Audit the remaining calibration enclosure from saved results

The regrouped construction passes all correctness controls but misses its usefulness target. Audit its sealed 48 generated sector results before adding another native experiment. Preserve all source freezes and previous failures.

Verify the completed source receipt, native input hashes and every saved training/future side artifact. For each sector, retain all 27 columns and project the necessary training polytope onto each parameter using the qualified exact support certificates. Report both endpoints, the previous row-contracted range, width ratio, multiplier/fallback status and the known generated truth check when that truth belongs to the sector. An LP infeasibility status alone cannot exclude a sector or mark a parameter known.

Use the existing certified future support to separate affine and nonlinear contributions to the computed range width. Reconcile the parts against the sealed future endpoints. Compare the retained nonlinear width with the original tenfold target; do not reinterpret a failed threshold as a pass. Show whether lever, orientation, velocity and timing ranges narrow, including every weak or unidentified direction. Do not rank physical candidates from this generated engineering audit.

Freeze 48 audit identities, 2,592 LP proposals and 5,184 exact certificates, with one thread and one cooperative second per LP, four total reserved numerical threads, and 120 cooperative seconds overall. Reserve each identity's 54 proposals and 108 certificates before calculation. No native evaluations, Earth-model fits, original-file reads, protected ESA outcomes, retries, extensions or process signals. Save exact certificate references and publish partial results honestly if the allowance ends.

If the dominant ranges remain effectively unrestricted, simply running a longer trajectory or tightening an arbitrary fitted box is not justified. The next mathematical candidate should preserve shared nonlinear calibration errors across epochs, with a separate proof/qualification. Any auxiliary relaxation must contain the original 27-parameter model; no physical direction may disappear. Differenced observations require their common reference error and temporal covariance to be retained. This audit itself changes no bound, model, confidence claim or eligibility policy.

One candidate decomposition is f(t,p) = f(t_ref,p) + displacement(t,t_ref,p): represent a proven common reference remainder once, rather than as an independent error at every epoch. The displacement bound would need to preserve initial lever/attitude and common time-offset cancellations algebraically. Subtracting two already widened intervals, or declaring similar-looking remainders constant, is insufficient. Auxiliary bounded reference-error coordinates may only relax the original physical feasible set, with all original parameters retained. This is a prospective mathematical design, not a qualification or authorization for new observed fits.

Regrouped outcome · Exact joint support · Canonical roadmap.