Leave-one-corpus-out transfer on tau2-bench
72.5AUCrun size (flat)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| run size (flat)Features=flat2026.06 | 72.5 | 66.5 | 0.001 | |
| flat plus dependency layer (f+d)Features=f+d2026.06 | 64.9 | 58.5 | 0.001 | |
| size-normalized dependency layer (dep)Features=dep2026.06 | 64.6 | 58.4 | 0.001 |