{
  "reviewer": "Independent AI reviewer; unblinded; not a human review",
  "reviewed_at_utc": "2026-11-16T03:55:47.747616+00:00",
  "scope": "Completed T01/T02 stages present at review snapshot. No candidate-generated code executed, no network or paid calls. Read source fixtures, actual CSVs and prose; arithmetic independently recomputed with reviewer-written code.",
  "limitations": [
    "Inherited parent context; model identities visible, so review is not blinded.",
    "Synthetic small fixtures; no population-level capability inference.",
    "This review does not prove candidate claimed tool checks occurred; those require transcript audit.",
    "Memo word-count checks use whitespace and regex counts; table/citation tokenization may vary.",
    "Task2 source packet lacks explicit discrete GPU flag and has correction/capture timestamp tension; apply intended scenario authority and do not penalize a disclosed reasonable GPU interpretation."
  ],
  "operator_findings": [
    {
      "type": "answer-key correction",
      "detail": "Task1 key corrected: RF05 affects Course and combined net/contribution/per-order values, not Workshop specifically. Unallocated RF04 still requires investigation."
    },
    {
      "type": "automated-check false negatives",
      "detail": "Opus `applied` refund column and Fable included-only CSV schema are valid. No score deduction for schema variation or data retained in exceptions."
    }
  ],
  "reviews": [
    {
      "run_id": "T01-R1-astra",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-astra/initial/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "83b9c4e548d691c98adc587d6302d797404aced42773a6e063ef0aba52662586",
        "summary.md": "5b2004843f1e23b92747ee2060d5e23bcfcb021d48822ac2fa588510004089cf",
        "cleaned-refunds.csv": "8f0f54755dc400466be7309d45eb54765ef0fdc847027d03276abb954136e3ad",
        "cleaned-sales.csv": "1f12c69a276bcc7a6407037bdc718e85a5583a80fff390f3dc365cb0bb904bdb",
        "cleaned-ads.csv": "6af3c470cc892e5acee8d04ec7344dfd8efbc1aeccbd778891c0f7c537835a71"
      }
    },
    {
      "run_id": "T01-R1-astra",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-astra/revision/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "83b9c4e548d691c98adc587d6302d797404aced42773a6e063ef0aba52662586",
        "summary.md": "385f1bd0c81a9f3add7698d9dfad7572ebf65172247fb2b7e0de488cede93624",
        "cleaned-refunds.csv": "8f0f54755dc400466be7309d45eb54765ef0fdc847027d03276abb954136e3ad",
        "cleaned-sales.csv": "1f12c69a276bcc7a6407037bdc718e85a5583a80fff390f3dc365cb0bb904bdb",
        "change-log.md": "e0455412b5092500655d086fdeb2818d5c3407fbef8dd452f43fc8c73dece5b2",
        "cleaned-ads.csv": "6af3c470cc892e5acee8d04ec7344dfd8efbc1aeccbd778891c0f7c537835a71",
        "review-checks.txt": "e8c2535ea90331a52d4711c526960bcbdb41107c584a2f0bcd49c831a81b455a"
      }
    },
    {
      "run_id": "T01-R1-sol",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-sol/initial/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "36e5cac4e32cc1dea25721a9e88669845c8a464d1e9123236ab772c51def2d98",
        "summary.md": "3e1d3837b6fc7fe665ee10cd86e0e808071ff27cb9fb1512a85087c60cd25f99",
        "cleaned-refunds.csv": "90de81c3c307df3587e06bd168065bb6763ca8a9240bc24b8268ed02d0534f88",
        "cleaned-sales.csv": "4e9bc704a2573cc9fb735535b152cae0b7c0747c04cb326f5590238c7b98fa85",
        "cleaned-ads.csv": "58ee52132426b390092de4687f6a8cd6f7d55479c1d928bd1f06698bde5bee59"
      }
    },
    {
      "run_id": "T01-R1-sol",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-sol/revision/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "36e5cac4e32cc1dea25721a9e88669845c8a464d1e9123236ab772c51def2d98",
        "summary.md": "f68aaf35812353a1b35b0be4d2fd41d72233d5b9d3e7e91b4bfd86eb80a7ee0c",
        "cleaned-refunds.csv": "90de81c3c307df3587e06bd168065bb6763ca8a9240bc24b8268ed02d0534f88",
        "cleaned-sales.csv": "4e9bc704a2573cc9fb735535b152cae0b7c0747c04cb326f5590238c7b98fa85",
        "validate.py": "82ec575663563ce07d8743292de863d56d295c3b3bc3e8a96c49f7bc786eae4b",
        "review-checks.md": "5829ab3427c633612a0028df5d6b7e98eee6733413a73049d2a7e602de5d56c1",
        "review-change-log.md": "ec749140abef0c050d421a464f6f4501cf14f2edfd185ba2fb3aeb4eeb18d26d",
        "cleaned-ads.csv": "58ee52132426b390092de4687f6a8cd6f7d55479c1d928bd1f06698bde5bee59"
      }
    },
    {
      "run_id": "T01-R1-fable",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        3
      ],
      "total": 19,
      "critical_failure": "No",
      "outcome": "minor edits",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 3,
          "evidence": "Correct no-reallocation recommendation and profit distinction, but summary §5 says policy attributes sales to ads; policy only assigns ad spend. Also states Workshop has larger refund exposure without qualifying unknown Course rate."
