The endpoint-certificate change did not pass the runtime gate
Only two of six timing pairs completed. In each globe pair, one call reached the unchanged 120-second ceiling. Both implementations were affected. The two complete flat-disc pairs retained exact report agreement, but their time ratios were 0.9673 and 1.0111; they do not demonstrate the required speed improvement.
The completed study used all twelve bound-call reservations and took 370.30 seconds. It cannot be retried or expanded. The earlier 24-case exact-report qualification still stands, but software equivalence alone does not establish runtime usefulness. No replacement future-bound rollout follows.
The saved source profile gives a more specific explanation: the two rotation certificates consumed about 13.3 seconds of the 25.4-second physical-value stream in its first case. Both rotation certificates remain necessary to preserve midpoint and endpoint state. Omitting the endpoint force certificate does not remove that work. Full prediction propagation contains additional costs that the source-only profile did not measure.
Next is one separately bounded, whole-bound profile on the original generated domain. It will retain timing even if the bound does not finish, distinguish wall time from process CPU time, and identify the remaining costly operations. It is a diagnostic, not a relaxed qualification or another aircraft fit.
The prediction-width gate remains four of six cases. All 53 aircraft model preferences and all earlier failures remain unchanged.