Combinatorial Reasoning on Sudoku
100AccuracyOracle policy
Evaluation Results
| Method | Links | |
|---|---|---|
| Oracle policyTraining protocol=Policy-only training2026.05 | 100 | |
| PolicyTraining protocol=Joint training2026.05 | 92.87 | |
| BaselineTraining protocol=Joint training2026.05 | 92.72 | |
| High conf.Training protocol=Joint training2026.05 | 92.68 | |
| MarginTraining protocol=Joint training2026.05 | 91 | |
| PolicyTraining protocol=Policy-only training2026.05 | 90.82 | |
| High conf.Training protocol=Policy-only training2026.05 | 89.84 | |
| MarginTraining protocol=Policy-only training2026.05 | 88.67 |