        }
      ],
      "issues": [
        {
          "status": "contradicted",
          "file": "summary.md",
          "passage": "Sales are also attributed to ads only at product level, as policy.md instructs",
          "correction": "Only ad spend is assigned to products; sales-to-ad attribution is not given."
        },
        {
          "status": "unsupported",
          "file": "summary.md §5",
          "passage": "Workshop carries the larger refund exposure",
          "correction": "Qualify larger known amount; relative refund rates cannot be determined with RF05 unknown."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-fable/initial/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "d469a991130c69287d9276b2e38cca88627e83c2ebb8719deb3e3980b1c914d8",
        "summary.md": "9c93825583c3eb149230838a08f12b4e0aeb778c959480432e58d0058cc1ad64",
        "cleaned-refunds.csv": "5e56267a64734948ffc95d1254e0837efd24c95f88b420d3ee142f59e0c1b4a6",
        "cleaned-sales.csv": "942b0a22f5735d31c4d3f2e27db6f51d0b971fcd35133099054eefabfd4cbca6",
        "reconcile.py": "9bfe4c01e171683c060aee9a7ad8215cb6dbce5aa0148a99624e3e132f7c1016",
        "cleaned-ads.csv": "3d1e29590f61402d78f80d771beba6cb11c4c214162d122c2d4e3817259a0bac"
      }
    },
    {
      "run_id": "T01-R1-fable",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-fable/revision/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "d469a991130c69287d9276b2e38cca88627e83c2ebb8719deb3e3980b1c914d8",
        "CHANGELOG.md": "a6fa9c6b4c90803438f4d681cb511adae6c7566aa8a072fb7dfd155e19044186",
        "summary.md": "3347aa2ce7d09f048aaf0dc39fc10c5a1c10ab15a6adf0e2c13019e30344c3f4",
        "cleaned-refunds.csv": "5e56267a64734948ffc95d1254e0837efd24c95f88b420d3ee142f59e0c1b4a6",
        "cleaned-sales.csv": "942b0a22f5735d31c4d3f2e27db6f51d0b971fcd35133099054eefabfd4cbca6",
        "verify.py": "88eee15ae5e02813052f16eb75503cb9fc3becdd389db7a54b5e25a11f1b8475",
        "reconcile.py": "ea50423bdd074ec691827a3e60baaddfc4a6d979d6dbcd3a4612f37d7221a9d8",
        "cleaned-ads.csv": "3d1e29590f61402d78f80d771beba6cb11c4c214162d122c2d4e3817259a0bac"
      }
    },
    {
      "run_id": "T01-R1-opus",
      "stage": "initial",
      "scores": [
        4,
        4,
        3,
        4,
        4
      ],
      "total": 19,
      "critical_failure": "No",
      "outcome": "minor edits",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 3,
          "evidence": "R06 is blank and results provisional; R05 is flagged but summary labels its assumption Refund on an out-of-period order, although no date/order evidence establishes this."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five files parse; `applied` is the correct inclusion column in cleaned-refunds.csv. It selects only R01/R02. An automated checker missing this column is an operator false negative, not a model failure."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [
        {
          "status": "unsupported",
          "file": "summary.md assumptions",
          "passage": "Refund on an out-of-period order",
          "correction": "Refund on an unknown order; N9999 is absent."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-opus/initial/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "2a79031371fd5fcfd93dcb88399b4d756515cffc8e3899a7ccda1c228d961e14",
        "summary.md": "d49bead6fdfdb019684c4ad7395d7c30ae8be7a6aa65e5b4ef4e0b5aa256f21c",
        "cleaned-refunds.csv": "5260b976bf5edc5e724933ea06345c0c6e4ec022decaf85080c2bcb03d212fc9",
        "cleaned-sales.csv": "8361903234adcc076a7e431b2b6a6e6facc9f456222cac611a0e59a7f43025b5",
        "reconcile.py": "d6bf4db610676540de47a9221a0ac793785573044a7c3350dca42d2253e5e951",
        "cleaned-ads.csv": "fda8b6f3dbbf4e78cce1d963bd7234ac05fc488d3ee244ecde9c2978cd8d7f49"
      }
    },
    {
