{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "name": "SponsorSkip local sponsor detection benchmark",
  "runDate": "2026-08-25",
  "status": "passed-release-gate",
  "command": "npm run test:architectures",
  "runtime": {
    "node": "22.23.1",
    "architecture": "arm64",
    "cpuFamily": "Apple M3"
  },
  "corpus": {
    "videos": 24,
    "transcriptHours": 17.2,
    "truthSponsorSegments": 29,
    "fullyReportedLanguages": ["English", "German", "French"],
    "scope": "Fixed hermetic regression corpus with multiple creator families and adversarial negative cases"
  },
  "architecture": {
    "id": "D",
    "name": "stable anchors + narrative-island change points"
  },
  "result": {
    "matchedSponsorSegments": 29,
    "falsePositiveSegments": 0,
    "missedSponsorSegments": 0,
    "derivedSegmentPrecision": 1,
    "derivedSegmentRecall": 1,
    "aggregateContentLossSeconds": 7.7,
    "aggregateSponsorResidualSeconds": 6.127,
    "weightedLoss": 776.127,
    "detectorCpuMilliseconds": 10598.2,
    "detectorCpuMillisecondsPerVideoHour": 616.3
  },
  "releaseGate": {
    "maximumFalsePositiveSegments": 0,
    "maximumMissedSponsorSegments": 0,
    "maximumAggregateContentLossSeconds": 8,
    "maximumAggregateSponsorResidualSeconds": 12,
    "maximumWeightedLoss": 812,
    "passed": true,
    "failedChecks": []
  },
  "observedBoundaryExamples": [
    {
      "case": "History · Silver",
      "type": "boundary under-skip",
      "truthSeconds": [41.1, 211],
      "predictionSeconds": [41.1, 211]
    },
    {
      "case": "SourSweet",
      "type": "boundary under-skip",
      "truthSeconds": [42, 128.3],
      "predictionSeconds": [40.7, 126.4]
    },
    {
      "case": "Riley Brown · HubSpot",
      "type": "coarse-caption boundary overrun",
      "truthSeconds": [275, 319],
      "predictionSeconds": [270, 319]
    }
  ],
  "interpretation": [
    "The result describes this fixed corpus and current worktree only.",
    "Derived precision and recall are corpus-level segment metrics, not calibrated per-skip probabilities.",
    "Runtime is a single local development measurement and should not be generalized to every device.",
    "Passing this fixed regression gate does not establish performance across every creator, language, or caption style."
  ]
}