尾巴保宽，动作按事件重编

[ { "id": "tail-control", "hypothesis": "Preserving generated alpha cross-section widths removes V4 curvature thinning.", "roi": "Direct cause of visible defect; reuse all textures and edge meshes.", "result": "397 quarter-frame samples: 0 flips, 0 strip overlaps; max centerline adjustment 4.85 px.", "visual": "Viewed source/V4/V5 at 13,56,81. Thickness no longer collapses; original 62 px texture scale still too thin.", "decision": "Keep controller; test fixed rest width calibration 62 -> 84." }, { "id": "tail-calibration", "result": "9 difficult-pose/transition samples: 0 flips and overlaps; max adjustment 8.18 px.", "visual": "Viewed all three calibrated difficult poses. Shaft becomes fuller without a transient pinch. Generated furry tip still differs from smooth source.", "decision": "Use constant 84 px texture scale. Run full final audit later." }, { "id": "body-pilot", "range": [ 75, 99 ], "roi": "Largest clear head/eye event; avoids spending effort on low-amplitude early motion first.", "evidence": "body-pilot-comparison.jpg and six rendered frames at 75,80,83,88,94,99; reviewed source anchor overlays.", "visual": "Left wrist and neck return toward source event targets; eye opening now explicitly timed around 81–85. No visible joint separation in these poses. Differences from V4 are modest; generated standing body remains unlike source seated pose.", "decision": "Extend accepted event scheme to full 100 frames. Do not claim source motion solved.", "correction": "Reject unreliable cheek tracks. Occluded right wrist follows observed elbow conservatively; not exact recovered motion." } ]