Web agent task completion on GoBrowse
90.4Success RateCATTS (∆, best τ)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CATTS (∆, best τ)Gating=Margin, N=102026.02 | 90.4 | 372,000 | |
| CATTS (H, best τ)Gating=Entropy, N=10, Threshold (τ)=0.52026.02 | 90.2 | 422,000 | |
| Always-arbitrateN=102026.02 | 88.3 | 443,000 | |
| Majority voteN=102026.02 | 88 | 481,000 |