Matched observed comparison: frozen scope

Question

Does the qualified wider numerical calculation improve completion and prediction on the same ESA development recording, with the scientific assumptions unchanged?

The software prerequisites are complete: all 48 long known-motion controls and all 24 training/forecast controls qualify. The latter used 268 native calls in 113.96 minutes. The observed adapter and saved-results audit pass 56 checks bound to the current source and environment.

Measurements and method

Reuse the original prepared twenty-minute interval from recording 187, with its saved measurements and hashes. Earlier 800 receiver positions determine the initial state, calibration, weights and gravity anchor. Later 400 positions score a continuous forecast after it is saved; they do not adjust it.

Keep all three candidates, both processing assumptions, both working receiver error policies and both declared starts: 24 fits. The height-gravity law, calibration bounds, initialization, stopping checks, accuracy limits and training-only primary selection are unchanged. Wider precision is the numerical change. No original aircraft file or reserved confirmation recording is opened.

Separate finite allowance

The time ceiling is a limit, not a runtime forecast or a promise of completion. It follows measurement of the qualified controls and matches the earlier observed pilot's ceiling. This is a fresh allowance; no unused allowance is transferred from earlier work. Starts and evaluations are charged before use. Interruption does not erase charges. A cooperative limit never signals or kills a running calculation.

Reporting

Preserve every failed or incomplete fit and compare each declared start with the earlier height-gravity result. Select primary starts using training results only. Assess horizontal and height accuracy together, separately from numerical completion. Compare shape before the specified-background-field alternatives.

The saved-results audit performs no refit or propagation. It verifies inputs, source and compiled-calculation provenance, complete calibration columns, training/forecast boundaries, physical errors, failures and all charges.

This single development recording cannot establish independent replication or calibrated confidence. Seven ESA confirmation groups remain protected. The 53-stretch findings and all original failure labels remain unchanged.

Prerequisite plan · Study overview · Qualified training controls.