Ranking Consistency Analysis on MMLU-Pro Anatomy health
0.62Spearman CorrelationSCICUSTOM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SCICUSTOMConstruction Method=Grounded data and ontology-driven2026.05 | 0.62 | 0.51 | |
| GPQAConstruction Method=General scientific benchmark2026.05 | 0.48 | 0.32 | |
| GPT-5Construction Method=Synthetic benchmark generation2026.05 | 0.11 | 0.06 | |
| MMLUConstruction Method=Domain-specific comparison2026.05 | -0.15 | 0 | |
| MedQAConstruction Method=Domain-specific comparison2026.05 | -0.19 | -0.1 | |
| IfBenchConstruction Method=General-purpose benchmark2026.05 | -0.26 | -0.21 | |
| EmbeddingConstruction Method=Embedding-based k-NN retrieval2026.05 | -0.32 | -0.16 | |
| SimpleQAConstruction Method=General-purpose benchmark2026.05 | -0.37 | -0.21 |