Four groups now meet the prediction-width target
All 48 sectors pass correctness; four of six groups meet the unchanged twofold width-reduction target. The preceding comparison passed three groups. Both specified-field globe groups still miss the target, so the complete method has not qualified for observed prediction work.
Complete 7.2-second position widths improve by about 3–5% for globe candidates and 1–2% for flat disc. The resulting widths span about 0.21–1.63 m. These are generated controls, not measured aircraft errors or calibrated intervals. The much larger improvement in the acceleration-change radius does not carry through proportionally: most remaining width comes from the affine uncertainty, with additional position and lever-arm remainders.
The study finished in 794.3 seconds. It used 48 enclosure calls and no new native predictions, original-file reads or observed fits. The independent accounting audit reconciles all 672 hashed reservation events with the frozen 576 LP / 2,304 exact support limits. No allowance was expanded.
Training regions, all eight sectors per group, all 51 coordinates and physical limits remain. Predictions were saved before known outcomes were checked. No aircraft finding, uncertainty claim or independent-flight count changes.
Next, qualify stronger relationships between matrix entries. Earlier diagnostics found that the linear uncertainty region admits matrices incompatible with a physical rotation. Separate column constraints barely helped; the next component must enforce necessary relationships between columns without adding physical calibration freedom or dropping any existing coordinate.