Local Level Lab

Airborne measurements. Open methods.

A missing rounding allowance at a fixed source point

This describes the preserved older failure. The separate rounded replacement now passes the physical and trained-source qualifications. Native integration is next.

A new local regression test fixes every physical parameter and evaluates the source immediately beside a packet boundary. The current prototype rejects the resulting interval as contradictory.

At this fixed point the old body bound reports zero change and zero remainder. Its literal floating-point centers nevertheless miss the exact norm identity by 5.33 × 10⁻¹⁴ m²/s⁴. Independently streamed source centers differ by up to 5.54 × 10⁻¹⁶ m/s². These are tiny numerical discrepancies, but a zero-width mathematical interval cannot silently absorb them.

This check blocks installation of the tighter source adapter. It does not change the 53-stretch aircraft findings or the separately qualified transport implementation. No aircraft fitting occurred.

The correction needs an explicit bound for source-reference arithmetic, including rotation evaluation, matrix products and removal of gravity from the rounded acceleration center. A rotation-center enclosure can be constructed with outward polynomial arithmetic and a proved series-tail bound, avoiding an assumed error limit for the system trigonometric library. Propagate these allowances through the source calculation and retain all reference-row differences. Then rerun point, tangent and clock-boundary checks.

Do not merely allow the measured discrepancy as a constant, widen physical calibration bounds, suppress the contradiction or increase comparison tolerances.

The reference-rounding work now includes an exact-rational rotation-center component. Propagation through the complete source remains open.