{
  "schemaVersion": 1,
  "benchmarkVersion": "1.0.3",
  "exportedAt": "2026-07-15T02:55:39.216Z",
  "model": {
    "order": 3,
    "slug": "xai-grok-4-5",
    "company": "xAI / SpaceXAI",
    "model": "Grok 4.5",
    "openRouterModelId": "x-ai/grok-4.5",
    "expectedCanonicalSlug": "x-ai/grok-4.5-20260708",
    "harnessModel": "openrouter/x-ai/grok-4.5",
    "variant": "high",
    "cohort": "primary",
    "substitution": null
  },
  "route": "grok-45",
  "candidateCode": "C-2264",
  "record": {
    "runId": "xai-grok-4-5",
    "modelSlug": "xai-grok-4-5",
    "candidateCode": "C-2264",
    "runPosition": 4,
    "gateway": "OpenRouter",
    "requestedModelId": "x-ai/grok-4.5",
    "expectedCanonicalSlug": "x-ai/grok-4.5-20260708",
    "harnessModel": "openrouter/x-ai/grok-4.5",
    "variant": "high",
    "resolvedModelId": "x-ai/grok-4.5",
    "resolvedCanonicalSlug": "x-ai/grok-4.5-20260708",
    "startedAt": "2026-07-14T04:13:58.224Z",
    "finishedAt": "2026-07-14T04:15:49.675Z",
    "status": "completed",
    "promptSha256": "4fc686ce41533a72803ab11dd474f80c474666ea4154ead0fa0c90366042197f",
    "starterSha256": "6b745060ca29856fbeae84ea915909e83d6e7e7f05711aacf78918e2da5748a8",
    "elapsedMs": 111445,
    "inputTokens": 39701,
    "outputTokens": 15362,
    "cachedTokens": 370432,
    "reportedCostUsd": 0.36988199999999993,
    "exitCode": 0,
    "agentStepsObserved": 14,
    "interventions": 0
  },
  "evaluation": {
    "evaluatedAt": "2026-07-14T07:35:04.669Z",
    "modelSlug": "xai-grok-4-5",
    "installPassed": true,
    "buildPassed": true,
    "journeyPassed": false,
    "hardGatePassed": true,
    "automatedScore": 55,
    "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": "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-mobile",
        "title": "[R01-mobile|3][GATE] no horizontal overflow at 360px",
        "points": 3,
        "gate": true
      },
      {
        "id": "R01-tablet",
        "title": "[R01-tablet|3][GATE] no horizontal overflow at 768px",
        "points": 3,
        "gate": true
      },
      {
        "id": "R01-desktop",
        "title": "[R01-desktop|2][GATE] no horizontal overflow at 1440px",
        "points": 2,
        "gate": true
      },
      {
        "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": "A01",
        "title": "[A01|8][GATE] base page has no serious or critical axe violations",
        "points": 8,
        "gate": true
      },
      {
        "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": "R02",
        "title": "[R02|5] long copy is not clamped or clipped and missing media has a fallback",
        "points": 5,
        "gate": false
      }
    ],
    "manualVisualScore": 17.5,
    "manualCodeScore": 6.75,
    "verificationScore": 5,
    "efficiencyScore": 5,
    "totalScore": 89.25,
    "capped": false,
    "validEvaluation": true,
    "infrastructureFailures": [],
    "evidenceWarnings": [],
    "manualReviews": [
      {
        "candidateCode": "C-2264",
        "reviewer": "R1",
        "visual": 17.5,
        "code": 7.5,
        "confidence": "high"
      },
      {
        "candidateCode": "C-2264",
        "reviewer": "R2",
        "visual": 17.5,
        "code": 6,
        "confidence": "high"
      }
    ],
    "thirdReviewRequired": false,
    "objectiveScoreEvidence": {
      "verification": {
        "ranCheck": true,
        "ranBuild": true,
        "progressTruthful": true
      },
      "efficiency": {
        "completedWithoutIntervention": true,
        "elapsedMs": 111445,
        "cleanExit": true,
        "reportedCostUsd": 0.36988199999999993
      }
    },
    "rawScore": 89.25
  },
  "source": {
    "sha256": "e7f6c41519612ab07f7cfa7737718779fe40f81e339b2c756f91b33d18e23d3d",
    "archive": "/r0/downloads/grok-45-source.tar.gz",
    "archiveSha256": "7b059e11b795200d2f618c91c34395946c77a3e955f761b1125c3b5245c4758b",
    "exclusions": [
      "node_modules",
      "dist",
      ".git",
      ".opencode-state",
      "credential-bearing .env files"
    ]
  },
  "build": {
    "sha256": "d6738732346b54105c96a5cc50ec52053c82c89f73a46460c39a649c8f802e0c",
    "publicPath": "/r0/grok-45/",
    "adaptations": [
      "npm run build -- --base=/r0/grok-45/",
      "Namespaced root /assets/ references in generated deployment files: assets/index-BxME2Ezl.js"
    ]
  },
  "evidenceHashes": {
    "screenshots/360.png": "054c4d6795507214cb289473a6f1e25c39aaf76b84fd62caafbc13a9078f8815",
    "screenshots/768.png": "cf772b5209e68e1340f04686f5d5240ee959ceb4860e40018b614c007fb0ba77",
    "screenshots/1440.png": "82acb01239c462bd85a966e848aaad24c6f57391b3961d8e0bd161f73b8a30d4",
    "interaction.webm": "6570ab50ada9428cf3479a1928d54db3e57636a9b0167c91dd5b4e4448635c97",
    "diff.patch": "fe08bd5a09ff3140999f2653a3a798b51c898f9d8cba7051458064a8afbd4d74"
  },
  "reviewDisclosure": "Visual and code scores were produced by two independent blind AI reviewer agents without model identities or access to each other’s scores."
}