{
  "runId": "4524afb8-11f9-4d79-b50f-a599f7b0109c",
  "generatedAt": "2026-07-10T15:49:32.186Z",
  "models": [
    {
      "adapterId": "agy",
      "model": "Gemini 3.1 Pro (Low)",
      "label": "agy/Gemini 3.1 Pro (Low)",
      "attempts": 38,
      "scored": 36,
      "pass": 33,
      "partial": 3,
      "incorrect": 0,
      "protocol_error": 2,
      "otherErrors": 0,
      "passRate": 0.9166666666666666,
      "quality": 0.9583333333333334,
      "coverage": 0.9473684210526315,
      "fixtureScore": 0.9078947368421053,
      "fixtureSolveRate": 0.868421052631579,
      "passAtKRate": 0.9166666666666666,
      "fixturesPassed": 33,
      "fixturesAttempted": 38,
      "avgDurationMs": 11490,
      "minDurationMs": 8109,
      "maxDurationMs": 21900,
      "totalDurationMs": 436610,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 1,
      "passRateLabel": "91.7%",
      "qualityLabel": "0.958",
      "coverageLabel": "94.7%",
      "fixtureScoreLabel": "90.8%",
      "fixtureSolveRateLabel": "86.8%",
      "passAtKLabel": "91.7%"
    },
    {
      "adapterId": "agy",
      "model": "Gemini 3.5 Flash (High)",
      "label": "agy/Gemini 3.5 Flash (High)",
      "attempts": 38,
      "scored": 37,
      "pass": 32,
      "partial": 4,
      "incorrect": 1,
      "protocol_error": 1,
      "otherErrors": 0,
      "passRate": 0.8648648648648649,
      "quality": 0.918918918918919,
      "coverage": 0.9736842105263158,
      "fixtureScore": 0.8947368421052632,
      "fixtureSolveRate": 0.8421052631578947,
      "passAtKRate": 0.8648648648648649,
      "fixturesPassed": 32,
      "fixturesAttempted": 38,
      "avgDurationMs": 6955,
      "minDurationMs": 4532,
      "maxDurationMs": 17900,
      "totalDurationMs": 264279,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 2,
      "passRateLabel": "86.5%",
      "qualityLabel": "0.919",
      "coverageLabel": "97.4%",
      "fixtureScoreLabel": "89.5%",
      "fixtureSolveRateLabel": "84.2%",
      "passAtKLabel": "86.5%"
    },
    {
      "adapterId": "agy",
      "model": "Gemini 3.1 Pro (High)",
      "label": "agy/Gemini 3.1 Pro (High)",
      "attempts": 38,
      "scored": 36,
      "pass": 32,
      "partial": 3,
      "incorrect": 1,
      "protocol_error": 2,
      "otherErrors": 0,
      "passRate": 0.8888888888888888,
      "quality": 0.9305555555555556,
      "coverage": 0.9473684210526315,
      "fixtureScore": 0.881578947368421,
      "fixtureSolveRate": 0.8421052631578947,
      "passAtKRate": 0.8888888888888888,
      "fixturesPassed": 32,
      "fixturesAttempted": 38,
      "avgDurationMs": 12759,
      "minDurationMs": 8589,
      "maxDurationMs": 24474,
      "totalDurationMs": 484841,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 3,
      "passRateLabel": "88.9%",
      "qualityLabel": "0.931",
      "coverageLabel": "94.7%",
      "fixtureScoreLabel": "88.2%",
      "fixtureSolveRateLabel": "84.2%",
      "passAtKLabel": "88.9%"
    },
    {
      "adapterId": "agy",
      "model": "Gemini 3.5 Flash (Medium)",
      "label": "agy/Gemini 3.5 Flash (Medium)",
      "attempts": 38,
      "scored": 37,
      "pass": 30,
      "partial": 4,
      "incorrect": 3,
      "protocol_error": 1,
      "otherErrors": 0,
      "passRate": 0.8108108108108109,
      "quality": 0.8648648648648649,
      "coverage": 0.9736842105263158,
      "fixtureScore": 0.8421052631578947,
      "fixtureSolveRate": 0.7894736842105263,
      "passAtKRate": 0.8108108108108109,
      "fixturesPassed": 30,
      "fixturesAttempted": 38,
      "avgDurationMs": 6292,
      "minDurationMs": 4645,
      "maxDurationMs": 25349,
      "totalDurationMs": 239081,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 4,
      "passRateLabel": "81.1%",
      "qualityLabel": "0.865",
      "coverageLabel": "97.4%",
      "fixtureScoreLabel": "84.2%",
      "fixtureSolveRateLabel": "78.9%",
      "passAtKLabel": "81.1%"
