Smaller orientation regions help, but not enough

All 48 generated sector controls pass. All six complete model/processing cases retain their known future truth. None of the original yaw range is discarded: every case keeps all eight sectors, along with all other calibration directions.

Compared with the saved broad-box construction, future range widths fall to about 50% north, 29% east and 23% vertically. That is roughly a twofold to fourfold improvement. None reaches the predeclared tenfold target. Typical globe ranges still span about 386 m north, 172 m east and 118 m in height after only 1.8 seconds of generated propagation. These are conservative calculated ranges, not measured aircraft errors or survey-IMU drift.

Both pieces remain important. In the first specified-field/no-removal case, the largest sector's nonlinear allowance alone contributes about 187 m north, 80 m east and 54 m vertically to its range width. Remaining affine uncertainty adds about 198 / 91 / 64 m. More precise linear optimization alone cannot remove the nonlinear contribution.

The completed study spent 192 native evaluations in 411 seconds. It performs no Earth-model fits, original IMU reads or protected ESA evaluations. Preserve its source freeze, all sectors and its failure to meet the improvement target; do not restart it or spend another study's unused allowance.

A concrete source of overestimation

The current bounds treat each packet's uncertain cutoff separately, even though every packet shares one time offset. Taking absolute remainders before combining them loses cancellations.

For this generated 400-packet clock and the 1.80 ± 0.15 second query, an exact rational calculation gives a sum of individual packet-time remainders of 0.445 seconds. Multiplied by a constant 150 m/s northward velocity, that alone can contribute 66.75 m to a conservative remainder. Yet the combined elapsed time is essentially linear: its start-tag rounding defect is below 2.06 × 10⁻¹⁴ seconds, permitting a combined remainder below 4.12 × 10⁻¹⁴ seconds for this interior query. This diagnoses an uncertainty-construction loss, not an actual instrument timing error.

The next method will preserve the shared clock and combine force/gravity terms before bounding them. Those changes require independent mathematical controls and a separate native qualification; the completed results will stay unchanged. Useful long-record bounds, global numerical treatment and observational-error adequacy remain open, so new qualified aircraft prediction fits remain blocked.

All six unions · Frozen allowance · Shared-clock next plan · Roadmap.