Wait for the grip.
Keep a way forward.

What changed

The authored supervisor allows approach and centering, then waits for supported secure contact. Its total wait budget is 200 ms. Exhausted budgets and loss of support produce an explicit abort; they do not force an insecure lift.

Bounded minus learned: +7.81 percentage points (adjusted interval −7.25 to +22.87). Bounded minus fixed: −1.56 points (−12.44 to +9.32). Both comparisons are inconclusive.

Four interruptions are unknown observations. The primary analysis conservatively imputes failure; a sensitivity analysis removes the entire affected XY group. Timing sessions occurred on different dates under shared load. Unequal pulse exposure prevents a slip-recovery advantage claim.

Every scheduled attempt

Inspect outcomes and source identities. Unknown means the session ended before complete evidence existed; it is not an observed controller failure.

Loading the compact evidence…

Read the boundaries

One simulated rigid 40 g block, two known goals, and a narrow starting envelope. Privileged simulator contacts and state, rather than cameras or human videos. The acquisition latch does not continuously certify retention. There is no hardware, real-time, or general video-to-skill claim.

The compact bundle omits native arrays, complete traces, and the full raw report. Those remain retained locally and are identified by hashes. This publication does not upload that raw archive.

• Research note
• All outcomes and both statistical analyses
• 556 audit identities
• Four interruption manifests
• Source lineage
• Release scope · Byte inventory · MIT License