Test the norm constraint against the actual saved source
This preflight tests a value range before any new native propagation candidate. Use all six existing generated candidate/processing groups, always original yaw sector 4 and its saved physical domain. The fixed selection does not depend on which interval looks favorable. Keep all 27 physical coordinates and the training-derived acceleration limits already in use.
Let R be the initial installation/correction map, P the accumulated body
rotation, f the calibrated first force sample, A=R f and B=R(P−I)f.
The frozen body-prefix operations provide the original bound on B. The saved
training-conditioned acceleration and gravity ranges provide the bound on A.
Rotation preserves vector length. Bound the small metric defect of the literal
installation matrix explicitly from R0ᵀR0−I; do not normalize the input.
Use that bounded norm relation to compute a vertical interval for B.
Report the original, raw geometric and intersected vertical ranges. Establish the sign branch over the complete supplied domain. An unsupported branch must remain unsupported. This is value-only: it does not qualify a Taylor remainder or establish equality of native reference coefficients.
After fixing every interval, evaluate the existing known in-sector control once
per identity using independent quaternion rotations and point calibration.
Check all three A components, all three original B components and the new
vertical interval. The declared point-comparison margin is 1e-10 m/s²; it is
not a measurement-noise model or a floating-execution proof.
Finite allowance: six identities, six reference setups, six bounded body prefixes, six norm intervals, six point oracles and 84 scalar containment checks. Precharge at most 1,600 packet visits per bounded prefix and per point oracle: 19,200 visits total. Record actual visits separately; reserved visits remain charged. Four workers use one numerical thread each; the cooperative overall ceiling is 120 seconds. No retries, extensions, signals, full native trajectory calls, propagation bounds, LPs, optimizer starts, original-aircraft reads or protected ESA data.
Successful containment and a tighter value interval would justify a source reference-coefficient investigation. They do not permit aircraft fitting or another complete prediction run. Preserve the failed earlier comparison and its unused allowance.