Fixed-point horizontal correction checks

Use all twelve endpoints from the completed horizontal full-calibration qualification. Freeze the parent and saved correction-audit hashes. At each endpoint evaluate exactly three points: the saved center, its original local bounded correction and its diagnostic column-scaled ridge correction.

The two correction vectors come from the completed saved-array audit; do not re-solve them or select a vector after seeing native results. Retain all 26/27 nuisance columns and the same independent generated observations, nominal clock, physical bounds and projected covariance.

Require the center to reproduce saved residuals and tangents. Report fresh stationarity, actual versus linear cost, retained-coordinate motion, unscored height errors and whether the unchanged 1e-4 stationarity /1e-8 relative-cost /1 mm axis-RMS /1 cm epoch-maximum numerical limits pass. These fixed-point checks cannot relabel fits, qualify a solver or satisfy paired-start fitting.

Frozen allowance

• Twelve endpoint identities, three points each: 36 native tangent calls.
• Zero nonlinear optimizer starts, original-recording reads or retries.
• Four total numerical threads; 120 seconds after kernel compilation.
• Freeze exact sources, environment and input hashes before launch. Use an exclusive lock, pre-call durable journal and atomic hashed result artifacts.
• Completed or charged work cannot restart; saved-only publication makes no calls.

Passing ridge probes would justify a separately tested candidate. They do not authorize observed refits, change the failed qualification or address its remaining zero-field globe cost-repeatability failure by themselves.