E9 / Post-release completion verification

01 / QUESTION

Apply single-capture and sustained-window checks to five actual trajectories, including forged observations and post-completion disturbance.

METHOD

Five physical trajectories scored by single-capture and 250 ms-window observers without control feedback.

RESULT

A forged capture fools the single rule, not the window. Later force invalidates final placement after both had accepted.

OBSERVATIONS

Clean completion at 6.842/7.102 s;20 ms sampling makes the actual window span 260 ms.

CONCLUSION

Completion events and current validity differ; a sustained window does not guarantee future stability.

LIMITATIONS

Privileged simulated observations and a synthetic capture fault; no realistic vision error or validity-revocation implementation.

REPRODUCE

See the pinned repository for code, frozen parameters, raw results and independent auditing.

1. Engineering cases

   Embodied AI IX: a release command is not verified placement

   Apply single-capture and sustained-window checks to five actual trajectories, including forged observations and post-completion disturbance.

   2026-10-039 MINPublished
2. PROJECT / 003

   Hohoo Embodied Experiments

   Two MuJoCo teaching scenes—pick-and-place and planar force control—record contracts, tuning, clearance and delayed-feedback outcomes, including failures.

   2026-09-28Experiment