{
  "arm": "ARM 1 / styles01",
  "suite": "Decode speed",
  "metrics": {
    "code": {
      "score": 91.2,
      "n": 3,
      "unit": "tok/s",
      "status": "verified"
    },
    "prose": {
      "score": 49.5,
      "n": 3,
      "unit": "tok/s",
      "status": "verified"
    },
    "structured": {
      "score": 102.0,
      "n": 3,
      "unit": "tok/s",
      "status": "verified"
    }
  },
  "config": {
    "thinking": true,
    "max_gen_toks": 512,
    "temperature": 0
  },
  "evidence": {
    "engine": "styles01 vLLM + TensorFlash",
    "image": "ghcr.io/styles01/dsv41-1017-exl3-2x-dgx-sparks:v2",
    "kv": "2.68 GB / rank",
    "context": "262K server context; vision on",
    "source": "Operator-verified summary supplied 2026-10-21",
    "raw": "/fixtures/styles01-speed.json",
    "note": "Median of 3 greedy 512-token trials. Original per-trial speed artifact not located; timestamp is archive date, not execution time."
  },
  "ts": "2026-10-21T00:00:00Z",
  "hardware": "2 \u00d7 NVIDIA DGX Spark / TP2 / EXL3 2.9bpw"
}