Collusion Detection on NARCBENCH CORE
100AUROCSorted Concatenation
Evaluation Results
| Method | Links | |
|---|---|---|
| Sorted ConcatenationTraining Split=CORE, Layers=26–30, Evaluation Protocol=In-distribution2026.04 | 100 | |
| Activation AlignmentTraining Split=CORE, Layers=26–30, Evaluation Protocol=In-distribution2026.04 | 100 | |
| Peak SuspicionTraining Split=CORE, Layers=26–30, Evaluation Protocol=In-distribution2026.04 | 100 | |
| Deception SplitTraining Split=CORE, Layers=26–30, Evaluation Protocol=In-distribution2026.04 | 100 | |
| Asymmetry ProbeTraining Split=CORE, Layers=26–30, Evaluation Protocol=In-distribution2026.04 | 100 |