{
  "version": 1,
  "recordedOn": "2026-08-05",
  "skill": "responsive-release-proof",
  "runner": "codex",
  "command": "skillbench challenge examples/generated/responsive-release-proof --runs 2 --seed 29",
  "verdict": "proven",
  "thresholds": {
    "minSkillScore": 0.9,
    "minDelta": 0.2
  },
  "metrics": {
    "pairedRuns": 2,
    "averageBaselineScore": 0.25,
    "averageSkillScore": 0.95,
    "averageDelta": 0.7,
    "deltaStdDev": 0.1,
    "averageBaselineDurationMs": 97977,
    "averageSkillDurationMs": 141627.5,
    "durationDeltaPercent": 0.4455,
    "averageBaselineTokens": 237568.5,
    "averageSkillTokens": 299183.5,
    "tokenDeltaPercent": 0.2594,
    "improved": 2,
    "unchanged": 0,
    "regressed": 0
  },
  "runs": [
    {
      "run": 1,
      "order": ["skill", "baseline"],
      "baselineScore": 0.3,
      "skillScore": 0.9,
      "delta": 0.6
    },
    {
      "run": 2,
      "order": ["baseline", "skill"],
      "baselineScore": 0.2,
      "skillScore": 1,
      "delta": 0.8
    }
  ],
  "note": "This is a small local dogfood result, not a universal benchmark. The recorded score is conservative: run 1 exposed that the candidate scorer selected an earlier failed QA command instead of the final rerun; v0.4 now scores the last matching command."
}