",
         "decision": "verified_fail"
        }
       ],
       "label": "additions column †",
       "requirement_source": "rule"
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "not_confirmed",
       "failed_requirements": [
        "C2",
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C2",
         "item": 21,
         "failure_type": "ADDS",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C3",
         "item": 39,
         "failure_type": "ADDS",
         "decision": "confirmed_fail"
        }
       ],
       "label": "additions column †",
       "requirement_source": "rule"
      }
     },
     {
      "task_id": "2ea2e5b5-257f-42e6-a7dc-93763f28b19d",
      "task": "Work-time study deck",
      "occupation": "Computer and Information Systems Managers",
      "opus": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C3",
         "item": 27,
         "failure_type": "M2",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C3",
         "item": 38,
         "failure_type": "M2",
         "decision": "confirmed_fail"
        }
       ],
       "label": "exact label",
       "requirement_source": "rubric"
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C3",
         "item": 27,
         "failure_type": "M2",
         "decision": "verified_fail"
        },
        {
         "requirement": "C3",
         "item": 38,
         "failure_type": "M2",
         "decision": "verified_fail"
        }
       ],
       "label": "exact label",
       "requirement_source": "rubric"
      }
     },
     {
      "task_id": "c357f0e2-963d-4eb7-a6fa-3078fe55b3ba",
      "task": "UAT test plan",
      "occupation": "Computer and Information Systems Managers",
      "opus": {
       "after_review": "pass",
       "before_review": "fail",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "fail",
       "before_review": "fail",
       "without_additions_rule": "fail",
       "failed_requirements": [
        "C2",
        "C3"
       ],
       "confirmed_items_in_failed_requirements": [
        {
         "requirement": "C2",
         "item": 12,
         "failure_type": "M2",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C2",
         "item": 50,
         "failure_type": "M5",
         "decision": "confirmed_fail"
        },
        {
         "requirement": "C3",
         "item": 15,
         "failure_type": "M5",
         "decision": "confirmed_fail"
        }
       ],
       "label": "label and missing test case",
       "requirement_source": "rubric"
      }
     },
     {
      "task_id": "a45bc83b-22f9-4def-8d89-9c5661b2b86f",
      "task": "GCP migration design",
      "occupation": "Computer and Information Systems Managers",
      "opus": {
       "after_review": "pass",
       "before_review": "pass",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      },
      "gpt": {
       "after_review": "pass",
       "before_review": "pass",
       "without_additions_rule": "pass",
       "failed_requirements": [],
       "confirmed_items_in_failed_requirements": [],
       "label": null,
       "requirement_source": null
      }
     }
    ],
    "after_review": {
     "opus": {
      "pass": 7,
      "fail": 3,
      "not_confirmed": 0
     },
     "gpt": {
      "pass": 3,
      "fail": 7,
      "not_confirmed": 0
     }
    },
    "before_review": {
     "opus": {
      "pass": 3,
      "fail": 7,
      "not_confirmed": 0
     },
     "gpt": {
      "pass": 1,
      "fail": 8,
      "not_confirmed": 1
     }
    },
    "failures_only_an_exact_label": {
     "opus": 2,
     "gpt": 2
    },
    "failed_tasks_by_requirement_source": {
     "opus": {
      "prompt": 0,
      "rubric": 2,
      "rule": 1
     },
     "gpt": {
      "prompt": 2,
      "rubric": 4,
      "rule": 1
     }
    },
    "failed_tasks_decided_by_a_grader_false_pass_alone": 1
   },
   "source": "human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3; human adjudication records (local run store, not in repo); see report §7.3 and §7.7; failure types from repository: analysis/human_taxonomy.json; requirement sources from repository: analysis/requirement_source.json; labels from site/content.json",
