Batched implementation qualification
The installed-source profiles identify repeated Python array checks, argument conversion and separate norm calls as substantial costs. Scalar geographic intervals already store binary64 endpoints; expanding rational denominators are not the cause of the observed cost.
This separate implementation batches the existing component/global-radius intersection inside C++, and passes Python-owned contiguous binary64 buffers after checking their lengths. C++ retains every finite-input, radius-sign, overflow and floating-point-mode check. It includes the unchanged arithmetic primitive. No bound formula, point operation order, calibration cap or physical assumption changes. Earlier candidates and completed studies remain untouched.
Local tests must show exact equality with separate operations, including scalar and matrix cases, rational fallback, small/large inputs, invalid inputs and immutable output views. Then repeat the original 24 generated native controls for all three models and both processing cases. Compare all 348 generated bounds with their sealed predecessor, including point predictions, the three coefficient matrices and physical parameter/domain bindings.
Separate finite allowance
- 24 identities; no retries or optimizer starts.
- At most 78 native evaluations and 19,632 native packet visits.
- At most 348 bounds and 30,144 world/source bound steps.
- Separately reserve 348 source setups, 7,536 source-reference steps and 7,536 rounded body bounds before execution.
- Four workers with one numerical thread each.
- Original cooperative 60-second bound limit and 600-second overall limit.
- Original center, stationary-pose, derivative and containment acceptance limits.
Freeze code, saved generated inputs, matched comparison values, environment and compiler before launch. Every native control and exact comparison must pass. This qualifies implementation equivalence, not useful complete prediction widths or observational confidence. A successful result permits a separately frozen long-history runtime/range comparison. No aircraft originals, protected ESA observations or new scientific fits enter this study.