{
  "evidence": {
    "runId": "4a83d910-4b81-420f-86a0-ad4d10d028fb",
    "evaluatedAt": "2026-09-27T07:56:21.635Z",
    "packageVersion": "0.7.0",
    "recipeFingerprint": "a6b4ea63f7fd7fa621bad8ebdac68421b2db9ccbf35efc7fd32ca34bc4f712c5",
    "datasetFingerprint": "76b23fbd70fbaf9b0347bf44ecc4f34198e1f6101335b8d79aec113f1b3bbad5",
    "split": "held-out",
    "mode": "live",
    "policy": {
      "minConfidence": 0.8,
      "selectedOn": {
        "runId": "9f74afdf-c99f-4dec-97fa-c572cb014bfd",
        "datasetFingerprint": "f114c3eafe3661a5abbe5cd4c127a5baba64004b0c3903066bb017bb9284443b",
        "recipeFingerprint": "a6b4ea63f7fd7fa621bad8ebdac68421b2db9ccbf35efc7fd32ca34bc4f712c5",
        "model": "jev-1.13.0",
        "split": "development"
      }
    },
    "provenance": [
      {
        "method": "author-synthetic",
        "source": "AI-authored synthetic boundary cases for jev-recipes; labels have not had independent human review.",
        "cases": 40
      }
    ]
  },
  "recipe": "handoff",
  "model": "jev-1.13.0",
  "models": [
    "jev-1.13.0"
  ],
  "cases": 40,
  "completed": 40,
  "failed": 0,
  "correct": 36,
  "wrong": 4,
  "ready": 28,
  "review": 12,
  "accuracy": 0.9,
  "completedAccuracy": 0.9,
  "accuracyInterval95": [
    0.7695,
    0.9604
  ],
  "readyAccuracy": 1,
  "readyAccuracyInterval95": [
    0.8794,
    1
  ],
  "reviewRate": 0.3,
  "contestedAccuracy": 0.8,
  "adversarialAccuracy": 0.9,
  "calibration": [
    {
      "confidence": "0-0.5",
      "cases": 5,
      "accuracy": 0.2
    },
    {
      "confidence": "0.5-0.6",
      "cases": 0,
      "accuracy": null
    },
    {
      "confidence": "0.6-0.7",
      "cases": 0,
      "accuracy": null
    },
    {
      "confidence": "0.7-0.8",
      "cases": 1,
      "accuracy": 1
    },
    {
      "confidence": "0.8-0.9",
      "cases": 2,
      "accuracy": 1
    },
    {
      "confidence": "0.9-1",
      "cases": 32,
      "accuracy": 1
    }
  ],
  "thresholdEvaluation": "frozen-policy",
  "thresholds": [],
  "suggestedMinConfidence": null,
  "latencyMs": {
    "p50": 108.85,
    "p95": 216.82,
    "totalCaseTime": 4989.93
  },
  "usage": {
    "requests": 40,
    "input_tokens": 18994,
    "output_tokens": 1769,
    "unknownResponses": 0,
    "failedRequests": 0,
    "models": [
      "jev-1.13.0"
    ]
  },
  "cost": {
    "model": "jev-1.13.0",
    "inputPerMillion": 0.042,
    "outputPerMillion": 0,
    "currency": "USD",
    "date": "2026-09-27",
    "source": "https://docs.typesafe.ai/models",
    "estimated": 0.00079775
  },
  "confusion": {
    "{\"checks.0.verdict\":\"does_not_match\"}": {
      "total": 20,
      "correct": 18,
      "misses": {
        "{\"checks.0.verdict\":\"matches\"}": 1,
        "{\"checks.0.verdict\":\"unclear\"}": 1
      }
    },
    "{\"checks.0.verdict\":\"matches\"}": {
      "total": 10,
      "correct": 10,
      "misses": {}
    },
    "{\"checks.0.verdict\":\"unclear\"}": {
      "total": 10,
      "correct": 8,
      "misses": {
        "{\"checks.0.verdict\":\"does_not_match\"}": 2
      }
    }
  },
  "failures": [
    {
      "id": "featured-18-missing",
      "expected": {
        "checks.0.verdict": "unclear"
      },
      "actual": {
        "checks.0.verdict": "does_not_match"
      },
      "confidence": 0.25,
      "contested": true
    },
    {
      "id": "featured-19-historical",
      "expected": {
        "checks.0.verdict": "does_not_match"
      },
      "actual": {
        "checks.0.verdict": "matches"
      },
      "confidence": 0.35,
      "adversarial": true
    },
    {
      "id": "featured-22-not-matched",
      "expected": {
        "checks.0.verdict": "does_not_match"
      },
      "actual": {
        "checks.0.verdict": "unclear"
      },
      "confidence": 0.44
    },
    {
      "id": "featured-23-missing",
      "expected": {
        "checks.0.verdict": "unclear"
      },
      "actual": {
        "checks.0.verdict": "does_not_match"
      },
      "confidence": 0.43,
      "contested": true
    }
  ],
  "acceptance": {
    "minConfidence": 0.8,
    "minimumHeldOutReadyCases": 20,
    "minimumReadyAccuracy": 0.95,
    "maximumProviderFailures": 0,
    "met": true,
    "label": "Measured on these synthetic cases"
  }
}