Recovering the remaining airborne comparisons

Complete: 30 globe preferences, zero flat preferences and 23 unresolved stretches. The run finished all 159 new fits. It preserved the original measurements, selected intervals, three physical candidates, calibration bounds, receiver weights and clock.

The original study found twelve complete globe preferences. The preceding 24-fit qualification recovered three more. This wider run brings the total recovered from the original unfinished population to eighteen, including those three, rather than counting the subset twice. All thirty resolved stretches favor the specified background-field globe over the zero-field globe under both processing assumptions. The comparison does not identify the field's cause or distinguish solar from sidereal rate.

Numerical results

The run used 22,355 native integration sweeps within its fixed 95,718-sweep ceiling, with no retries or execution errors. The entire run did not qualify every comparison: twenty-one stretches have at least one unfinished primary fit, and two more have complete primaries but failed scheduled repeat checks. Those 23 remain unresolved, rather than becoming flat preferences or receiving automatic retries.

The thirty resolved stretches total 234.9 minutes: 29 IMU21 stretches and one IMU8 stretch, drawn from thirty recording hashes across nineteen dates. Overlapping recorders and shared installations prevent treating them as thirty independent flights. See current findings, all 53 stretches, and costs and stopping reasons.

The frozen plan applies the same method to the remaining original unfinished calculations. It reuses 18 qualified results and permits 159 new starts: 115 primary fits, 38 repeated starts and six fresh controls. All physical models, observations and calibration bounds remain unchanged. The stretch inventory records source filenames, hashes, UTC intervals, duration, speed and recorded IMU type; the task ledger identifies new and reused calculations.

The worker uses two numerical threads, records each attempt before execution and allows no retries. Its ceiling is 95,718 native sweeps, estimated at 51–76 hours if difficult fits consume their full allowances. Actual time may be shorter.

Every recovered stretch passed six primary model fits, its scheduled repeat and the fresh controls. Globe/flat is reported first; the background-field comparison follows only for consistent globe preferences. These are numerical model preferences, not calibrated statistical detections.

Reproducibility and next investigation

env PYTHONPATH=analysis OPENBLAS_NUM_THREADS=2 OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 ~/venv/bin/python analysis/tests/ilvis0_position_recovery_worker.py --package-only

The completed study is frozen. Packaging verifies and copies existing evidence without fitting; its unused allowance is closed. The next investigation compares overlapping flight legs whose recorders have different numerical completion outcomes, starting with the three marked pairs in the matched-recorder inventory. Use saved fits and common UTC intervals before preparing a separate finite numerical study.