Preserve relationships between calibration parameters
The saved projection narrows only two individual parameter intervals. Repeating the broad mounting-transform bound would therefore leave that allowance unchanged. The next diagnostic should test the relationships between parameters directly.
The shared 51-column adapter now passes known-answer and wrong-reference checks. A second check reconstructs the static calibration reference from verified start fields, without loading the entire trajectory. Compatibility with the saved long-history cases still needs its own finite diagnostic.
- Reuse the checked static-calibration map in
lll.ilvis0_coupled_mapand its exact coupling-certificate audit. Its 24 shared error terms describe one calibration, not 24 independent adjustments. Build new certificates from the long-history reference; do not transplant a short-history source binding. - Add the necessary coupling constraints to the saved training region while retaining all 51 columns, all 30 training inequalities and the original limits. Qualify that adapter on known answers, wrong-reference rejection and unchanged feasible witnesses before using it for this diagnostic.
- Compare the full box with a complete eight-part orientation cover. Preserve every part, including unsuccessful calculations. Numerical infeasibility alone must not discard a part of the physical domain. Freeze the finite map/support allowance and cost it before launch; this is a static saved-input calculation, with no trajectory integration or aircraft model fitting.
- Assess whether the resulting constraints tighten acceleration or transform allowances enough to justify a separately frozen trajectory-bound test. Static success alone cannot qualify the long-history prediction ranges.
The original prediction-width targets and physical bounds stay fixed. The aircraft observation-error model and full prediction-period coverage remain separate requirements.