{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-06T14:14:55Z",
 "ollaya": "0.11.0",
 "machine": "rtx-4090",
 "machine_detail": {
  "id": "rtx-4090",
  "cpu": "13th Gen Intel(R) Core(TM) i9-13900K",
  "threads": 24,
  "gpus": [
   {
    "name": "NVIDIA GeForce RTX 4090",
    "driver": "616.92",
    "memory": "24564 MiB"
   }
  ],
  "os": "Linux 6.18.33.2-microsoft-standard-WSL2"
 },
 "device": "cuda",
 "protocol": {
  "preset": "triage",
  "questions": 5,
  "states": 30,
  "warm": 5,
  "n": 20,
  "client": "one request at a time over HTTP to /api/decide, timed by the client; each request a different short message, in turn"
 },
 "note": "snap:2b-q4_k_m, snap's own default file, measured with the Ollaya 0.11.0 release binaries before its registry manifest shipped; the same protocol and machine as the 2026-10-01 RTX 4090 sweep.",
 "results": [
  {
   "model": "snap:2b-q4_k_m",
   "questions": "triage",
   "load_ms": 569.4,
   "first_request_ms": 634.8,
   "answered_by": "snap:2b-q4_k_m",
   "p50_ms": 62.7,
   "p90_ms": 63.6,
   "min_ms": 61.7,
   "eval_p50_ms": 62.1,
   "input_tokens": 606,
   "device": "cuda:0",
   "precision": "Q4_K - MEDIUM"
  }
 ]
}