The directional component passes native controls
All 24 matched controls pass under the original acceptance limits. The study used 78 native evaluations and 348 propagation bounds, preserving a source-bound directional acceleration certificate for each bound. It completed in 79.11 seconds; all work and source checks are recorded.
The new calculation keeps the physical predictions unchanged. It tightens the range assigned to acceleration, while preserving the original nonlinear remainder and all 27 physical plus 24 shared-error columns. The checks cover native centers, known stationary positions, all 162 derivative columns and nonlinear probe containment across all three candidates and both processing assumptions.
The static proof covers its supplied physical domain. In a prediction study, that domain can itself come from earlier training constraints; the proof does not inspect future receiver values or select a favorable calibration.
This result qualifies the component on the tested generated cases. The subsequent prediction comparison did not improve the complete globe groups: their existing training-derived range was tighter. Recording-scale uncertainty and observational coverage remain open.