Local Level Lab

Airborne measurements. Open methods.

Direct source bound in the complete calculation

All 24 native controls pass. Across 348 checked bounds, the numerical predictions, shared coefficient matrices, physical bindings and shared caps are unchanged. No coordinate, position, force or lever approximation allowance widens. Some position allowances decrease; most final-coordinate allowances are unchanged in these short controls.

The study charged its 78 native oracle calls and 348 bounds, using 67,824 logical bound steps in 232.83 seconds. A separate saved-only audit verifies 6,726 actual point-derivative steps and 6,726 direct-formula steps against their frozen reservations. No aircraft observations or fits were used.

The direct bound is now qualified for these generated native paths. The next gate is the separately frozen six-source full-history screen. Only after that passes can a new complete prediction-width comparison be prepared. The last complete result remains four of six generated groups.

These controls do not establish recording-scale prediction ranges, whole-program floating execution bounds or observational confidence.

Results · Frozen scope · Saved-only audit code · Roadmap.