   "verification": "a task passes when all its critical requirements pass (after review); each failing cell's label is checked against the failure types of the confirmed items in its failing requirements; a failed task's source is 'prompt' if any confirmed item in its failing requirements is stated in the prompt, else 'rubric', else 'rule'"
  },
  "requirement_source_sensitivity": {
   "value": {
    "as_tagged": {
     "prompt_items": {
      "opus": 0,
      "gpt": 12
     },
     "failed_tasks_on_prompt": {
      "opus": 0,
      "gpt": 2
     }
    },
    "judgment_calls_as_rubric": {
     "prompt_items": {
      "opus": 0,
      "gpt": 3
     },
     "failed_tasks_on_prompt": {
      "opus": 0,
      "gpt": 1
     }
    },
    "judgment_calls_as_prompt": {
     "prompt_items": {
      "opus": 0,
      "gpt": 15
     },
     "failed_tasks_on_prompt": {
      "opus": 0,
      "gpt": 4
     }
    }
   },
   "source": "derived from repository: analysis/requirement_source.json and the task verdicts",
   "verification": "the tally recomputed with every judgment call (debatable tag) read as rubric, and as prompt"
  },
  "opus_failure_split": {
   "value": {
    "rubric_detail": 5,
    "exact_wording": 3,
    "added_rule": 2
   },
   "source": "derived from repository: analysis/requirement_source.json (source) and analysis/human_taxonomy.json (M2 = exact wording)",
   "verification": "rubric-tagged items in M2 count as exact wording, the other rubric-tagged items as detail the rubric adds"
  },
  "task_verdicts_without_additions_rule": {
   "value": {
    "opus": {
     "pass": 7,
     "fail": 2,
     "not_confirmed": 1
    },
    "gpt": {
     "pass": 3,
     "fail": 6,
     "not_confirmed": 1
    }
   },
   "source": "human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "after review, with the per-line additions rubric items set to cannot assess (the pre-registered additions rule not binding)"
  },
  "weekly_sales_example": {
   "value": {
    "critical_requirements": 4,
    "names_recorded": {
     "C1": "Structure",
     "C2": "Brand totals",
     "C3": "PivotTable",
     "C4": "ST% logic"
    },
    "store_requirement": "C3",
    "store_requirement_items": 7,
    "failed_subtotal_items_gpt": [
     16,
     25,
     26
    ],
    "item_16_points": 2,
    "subtotal_items_grader": [
     "deterministic"
    ]
   },
   "source": "repository: private/rubric (the task's GDPval rubric, not published) and failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7; membership from human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "counted from the records; the page's names of the requirements and the quoted rubric items are descriptive paraphrases, the recorded names are listed here"
  },
  "expert_file_weekly_sales": {
   "value": {
    "st_pct_method": "average of row ratios (expert error in the audit)",
    "clinique_wtd_st_expert": "0.2467",
    "clinique_wtd_st_ratio_of_sums": "0.2309",
    "data_tab_cells_differing_from_source": 557,
    "data_tab_difference": "unexplained in the audit: an undocumented expert edit or another input version"
   },
   "source": "repository: docs/expert_gold_fail_audit.md (weekly sales rows); the task input (local run store, not in repo) for the ST% recomputation",
   "verification": "pre-run material audit; the Clinique WTD values recomputed from the input at build time"
  },
  "pilot_estimate": {
   "value": {
    "families": [
     "weekly sales reporting",
     "weeks of supply",
     "PO entry audit"
    ],
    "tasks_per_family": 20,
    "runs_per_task": 3,
    "accept_rise_points": 10,
    "hold_within_points": 2,
    "gdpval_hours_per_task": 7,
    "gdpval_source": "GDPval paper (arXiv:2510.04374v1), section 1: \"Tasks require an average of 7 hours of work for an expert professional to complete.\"",
    "extra_hours_per_sample": 4,
    "train_per_family": 15,
