List your company

False Success Detection on tau2-bench v1 v2 (test)

89RecallTF-IDF + XGBoost
2.6825.0947.569.91Jun 1, 2026

Evaluation results

MethodLinks
TF-IDF + XGBoostFlag rate=20%2026.068931—
TF-IDF + XGBoostFlag rate=10%2026.067250—
TF-IDF + XGBoostFlag rate=5%2026.065172—
Judge (Sonnet 4.5)Flag rate=20%, Condition=no closing2026.062555—
Judge (Sonnet 4.5)Flag rate=10%, Condition=no closing2026.061358—
Judge (Sonnet 4.5)Flag rate=5%, Condition=no closing2026.06656—
Judge (Sonnet 4.5)Interface=LLM API call2026.06——4,000
TF-IDF + XGBoostHardware=CPU2026.06——1.19