# Stable-motion revision review This report replaces v9 acceptance for the new picture. The original reference remains unchanged. The requested change is fixed contours and settled poses, not reference motion equivalence. ## Repairs - Fixed ink phase; seeded and resampled paths in object space to prevent texture movement under camera transforms. Output stroke widths remain controlled. - Removed dialogue camera drift and reaction shake. - Removed breathing scale loops, panic body displacement and rapid pupil jitter. - Changed repeated gesture/head accents into a single entrance, hold and release. - Fixed portal silhouettes, floating labels and the decorative debt bounce; kept transport, internal swirl, running, explanatory props, mouths and blinks. - Preserved all 7,889 frame positions, 95 shots, dialogue and captions. ## Source checks completed Native-size source images opened: 190, 466, 1380, 2367, 2840, 3500, 4190, 4780, 6560, 7480, 7840. Faces, hands, labels, captions and source-framing continuity were inspected at these samples. They do not establish every-frame visual coverage. A seven-second encoded preview of frames [2280,2448) played to its natural end in Ego Lite, with a screenshot taken during playback. This verifies playback and sampled appearance; it is not a continuous subjective watch of every frame. The source diagnostic renders [24,48) with fixed geometry: eight background layers and six torso cases have zero differing pixels across all 23 comparisons. In the same test the previous implementation changes 16 background frames and 22 torso frames. Mouths, blinks and intentional action are excluded from the torso comparison. This detects unwanted layer motion; it is not a style-match score. No new subjective audio-listening acceptance is claimed. ## Encoded delivery The complete H.264/AAC file is 1920×1080 at 24 fps, 7,889 decoded frames, 328.708333 seconds, YUV420P / BT.709. Full video/audio decoding completed without errors. The published web file is byte-identical to the local master. Master/web SHA-256: `2669522be2d3ce0aad29f18571f87db5466c45c827b67e5d74cc012a6edc6f37`. Size: 39553875 bytes. AAC packet hashes and decoded PCM hashes match the v9 soundtrack, and audio start time, duration and sample rate are unchanged. Script and timeline files are identical to the retained v9 source. No TTS regeneration or audio time-stretch was performed. All 17 selected decoded samples were opened as a contact sheet (approximately 448×252 per thumbnail in the viewer). Native-size decoded samples opened: 1380, 2367, 4190, 7461 and 7462. The two frames around the recovered segment join retain the same coffee-sign geometry and framing. The selected reaction, insert, dense label, travel boundary and ending samples retain legible text and coherent figures. This is sampled visual coverage, not a watch of every frame. Ego Lite played the final local file at normal speed over 7–11, 96–101, 310–313 and 325–328.708333 seconds, with successful seeking and natural end. Playback state reported 1920×1080 and no media error. Screenshots were captured for all four intervals; reaction and ending screenshots were opened. These checks show playback and sampled appearance; continuous subjective motion and listening coverage is not claimed. The source stability diagnostic above independently checks that fixed backgrounds and torso geometry no longer oscillate. ## Rendering recovery High host memory pressure required saving the first 7,462 encoded frames and rendering the remaining 427 frames separately. The ranges are contiguous and the combined file passes the full 7,889-frame decode. Finalization used shared-lock stream copies and single-thread decoding. The 1080p file is compact enough to serve directly, so no additional web encoding was required. The old v9 export, its historical review and the supplied reference remain intact. The current stable edition supersedes the old line-boil motion direction.