{
  "date": "2026-10-24",
  "hardware": "NVIDIA A10G 24GB",
  "weightsChanged": false,
  "referenceDefault": true,
  "executionOptions": {
    "compile": "layer",
    "cuda_graphs": false,
    "prune_last": false,
    "fp8": false,
    "prefetch": false,
    "token_budget": 16384
  },
  "finalDecisionRegression": {
    "complete": true,
    "cases": 303,
    "finalMatches": 303,
    "candidateSetChanges": 33,
    "sameJobControls": 4,
    "allSameJobControlsMatch": true,
    "firstFinalMismatch": null,
    "error": null,
    "timings": {
      "reference": {
        "n": 15,
        "medianMs": 2451.9707610002115,
        "p95Ms": 2539.2162020002615
      },
      "compiled": {
        "n": 15,
        "medianMs": 1690.3312350000306,
        "p95Ms": 1733.433431999856
      }
    },
    "medianReductionPercent": 31.062341285403082,
    "p95Ratio": 0.6826647650697676,
    "peakAllocatedBytes": {
      "reference": 9997422592,
      "compiled": 10086032384
    },
    "allocatedMemoryRatio": 1.0088632636246573,
    "localAdoptionEligible": true
  },
  "packagedIntegration": {
    "routingControls": 6,
    "routingMatches": 6,
    "typedRequests": 4,
    "typedQuestions": 12,
    "allTypedWinnersMatch": true,
    "maxProbabilityDelta": 0.0016201982392612946,
    "realAuthenticatedSdkPlainAndZstdPassed": true,
    "compiledPeakAllocatedBytes": 10086030848
  },
  "limitations": [
    "15 warm timing samples per mode on three selected workloads; excludes loading, compilation, HTTP, queuing and concurrent throughput",
    "Final decisions match on the regression; internal candidates and probabilities can differ",
    "Packaged integration is a smaller additional check, not a second full regression",
    "No production CPU capacity or operating-cost measurement"
  ]
}