Final selected-stretch analysis summary

Of 53 selected stretches, the completed conditional shape comparisons favored globe in 12 and flat in 0. 41 remained unknown; 0 were unsupported.

The shape question comes first. These are conditional comparisons under assumed calibration and processing, not calibrated scientific detections. Overlapping instrument streams are not independent flights.

Shape and rotation

Rotation is exposed only for resolved globe stretches. Across their processing cases, 24 favored rotating globe, 0 favored still globe and 0 tied. These are matched case comparisons, not independent votes.

Which calculations finished?

The following counts measure solver completion, not model support. A flat-disc calculation can finish while describing the flight much worse than a globe. The quality of matched fits determines the comparison; counting finished calculations does not.

188 calculations passed the stopping checks, 130 remained numerically unresolved and 0 returned an explicit error. 318 starts were charged against the 424-start limit.

Convergence requires the numerical stopping and stationarity checks. A low residual alone is insufficient; an unresolved optimization cannot rule out its model.

Fully converged versus unresolved stretches

A fully converged stretch has all six required fits. Any unresolved or failed fit leaves its stretch in the unresolved group. Geometry was selected before any model comparison.

These are descriptive comparisons, not evidence that speed, duration or instrument type caused convergence. All selected stretches remain in the report.

Source recordings and tracks

Every stretch uses Applanix Group-4 IMU increments at nominal 200 Hz. IMU type is the logged type code, not a confirmed manufacturer/model. Exact original filenames, SHA-256 fingerprints, UTC bounds, mounting epochs and conversion details are in the stretch ledger. Receiver track points retain the dated positions without interpolation.

Track distance sums receiver-position steps using a 6,371 km spherical coordinate convention. It is a route description, not evidence for a globe. Ground speed is receiver speed; it is not true airspeed.

Partial comparisons in indeterminate stretches

These rows show only matched globe/disc pairs whose two fits converged. An unfinished fit is never treated as a losing model. A pairwise lean does not resolve the whole stretch or unlock a rotation decision. Numerical costs have no calibrated significance here.

Case-by-case details preserve the available cost differences and unresolved models; fit details also preserve unfinished costs as diagnostics. They are not used to rank models whose fits have not converged.

Residuals and limits

Residuals from unresolved fits describe unfinished solutions; they do not support model rejection. A converged fit at a parameter limit can still be sensitive to the assumed bounds. Smaller sections are fixed-parameter consistency checks, not extra trials.

Calibration offsets, onboard corrections, the integration clock and receiver uncertainty remain partly assumed. Numerical convergence does not establish model adequacy, a global optimum or a validated error rate. All scientific decisions still abstain.

Every stretch and every fit retain unresolved outcomes, solver messages and diagnostics. See selection criteria, methods and pilot findings.