Fact Verification on FEVER (EM, $Δ$NF)
93.6Exact Match (EM)MADARA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MADARAModel=Gemma-2-9B, RSC=Quality-Ordered, ρ̂1=0.58, Strategy=ATF2026.06 | 93.6 | 1.4 | |
| CoTModel=Mistral-7B-v0.32026.06 | 92.6 | — | |
| MADARAModel=Mistral-7B-v0.3, RSC=Quality-Ordered, ρ̂1=0.28, Strategy=CoT2026.06 | 92.6 | 1.9 | |
| CoTModel=Gemma-2-9B2026.06 | 92.5 | — | |
| SDAModel=Gemma-2-9B2026.06 | 92.5 | — | |
| 3-AgentModel=Gemma-2-9B2026.06 | 92.4 | — | |
| NFModel=Gemma-2-9B2026.06 | 92.2 | — | |
| MADARA (Average)2026.06 | 91.2 | 1.5 | |
| MADARAModel=Llama-3.1-8B, RSC=Quality-Ordered, ρ̂1=0.65, Strategy=ATF2026.06 | 90.9 | 2.2 | |
| 3-AgentModel=Mistral-7B-v0.32026.06 | 90.9 | — | |
| MADARAModel=Qwen3-8B, RSC=Quality-Ordered, ρ̂1=0.63, Strategy=ATF2026.06 | 90.9 | 1.3 | |
| NFModel=Mistral-7B-v0.32026.06 | 90.7 | — | |
| SDAModel=Mistral-7B-v0.32026.06 | 90.7 | — | |
| SDAModel=Qwen2.5-7B2026.06 | 90.5 | — | |
| AverageMethod_Group=SDA2026.06 | 90.4 | — | |
| AverageMethod_Group=CoT2026.06 | 90.1 | — | |
| SDAModel=Qwen3-8B2026.06 | 90 | — | |
| AverageMethod_Group=3-Agent2026.06 | 90 | — | |
| 3-AgentModel=Qwen3-8B2026.06 | 89.8 | — | |
| AverageMethod_Group=NF2026.06 | 89.7 | — | |
| NFModel=Qwen3-8B2026.06 | 89.6 | — | |
| CoTModel=Qwen3-8B2026.06 | 89.5 | — | |
| 3-AgentModel=Qwen2.5-7B2026.06 | 88.9 | — | |
| NFModel=Llama-3.1-8B2026.06 | 88.7 | — | |
| CoTModel=Llama-3.1-8B2026.06 | 88.4 | — | |
| SDAModel=Llama-3.1-8B2026.06 | 88.2 | — | |
| MADARAModel=Qwen2.5-7B, RSC=Quality-Ordered, ρ̂1=0.62, Strategy=ATF†2026.06 | 88.2 | 1.1 | |
| 3-AgentModel=Llama-3.1-8B2026.06 | 87.8 | — | |
| CoTModel=Qwen2.5-7B2026.06 | 87.5 | — | |
| NFModel=Qwen2.5-7B2026.06 | 87.1 | — |