{
  "model": "banyan-sapling",
  "release_sha256": "245970b0e5b0b22d64c1434ee21dd221b1f211f06bda5465f8fc108f70cd5c3f",
  "scope": "Postrun HTTP transport and exact reference parity only; no model quality or nominal coverage claim.",
  "passed": true,
  "production_ready": false,
  "fresh_reload": {
    "release_sha256": "245970b0e5b0b22d64c1434ee21dd221b1f211f06bda5465f8fc108f70cd5c3f",
    "fresh_process": true,
    "exact": true,
    "max_absolute_drift": 0.0,
    "rows": 53,
    "reverse_exact": true,
    "mixed_exact": true
  },
  "http_sequential": {
    "scope": "All 53 fixed mixed reload references, one request at a time; includes known diagnostics and short/long held-out cases. Not a general latency benchmark.",
    "http_latency_ms": {
      "n": 53,
      "median": 97.08157597924583,
      "p95": 111.32656197878532,
      "min": 95.18892501364462,
      "max": 447.7676579845138,
      "p95_method": "nearest_rank"
    },
    "model_latency_ms": {
      "n": 53,
      "median": 95.19620198989287,
      "p95": 107.31824999675155,
      "min": 93.50682701915503,
      "max": 426.2440019811038,
      "p95_method": "nearest_rank"
    }
  },
  "http_concurrent": {
    "clients": 8,
    "effective_model_batch_size": 1,
    "scope": "One synchronized wave of the eight shortest references. HTTP clients overlap; model executes serially. Not throughput capacity or full-suite concurrency coverage.",
    "wave_elapsed_ms": 771.4673829905223,
    "http_latency_ms": {
      "n": 8,
      "median": 433.0706769833341,
      "p95": 767.8734929941129,
      "min": 103.20778499590233,
      "max": 767.8734929941129,
      "p95_method": "nearest_rank"
    },
    "model_latency_ms": {
      "n": 8,
      "median": 94.33105499192607,
      "p95": 99.82317499816418,
      "min": 93.90006499597803,
      "max": 99.82317499816418,
      "p95_method": "nearest_rank"
    }
  }
}