Next check: preserve vector-norm information

The calibration-consistency comparison is complete. It passes correctness but narrows complete ranges by less than one percent. Do not repeat it or enlarge its allowance. Further duration expansion remains deferred.

Several shared-error caps were derived as lengths of three-component vectors, then relaxed to separate coordinate boxes for linear support calculations. Audit the derivations before treating any block as a unit ball. Candidate blocks are the three force-matrix columns, the three acceleration/clock remainder vectors and the initial lever remainder. Velocity/clock component bounds must not be assigned a common norm without a separate proof.

For a verified vector cap C and any declared direction d, the necessary bound is d·e <= ||d|| C. With equal normalized component caps, this becomes d·gamma <= ||d||. Integer directions from {-1,0,1} provide a small outer polytope; coordinate directions are already present in the box. Use outward rational/square-root endpoints. Never infer a norm bound merely because three reported component caps happen to have the same value.

First qualify the source-bound block definitions, exact zero-cap behavior, signs, coordinate ordering, rejected unsupported blocks and retention of known physical calibrations. Include independently computed vector norms over the six generated configurations and the original domain families. Record a finite allowance and freeze before execution; these are component controls, with no native integration or observed fitting.

Only after that qualification passes, prepare a separately frozen saved-only comparison covering every original sector and processing group. Keep all 51 coordinates, physical bounds and error allowances. Preserve earlier valid certificates as fallbacks, withhold future truth until predictions are saved, and independently reconcile the new support certificates. Report the same unchanged twofold target and any legitimate lack of improvement.

No new aircraft fits, protected confirmation outcomes or precision-field claims follow from these plans. Recording-scale and observation-error gates remain.