{
  "schema": "phntm-public-latency/1",
  "measured_date": "2026-10-30",
  "software": "0.9.0",
  "build_commit": "9339b2915a0c456340c388933d06b36aea68aaae",
  "unit": "review unit, Raspberry Pi 5 8 GB, 256 GB microSD",
  "release_identity_source": "Installed release identity read from the device API (/api/status release.commit, source BUILD_COMMIT stamp) and the BUILD_COMMIT file before and after each run; the benchmark header of both receipts records the same commit (9339b2915a0c456340c388933d06b36aea68aaae), so release identity and benchmark header agree for this set.",
  "source_receipt_sha256": {
    "warm": "638339012d550a19ee81956059799d3f1ba700abcd608395bf64deaf2cbdb500",
    "after_model_eviction": "568f6f59e9a541c18d3db887292715b3ed25607b0674256e876857976b4dc386"
  },
  "method": "Scripted local chat on one Raspberry Pi 5 review unit with 8 GB RAM and Gemma 3 4B, Private mode, owner language English, 30-second pauses between turns; timing from the request until the first nonempty text received from the chat stream and until the completed answer. Screen rendering time was not measured. No cloud turns. Exploratory observations, not a customer journey, retail qualification or response-time promise.",
  "warm": {
    "total_conversation_turns": 10,
    "model_turns_in_statistics": 8,
    "excluded": "The first uncached-prompt turn (reported separately below) and 1 deterministic device reply (intercept, answered without the model).",
    "percentile_method": "nearest rank over model turns after the first; exploratory",
    "first_text_seconds": {
      "median": 19.3485,
      "p95": 29.039
    },
    "completion_seconds": {
      "median": 27.7315,
      "p95": 35.858
    },
    "answer_validity": "Every answer was read: all 8 model answers in the statistics were real answers in the language of the question.",
    "all_ten_turns": {
      "first_text_seconds": {
        "median": 19.3485,
        "p95": 86.689
      },
      "completion_seconds": {
        "median": 27.7315,
        "p95": 94.395
      },
      "note": "Includes the first uncached turn and the deterministic reply; not a model speed figure."
    }
  },
  "first_uncached_prompt": {
    "observations": 1,
    "first_text_seconds": 86.689,
    "completion_seconds": 94.395
  },
  "after_model_eviction": {
    "observations": 1,
    "first_text_seconds": 126.605,
    "completion_seconds": 130.684,
    "note": "Chat model unloaded from memory, then one question; not a power-on or reboot."
  }
}
