Compare installed source bounds with the best saved allowances
The revised rounded-source candidate has passed all 24 native controls under the original acceptance limits. This new finite screen evaluates the same six 1,600-packet generated histories, sector 4, at the original 7.2-second query.
Retain the original physical domain, reference point, verified six-support training certificate and all 51 columns. First verify that the original and previous compiled-transport bounds have identical centers, coefficient matrices, source bindings and conditioning. Their per-axis minimum supplies the saved baseline. Compare the newly installed source bound with that baseline only after the same exact compatibility checks pass.
Report old and new allowances on every axis, including any widening. Retain the per-axis minimum only for compatible maps. This is not a complete eight-sector union or a prediction-width qualification. The current complete target remains four of six groups until a separate complete comparison establishes otherwise.
Finite allowance
- Six identities, six setups and 36 exact training-support verifications.
- Six complete bounds, 38,400 world/source packet steps.
- Separately reserve 9,600 compiled source-reference steps and 9,600 rounded body bounds before execution.
- Four single-thread workers, four total numerical threads.
- 120 seconds per bound, checked cooperatively; 600 seconds overall.
- No retries, expanded allowances, full native trajectory evaluations, LP proposals, optimizer starts, aircraft originals or protected ESA data.
Freeze inputs, sources, compiler and numerical environment. The changed worker allocation is recorded; aggregate elapsed time is not a matched speed benchmark. Preserve failed attempts and all source-work reservations. Passing compatibility does not establish useful full-recording ranges or calibrated coverage.