All 53 selected flight stretches

41 globe preferences, 0 flat, 12 unresolved. Current dispositions combine the preserved first study with the completed, qualified recovery run. Each recovered result passed all six primary fits, its scheduled repeat and the fresh controls. Failed or partial comparisons remain unresolved. An unresolved comparison supplies no conclusion about shape. These are numerical preferences; statistical significance has not yet been calibrated.

An inertial measurement unit (IMU) records angular and acceleration increments. Every stretch below uses Applanix Group-4 increments and the primary receiver positions. Recorded IMU types identify configurations, not individual serial numbers. Overlapping tracks are not independent flights.

Selection requires at least four continuous minutes, every supported receiver ground-speed sample at least 700 km/h, and the frozen motion, completeness and configuration checks. Intervals were selected before comparing model residuals. Full eligibility criteria and methodology describe those checks.

Course is direction of travel over the ground, clockwise from true north (0° north, 90° east). It is the time-weighted circular mean of bearings between successive saved receiver positions, not aircraft nose heading. Latitude is the time-weighted mean of receiver latitude, followed by its minimum–maximum range; negative values are south. Track distances use a 6,371 km spherical radius for route description only; they are not model-comparison evidence.

All resolved stretches also prefer the specified background-field globe over the zero-field globe under both processing assumptions. This does not establish the field's cause or distinguish solar from sidereal rate. Findings and recovery checks explain the comparisons. The downloadable inventory includes exact source hashes, intervals, course, latitude and result origins.

For deeper analysis, download the comprehensive metadata CSV. It adds raw gyro means, explicitly labeled motion-correction diagnostics, north/east velocities, transport predictions, all six saved model costs and stopping reasons, and shared-leg IDs. Definitions and limitations, fit history and matched recorder differences keep those quantities and their provenance separate.