Retrieval-Augmented Generation on Versioned-policy benchmark (138 held-out queries)
66.7Overall AccuracyRAG
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RAGaugmentation=Temporal2026.04 | 66.7 | 60 | 71.7 | 71.4 | 6.7 | 0.259 | |
| SmartVectorweight_config=tuned2026.04 | 64.8 | — | — | — | — | — | |
| SmartVectorweight_config=default2026.04 | 61.7 | 53.3 | 70 | 61.9 | 13.3 | 0.244 | |
| RAGaugmentation=Confidence2026.04 | 36.9 | 58.3 | 5 | 66.7 | 13.3 | 0.572 | |
| RAGvariant=plain2026.04 | 31.2 | 50 | 1.7 | 61.9 | 35 | 0.47 |