Saved shared-source map audit

Use all twelve sealed nonlinear probes from the completed generated-data diagnostic. Compare six saved epochs per point: 2, 10, 30, 180, 290 and 438 seconds. Keep the two processing cases separate.

Rebuild one source-bound reference per case and require exact agreement with the original saved reference. Verify that the old six native centers and all physical Jacobians match the corresponding newer batch outputs. Reuse only coefficient maps with the same source, reference and shared normalization.

For each fixed point, evaluate its 24 shared source coordinates once. Compare the saved native position changes with the 27-column linear expression and the full 51-column expression. Report residuals and the three shared-block contributions. Do not optimize or adjust the points. Do not transfer old training-conditioned radii: these probes need not satisfy that training region.

Finite scope: two identities, two reference setups, twelve source-coordinate oracles and twelve map comparisons covering 72 saved epochs / 216 coordinate residuals. Zero native integration, LP proposals, aircraft reads or model fits. Use one numerical thread with other allocations reserved; cooperative wall limit 120 seconds. No retries, process signals or allowance increases. Freeze the sources, saved inputs, environment and hash references before launch. Publish only from sealed results.

This is discrepancy attribution, not prediction coverage, a check of unobserved calibrations, an aircraft error model or scientific significance.