{
  "project": "entropy",
  "sha": "901ddb7d2f2c92dca1b3bf57ea9fbf83819e5ad7",
  "baselineName": "SQLite (system libsqlite3, identical Google Benchmark harness)",
  "conditions": {
    "n": "1k and 10k rows per case; 1 repetition (Google Benchmark auto-iterations)",
    "machine": "Shrees-MacBook.local, 8 CPUs, release build",
    "compiler": "C++20, CMake Release (Google Benchmark harness, ENTROPY_BENCH_COMPARE_SQLITE=ON)"
  },
  "series": [
    {
      "workload": "Insert batch (1,000 rows, single txn)",
      "subject_value": 940970,
      "baseline_value": 1048270,
      "ratio": 0.897638,
      "unit": "ns/op",
      "lowerIsBetter": true
    },
    {
      "workload": "Insert batch (10,000 rows, single txn)",
      "subject_value": 9693950,
      "baseline_value": 6992260,
      "ratio": 1.38638,
      "unit": "ns/op",
      "lowerIsBetter": true
    },
    {
      "workload": "Point select (1,000 rows)",
      "subject_value": 46041,
      "baseline_value": 23019.7,
      "ratio": 2.00007,
      "unit": "ns/op",
      "lowerIsBetter": true
    },
    {
      "workload": "Point select (10,000 rows)",
      "subject_value": 459723,
      "baseline_value": 179783,
      "ratio": 2.5571,
      "unit": "ns/op",
      "lowerIsBetter": true
    }
  ],
  "notes": "ratio = entropy / sqlite (lower is better). Entropy is slightly faster on 1k batch insert (0.90x) but is honestly reported as slower on 10k batch insert (1.39x) and on point selects (2.00x at 1k, 2.56x at 10k) — the Volcano one-tuple-at-a-time model trades point-lookup latency for composability (ADR-005). Single-repetition results; rerun with --benchmark_repetitions for p50/p95. SHA is the pinned published commit from facts.yaml (run JSON carries no git_sha)."
}