Global calibration and rotating-force bound controls

30/30 identities passed in 4.51 seconds. The study charged 12,726 software reference/value evaluations, with no original reads or native-engine fits.

The bounds come from global analytic inequalities, not the largest sampled error. Probes test their implementation. Fixed-time body preintegration is covered; candidate-specific geographic dynamics, timing across packet boundaries and a whole-record floating-point enclosure remain outside this result. No archive prediction fit is authorized by these controls.

Proof outline and frozen limits · Prediction interface · Roadmap.