{
  "turns": [
    {
      "turn": 1,
      "transcript": "I have to tell you something. Something really sad just happened to me today.",
      "reply": "I am so sorry to hear you're going through a tough time today. Please know I am here to listen whenever you feel ready to share.",
      "mode": "fresh",
      "phone_request_to_first_pcm_ms": 4794.748956,
      "phone_request_to_visible_ms": 5577.661039,
      "phone_pcm_to_model_ms": 496.0628649999999,
      "phone_model_to_rgb_ms": 171.15552099999968,
      "phone_rgb_to_visible_ms": 115.6936969999997,
      "server_stt_ms": 1186.3531099999989,
      "server_llm_first_text_ms": 1068.237244999999,
      "server_llm_first_text_to_fish_audio_ms": 511.46673299999884,
      "server_handler_to_first_audio_ms": 2877.2048690000033,
      "unattributed_network_startup_residual_ms": 1917.544086999997,
      "headers_ms": 2081.577656,
      "first_event_ms": 3210.947968,
      "generated_presented": 89,
      "audio_underruns": 0,
      "compute_fps": 17.513797386799943
    },
    {
      "turn": 2,
      "transcript": "Okay, now please cheer me up. Tell me a joke and make me laugh out loud.",
      "reply": "Why don't scientists trust atoms? Because they make up everything!",
      "mode": "pooled",
      "phone_request_to_first_pcm_ms": 3942.924218,
      "phone_request_to_visible_ms": 4383.838124,
      "phone_pcm_to_model_ms": 129.68926999999985,
      "phone_model_to_rgb_ms": 151.15812499999993,
      "phone_rgb_to_visible_ms": 160.066511,
      "server_stt_ms": 1473.2533060000037,
      "server_llm_first_text_ms": 760.1281190000009,
      "server_llm_first_text_to_fish_audio_ms": 499.1530409999996,
      "server_handler_to_first_audio_ms": 2789.2732359999995,
      "unattributed_network_startup_residual_ms": 1153.6509820000006,
      "headers_ms": 1090.898072,
      "first_event_ms": 2605.408645,
      "generated_presented": 51,
      "audio_underruns": 0,
      "compute_fps": 18.17920894361416
    }
  ],
  "recording": {
    "filename": "conversation-device-av.mkv",
    "duration": "81.131000",
    "streams": [
      {
        "type": "video",
        "codec": "h264"
      },
      {
        "type": "audio",
        "codec": "opus"
      }
    ]
  },
  "limitations": [
    "Question audio is a prerecorded test fixture played by the phone, not a live microphone",
    "Replies are generated by STT + Gemini + Fish with actual prior-turn history",
    "Two conversation turns differ in input text; fresh vs pooled comparison here is not controlled A/B",
    "Recording audio is captured from device output, not post-muxed; content must still be inspected"
  ]
}