Qualify omission of the unused endpoint force certificate
The unchanged-source profile completed all six cases. Physical-value calculation accounts for most of the measured source-stream time. The candidate omits only the physical endpoint departure/force certificate, which the integrator never consumes. It keeps endpoint rotation state, both derivative/direct phases, clipped clock handling and every midpoint value and remainder.
The 12 focused source checks preserve exact midpoint certificates, auxiliary frames and subsequent packet state across all models, both processing cases, point and nonpoint domains, and clock boundaries. Seven native checks confirm the wrapper changes only its explicit source-stream import and preserves each tested complete report exactly. These tests do not establish runtime usefulness.
Frozen qualification matrix
Reuse the same 24 known-answer inputs from the completed direct-native controls. Compare every saved report field, including all four remainder arrays, exactly against that study. Retain its independent native predictions, all 27 physical columns, stationary-pose, derivative and nonlinear-containment criteria. Do not relabel or restart the earlier study. The new candidate has its own freeze.
- 24 identities; at most 78 native trajectory evaluations and 348 bounds.
- 19,632 maximum native packet visits and 67,824 logical bound steps.
- Keep the predecessor's separate per-task source setup/prefix reservations.
- Four workers with one numerical thread each; four total threads.
- 600 cooperative seconds overall; 60 seconds per bound, no retries or signals.
- Zero LP proposals, optimizer starts, original-aircraft reads or protected ESA outcomes. Preserve all failures and charged work.
Passing qualifies equivalence on this software matrix only. A separately frozen matched performance test on complete input domains must demonstrate runtime usefulness before generating replacement future bounds. Keep the original width target and physical assumptions unchanged. Recording-scale uncertainty and the observation-error model remain open.