[Tarzelf]

A goal image can look finished while a shoe is still moving or held by the robot. This adds 32 native PiPER fixtures, their physical references, an offline current/goal viewer and a frozen 64-trial Jev evaluation protocol. Four settled/moving pairs have identical complete RGB request hashes and different physical labels.

Jev has not been run. All provider trials remain not_run; model accuracy, false-completion rates, latency and cost are explicitly unmeasured. Native validation recomputes every physical label and full-window telemetry extrema. The work is a passive synthetic diagnostic, with no model weights, learned controller or hardware qualification.

The complete evidence package includes the original native inventory, evaluation plan, independent validation receipt, runnable verifier, MIT notices and separate viewer/publication checksums. Existing research remains available; all 25 earlier paper bodies and prior source evidence are preserved.

Validation: 69 research tests; catalog build and link validation; three focused evidence checks; native independent reproduction (32 cases, 67 original files); 42 training/report tests and nine publisher tests. Reviewed viewer bytes and frozen native/evaluation files remain unchanged. Hugging Face publication retains historical files and verifies all uploaded bytes.

Source: Dvidia-Inference/dvidia-training#23, native core 19acf0734d63ecebc07140c8d84302a54baee4c4, publication code 3d3c8f660f6f9544d6a58c654bab143dd1eddc00.