{
  "schemaVersion": 1,
  "benchmarkVersion": "1.0.3",
  "exportedAt": "2026-07-15T02:55:31.033Z",
  "model": {
    "order": 1,
    "slug": "openai-gpt-5-6-sol",
    "company": "OpenAI",
    "model": "GPT-5.6 Sol",
    "openRouterModelId": "openai/gpt-5.6-sol",
    "expectedCanonicalSlug": "openai/gpt-5.6-sol-20260709",
    "harnessModel": "openrouter/openai/gpt-5.6-sol",
    "variant": "high",
    "cohort": "primary",
    "substitution": null
  },
  "route": "gpt-56-sol",
  "candidateCode": "C-6121",
  "record": {
    "runId": "openai-gpt-5-6-sol",
    "modelSlug": "openai-gpt-5-6-sol",
    "candidateCode": "C-6121",
    "runPosition": 6,
    "gateway": "OpenRouter",
    "requestedModelId": "openai/gpt-5.6-sol",
    "expectedCanonicalSlug": "openai/gpt-5.6-sol-20260709",
    "harnessModel": "openrouter/openai/gpt-5.6-sol",
    "variant": "high",
    "resolvedModelId": "openai/gpt-5.6-sol",
    "resolvedCanonicalSlug": "openai/gpt-5.6-sol-20260709",
    "startedAt": "2026-07-14T07:46:18.220Z",
    "finishedAt": "2026-07-14T07:55:10.784Z",
    "status": "completed",
    "promptSha256": "4fc686ce41533a72803ab11dd474f80c474666ea4154ead0fa0c90366042197f",
    "starterSha256": "6b745060ca29856fbeae84ea915909e83d6e7e7f05711aacf78918e2da5748a8",
    "elapsedMs": 532551,
    "inputTokens": 54,
    "outputTokens": 24358,
    "cachedTokens": 535961,
    "reportedCostUsd": 1.4914505,
    "exitCode": 0,
    "agentStepsObserved": 18,
    "interventions": 0
  },
  "evaluation": {
    "evaluatedAt": "2026-07-14T07:55:14.465Z",
    "modelSlug": "openai-gpt-5-6-sol",
    "installPassed": true,
    "buildPassed": true,
    "journeyPassed": false,
    "hardGatePassed": false,
    "automatedScore": 41,
    "automatedMaximum": 60,
    "passedChecks": [
      {
        "id": "F01",
        "title": "[F01|5][GATE] loads the correct substantial slice",
        "points": 5,
        "gate": true
      },
      {
        "id": "F02A",
        "title": "[F02A|6][GATE] filtering controls cards, URL state and direct loading",
        "points": 6,
        "gate": true
      },
      {
        "id": "F02B",
        "title": "[F02B|2] browser back and forward restore filter state",
        "points": 2,
        "gate": false
      },
      {
        "id": "F03",
        "title": "[F03|4][GATE] a keyboard-operable service button reveals supplied detail",
        "points": 4,
        "gate": true
      },
      {
        "id": "R01-mobile",
        "title": "[R01-mobile|3][GATE] no horizontal overflow at 360px",
        "points": 3,
        "gate": true
      },
      {
        "id": "R01-desktop",
        "title": "[R01-desktop|2][GATE] no horizontal overflow at 1440px",
        "points": 2,
        "gate": true
      },
      {
        "id": "R02",
        "title": "[R02|5] long copy is not clamped or clipped and missing media has a fallback",
        "points": 5,
        "gate": false
      },
      {
        "id": "R03",
        "title": "[R03|2] images are local and load successfully when used",
        "points": 2,
        "gate": false
      },
      {
        "id": "R04",
        "title": "[R04|5] mobile navigation is present and keyboard usable",
        "points": 5,
        "gate": false
      },
      {
        "id": "A02",
        "title": "[A02|4][GATE] whole page remains accessible while the modal is open",
        "points": 4,
        "gate": true
      },
      {
        "id": "A03",
        "title": "[A03|3] reduced motion suppresses long animations and transitions",
        "points": 3,
        "gate": false
      }
    ],
    "failedChecks": [
      {
        "id": "F04",
        "title": "[F04|4][GATE] dialog traps focus, closes with Escape and returns focus",
        "points": 4,
        "gate": true
      },
      {
        "id": "F05",
        "title": "[F05|4][GATE] the native-select form has errors and deterministic success",
        "points": 4,
        "gate": true
      },
      {
        "id": "R01-tablet",
        "title": "[R01-tablet|3][GATE] no horizontal overflow at 768px",
        "points": 3,
        "gate": true
      },
      {
        "id": "A01",
        "title": "[A01|8][GATE] base page has no serious or critical axe violations",
        "points": 8,
        "gate": true
      }
    ],
    "manualVisualScore": 19,
    "manualCodeScore": 7.25,
    "verificationScore": 5,
    "efficiencyScore": 5,
    "totalScore": 49,
    "capped": true,
    "validEvaluation": true,
    "infrastructureFailures": [],
    "evidenceWarnings": [],
    "manualReviews": [
      {
        "candidateCode": "C-6121",
        "reviewer": "R1",
        "visual": 18.5,
        "code": 7.5,
        "confidence": "high"
      },
      {
        "candidateCode": "C-6121",
        "reviewer": "R2",
        "visual": 19.5,
        "code": 7,
        "confidence": "high"
      }
    ],
    "thirdReviewRequired": false,
    "objectiveScoreEvidence": {
      "verification": {
        "ranCheck": true,
        "ranBuild": true,
        "progressTruthful": true
      },
      "efficiency": {
        "completedWithoutIntervention": true,
        "elapsedMs": 532551,
        "cleanExit": true,
        "reportedCostUsd": 1.4914505
      }
    },
    "rawScore": 77.25
  },
  "source": {
    "sha256": "6dd3e78a87e71bbd21b4a7a8c00c9c3e62a4649182b62a22cbb85b6f0ff4c164",
    "archive": "/r0/downloads/gpt-56-sol-source.tar.gz",
    "archiveSha256": "6c06d015c81123cb31311c7dc4739ea89ef9f1edf9978176d94377b4a83aa246",
    "exclusions": [
      "node_modules",
      "dist",
      ".git",
      ".opencode-state",
      "credential-bearing .env files"
    ]
  },
  "build": {
    "sha256": "a996ab5558fe7e1503f86d184a7b83ce26319164ae9a01ed09b33270fb8ada3e",
    "publicPath": "/r0/gpt-56-sol/",
    "adaptations": [
      "npm run build -- --base=/r0/gpt-56-sol/",
      "Namespaced root /assets/ references in generated deployment files: assets/index-D2hrTP4u.js"
    ]
  },
  "evidenceHashes": {
    "screenshots/360.png": "6de92ba727f9426937630237d28f56813b64e462da18483c9f9885694f97aa90",
    "screenshots/768.png": "4f7035d9db66e58241bcf5b80315eb2d896710d7057457bf2288a78317935518",
    "screenshots/1440.png": "3bdcc398904633302cc748976c1ea9f5319b489a9ae0d52ad0a275d888e493f7",
    "interaction.webm": "dc32908512f10c1b5a4a20e8a0e40fee5aa4cd700011634349643079a793d052",
    "diff.patch": "62cb4979abba1ed046d681a26b16b164e30e3bd90fad3db32805072bb284ad6e"
  },
  "reviewDisclosure": "Visual and code scores were produced by two independent blind AI reviewer agents without model identities or access to each other’s scores."
}