Knowledge Boundary Fidelity on KBF-QA
68.1KRF AccuracyREVERIEMEM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| REVERIEMEMBackbone=gpt-5-mini2026.06 | 68.1 | 81.2 | 73.3 | |
| RAPTORBackbone=gpt-5-mini2026.06 | 66.4 | 8.7 | 16.8 | |
| HippoRAGBackbone=gpt-5-mini2026.06 | 65.4 | 11.3 | 20.9 | |
| naive RAGBackbone=gpt-5-mini2026.06 | 63.8 | 10 | 18.9 | |
| BOOKWORLDBackbone=gpt-5-mini2026.06 | 41.6 | 35.5 | 38.7 | |
| directBackbone=gpt-5-mini, Retrieval=None2026.06 | 36.4 | 38.2 | 37.2 | |
| ComoRAGBackbone=gpt-5-mini2026.06 | 33 | 45.9 | 37.7 |