Native qualification using the actual calibration box
The first integrated source study is complete and failed qualification at 18/24 controls. All six broad cases stopped at a nonlinear-cap check. Its results and unused allowance remain preserved; this is a separate candidate and finite study.
The diagnosis shows that the proof replaced the asymmetric bias/gain box with a larger symmetric box. The revised proof retains the actual intervals, including the reference point for the Taylor path. It leaves all physical bounds, existing nonlinear caps, reference centers, rows and shared columns unchanged. Local checks confirm the previously failed cap comparison and all six model/processing native cases.
Run the same original 24 generated controls. Acceptance remains 1 micrometre for reference/stationary predictions, 0.002 normalized derivative steps with 20 micrometres tolerance, and 10 micrometres numerical allowance for nonlinear containment. Every registered control must complete and pass.
Separate finite allowance
- 24 identities, no retries or optimizer starts.
- 78 full native evaluations and 19,632 native packet visits maximum.
- 348 complete enclosure bounds and 30,144 world/source bound steps maximum.
- Separately reserve at most 348 source setups, 7,536 compiled reference steps and 7,536 rounded body bounds before their enclosures run.
- Four single-thread workers; four total numerical threads.
- 60 seconds per bound, checked cooperatively; 600 seconds overall.
Freeze implementation, generated inputs, compiler and numerical environment. Preserve failures and charges. No aircraft observations or protected ESA data enter this qualification. A successful result permits a new prediction comparison; it does not establish useful complete ranges or calibrated coverage.