Some allowed error combinations are not physical calibrations
All 36 checked extremes fail when their common-error terms are recomputed from a single calibration. The known generated controls reproduce exactly in all six groups. No witness collapsed to the known answer during exact repair.
The current prediction calculation contains 27 physical calibration values and 24 derived error terms. Its outer region lets those terms vary more independently than a real calibration does. Necessary matrix relationships did little to remove that freedom. This diagnostic now tests the complete calibration relationship.
Recomputing it changes the saved affine future position by as much as 81.86 m. The largest individual block changes are 72.14 m from acceleration, 12.46 m from the lever arm, 3.63 m from velocity/clock coupling and 0.22 m from the force matrix. Binary conversion contributes less than 7e-15 m. Block changes can cancel; these are not percentages of evidence or errors measured on an aircraft.
The result supports replacing independent error variables with a coupled calibration treatment. It does not supply a narrower certified interval: rejecting 36 proposed extremes does not exclude every other point. The existing prediction gate remains unmet, and the 53 aircraft comparisons are unchanged.
The six-case run took 13.2 seconds. It used all 36 LP proposals and 42 map calls, plus 142,878 of 246,000 allowed scalar checks. Reservations reconcile with the sealed results. There were no new native trajectories, bounds, original-aircraft reads or observed fits. All allowances are now closed.