More positions help, but do not qualify prediction yet
All six numerical diagnostics passed their native and exact-certificate checks. Using 292 earlier positions instead of five narrowed every reported coordinate range. None met the fixed 20 / 20 / 50 m prediction-width target.
The dense horizontal outer widths are about 123–421 m. These describe a simplified linear calculation over generated data. They are not measured aircraft errors or evidence that the actual nonlinear prediction is that uncertain.
Two improvements are now established separately:
- The tighter force remainder passes all twelve short integration controls.
- Denser training reduces the certified linear ranges over the complete history.
The next check is whether calibration choices producing large linear forecast changes also satisfy the full nonlinear training constraints. If they do, they demonstrate genuine ambiguity in this generated test. If they do not, the linear relaxation needs stronger physical coupling before another expensive long-bound run is justified. Rejecting sampled choices would not prove that all remaining predictions are narrow.
The original long-record run continues unchanged. No aircraft prediction fits have started, and the 53-stretch findings are unchanged.