Final-stretch specified-field boundary checks

The remaining stretch b4e44ddc0f130ecd45674c78 has two separate numerical problems. Its specified-field-globe primary/repeat solutions agree closely in cost and predictions, but their boundary checks include unstable calibration corrections and unchecked neighboring boundaries. Its zero-field-globe solutions differ by about 57 ms and exceed the required repeat-cost tolerance.

Apply the already qualified range-scaled affine correction and representable-clock checks to the four complete original specified-field endpoints, under both processing assumptions. This tests a known local numerical defect before the larger nonlinear timing-profile problem. It does not retry the earlier algorithm, resume an old identity or spend a completed allowance.

Exact scope

Only sphere_rotating, bias1_unsubtracted and bias1_profiled_removal, primary and repeat. Preserve each original full endpoint, observation interval, clock, covariance, physical bounds and unknown-processing assumptions. Reuse the qualified saved-context adapter, checking exact receiver rows and relative times before extraction. Two fresh smooth controls run first.

The preceding matched study used 54 calls in 1,728 seconds, about 32 seconds per call including extraction and checks. At that rate this ceiling is approximately 59 minutes; runtime can differ with the recording and early stopping. It is an estimate, not permission to exceed the fixed evaluation allowance.

Qualification is one 34-test batch with a 120-second ceiling: the existing 30 scaled-candidate controls and four exact-scope/worker checks. The existing candidate controls include both saved failure systems, three independent compiled-motion controls and six compiled clock comparisons. No original-file reads or observed native calls occur during qualification.

Seal the exact tasks, original hashes, source snapshots, qualification, kernel, numerical environment and this plan before launch. Check locks and receipts; charge before execution, preserve failures and block any incomplete transaction. Never restart completed work or retry a charged identity. Finite limits cannot be expanded while running or after completion.

All targets require unchanged full two-sided native gates and cumulative source-to-final movement checks. Compare primary and repeat separately for each processing assumption, retaining the relative-cost limit 10^-6 and per-axis track RMS limit 0.01 m. Both controls and both pairs must pass full qualification.

Even a complete pass cannot resolve the stretch: the zero-field-globe timing profiles, required repeat comparisons and all other model/case companions remain necessary. Keep the current 52 globe / 0 flat / 1 unresolved count unchanged. The subsequent zero-field profile requires its own finite freeze.