Completed comparison: verified numerical stopping

Completed: five of six fits converged; qualification did not pass. The comparison preserved the five completed fits but did not finish the remaining flat-disc fit. Its local correction exceeded the declared residual-step stability limit by a factor of 1.55. The physical data and assumptions stayed the same. See the matched results and saved failure diagnosis.

An IMU (inertial measurement unit) records turning and acceleration. This trial uses one saved airborne recording to qualify a numerical revision; the archive still contains 53 selected stretches. The diagnosis explains why the previous attempt remains unfinished. The plan gives the unchanged science, stopping tests, six-start ceiling and estimated 55–65 minute budget.

Preparation and ordinary software tests start no airborne fits. The separate six-start allowance is spent; no replication or full-corpus starts are included. Replication over the two preselected recordings remains deferred pending numerical qualification. No frozen failure is relabeled and no retries are authorized.

The tested watch_report.py completed publication automatically after the worker finished and released its lock. Both processes are complete; do not relaunch them. The report can be regenerated from saved metadata using build_report.py with PYTHONPATH=analysis; it performs no fits or retries.