Separate coordinate bounds now survive the native controls

All 24 generated native controls pass: six derivative, six packet-boundary, six broad-domain and six stationary-motion cases. The study used its 78 native evaluations and 348 propagation bounds, completing in 348.86 seconds. All physical and shared-error columns remain present.

The candidate keeps both separate coordinate bounds and useful global norm bounds through force, velocity and position calculations. It reuses immutable validated views and caches repeated exact norms. Local tests compare every returned bound bit for bit with the earlier rational implementation and check the training-conditioning path independently.

The earlier implementation passed 20/24 controls and missed four per-bound time limits. Its failures remain preserved. The new freeze uses two workers with two numerical threads each, reducing competition within each bound's unchanged 60-second limit. Its numerical limits and total four-thread ceiling are unchanged. The completed studies are not a controlled wall-time benchmark of caching alone.

This qualifies transport of coordinate bounds. The tighter norm-constrained body-source interval has not yet been installed. Connecting it at the correct packet midpoints and endpoints is next; complete prediction-width testing and aircraft validation remain downstream. No aircraft fit or protected ESA data was used, and the existing aircraft findings are unchanged.

Results and evidence · Next integration task · Earlier failed implementation · Roadmap.