Prediction with training-conditioned acceleration
48/48 cases pass correctness.
Every group meets the unchanged twofold prediction-width target: no.
| Candidate / removal | Complete | Width N / E / height (m) | Fraction of previous width | Twofold target |
|---|---|---|---|---|
| Globe with specified field / zero | Yes | 1.1964 / 1.5584 / 1.0339 | 0.9570 / 0.9595 / 0.9634 | No |
| Globe with specified field / profiled | Yes | 1.2818 / 1.6334 / 1.1203 | 0.9597 / 0.9613 / 0.9661 | No |
| Globe with zero field / zero | Yes | 0.9107 / 0.9761 / 0.7291 | 0.9571 / 0.9485 / 0.9651 | Yes |
| Globe with zero field / profiled | Yes | 1.0659 / 1.1221 / 0.8852 | 0.9631 / 0.9549 / 0.9711 | Yes |
| Flat disc / zero | Yes | 0.6692 / 0.2096 / 0.5155 | 0.9913 / 0.9840 / 0.9885 | Yes |
| Flat disc / profiled | Yes | 0.8588 / 0.2688 / 0.6987 | 0.9932 / 0.9864 / 0.9915 | Yes |
Elapsed 794.3 seconds; 48/48 enclosure calls. Zero new native predictions, original reads or observed fits. The comparison retains every source-bound coefficient, all 51 coordinates and the original training regions. Saved outcomes are checked only after prediction. These are generated software controls, not measured aircraft errors or calibrated intervals.
Frozen scope · Native qualification · Previous widths · Roadmap.