Updated airborne findings

41 globe preferences, zero flat, 12 unresolved among the same 53 preselected stretches. The original study supplied 12 complete comparisons; numerical recovery adds 29, including the three previously recovered in the qualification trial. All forty-one favor the specified background-field globe over the zero-field globe in both processing cases. This does not identify the field's physical cause or measure a solar-versus-sidereal difference.

These are numerical model preferences. Statistical significance and uncertainty remain uncalibrated. The original observations, selected intervals and first-pass results are preserved.

The twenty-nine recovered stretches

The twenty-nine stretches add 237.91 minutes of completed comparisons. Completed findings include 2 IMU8 and 39 IMU21 stretches. These codes identify instrument/data configurations, not unique unit serial numbers or different experiment protocols. Their conversion scales are checked separately; the selection and physical model comparisons are the same.

The two October 20 recordings overlap in time and must not be counted as two independent flights. The forty-one resolved stretches contain 41 distinct recording hashes across 23 dates, not forty-one established independent replications. All 53 stretches list current dispositions, courses and latitudes. The current inventory preserves exact sources, hashes, intervals, speeds and result origins.

What changed in the calculation

The physical models, gyro-offset bounds, receiver weights and integration clock are unchanged. The revised solver checks its final correction with a fresh full-model prediction in metres, rather than rejecting it solely from a local residual approximation. The gradient and objective-change limits remain; predicted movement must be at most 1 mm RMS per axis and 1 cm at any receiver epoch. These are numerical stopping tolerances, not sensor accuracy claims.

The trial passed 24/24 fits, including 15 primary recoveries, six controls and three predetermined perturbed-start repeats, using 393 native sweeps. Each newly completed stretch has all six model/case fits and a passing repeat. Repeat agreement demonstrates numerical repeatability under these starts; it does not prove a global minimum. Individual results and the finite plan retain the evidence.

Completed wider recovery and remaining failures

The wider recovery completed all 159 new starts and 22,355 native integration sweeps, with 117 passing fits: 84/115 primary recoveries, 27/38 repeats and all six controls. There were no execution errors. The combined tally improves from 15/0/38 after the subset to 30/0/23. Some primary and repeat checks failed, so the entire run did not qualify every comparison; only individually complete comparisons contribute to the findings. Incomplete calculations remain unknown, rather than evidence for either shape.

Qualified wider-arithmetic recovery

The paired precision trial passed all 16 fits, including both controls and both perturbed repeats, in 129 native sweeps. It adds the October 24, 2009 IMU8 and October 7, 2015 DMS stretches, bringing the tally to 32/0/21. Wider internal arithmetic removes rounding sensitivity in these cost-only stopping failures; measurements, physical models, calibration bounds, clock, receiver weights and stopping tolerances are unchanged. Neither repeats nor probes replace failed original fits. All earlier outcomes remain available.

The completed cost-only recovery passed 83/86 fits in 891 sweeps and about four hours: 69/72 primaries, both controls and all twelve repeats. Nine additional stretches pass their complete comparisons, giving 41/0/12. Three primary fits exhausted their forty-evaluation allowance, so those stretches remain unresolved; the nine excluded stationarity/position-failure stretches remain unchanged. The run did not qualify every target. Only individually complete comparisons are reported.

Next are bounded continuations of the three budget failures, followed by a separate investigation of the other nine. All stretch metadata retains last costs, stopping reasons and recorder-pair differences. No recorded interval or physical bound was changed to obtain these recoveries.