Patient Simulation on Client-Therapist Interactions CBT Therapist persona
4.8Factual ConsistencyClientCAST
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ClientCASTEvaluation Protocol=LLM judge, 5-point Likert scale2026.02 | 4.8 | 4.32 | 4.15 | 3.62 | 3.78 | 3.11 | 3.36 | 3.82 | 3.84 | 32.87 | 33.87 | 0.07 | |
| AnnaAgentEvaluation Protocol=LLM judge, 5-point Likert scale2026.02 | 4.78 | 4.25 | 4.15 | 3.62 | 3.8 | 3.17 | 3.33 | 3.78 | 3.87 | 172.1 | 317.13 | 0.24 | |
| PATIENT-ΨEvaluation Protocol=LLM judge, 5-point Likert scale2026.02 | 4.75 | 4.45 | 4.45 | 3.6 | 3.9 | 3.15 | 3.5 | 4 | 4.05 | 97.03 | 98.03 | 0.07 | |
| EeyoreEvaluation Protocol=LLM judge, 5-point Likert scale2026.02 | 4.7 | 4.25 | 4.25 | 3.52 | 3.68 | 3.08 | 3.28 | 3.72 | 3.9 | 26.01 | 28.01 | — | |
| Roleplay-dohEvaluation Protocol=LLM judge, 5-point Likert scale2026.02 | 4.67 | 4.22 | 4.1 | 3.57 | 3.75 | 3.05 | 3.32 | 3.79 | 3.8 | 103.97 | 502.46 | 0.18 |