E4 / Observation freshness and sampling

01 / QUESTION

Separate delivery from fresh evidence and test a fixed count at different sampling periods.

METHOD

Three periods (10/20/50 ms) × five conditions × three policies = 45 deterministic rollouts, one per cell. Physics/control: 2 ms; age limit: 60 ms; bad-evidence span: 40 ms.

RESULT

After forced opening, count alarms take 22/42/102 ms and duration alarms 42/42/52 ms. Only the age check alarms on replayed good packets and silence.

OBSERVATIONS

Only 10ms-count3 falsely cancels a 40 ms empty-report pulse. Age alarms occur after 52/42/12 ms because captures have different ages at fault onset; slower sampling is not established as safer.

CONCLUSION

Check packet identity, evidence content and capture age separately. Ongoing delivery does not guarantee freshness, and a stale alarm alone does not prove a drop.

LIMITATIONS

One simulation clock, fixed fault phase, one rollout per cell, transfer only. No clock skew, real network jitter, vision, recovery or hardware safety validation. Every forced-opening condition still fails placement.

REPRODUCE

17 new and 24 existing tests passed; 45 independently audited rollouts, 45 matching prefixes and nine reproduced baselines. 180 committed raw-file hashes match.

1. Engineering cases

   Embodied practice IV: messages arrive, but observations expire

   45 MuJoCo rollouts separate sampling intervals, bad-sample counts and capture age, showing how old messages can hide grip loss and why duration thresholds still have limits.

   2026-09-2914 MINPublished
2. Engineering cases

   Embodied practice III: a confirmed grasp still needs monitoring

   A 27-rollout MuJoCo comparison of transient observation loss and forced opening, with false cancellations, a 40 ms confirmation delay, code and recorded trajectories.

   2026-09-2914 MINPublished
3. PROJECT / 003

   Hohoo Embodied Experiments

   Seven MuJoCo teaching experiments, from pickup bias to stage contracts, with raw trajectories, decisions and independent audits.

   2026-09-28Experiment
4. LAB / 007

   E3 / Contact monitoring during transfer

   Separate missed loss, false cancellation and confirmation delay using forced opening and one empty contact report.

   2026-09-29Completed