      "run_id": "T01-R1-opus",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "R05 orphan and R06 missing amount are flagged; missing source amount remains blank and Course/combined net results are explicitly provisional."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five files parse; `applied` is the correct inclusion column in cleaned-refunds.csv. It selects only R01/R02. An automated checker missing this column is an operator false negative, not a model failure."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Recommendation resolves refund gaps before final close, distinguishes contribution from profit and does not claim ad spending caused revenue."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-opus/revision/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "5b2333141174fa8b521d4b29c6d3be74beadeba355a53c708bb996283f89bcf6",
        "CHANGELOG.md": "458a1bf491e2488efbb353f66030f8a1c972be942e427c015f061a9812ca9557",
        "summary.md": "c79fb2a0ad8d6022abe9226d8f8b2e3b2fc6a2609bd512cb99b14954a1f45a11",
        "cleaned-refunds.csv": "5260b976bf5edc5e724933ea06345c0c6e4ec022decaf85080c2bcb03d212fc9",
        "cleaned-sales.csv": "8361903234adcc076a7e431b2b6a6e6facc9f456222cac611a0e59a7f43025b5",
        "verify.py": "05a7c28067e2edf136c7b101900e415e77d966868aa080125dcc8e84c8dfc4f5",
        "reconcile.py": "9ebeacbc7cc5075fb280322d6c24313bf1fc60dec77d6a836ba08ff07a6d5226",
        "cleaned-ads.csv": "fda8b6f3dbbf4e78cce1d963bd7234ac05fc488d3ee244ecde9c2978cd8d7f49"
      }
    },
    {
      "run_id": "T01-R1-qwen4",
      "stage": "initial",
      "scores": [
        4,
        4,
        3,
        4,
        3
      ],
      "total": 18,
      "critical_failure": "No",
      "outcome": "minor edits",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 3,
          "evidence": "R05 and R06 are correctly flagged and unknown amount blank, but summary final limitation says all other exclusions are policy-driven and final, overstating certainty around unresolved orphan RF04."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 3,
          "evidence": "Correct contribution/profit distinction and notes equal contribution/ad dollar; nevertheless suggests prioritizing Workshop capacity and modest spend increases from nine orders without marginal return or capacity evidence. Treat as a proposed experiment, not measured superiority."
        }
      ],
      "issues": [
        {
          "status": "unsupported",
          "file": "summary.md assumptions",
          "passage": "all other exclusions are policy-driven and final",
          "correction": "RF04 is currently excluded but still requires reconciliation and could change final results."
        },
        {
          "status": "ambiguous wording, not critical failure",
          "file": "summary.md assumptions",
          "passage": "it contributes 0 to the calculable totals because its amount is unknown",
          "correction": "Excluded from calculable totals; unknown is not zero. Context explicitly states blank/not zero/provisional, so no critical failure."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-qwen4/initial/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "2468d6b2c2f694e0121848682f7e28379d8e21be582e7dcfc0ef3b1968fe4922",
        "summary.md": "8b0c1aec3a304a5291f2c40f92681977ead3a29cdec5b0a4ea6cc62293346a57",
        "cleaned-refunds.csv": "6ee8917b35497e8dafe1e8df397f3c53ffa68c0ae23955779dda2990a6db1def",
        "cleaned-sales.csv": "4246eeec8e33b98226c110ed50abaf89e5171e1fa88614b448f75d2d675ee5c4",
        "reconcile.py": "ba257e73c4d7231715b8d4cf35e8f333bb714ccc7509fb0e4a36a30e49c8cb12",
        "cleaned-ads.csv": "0e3c9e188124bd5ffc64327ae19741f6470f964988f1ff0403b464ed6169d7d7"
      }
    },
    {
      "run_id": "T01-R1-qwen4",
      "stage": "revision",
      "scores": [
        4,
        4,
        3,
        4,
        3
      ],
      "total": 18,
      "critical_failure": "No",
      "outcome": "minor edits",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Cleaned CSVs select N1001/2/3/5/6/8/10/11/12 once, retain S10 gross180 instead of S09 and exclude non-October/nonpaid records; exceptions.csv records all13 excluded IDs."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "Reviewer recomputation from CSVs:9 paid orders; gross2820; refunds300; ads450; net2520; contribution2070. Summary product values1920/900 gross,1680/840 net,1380/690 contribution,336/210 per order and280 combined match."
