Final-stretch zero-field timing profiles

The specified-field endpoint study is complete: all six checks passed in 100 native calls, including both processing-case primary/repeat pairs. The final stretch b4e44ddc0f130ecd45674c78 still has different zero-field-globe time alignments: its saved repeat is at the +0.15 s physical bound, about 57 ms from the primary. Their costs differ beyond the required repeat tolerance.

Hold each of the two exact saved offsets fixed, separately under unsubtracted and profiled-removal assumptions. At every offset, fit every other active parameter twice: once from the original primary and once from the original repeat. Both original starts and both offsets remain represented regardless of which fits better. Passing source repeats are diagnostic inputs, not automatic qualification of the other endpoint.

Use the existing fixed-time adapter and bounded trust solver unchanged. Removing the time column here defines a conditional diagnostic; it does not fix timing in the final scientific comparison. Keep the raw observations, receiver weights, clock, physical bounds and processing hypotheses fixed. No instrument drift model, extra calibration freedom or favorable interval selection is introduced.

Qualification and freeze

One 26-test software batch, at most 120 seconds and 1,024 callbacks: the existing 15 fixed-time adapter checks, including three compiled-motion controls, and eleven scope/worker checks. Tests cover parameter lifting/reduction, exact-bound timing anchors, both processing cases, both nuisance starts, unrelated model exclusion, failure propagation and private-template isolation. No archive originals or observed fits occur during qualification.

Require the completed field-endpoint evidence before preparing the observed freeze. Seal its completion and receipts alongside the exact source endpoints, selected intervals, original hashes, qualified sources, kernels and numerical environment. Preserve every prior study and failure. The worker privately adapts the completed executor; it does not edit or resume it.

Finite observed allowance

At the preceding matched study's 32 seconds per native call, exhausting this ceiling would take about 5.8 hours. Earlier conditional profiles stopped well below their ceilings; that does not authorize increasing this one. Charge before execution, require fresh controls first, preserve failed starts and block any incomplete receipt transaction. Never restart completed or charged identities, reset counters or reuse an earlier allowance.

At each fixed offset, require both conditional fits and compare their costs and tracks using the existing relative-cost limit 10^-6 and 0.01 m per-axis RMS limit. Save the complete full-parameter residual/Jacobian, receiver epochs and timing derivative for each final diagnostic, including failures.

Report every result. Fixed-time completion does not establish free-time stationarity, repeat agreement across different offsets or a global optimum. The next full free-time test requires its own finite freeze, followed by all model/case companions and scheduled repeat checks. Published findings remain 52 globe / 0 flat / 1 unresolved throughout this conditional study.