Compare with the baseline actually in use
The existing training-conditioned acceleration bound is at least as tight in all three components and in vector norm for 39/39 completed cases. The saved reference and domain checks establish that it applies to the new domain. Intersecting these particular static bounds would leave the old acceleration limits unchanged.
All 4/4 complete candidate/processing groups retain exactly their inherited combined prediction widths. Both specified-field globe groups still miss the fixed width target. Completing the unstarted or interrupted disc calculations cannot change that failed all-group gate.
The earlier static comparison used the broad physical prior as its baseline. The complete prediction method already had a much tighter training-derived acceleration range. Replacing it with the new physical-only range lost useful information. The comparison with the actual prediction baseline should have preceded this run.
This diagnosis uses saved proof radii and qualified range endpoints only. It performs no native prediction, propagation bound, support calculation or fit. It does not reinterpret the numerical failures or change their labels.