What the global increment controls establish
All 30 identities passed, covering 1,920 generated probes and 12,726 charged reference/value evaluations. Twelve calibration cases check complete bias/gain boxes; twelve rotation cases check rotation and integrated-force maps; six histories check shared calibration over 1, 7 and 31 packets. No archive originals, optimizer starts or native-engine evaluations were used.
The largest sampled remainder reaches the analytic calibration bound, rather than exceeding it. Packet and history remainder/bound ratios are at most about 0.948 and 0.894. Repository/reference discrepancies are below 6e-14 in these short controls. The comparison tolerance does not prove a long-record rounding bound.
The useful change is methodological: the implementation now has global bounds derived from the declared calibration region and rotation integrals. Their size is not chosen from a favorable sample of local perturbations. The bounds retain joint angle/velocity effects and calibration shared across packets.
Important distinction from the full prediction engine
The body-preintegration helper integrates constant rotating body force with rotation/J/K factors. The main native prediction engine instead uses midpoint force rotation, local-frame transport, Coriolis dynamics and candidate-dependent geographic rates. This study does not assert that those are the same map.
Reciprocal calibration and exponential rotation bounds transfer as components; the complete native local step, geographic conversion, processing/removal term, lever arms and uncertain receiver timing still need their own propagation. The initial-attitude region is also broad, including a full yaw range. A valid global bound can be wide; passing these controls does not establish useful prediction precision.
The new component therefore remains outside the archive prediction gate until all required terms and numerical errors are covered. Missing terms continue to produce unsupported uncertainty. No original finding or failure is relabeled.