Preserve coordinate bounds through the native calculation

Build a separate native-enclosure candidate that carries both an overall vector-norm bound and separate coordinate bounds through addition, scaling, cross products, matrix actions, velocity updates and coordinate extraction. Retain the tighter applicable bound at each operation without converting one norm radius into three independent equal coordinate uncertainties.

Preserve the native equations, centers, physical assumptions, all 27 physical and 24 shared columns, calibration remainder and previously verified training conditioning. Physical-domain force proofs may be used by generated native controls; the prediction path must retain its training-derived certificate. Never replace the latter with a weaker physical-only range.

This stage qualifies transport of component bounds. It does not yet install the new norm-constrained body-source interval into the full engine or authorize a complete prediction rerun. That source integration remains a separate gate.

Keep the original 24-case native control matrix and numerical limits: center and stationary-pose agreement within 1 micrometre, 162 finite-difference columns at step .002 within 20 micrometres per normalized coordinate, and all nonlinear probes within the computed remainder plus 10 micrometres arithmetic margin.

Finite study allowance: 24 identities, 78 full native evaluations, 19,632 native packet visits, 348 bounds, 348 physical source certificates and 15,072 bound packet-steps (two passes per packet). Four single-thread workers; each bound has a cooperative 60-second ceiling and the study a 600-second ceiling. Freeze sources, inputs, compiler and environment before launch. No retries, extensions, observed fits, original-aircraft reads or protected ESA data.

Passing these controls qualifies this implementation component, not aircraft prediction coverage or statistical significance. Preserve every earlier freeze.