Joint linear constraints can be retained safely

All 24 generated identities pass. Twelve polygon tests compare the bounds with 57 independently enumerated exact feasible vertices. Eight larger tests retain all 27 parameters, including 25 unidentified coordinates. Four weak systems retain their full box ranges.

For example, knowing that two uncertain coordinates are equal may leave each coordinate's range unchanged, while fixing their difference exactly at zero. A coordinate box discards that information. The new support calculation keeps the relationship when bounding a prediction.

The numerical optimizer only proposes multipliers. Exact arithmetic checks the resulting inequality. Poor proposals cannot exclude feasible points; failed or invalid proposals fall back to a full-box bound. Solver failure never silently means an empty calibration region.

This qualifies a mathematical component, not actual aircraft uncertainty. The previous generated native bounds still include large independent nonlinear remainders. Keeping linear correlations alone cannot remove them. The next separate feasibility study will cover the entire original orientation range with smaller sectors, retaining every sector that has not been validly excluded.

No native evaluations, original IMU reads, observed Earth-model fits or protected ESA evaluations were used. The completed allowance cannot be restarted.

Frozen controls and results · Next generated feasibility plan · Joint calibration method · Roadmap.