Win probability forecasting on Held-out 2024 n=5185 (held-out)
0.1355Brier ScoreBetting market
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Betting market2026.06 | 0.1355 | 0.0273 | 0.0824 | 79.9 | 0.1148 | |
| Empirical rate p^2026.06 | 0.1432 | 0.0437 | 0.0993 | 78.3 | 0.1083 | |
| DeepSeek-V4Evaluation Protocol=zero-shot2026.06 | 0.1438 | 0.043 | 0.0716 | 79 | 0.1078 | |
| Direct RLVRTraining Method=RLVR2026.06 | 0.1443 | 0.0292 | 0.0596 | 78.4 | 0.1058 | |
| Masked-CoT RLVRTraining Method=RLVR2026.06 | 0.1522 | 0.0293 | 0.0684 | 77.7 | 0.0986 | |
| Qwen2.5-7BEvaluation Protocol=zero-shot, Prompting Strategy=CoT2026.06 | 0.1681 | 0.0687 | 0.1493 | 75.6 | 0.0875 | |
| Qwen2.5-7BEvaluation Protocol=zero-shot, Prompting Strategy=direct2026.06 | 0.2057 | 0.0569 | 0.2539 | 64.4 | 0.0388 |