Local Level Lab

Airborne measurements. Open methods.

Saved horizontal correction audit

Question

The completed full-calibration horizontal qualification passed eight of twelve generated fits and three of six paired-start comparisons. Four flat-model fits passed endpoint stationarity but failed fresh local stability. One zero-field globe pair missed objective repeatability. Preserve every original outcome.

Inspect all twelve saved residual/Jacobian systems, not just failures. Reproduce the existing bounded linear probe and compare two diagnostic formulations: column scaling of the same feasible problem, and a column-scaled ridge penalty of 1e-6 on normalized parameter displacement. The ridge formulation is a different diagnostic objective, not a qualified replacement or a new gate.

Retain all 26/27 columns and physical step bounds. Report endpoint rank, stationarity, predicted versus saved actual probe changes, correction sizes, largest parameter moves and affine-solve completion. Compare the affine cost with the feasible zero step independently of the solver's success flag. An SVD reports rank only; it does not remove columns from any solve.

Finite scope

The parent report inherited an incorrect display ceiling of 3,888 calls; its frozen allowance was 1,944, of which 1,294 were used. Preserve the parent report and correct this presentation explicitly here. These software controls do not alter the archive's 53 globe / zero flat / zero unresolved finding.