    "heldout_per_family": 5,
    "samples": 60,
    "hours_per_sample": 11,
    "hours_total": 660
   },
   "source": "repository: site/content.json (data.pilot_plan, the proposal's parameters); GDPval paper (arXiv:2510.04374v1), section 1: \"Tasks require an average of 7 hours of work for an expert professional to complete.\"",
   "verification": "estimate for a proposal, not a measured result; the expert time per task is GDPval's reported average, the extra hours are this proposal's estimate"
  },
  "charts": {
   "value": {
    "review_change": {
     "per_model": 35,
     "opus": {
      "before": {
       "pass": 25,
       "fail": 7,
       "not_confirmed": 3
      },
      "after": {
       "pass": 31,
       "fail": 4,
       "not_confirmed": 0
      },
      "after_pass_pct": 89
     },
     "gpt": {
      "before": {
       "pass": 16,
       "fail": 14,
       "not_confirmed": 5
      },
      "after": {
       "pass": 26,
       "fail": 9,
       "not_confirmed": 0
      },
      "after_pass_pct": 74
     }
    },
    "review_result": {
     "n": 170,
     "segments": [
      {
       "key": "confirmed_fail+verified_fail+grader_false_pass",
       "name": "confirmed failure",
       "count": 40,
       "cls": "seg-accent",
       "pct": 24
      },
      {
       "key": "grader_false_fail",
       "name": "grader false fail",
       "count": 67,
       "cls": "seg-g1",
       "pct": 39
      },
      {
       "key": "verified_pass",
       "name": "cannot assess, then pass",
       "count": 21,
       "cls": "seg-g2",
       "pct": 12
      },
      {
       "key": "leave",
       "name": "cannot assess, left",
       "count": 42,
       "cls": "seg-g3",
       "pct": 25
      }
     ]
    },
    "review_result_graded_verdicts": {
     "grader_false_fail|fail": 67,
     "verified_pass|cannot_assess": 21,
     "confirmed_fail|fail": 33,
     "verified_fail|cannot_assess": 6,
     "leave|cannot_assess": 42,
     "grader_false_pass|pass": 1
    }
   },
   "source": "derived at build time from critical_counts (before review = as graded, after review = adjudicated) and worksheet_breakdown",
   "verification": "before review = frozen verdicts plus the pre-registered visual review; after review = plus the author's review decisions (final)"
  },
  "review_count_flow": {
   "value": {
    "graded_fail_or_cannot_assess": 239,
    "reviewed_of_those": 169,
    "critical_members": 145,
    "draft_suspected": 24,
    "graded_pass_from_blind_audit": 1,
    "reviewed": 170,
    "reviewed_graded_fail": 100,
    "reviewed_graded_fail_critical": 78,
    "reviewed_graded_fail_draft_suspected": 22,
    "reviewed_cannot_assess": 69,
    "not_reviewed": 96,
    "not_reviewed_fail": 60,
    "not_reviewed_cannot_assess": 10,
    "no_verdict": 26
   },
   "source": "author adjudication worksheet (local run store, not in repo); see report §7.7; failure-analysis bundle: every graded item of the formal runs (local run store, not in repo); see report §7.7",
   "verification": "reconciled at build time: graded fail or cannot assess = reviewed + not reviewed; reviewed = those + the blind-audit pass"
  },
  "rule_dependence_tasks": {
   "value": {
    "opus": 1,
    "gpt": 1
   },
   "source": "human-layer status recomputed with the frozen scoring on copies (local run store, not in repo); see report §7.3",
   "verification": "failed tasks after review minus failed tasks without the additions rule, per model"
  },
  "term_check": {
   "value": {
    "forbidden": [
     "R-2.7-21",
     "as graded",
     "adjudicated",
     "unresolved",
     "re-checked",
     "criteria",
     "code check",
     "not confirmed",
     "grader couldn't read",
     "added rule",
     "requirement source",
     "prompt-stated",
     "stated in the prompt",
     "deadline exception",
     "evidence pack",
     "source-only",
     "formal"
    ],
    "hits": []
   },
   "source": "site/index.html (text, alt and aria labels; link targets excluded)",
   "verification": "computed at build time"
  }
 }
}