Prediction bounds at every training epoch
This software study keeps the complete physical calibration range and calculates 289 training envelopes for each of six generated histories. No earlier one-metre training proof, aircraft original or withheld observation enters the calculation.
| Candidate | Processing | Saved epochs | Result |
|---|---|---|---|
| Globe + field | Zero removal | 58 | Not qualified |
| Globe + field | Profiled removal | 58 | Not qualified |
| Globe, zero field | Zero removal | 58 | Not qualified |
| Globe, zero field | Profiled removal | 58 | Not qualified |
| Stationary disc | Zero removal | 58 | Not qualified |
| Stationary disc | Profiled removal | 58 | Not qualified |
0/6 complete batches passed; 348/1,734 reports preserved.
These are training bounds, not withheld forecasts or scientific detections. The next stage connects each of the four declared noise laws to fresh training constraints, then checks every later epoch under a separate finite allowance. Observed error-law adequacy and whole-program numerical coverage remain separate gates.