Faster interface, identical checked results
All 24 native controls passed. Every one of the 348 prediction bounds exactly matches its sealed predecessor, including point predictions, error allowances, all three coefficient matrices and physical-domain bindings.
The study completed in 91.75 seconds with four single-thread workers. The predecessor used the same allocation and took 201.31 seconds. These are recorded study runtimes, not a guarantee for long recordings.
The change combines repeated interface calls. It does not change the physical models, calibration caps, source rounding allowances or acceptance limits. Floating-point mode and invalid-input rejection remain checked.
The next gate is a separate six-source long-history screen under the original time limits. The complete prediction-width target remains four of six groups; no aircraft fit or finding changes.