{
  "schema": 1,
  "artifactSha256": "d14ba9edee1bb4c4996a726deb81e49ae81800a3216f0774634238c380aee496",
  "privateHoldout": false,
  "benchmarks": [
    {
      "name": "s46-mvp-gsm8k-slice-v0",
      "datasetSource": "https://github.com/openai/grade-school-math",
      "datasetVersion": "mvp-public-slice-2026-05",
      "config": {
        "name": "s46-mvp-gsm8k-slice-v0-config",
        "path": "public/benchmark-configs/s46-mvp-gsm8k-slice-v0-config.json",
        "mediaType": "application/json",
        "sha256": "252decb7736c3956e54fff6a24dbb0190ac9f146856b925d09d457cd3b3f3bf5",
        "sizeBytes": 1311
      },
      "public": true,
      "score": 0.3333333333333333,
      "tolerance": {
        "metric": "exact_match_final_integer",
        "value": 0,
        "reason": "MVP public smoke benchmark records the score and proves the evidence path; it does not enforce a release-quality score threshold. TODO: replace with the Sovereign46 production benchmark set."
      }
    }
  ],
  "rawResults": [
    {
      "name": "s46-mvp-gsm8k-slice-v0-raw-results",
      "path": "raw/s46-mvp-gsm8k-slice-v0-raw-results.json",
      "mediaType": "application/json",
      "sha256": "b778ae76971f12538c74b487a8564be954be51a7efd909c2f2fd883b53cddf45",
      "sizeBytes": 2058
    }
  ],
  "summary": {
    "backendIdentity": {
      "artifactSha256": "d14ba9edee1bb4c4996a726deb81e49ae81800a3216f0774634238c380aee496",
      "field": "model_path",
      "method": "reported_model_path_matches_artifact_path",
      "source": "/props",
      "verified": true
    },
    "benchmark": "s46-mvp-gsm8k-slice-v0",
    "correct": 1,
    "generator": "s46-evidence-eval",
    "generatorVersion": "0.1.0",
    "score": 0.3333333333333333,
    "todo": "Replace MVP slice with Sovereign46 production benchmark set: LiveCodeBench date-filtered, GPQA Diamond, SWE-bench Verified, AIME, IFEval, FrontierMath, HLE.",
    "total": 3
  }
}