Saved linear-system diagnosis of the right-side corrections

Use only the two completed target endpoints from the corrected-context trial. Preserve its failed full gates, original source arrays and all earlier studies. No result from this audit can change a stretch disposition.

The question is whether the near-zero permitted timing interval destabilizes the bounded local correction, or whether the center tangent does not reproduce the failing side tangent. Compare three affine solves per saved center: timing restricted to the left interval, restricted to the right interval, and fixed at zero correction with every other physical bound retained. The fixed-time solve is diagnostic only; it is not a proposed replacement for a full free-time gate.

Compare their predicted costs and parameter steps with the saved native branch corrections. Compute singular spectra for the full center Jacobian and its non-timing columns. Keep small singular directions in the report; rank here is a numerical diagnostic, not a new scientific acceptance threshold.

Limits: one batch, two saved endpoints, six bounded affine solves, at most 100 iterations per solve, four singular decompositions, four total numerical threads, zero native evaluations, original-file reads or optimizer starts. The six affine solves do not fit the archival observation model. Freeze source, input and numerical-environment hashes before the batch; charge before numerical work and prohibit rerun. Publish/render from saved outputs after completion.

Software checks cover independently known constrained affine optima, fixed-time elimination, rank deficiency and malformed/nonfinite problems. If the center Jacobian does not reproduce the observed side behavior, record that limit rather than assigning a cause. A new native tangent-capture diagnostic would then need its own bounded plan. Scientific stationarity/cost/position limits remain fixed.