        },
        {
          "criterion": 3,
          "score": 3,
          "evidence": "R05 and R06 are correctly flagged and unknown amount blank, but summary final limitation says all other exclusions are policy-driven and final, overstating certainty around unresolved orphan RF04."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "All five required files parse and preserve source-row traceability through retained rows and exceptions; dates/money normalized. Included-only files and all-row audit files are both acceptable."
        },
        {
          "criterion": 5,
          "score": 3,
          "evidence": "Correct contribution/profit distinction and notes equal contribution/ad dollar; nevertheless suggests prioritizing Workshop capacity and modest spend increases from nine orders without marginal return or capacity evidence. Treat as a proposed experiment, not measured superiority."
        }
      ],
      "issues": [
        {
          "status": "unsupported",
          "file": "summary.md assumptions",
          "passage": "all other exclusions are policy-driven and final",
          "correction": "RF04 is currently excluded but still requires reconciliation and could change final results."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T01-R1-qwen4/revision/output",
      "reviewed_file_sha256": {
        "exceptions.csv": "2468d6b2c2f694e0121848682f7e28379d8e21be582e7dcfc0ef3b1968fe4922",
        "CHANGELOG.md": "f595f36e8266989d6723e9cb347d13b44a7c309b52ffac4304af46645f9cd690",
        "summary.md": "1fc3c35cc3eab31b380d3b950b48aefdfa48718446cde0758ab10836371643d2",
        "cleaned-refunds.csv": "8027491e4874698b8d8db5397d8f854bd670b839a695afd6eeb7451ee564f916",
        "cleaned-sales.csv": "d178d5359559ab4ad5c8fe62b7adcd33bbe2acf7b3093767720727fa57feb482",
        "reconcile.py": "1c202eaedcf0790e4ae81874fd0394beac8d3003edc1ad8a4eb732a2e59fd5c5",
        "cleaned-ads.csv": "60be8d0c1a2a43923528b39e0f416a0433120606d2ed3ae616a88e356ddd58d0"
      }
    },
    {
      "run_id": "T02-R1-astra",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-astra/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "cb4d3add5455f4f25686a017b5abbd0aa0c1c969e0c7507f9ecb11a652e6f02e",
        "build_comparison.py": "90a9aa341dbc5b1b28ba8c7a1e83e726744ae44f0235425d0e290eee63c3ff0b",
        "comparison.csv": "ee05a98ec90f3b5976f538b7a39122f031cbde7c512fef5da93b220054569101"
      }
    },
    {
      "run_id": "T02-R1-astra",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-astra/revision/output",
      "reviewed_file_sha256": {
        "review_submission.py": "fb46796500022b10852ae5beb7328a8694b19272c3ba38a45c7018f6552a8359",
        "recommendation.md": "741b3aadc36847bd1cfa61ecbd90170a4a19b1d087632cc9875e20afd892aaac",
        "build_comparison.py": "c1b9950ba1b3ec6fd2c867ba0ecca0a7c050d11a2a28b2594482960d6ecc5ca7",
        "change_log.md": "c61356aabf106e3147a3798ab311f116f925f375fe3ab021e9ffe6f28cf4de26",
        "comparison.csv": "77713a65dafddd64ac0bfc567c57e4284a54d5ffc2c8811a36ea174e0448d944",
        "review_checks.txt": "1d9c7cf35fdaabd798098270744f7a557393d7f1b7cce7eb28ad412434681735"
      }
    },
    {
      "run_id": "T02-R1-sol",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-sol/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "dec3a70d2b835e23e6154f71eb8607f0f40cef3c10d2887a91296e28da05d9f7",
        "comparison.csv": "3422773eb08c8d0c117a7997eb2747559f62da3f7d70a3e5215bf312076cecae"
      }
    },
    {
      "run_id": "T02-R1-sol",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-sol/revision/output",
      "reviewed_file_sha256": {
        "recommendation.md": "dfa5175949cf79121428aae8b744c6bacba635ad7ba424f42a7ef618d20f8f78",
        "change_log.md": "823e625dc4ce295b9e32f111d0c3cce8e622d057be87d451af68b942f0287cc8",
        "comparison.csv": "b1e3f7895d851d7bb5ecdbe91adfb94f02cdbec79a092197bc402c6d7e6870ad"
      }
    },
    {
      "run_id": "T02-R1-opus",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [
        {
          "status": "minor factual imprecision",
          "file": "comparison.csv O2",
          "passage": "heaviest-but-one at 2.3 kg",
          "correction": "Tied at2.3kg with the other Forge SKU; P16-32 is heavier at2.5kg. Does not affect purchase decision."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-opus/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "324760ec8a25e3de71c5450c09e65a5e4f278f2319ab31ca6afb685c9ca9d1b7",
