Wider-precision training and prediction controls

24/24 generated fits pass all frozen checks.

Earlier 800 receiver positions determine initialization, calibration and the gravity anchor. Later 400 positions score a continuous forecast only after it is saved. True initial state and calibration are not supplied. Both starts, both processing cases and both working error assumptions remain in the test.

The existing procedure supplies the same fit allowance and stopping checks. Only its explicitly bound problem factory uses the wider calculation.

These are generated software controls, not aircraft findings or calibrated confidence. No protected confirmation flight is opened. All original fit labels remain unchanged. A revised observed study needs its own freeze.

The run charged 268 native calls in 113.96 minutes. Frozen plan · Known-motion gate · Roadmap.