Collusion Detection on NARCBENCH Research Pipeline
0.94AUROCDeception Split
Evaluation Results
| Method | Links | |
|---|---|---|
| Deception SplitTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.04 | 0.94 | |
| Asymmetry ProbeTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.04 | 0.92 | |
| Peak SuspicionTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.04 | 0.83 | |
| Sorted ConcatenationTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.04 | 0.78 | |
| Activation AlignmentTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.04 | 0.56 |