Vector norms do not resolve the prediction bottleneck
All 48 saved cases passed correctness and all 288 independent support checks passed. Complete prediction widths narrowed by less than 0.1% relative to the preceding calibration-consistency calculation. Only three of six model/processing groups meet the unchanged twofold target on every axis.
The extra constraints are valid but have little effect on this prediction construction. More directional facets of the same vector balls are a low priority. The calculation took 36.9 seconds and performed no native integration, aircraft fitting or original-file reads. Earlier evidence remains unchanged.
The next candidate preserves a stronger known property: the shared force matrix comes from a rotation and three reciprocal gains. Its columns cannot expand arbitrarily along their own directions. A separate component qualification and saved-case comparison must test the resulting necessary constraints before use.