Direct-source comparison: narrower ranges, incomplete coverage

36 of 48 regions pass; 12 reach the per-bound time limit. All 48 outcomes, 336 charged bounds and their frozen inputs and sources are preserved. The study finishes in 84.9 minutes and does not qualify the full six-group matrix. It cannot be restarted or donate unused allowance.

Three groups have all eight regions complete: globe with zero field and fixed-zero removal, and both flat-disc processing cases. Their fresh north / east / height widths are respectively:

Both specified-field globe groups and the zero-field globe with profiled removal remain incomplete. Their partial regions do not define a complete range and cannot qualify those groups. Every failed task reached the time limit on its seventh, future-bound call; its six completed training bounds remain distinct saved artifacts.

These are generated 7.2-second software cases. No aircraft recording, native trajectory oracle or optimizer was used. The 53 empirical model comparisons are unchanged. A narrow interval here is not a measured aircraft error or a calibrated confidence interval.

Next gate

Check compatible saved certificates before repeating integration. A preliminary inspection finds matching earlier future bounds and newer completed training bounds for all 16 specified-field regions. Their reuse needs its own frozen policy, known-answer controls and support allowance. It cannot change either original study's failed labels or permit a partial new union. If that approach does not meet the target, profile and qualify an implementation improvement.

Recording-scale error bounds, floating-execution assumptions and observational error assessment remain ahead of withheld-flight fitting.

Results · Frozen scope · Continuation plan · Certificate-reuse proposal · Roadmap.