The go/no-go gate (test 0.8) already cleared: only 1.28% of stripes double back, well under the bar. The training dataset was built and the fine-tune was configured, but training/scoring did not complete.
| Check | Result |
|---|---|
| Gate 0.8 (double-back audit) | cleared (1.28%) |
| Training set built | 1500 train / 150 val, 0 holdout overlap |
| Rejected from training | 43 non-monotone, 633 rated bad, 23 aspect mismatch |
Source: model-clrernet.json.