Qualify direction-preserving calibrated-force bounds
Use the six existing generated candidate/processing inputs and four fixed domains: the full calibration box, original yaw sector 4, a fixed calibration, and the existing acceleration-gain boundary control. The same rule applies to every candidate. There are 24 identities. This tests a static force action, not integrated aircraft motion.
Convert the first recorded velocity increment into a force interval with its actual period and the supplied accelerometer bias/gain ranges. Bound the rotation-vector action component by component; apply the exact supplied installation matrix without renormalizing it. Add the declared gravity range. Compare with the original norm-based acceleration bound and preserve wider outcomes. No bound is reduced by observing a probe.
Validate each interval against an independent point calculation using the existing source oracle and matrix exponential. Use 64 fixed probes per nonfixed domain: lower corner, upper corner, midpoint and 61 seeded interior points. The fixed domain uses one probe. All three axes must contain every probe within an explicitly declared 1e-10 m/s² point-comparison margin. That margin is not a sensor-noise estimate or a floating-execution enclosure.
Finite allowance: 24 reference setups, 24 interval actions, 1,158 point oracles and 6,948 scalar containment checks. Reserve before execution and retain all failures. Four workers use one numerical thread each; the cooperative overall ceiling is 120 seconds. No retries, expansion, process signals, native trajectory integrations, original-aircraft reads, optimization or protected ESA data. These controls do not qualify a full native propagation replacement.
Keep all 27 physical coordinates in domains and probes. This component depends only on mounting, accelerometer calibration and gravity; it does not remove other parameters from any subsequent model. Preserve raw input hashes, source and numerical-environment versions, journals and atomic evidence.