elizaos--eliza/packages/benchmarks/lifeops-quality/baseline.json

{
  "_comment": "Recorded reference run for the LifeOps quality benchmarks (#10723). TRIAGE: the committed fixed-quality mock model (triage/fixtures.ts, 7 deliberate errors over the 56-item corpus) driven through the REAL plugin-inbox classifier parse/validate path — fully deterministic, so the gate asserts an EXACT match against this block; editing the corpus or fixtures requires re-recording it in the same change (run `bun run bench:triage`, copy results/triage-results.json). TIMELINESS: the REAL plugin-personal-assistant scheduled-task tick replayed over two 4-day 2026 DST windows at a 5-minute cadence (2×1153 ticks); the gate enforces budgets.json (counts exactly zero, deviation ceilings) rather than exact equality because cron-next-fire walks Intl tzdata. The unit lane cross-checks this file against budgets.json so the recorded reference always satisfies the committed floors.",
  "triage": {
    "corpusSize": 56,
    "correct": 49,
    "accuracy": 0.875,
    "macroF1": 0.875545002618308,
    "perClass": {
      "ignore": {
        "precision": 1.0,
        "recall": 0.8333333333333334,
        "f1": 0.9090909090909091,
        "goldCount": 12,
        "predictedCount": 10
      },
      "info": {
        "precision": 0.6923076923076923,
        "recall": 0.9,
        "f1": 0.7826086956521738,
        "goldCount": 10,
        "predictedCount": 13
      },
      "notify": {
        "precision": 0.8888888888888888,
        "recall": 0.8,
        "f1": 0.8421052631578948,
        "goldCount": 10,
        "predictedCount": 9
      },
      "needs_reply": {
        "precision": 0.8666666666666667,
        "recall": 0.9285714285714286,
        "f1": 0.896551724137931,
        "goldCount": 14,
        "predictedCount": 15
      },
      "urgent": {
        "precision": 1.0,
        "recall": 0.9,
        "f1": 0.9473684210526316,
        "goldCount": 10,
        "predictedCount": 9
      }
    }
  },
  "timeliness": {
    "windows": {
      "spring-forward": {
        "tickCount": 1153,
        "expectedFires": 159,
        "actualFires": 159,
        "missedFireCount": 0,
        "duplicateFireCount": 0,
        "earlyFireCount": 0,
        "occurrenceMismatchCount": 0,
        "maxDeviationMs": 180000,
        "meanDeviationMs": 140377.35849056602
      },
      "fall-back": {
        "tickCount": 1153,
        "expectedFires": 160,
        "actualFires": 160,
        "missedFireCount": 0,
        "duplicateFireCount": 0,
        "earlyFireCount": 0,
        "occurrenceMismatchCount": 0,
        "maxDeviationMs": 180000,
        "meanDeviationMs": 139125
      }
    }
  }
}