Web agent task completion on WebArena Lite
0.479Success RateCATTS (H, best τ)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CATTS (H, best τ)Gating=Entropy, N=10, Threshold (τ)=0.22026.02 | 0.479 | 745,000 | |
| CATTS (∆, best τ)Gating=Margin, N=102026.02 | 0.479 | 405,000 | |
| Always-arbitrateN=102026.02 | 0.44 | 762,000 | |
| Majority voteN=102026.02 | 0.432 | 920,000 |