Source derivatives improve some horizontal bounds

All six eight-second generated-source comparisons pass with unchanged reference maps and original limits. The prediction endpoint is at 7.2 seconds. The study took 161.74 seconds; each bound completed within 99 seconds of its 120-second allowance. These are software controls, not aircraft recordings.

For the three profiled-processing cases, the fresh north/east error allowances are 6–11% smaller than the best compatible saved bounds. Improvements in the fixed-zero cases are limited: several old horizontal bounds remain tighter and are retained. Height allowances improve slightly beyond the previous source screen. This result does not establish a useful complete prediction range across all mounting regions.

The next comparison must preserve the training-derived acceleration bounds. A saved-only audit verifies that all 48 descendant regions retain their parent constraints, coordinate frames and calibration basis. Every physical domain has narrowed, so the existing exact-domain proof interface needs an explicit, checked transfer to those subdomains. Replacing it with a weaker physical-only bound would lose information already established from training.

The last complete prediction-width result remains four of six groups on the generated 7.2-second example. No aircraft fit, selection or finding changes.

Six-case results · Frozen scope · Next gates · Saved ancestry audit · Roadmap.