Factual Reasoning on FreshQA v2
16Baseline WinsDomain-Grounded Tiered Retrieval
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Domain-Grounded Tiered RetrievalN=1502026.03 | 16 | 97 | 37 | 64.7 | 3.5 | 19.2 | |
| Baseline (Llama 3.1 8B)N=1502026.03 | 16 | — | — | — | — | — | |
| Domain-Grounded Tiered RetrievalN=34, subset=factual-answer2026.03 | — | — | — | — | 15.5 | 84.5 |