        "check.py": "b5beea6b3bd057a8b9490e15f25cf2b7e7d51f3cc5b26796ae5ae023ef3f88d1",
        "calc.py": "da6520af8303217b056102b6b73af1b1b6e2a856cb500df03c9307b39f61f4df",
        "comparison.csv": "991455745cacd1eec9cb17e1627b44572f4dd43fa8a1e4c3a6a7590bde80eb7d"
      }
    },
    {
      "run_id": "T02-R1-opus",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [
        {
          "status": "minor factual imprecision",
          "file": "comparison.csv O2",
          "passage": "heaviest-but-one at 2.3 kg",
          "correction": "Tied at2.3kg with the other Forge SKU; P16-32 is heavier at2.5kg. Does not affect purchase decision."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-opus/revision/output",
      "reviewed_file_sha256": {
        "recommendation.md": "9fc3e1ab15ab47953fee70d21e3d1893369e4c83ccb17b08b66b91a14ece99b8",
        "check.py": "f52c124006fe53bb4268008c5b5e71faf3b463346c304a8847f74af6e1db9a39",
        "CHANGELOG.md": "972908454116325ec293b8e82dc7b36abb02fffa31125bd77fa760a94b77cff6",
        "calc.py": "da6520af8303217b056102b6b73af1b1b6e2a856cb500df03c9307b39f61f4df",
        "comparison.csv": "8522b7460b0621ea092b2745c5e859f7307b808e5f32a0499c547f2a4278c3c1",
        "preserve.py": "252958248d0a5bd9de479ceca4c54e75b976c396d841d862245c70793e124b64"
      }
    },
    {
      "run_id": "T02-R1-fable",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-fable/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "67e209b934069dbfcbf4e8e980dadf8dd1b66e7498542a3fd017bd4af8683bc0",
        "check.py": "4010870cdb596a49040814274aeae28ec34d95160db50173599212f90122d615",
        "comparison.csv": "48e74efe27d869a6575c4e88fc687af850542f11b46783049b49d5d4f2db841d"
      }
    },
    {
      "run_id": "T02-R1-qwen4",
      "stage": "initial",
      "scores": [
        4,
        4,
        4,
        3,
        4
      ],
      "total": 19,
      "critical_failure": "No",
      "outcome": "minor edits",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 3,
          "evidence": "Recommends qualifying O2 and treats O5 as conditional, but calls GX4070 the stronger-GPU/spec-headroom option without benchmark/spec evidence establishing performance beyond model name and equal8GB VRAM."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [
        {
          "status": "unsupported",
          "file": "recommendation.md fallback",
          "passage": "is the stronger-GPU option",
          "correction": "O5 has a differently named GPU and no supplied performance test; omit comparative strength unless source verifies it."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-qwen4/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "4aad9ede5b1b6718c44cc2d239ad953f1e7e55b5ea231abd060ef05169d37555",
        "comparison.csv": "49fe86e27593834a68e1d3ad58f00130d4f02eb3ec589f8135e622c34449f619"
      }
    },
    {
      "run_id": "T02-R1-fable",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        4,
        4
      ],
      "total": 20,
      "critical_failure": "No",
      "outcome": "usable",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 4,
          "evidence": "O2 recommended; O5 only conditional on a new written delivery commitment. S4 is limited to one exact-SKU run, not a comparative benchmark."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-fable/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "425cfd8ea8760f2a8805e37da874f8c4739f95635cd6dee495dae203a5d83aea",
        "check.py": "3da4c1657c549652afe2bd641891a796052687bd81007ba3c61e165be5efe905",
        "CHANGELOG.md": "e73fcf12fe3b7f92bfab8e110d4919e581a94c4c68c4c32241b0d1dca4bee109",
        "comparison.csv": "b18b92b7c367356e84268eabfc44ff3a5e559f5cdb323809a40ae15aaafa3c6f",
        "preserve_v1.py": "76fd918c927bd3bf84cff416cb7f8834b9e40c4813e07de7f25e56ee110f26c0"
      }
    },
    {
      "run_id": "T02-R1-qwen4",
      "stage": "revision",
      "scores": [
        4,
        4,
        4,
        3,
        4
      ],
      "total": 19,
      "critical_failure": "No",
      "outcome": "minor edits",
      "criterion_evidence": [
        {
          "criterion": 1,
          "score": 4,
          "evidence": "Reviewer recomputed (price+shipping)*1.12:O1=2238.88,O2=2098.88,O3=1790.88,O4=1902.88,O5=2126.88,O6=2037.28. Comparison CSV and recommendation agree; O2 leaves101.12."
