Any-violation monitoring on Safety conflict monitor dataset (edge-covered)
82.9Positive RateBT margins
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BT marginsModel=3B2026.06 | 82.9 | 98.3 | — | 70.6 | 78.4 | 11.1 | |
| full-hidden baselineModel=3B2026.06 | 82.9 | — | 96.4 | — | — | — | |
| BT marginsModel=14B2026.06 | 81.2 | 98.7 | — | 77.8 | 88 | 5.9 | |
| full-hidden baselineModel=14B2026.06 | 81.2 | — | 97.3 | — | — | — | |
| BT marginsModel=7B2026.06 | 52.1 | 94.9 | — | 73.8 | 80.6 | 9.5 | |
| full-hidden baselineModel=7B2026.06 | 52.1 | — | 86 | — | — | — |