Agent Verification on MIMIC-IV Pancreatitis
94.71AUROCGLEAN + Active (εu = 0.0)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GLEAN + Active (εu = 0.0)Agent Backbone=Qwen2.5-7B-Instruct2026.03 | 94.71 | 32.11 | 7.6 | 9.53 | |
| GLEAN + Active (εu = 0.5)Agent Backbone=Qwen2.5-7B-Instruct2026.03 | 93.25 | 33.43 | 7.01 | 10.31 | |
| GLEAN + Active (εu = 0.0)Agent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 91.19 | 29.15 | 7.22 | 11.75 | |
| GLEAN (K = 3)Agent Backbone=Qwen2.5-7B-Instruct2026.03 | 90.78 | 35.82 | 9.94 | 13.18 | |
| GLEAN + Active (εu = 0.5)Agent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 90.61 | 28.28 | 7.47 | 11.94 | |
| GLEAN (K = 3)Agent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 88.99 | 30.03 | 7.77 | 13.5 | |
| RAG-AugmentedAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 85.9 | 33.53 | 16.71 | 18.41 | |
| RAG-AugmentedAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 84.61 | 41.86 | 24.84 | 25.31 | |
| GLEAN (K = 1)Agent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 83.67 | 35.86 | 9.53 | 16.96 | |
| GLEAN (K = 1)Agent Backbone=Qwen2.5-7B-Instruct2026.03 | 82.48 | 43.31 | 9.33 | 17.14 | |
| Self-ConsistencyAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 82.32 | 37.32 | 20.88 | 20.96 | |
| Semantic EntropyAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 79.91 | 36.44 | — | — | |
| LLM-as-a-JudgeAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 76.63 | 41.69 | 14.88 | 20.83 | |
| Self-ConsistencyAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 76.21 | 47.09 | 25.82 | 24.46 | |
| Self-VerificationAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 75.67 | 43.44 | 34.19 | 32.41 | |
| P(TRUE)Agent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 75.1 | 45.19 | 27.38 | 28.15 | |
| Semantic EntropyAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 71.74 | 48.84 | — | — | |
| ORMAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 71.46 | 44.92 | 21.09 | 26.7 | |
| ORMAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 70.02 | 51.08 | 19.64 | 24.34 | |
| Self-VerificationAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 68.09 | 52.62 | 10.71 | 22.98 | |
| P(TRUE)Agent Backbone=Qwen2.5-7B-Instruct2026.03 | 65.65 | 56.1 | 40.09 | 41.03 | |
| LLM-as-a-JudgeAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 65.28 | 55.23 | 37.01 | 35.49 | |
| Med-PRMAgent Backbone=Qwen3-30B-A3B-Instruct2026.03 | 61.13 | 52.77 | 56.54 | 55.15 | |
| Med-PRMAgent Backbone=Qwen2.5-7B-Instruct2026.03 | 58.77 | 59.3 | 57.97 | 56.08 |