Calibration stopped at a numerical check

348 of 7,176 planned starts returned results in 6.82 hours. Of these, 344 passed the numerical checks, two failed and two stopped cooperatively. The remaining starts were not run. No calibration threshold was established, so independent validation was not launched.

All 346 completed forecasts met the unchanged position-error limits. These are generated measurements: each candidate fits data generated for that same candidate. They are neither additional aircraft observations nor votes for an Earth model.

Both prescribed starts failed for the same generated case. They reached nearly identical fitted values and later prediction errors: about 0.70 m horizontally and 1.60 m vertically. However, eight of nine timing-cell verification probes failed the required stability checks. Good prediction errors alone cannot replace numerical qualification.

The saved records identify a specific diagnostic lead: those eight linear verification proposals have higher predicted costs than their own anchor points. The ninth proposal passes. The next step is to check the bounded linear calculation and its handling of extremely narrow timing intervals. This is a lead, not a demonstrated cause or permission to relabel the failed fits.

The complete cohort, original acceptance limits, failed results and unused allowance remain closed. The 53-stretch airborne findings are unchanged.

Phase report · Every planned start · Frozen plan · Next analyses.