    },
    {
      "adapterId": "agy",
      "model": "Gemini 3.5 Flash (Low)",
      "label": "agy/Gemini 3.5 Flash (Low)",
      "attempts": 38,
      "scored": 37,
      "pass": 29,
      "partial": 5,
      "incorrect": 3,
      "protocol_error": 1,
      "otherErrors": 0,
      "passRate": 0.7837837837837838,
      "quality": 0.8513513513513513,
      "coverage": 0.9736842105263158,
      "fixtureScore": 0.8289473684210527,
      "fixtureSolveRate": 0.7631578947368421,
      "passAtKRate": 0.7837837837837838,
      "fixturesPassed": 29,
      "fixturesAttempted": 38,
      "avgDurationMs": 4696,
      "minDurationMs": 3730,
      "maxDurationMs": 7343,
      "totalDurationMs": 178459,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 5,
      "passRateLabel": "78.4%",
      "qualityLabel": "0.851",
      "coverageLabel": "97.4%",
      "fixtureScoreLabel": "82.9%",
      "fixtureSolveRateLabel": "76.3%",
      "passAtKLabel": "78.4%"
    },
    {
      "adapterId": "agy",
      "model": "Claude Sonnet 4.6 (Thinking)",
      "label": "agy/Claude Sonnet 4.6 (Thinking)",
      "attempts": 38,
      "scored": 34,
      "pass": 28,
      "partial": 2,
      "incorrect": 4,
      "protocol_error": 4,
      "otherErrors": 0,
      "passRate": 0.8235294117647058,
      "quality": 0.8529411764705882,
      "coverage": 0.8947368421052632,
      "fixtureScore": 0.7631578947368421,
      "fixtureSolveRate": 0.7368421052631579,
      "passAtKRate": 0.8235294117647058,
      "fixturesPassed": 28,
      "fixturesAttempted": 38,
      "avgDurationMs": 10101,
      "minDurationMs": 6711,
      "maxDurationMs": 18925,
      "totalDurationMs": 383854,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 6,
      "passRateLabel": "82.4%",
      "qualityLabel": "0.853",
      "coverageLabel": "89.5%",
      "fixtureScoreLabel": "76.3%",
      "fixtureSolveRateLabel": "73.7%",
      "passAtKLabel": "82.4%"
    },
    {
      "adapterId": "agy",
      "model": "Claude Opus 4.6 (Thinking)",
      "label": "agy/Claude Opus 4.6 (Thinking)",
      "attempts": 38,
      "scored": 17,
      "pass": 14,
      "partial": 2,
      "incorrect": 1,
      "protocol_error": 21,
      "otherErrors": 0,
      "passRate": 0.8235294117647058,
      "quality": 0.8823529411764706,
      "coverage": 0.4473684210526316,
      "fixtureScore": 0.39473684210526316,
      "fixtureSolveRate": 0.3684210526315789,
      "passAtKRate": 0.8235294117647058,
      "fixturesPassed": 14,
      "fixturesAttempted": 38,
      "avgDurationMs": 7744,
      "minDurationMs": 5744,
      "maxDurationMs": 15371,
      "totalDurationMs": 294285,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 7,
      "passRateLabel": "82.4%",
      "qualityLabel": "0.882",
      "coverageLabel": "44.7%",
      "fixtureScoreLabel": "39.5%",
      "fixtureSolveRateLabel": "36.8%",
      "passAtKLabel": "82.4%"
    },
    {
      "adapterId": "agy",
      "model": "GPT-OSS 120B (Medium)",
      "label": "agy/GPT-OSS 120B (Medium)",
      "attempts": 38,
      "scored": 0,
      "pass": 0,
      "partial": 0,
      "incorrect": 0,
      "protocol_error": 38,
      "otherErrors": 0,
      "passRate": 0,
      "quality": 0,
      "coverage": 0,
      "fixtureScore": 0,
      "fixtureSolveRate": 0,
      "passAtKRate": 0,
      "fixturesPassed": 0,
      "fixturesAttempted": 38,
      "avgDurationMs": 6574,
      "minDurationMs": 5852,
      "maxDurationMs": 7833,
      "totalDurationMs": 249823,
      "totalPromptTokens": 0,
      "totalCompletionTokens": 0,
      "totalReasoningTokens": 0,
      "totalTokens": 0,
      "totalCachedTokens": 0,
      "totalCost": 0,
      "totalAnswerBytes": 0,
      "rank": 8,
      "passRateLabel": "0.0%",
      "qualityLabel": "0.000",
      "coverageLabel": "0.0%",
      "fixtureScoreLabel": "0.0%",
      "fixtureSolveRateLabel": "0.0%",
      "passAtKLabel": "0.0%"
    }
  ]
}