Focused boundary check for the nearby timing pair
The saved audit at the final-two diagnosis shows one complete stretch missing only its profiled-removal specified-field primary/repeat checks. Its original primary has a stable left branch but an unstable joint correction on the right. The old repeat diagnostic was interrupted; that failure and its charges remain untouched.
Use the already qualified source-preserving complete-cluster search and explicit representable-clock guard. This is a separately identified test of that method on a different target, not a restart or retry of the old diagnostic algorithm. Both sources are complete original hybrid-fit endpoints with saved full residuals and Jacobians; neither comes from the interrupted diagnostic.
Scope and fixed limits
| Item | Maximum |
|---|---|
| Fresh smooth controls | 2 × 3 native evaluations |
| Target primary/repeat checks | 2 × 26 native evaluations |
| Charged identities | 4 |
| Native evaluations | 58 |
| Nonlinear optimizer starts | 0 |
| Clock queries | 402,048 |
| Bounded affine solves | 6 |
| Total numerical threads | 4, one worker |
| Retries | 0 |
Target: 370cec66878c4c1cf84b4ceb, bias1_profiled_removal,
sphere_rotating, both original primary and repeat. Reuse the latest passing
smooth-control sources, but evaluate their checks freshly. The saved source
observations, parameters, nominal integration clock, physical bounds, covariance
and processing hypothesis remain fixed. There is no permission to fit the other
remaining stretch or expand this allowance.
The candidate's 31 software checks are already complete. A separate software qualification reuses its 21 candidate controls and adds twelve scope/worker checks: 33 tests, one batch, 120-second timeout, at most 1,024 software callbacks, 81 compiled-motion calls and six compiled clock comparisons. No archive reads. Freeze the qualified sources, test result, exact plan, original file hashes, native kernel and numerical environment before launching observed work.
Require both fresh controls first. Each target must reproduce its saved source, pass the full two-sided stationarity/cost/physical-position/continuity gate and remain within source-to-final movement limits. Compare the target predictions under the unchanged relative-cost limit 10^-6 and per-axis RMS limit 0.01 m. Retain all outcomes, native-call receipts and failures. A pending call transaction blocks resume. Completed or charged identities cannot be restarted.
The durable worker and status metadata retain the four-thread allocation while active. Publication performs no native calls. Passing this study still requires a saved review of all six model/case primaries, both required case repeat pairs and controls before any stretch disposition changes. Current findings are 51/0/2.