Any-violation detection on Priority Graph (held-out)
94.6AUROC (BT)BT min-margin monitor
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BT min-margin monitorModel=7B2026.06 | 94.6 | — | 11.5 | 77.3 | 6.4 | |
| BT min-margin monitorModel=14B2026.06 | 91.9 | — | 11.5 | 63.9 | 16.2 | |
| BT min-margin monitorModel=3B2026.06 | 84.1 | — | 13.9 | 38.9 | 31.1 | |
| full-hidden linear probeModel=3B2026.06 | — | 70.3 | — | — | — | |
| full-hidden linear probeModel=7B2026.06 | — | 83.1 | — | — | — | |
| full-hidden linear probeModel=14B2026.06 | — | 80.4 | — | — | — |