Exact agreement across the complete native control matrix
All 24 controls passed. The candidate preserves all 348 saved predictions, coefficient matrices, physical bindings and four error-remainder arrays exactly. The study spent its 78 native calls, 348 bounds and 60,288 logical bound steps. The completed freeze and every charge remain preserved; it will not be restarted.
The result qualifies the arithmetic substitution on this generated control matrix. It does not demonstrate a speedup: the 243-second total includes native oracle calls, compilation, worker setup and publication-independent checks. Earlier runs occurred separately and are not a matched timing benchmark.
Next, compare the unchanged and integer implementations on the same shortened sources in one serial study. Require identical outputs and record propagation time without profiler instrumentation. Only a measured benefit justifies a new full-history performance screen. The existing width target, aircraft results and observational-error requirements remain unchanged.