        },
        {
          "criterion": 2,
          "score": 4,
          "evidence": "All six offers classified correctly:O1 budget;O3 RAM/SSD;O4 VRAM;O5 unconfirmed estimate;O6 late;O2 qualifies within scenario assumptions."
        },
        {
          "criterion": 3,
          "score": 4,
          "evidence": "Exact SKU manufacturer values override O3/O4 titles and S3 explicitly overrides O6 delivery to October27 despite ambiguous source chronology."
        },
        {
          "criterion": 4,
          "score": 3,
          "evidence": "Recommends qualifying O2 and treats O5 as conditional, but calls GX4070 the stronger-GPU/spec-headroom option without benchmark/spec evidence establishing performance beyond model name and equal8GB VRAM."
        },
        {
          "criterion": 5,
          "score": 4,
          "evidence": "Both required files present and parse; memo within650 words and identifies sources/rows for prices, specifications, conflicts and performance evidence."
        }
      ],
      "issues": [
        {
          "status": "unsupported",
          "file": "recommendation.md fallback",
          "passage": "is the stronger-GPU option",
          "correction": "O5 has a differently named GPU and no supplied performance test; omit comparative strength unless source verifies it."
        }
      ],
      "evidence_directory": "work/qwen4-real-world-test-kit/execution/runs/T02-R1-qwen4/initial/output",
      "reviewed_file_sha256": {
        "recommendation.md": "fefb8b02c6bc8df501d0c38bfe7a83a4a4b8a80be702ee4aa61ce1a145355b6c",
        "REVIEW-CHANGELOG.md": "353921ccbf44be0ae7d378a28902d9b4d927795a89ddef9e8f41a63574505bff",
        "comparison.csv": "49fe86e27593834a68e1d3ad58f00130d4f02eb3ec589f8135e622c34449f619"
      }
    }
  ],
  "complete_at_final_snapshot": [
    {
      "run_id": "T01-R1-astra",
      "stage": "initial"
    },
    {
      "run_id": "T01-R1-astra",
      "stage": "revision"
    },
    {
      "run_id": "T02-R1-opus",
      "stage": "initial"
    },
    {
      "run_id": "T02-R1-opus",
      "stage": "revision"
    },
    {
      "run_id": "T02-R1-qwen4",
      "stage": "initial"
    },
    {
      "run_id": "T02-R1-qwen4",
      "stage": "revision"
    },
    {
      "run_id": "T01-R1-sol",
      "stage": "initial"
    },
    {
      "run_id": "T01-R1-sol",
      "stage": "revision"
    },
    {
      "run_id": "T01-R1-opus",
      "stage": "initial"
    },
    {
      "run_id": "T01-R1-opus",
      "stage": "revision"
    },
    {
      "run_id": "T01-R1-fable",
      "stage": "initial"
    },
    {
      "run_id": "T01-R1-fable",
      "stage": "revision"
    },
    {
      "run_id": "T01-R1-qwen4",
      "stage": "initial"
    },
    {
      "run_id": "T01-R1-qwen4",
      "stage": "revision"
    },
    {
      "run_id": "T02-R1-astra",
      "stage": "initial"
    },
    {
      "run_id": "T02-R1-astra",
      "stage": "revision"
    },
    {
      "run_id": "T02-R1-fable",
      "stage": "initial"
    },
    {
      "run_id": "T02-R1-fable",
      "stage": "revision"
    },
    {
      "run_id": "T02-R1-sol",
      "stage": "initial"
    },
    {
      "run_id": "T02-R1-sol",
      "stage": "revision"
    }
  ],
  "completed_but_not_reviewed": []
}
