{
  "id": "lazy-speech",
  "story": "moss-tts-hq4-int8-codec-lazy-server-2026-09-19",
  "title": "First request and warm request are different states",
  "unit": "seconds",
  "rows": [
    {
      "label": "Cold first render",
      "value": 14.8,
      "highlight": false
    },
    {
      "label": "Warm render",
      "value": 5.7,
      "highlight": true
    }
  ],
  "sources": [
    {
      "path": "moss-tts-hq4/int8-codec-lazy-server-2026-09-19/README.md",
      "url": "https://github.com/groxaxo/experimentos/blob/6550ead3945b7ecacbaac1e0767edc49a9de9747/moss-tts-hq4/int8-codec-lazy-server-2026-09-19/README.md",
      "sha256": "0938fd62f3256af814035b581fca2bbca6f746131f757c231c55be82542441b5"
    }
  ],
  "revision": "6550ead3945b7ecacbaac1e0767edc49a9de9747",
  "scope": "HQ4 + INT8 codec lazy service · recorded request timings",
  "caveat": "Cold start includes service wake-up. This is not two equivalent decoding workloads.",
  "kind": "bar",
  "maximum": null,
  "threshold": null,
  "decimals": 1,
  "origin": "reported live-service timings"
}