Readable motion needs more than clean geometry

The separate October 21 service candidate

CRUMB’s independent tray props changed support only after the upper tray reached the existing console. The served tray remained while the droid withdrew, and persisted through the combined cleaning scene. Source limbs were not stretched or detached to invent a reach.

A moving-frame sweep followed an actual neutral-matte edge failure that a single source-pixel alpha probe had missed. The candidate recorded core-gap and contaminated-boundary checks as well as actual scene playback. MOPS travel used one floor projection for depth, scale, contact, shadow and deposited marks.

The DDS cutout’s lowered paw could not make the required delivery contact. Its full exchange stayed blocked. This local candidate was not deployed to the public Command Center or Pi runtime.

Model studies stayed review candidates

CRUMB’s secondary-hat study used a damped approximation with a fixed lower band, not a live cloth-collision simulation. A copied baseline action binding had been wrong in an early comparison harness; matched-yaw comparisons and saved-candidate evaluation corrected the evidence.

DDS hand and wheel checks sampled the complete turn and travel, with no intersecting hand/wheel triangle pairs in the recorded bounded audit. A modest decoded preview showed assembled forms and a gentle amber beacon. CODD’s requested 01 marking stayed attached through its turn review. These tests did not establish final artwork fidelity or accepted service animation.

MOPS modeling also exposed an export defect: an unsupported material-node pattern lost the blue enamel color. Native reimport rendering caught it; direct PBR texture inputs corrected that export path. This did not certify browser integration or a finished production mesh.

The letter action: contact, then visual continuity

The October 22 v73 approval covered the narrow two-view reinsertion test. On October 23, the v81 flap correction was accepted after actual rendered views exposed continuity errors missed by earlier geometric checks.

The later shadow twitch was real baked head yaw, caused by a close-target iterative gaze solution that could branch between opposite yaw values. A stable bounded attention bearing and smoothing corrected the head and iris channels while preserving other actor and prop animation.

The accepted v84 review removed 36 camera frames that cut away during reading. It reused the other 1,962 frames from the 1,998-frame, 83.25-second review. The camera change preserved non-camera animation channels. Actual encoded frames were decoded and inspected; that evidence is not described as real-time player viewing.

Scoped face and fur progress

The furred blink/seal and calm/listening checkpoints were approved for carry-forward at 00:07 UTC on October 24, which was the evening of October 23 in New York. Combined movement received positive feedback later that same Eastern evening. Final likeness and production adoption remained excluded.

A later cheek-groom study was a static neutral-frame comparison. It was not rebaked into the accepted motion cache. A clean static fur result cannot stand in for deformation or animated playback QA.

The boundary that stayed in place

The letter review used an explicitly labeled Ash stand-in. Its acceptance did not approve final Ash artwork, a full speech/viseme rig or production integration. All earlier candidates, failures and accepted assets remained preserved. The recurring lesson is to inspect what the audience actually sees, after checking the mechanism.