{
  "experiment": "2024-dec-31-en-lean-design-bakeoff",
  "commonInputManifestSha256": "d3040b9a95baf4b6c0b11d0c96b5b4694e8eaa16d8587303689e1c0ad570b0bd",
  "commonInputAfterRuns": "UNCHANGED",
  "results": [
    {
      "id": "fable",
      "requestedModel": "claude-fable-5-1",
      "actualModel": "claude-fable-5-1",
      "status": "CONTENT_CONTRACT_PASS_RESPONSIVE_GATE_FAIL",
      "firstAttempt": {
        "status": "ABORTED_HUNG_NO_FILES",
        "durationSeconds": 1807,
        "costUsd": 7.41692075
      },
      "authorizedReplacementAttempt": {
        "status": "COMPLETE",
        "durationSeconds": 937,
        "costUsd": 6.87705925
      },
      "totalCostUsd": 14.29398,
      "technicalChecks": {
        "orderedTextBlocks": "51/51",
        "unchangedAssetsExactlyOnce": "26/26",
        "declaredContentLinksExactlyOnce": "21/21",
        "shellLinksExactlyOnce": "5/5",
        "sourcePdfLinksExactlyOnce": "1/1"
      },
      "visualEvidence": "The 390px first-viewport capture clips edition metadata, the main heading, and body text on the right.",
      "humanVisualApproval": "NOT_REVIEWED"
    },
    {
      "id": "sol",
      "requestedModel": "gpt-5.6-sol",
      "actualModel": "gpt-5.6-sol",
      "status": "CONTENT_CONTRACT_PASS_RESPONSIVE_GATE_FAIL",
      "route": "Codex task; no paid OpenAI API",
      "technicalChecks": {
        "orderedTextBlocks": "51/51",
        "unchangedAssetsExactlyOnce": "26/26",
        "declaredContentLinksExactlyOnce": "21/21",
        "shellLinksExactlyOnce": "5/5",
        "sourcePdfLinksExactlyOnce": "1/1"
      },
      "visualEvidence": "The 390px first-viewport capture severely clips the download control, main heading, and body text on the right.",
      "humanVisualApproval": "NOT_REVIEWED"
    },
    {
      "id": "gemini",
      "requestedModel": "gemini-3.8-flash",
      "actualMainModelReportedByCli": "gemini-3.5-flash",
      "status": "REJECTED",
      "costUsd": null,
      "costNote": "Gemini CLI did not report currency cost.",
      "provenanceFailure": "Requested and reported main models differ.",
      "callGuardFailure": "One CLI invocation reported 60 main-model requests and 3 utility-model requests.",
      "technicalFailures": [
        "source/newsletter.pdf is linked twice; the frozen contract requires exactly once.",
        "Supplied logical text is duplicated and one supplied Rahul Gandhi attribution is not represented as its own exact logical block.",
        "The masthead asset is referenced five times rather than once.",
        "Invented visible copy includes CLICK TO WATCH, VIEW MORE, MAJOR SPEECHES, and an unsupported 2026 copyright statement."
      ],
      "humanVisualApproval": "NOT_APPLICABLE_TO_REJECTED_ARM"
    },
    {
      "id": "gemini-corrective-v2",
      "purpose": "User-authorized corrective retry focused on model provenance and the duplicated source-PDF link; it does not replace the original one-shot evidence.",
      "cliVersion": "0.58.0",
      "requestedModel": "gemini-3.8-flash",
      "actualMainModelReportedByCli": "gemini-3.5-flash",
      "status": "CONTENT_CONTRACT_PASS_RESPONSIVE_GATE_FAIL_PROVENANCE_FAIL",
      "costUsd": null,
      "costNote": "Gemini CLI did not report currency cost.",
      "provenanceFailure": "The current CLI still reported gemini-3.5-flash for the main model despite the explicit gemini-3.8-flash flag.",
      "callGuardObservation": "One CLI invocation reported 69 main-model requests and 3 utility-model requests; the utility model was gemini-3-flash-preview.",
      "technicalChecks": {
        "orderedTextBlocks": "51/51",
        "unchangedAssetsExactlyOnce": "26/26",
        "declaredContentLinksExactlyOnce": "21/21",
        "shellLinksExactlyOnce": "5/5",
        "sourcePdfLinksExactlyOnce": "1/1"
      },
      "correctedDefects": [
        "The duplicated source/newsletter.pdf link was reduced to exactly one.",
        "All supplied logical text blocks are present exactly once and in order, including the standalone Rahul Gandhi attribution.",
        "The masthead is referenced exactly once.",
        "Previously invented visible labels and unsupported copyright copy were removed."
      ],
      "visualEvidence": "The 390px capture remains horizontally clipped: the header control, masthead, headline, and body text extend beyond the viewport.",
      "humanVisualApproval": "NOT_APPLICABLE_TO_REJECTED_CORRECTIVE_RUN"
    }
  ],
  "decision": "NOT_PROVEN. Fable and Sol pass the content contract but both fail responsive containment at 390 px; Gemini is excluded on provenance, call-guard, and content-contract failures. No winner was selected.",
  "scaleOutAuthorized": false,
  "mergeAuthorized": false,
  "deployAuthorized": false,
  "publishAuthorized": false
}