{
  "status": "conventional restoration baseline completed; integration unresolved",
  "reviewer": "agent; not independent artistic evaluation",
  "scope": "Same 92-frame development shot; no additional generation or paid API call.",
  "authorship": "User agreed to proceed with selective source restoration; the agent selected the existing character-and-prop diagnostic masks for the baseline. Exact preservation was an agent-authored probe, not a universal aesthetic requirement.",
  "reviewed_frames": [
    83,
    101,
    119,
    125,
    143,
    161,
    171,
    174
  ],
  "findings": [
    {
      "item": "Original appearance",
      "finding": "Source hair, face, body and small prop visibly return in sampled frames. Opaque mask interiors equal native source PNGs before video encoding by construction."
    },
    {
      "item": "Generated environment",
      "finding": "Frozen-world appearance remains outside the source coverage. The raw model output is untouched and separately retained."
    },
    {
      "item": "Changed silhouettes",
      "finding": "The redrawn character is not removed before source-over restoration. Exposed generated contours and limbs remain where it extends outside native coverage; inspect recovery/jump frames 161\u2013174."
    },
    {
      "item": "Optical integration",
      "finding": "Source render detail and warm edge colors differ from the smooth, cool generated world. Diagnostic masks do not reconstruct correct optical alpha."
    },
    {
      "item": "Contact and shadows",
      "finding": "Generated surroundings and shadows remain while native pose pixels are restored. Correct new-world contact, shadow correspondence and occlusion have not been established."
    },
    {
      "item": "Prop motion",
      "finding": "Source prop pixels are copied inside diagnostic coverage. This does not establish removal of all generated prop traces or preservation of surrounding effects."
    }
  ],
  "interpretation": "Redraw is acceptable as a model interpretation. Whether it is desirable is an authorial choice. This baseline tests the separate ability to keep selected source decisions; its copy equalities do not prove model obedience or a new technical capability.",
  "next_experiment": "Before another paid generation, construct and review a removal/reconstruction region covering the generated character as well as the native one. Restore the selected native character only after clearing displaced generated pixels; then evaluate edges, contact and illumination. Ordinary compositing remains the comparison to beat.",
  "not_claimed": [
    "Production-ready composite",
    "Temporal or optical certification",
    "Improvement over a skilled compositor",
    "Novel preservation algorithm"
  ]
}