Local Level Lab

Airborne measurements. Open methods.

Install the source bound only after transport qualifies

The cached-transport study is complete and qualified: all 24 native controls pass. Do not restart it or enlarge its allowance. This candidate still does not install the tighter norm-constrained body-source interval.

The next adapter must compute source references at actual packet midpoints and endpoints. Stream the body-rotation state and its 27 reference columns instead of storing a full-recording tensor or recomputing every prefix from scratch. Keep clipped clock fractions and the native one-sided boundary convention.

For each prefix, keep the verified training range for A=R f, the original gravity range, and the literal installation matrix's bounded norm defect. Form the vertical interval for B=R(P−I)f and its affine error around the specified native row. Include center and coefficient discrepancies. Retain each later force sample's calibrated difference from the reference force.

Carry (Q−I)(A+B)+B through the component-aware transport path. Unsupported source branches must remain explicit; do not replace them with an unjustified narrow interval. Preserve all physical and shared-error columns and limits.

Before another complete prediction comparison, freeze a finite source/native qualification, check independent generated motions and compare actual baseline allowances. Only a correctly qualified and useful candidate advances to the separate prediction-width gate. Aircraft fits and protected ESA confirmation remain downstream.

Detailed source tasks · Source reference evidence · Roadmap.