Hint-sensitivity forecasting on FEVEROUS 610 trajectories (held-out)
0.8Spearman CorrelationBehavior Forecaster
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Behavior ForecasterType=transferred2026.06 | 0.8 | 0.817 | |
| Claude Opus 4.6 reader2026.06 | 0.768 | 0.779 | |
| Behavior ForecasterType=direct2026.06 | 0.732 | 0.814 | |
| GPT-5.4 reader2026.06 | 0.668 | 0.68 |