Tool-use Policy Auditing on τ2-BENCH Telecom (test)
30PASS^4Baseline
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BaselineAgent and Verifier Model=GPT 5.4, n=4, Verifier type=paired-verifier2026.06 | 30 | 47.6 | 10.5 | |
| PG-CHECKLISTVariant=CHECKLIST, Agent and Verifier Model=GPT 5.4, n=4, Verifier type=paired-verifier2026.06 | 30 | 42.9 | 15.8 | |
| PG-RAWVariant=RAW, Agent and Verifier Model=GPT 5.4, n=4, Verifier type=paired-verifier2026.06 | 25 | 47.6 | 0 |