Recording-scale outcome: the bounds are too wide
Three four-minute histories completed their checks. A fourth was partial, and two were not started. The worker reached its fixed 900-second cooperative limit and sealed its results without signals or retries. The generated study did not qualify the full six-case matrix. Its allowance cannot be reused.
The automatic result listing contains four terminal result records; that count includes the partial history. It does not mean four complete histories. All 49 published evidence artifacts pass their SHA-256 checks.
| Completed history | Broad or narrow region | Four-minute largest remainder allowance |
|---|---|---|
| Specified-field globe | Broad | 1,791 km |
| Specified-field globe | Narrow sensitivity | 477 m |
| Zero-field globe | Broad | 1,792 km |
The zero-field narrow history completed its saved 0.5-, 5- and 50-second checks before the time limit. Both flat histories were unstarted. Missing outcomes remain missing. All completed checkpoints passed their numerical comparisons; the largest completed center discrepancy was below 0.11 micrometre, within the declared 1 mm arithmetic allowance. That is a comparison of calculations, not a measurement of instrument accuracy or a global rounding proof.
What this means
The bounds retain the entire nuisance region while repeatedly discarding some correlations to remain conservative. They cover possibilities far removed from the fitted trajectory. Their enormous width is a limitation of this construction, not observed aircraft error, a claim of poor survey hardware, or evidence against the completed model preferences.
Even the separately declared narrower box leaves a 477 m approximation allowance after four minutes. It is not measured calibration and cannot replace the main region. The present bounds therefore do not supply useful prediction intervals for discriminating the models at recording scale. Passing containment alone would not have established that usefulness.
The next nonlinear step needs tighter propagation or a supported training-feasible region, preserving every nuisance direction and uncertainty. The separately planned shared-reference error assessment may proceed, but it cannot make these nonlinear bounds narrower. No new aircraft prediction fit or protected ESA model evaluation follows from this study. The original 53 comparisons are unchanged.