{
  "recorded_date": "2026-10-09",
  "run": "clean-034203",
  "hardware": {
    "soc": "Apple M5 Max",
    "gpu_cores": 40,
    "unified_memory_gib": 128,
    "os": "macOS 27.2"
  },
  "engines": {
    "fastkernel": "1.1.3",
    "lithos-metal": "0.1.2"
  },
  "configuration": {
    "temperature": 0,
    "thinking": false,
    "output_cap_tokens": 1024,
    "order": [
      "fastkernel",
      "lithos-metal",
      "lithos-metal",
      "fastkernel"
    ],
    "prompts": 10,
    "requests": 40,
    "repetitions_per_engine": 2,
    "other_model_server": "stopped"
  },
  "metric": "(completion_tokens - 1) / (last content event - first content event), client-observed",
  "aggregation": "geometric mean of paired per-prompt rate ratios",
  "ratio": 1.3647826479214478,
  "prompt_bootstrap_interval": [
    1.262,
    1.479
  ],
  "identity": {
    "fastkernel_binary_sha256": "400a6adc15589296f48a939b6373d989a2bee3489223f5253faad8b35aafa086",
    "fastkernel_metallib_sha256": "232e2bb1fc253e173e119dae66267480b223ff920604e61bf56b2c44d9028e7d",
    "lithos_package_sha256": "269ccccef3512f2ef3acbae5d3a57030dc8641ba93cc223ec7d34dfb5b934fef",
    "lithos_target_revision": "482ca0f3832238542f8f5295dde86b5f22711d80",
    "lithos_draft_revision": "b169bc4c8dc805a9e6178fa82f6a7b84b169a457"
  },
  "limits": [
    "Different quantized checkpoints and drafters; task-quality parity not evaluated.",
    "One session; interval resamples prompts, not sessions.",
    "Four capped prompts in each engine; not completed-answer latency.",
    "Long-context prompts excluded.",
    "Lithos warm-up plateau label unsupported with two observations.",
    "Reported demo settings are not fully known; this is not an exact reproduction of its complete response."
  ]
}