Carrying uncertainty through the motion step
The new derivative arithmetic now covers the geographic metrics, background and transport terms, midpoint rotations, velocity update and position update used by the flight reconstruction. It follows the native calculation's two local solves and revised-rate position update.
48 focused tests pass across this component and its arithmetic prerequisites. Known examples check acceleration and height derivatives through third order, globe/disc metric differences, rotation signs and the common upstream-removal convention for all three physical candidates. These are software controls, not new aircraft results.
The calculation retains rational interval endpoints and rounds them outward to a fixed binary grid to limit denominator growth. It preserves the native small-angle rotation polynomial. A region crossing that polynomial's switch is rejected for now, rather than silently treated as smooth.
The packet-to-position adapter also passes four initial known-answer checks for clock sensitivity, preceding history, initial offsets and the receiver lever arm. The native comparison now passes all six generated cases and all 27 physical columns, using six native queries and 162 interval steps. The comparison took about five minutes.
This supplies the motion-step expressions. It does not yet qualify useful full-history bounds, clock-boundary changes, native floating-point execution or the receiver/IMU error model. No new aircraft prediction fits follow from these tests.