The compiled calculation passes the native controls
All 24/24 generated controls passed. The study charged 78 native evaluations, 348 bounds and 15,072 two-pass bound steps, completing in 98.36 seconds with four single-thread workers. Every result records the frozen arithmetic source hash.
The original reference, derivative, stationary-pose and nonlinear-containment limits were retained. The calculation preserves reference centers, training constraints and all 27 physical plus 24 shared-error columns. Separate local checks compare primitive bounds with exact arithmetic and reject unsupported rounding or subnormal modes.
This qualifies the component on the declared generated controls. It does not prove useful error ranges over a whole recording, certify every floating-point operation in the solver or establish statistical coverage. The tighter norm-constrained body-source interval is not installed. No aircraft data was fitted and no protected ESA record was used.