Honest accuracy tracking
Accuracy report
Every prediction is stored with a timestamp before kick-off and graded when the result is final. Nothing is deleted or re-scored. Predictions created after kick-off are excluded (56 rows seeded on 2026-06-27, the first day of operation).
AI model leaderboard
Same fixtures, seven models
| # | Model | Overall | Match winner | Last 100 | Graded |
|---|---|---|---|---|---|
| 1 | Scorebase Model Elo + Dixon-Coles + market blend | 58% | 55% (2544) | 59% | 4174/7136 |
| 2 | Kimi K3 Moonshot | 58% | 56% (1793) | 62% | 2844/4863 |
| 3 | GPT-5.6 OpenAI | 58% | 56% (2350) | 62% | 3702/6336 |
| 4 | Gemini Google | 58% | 55% (2049) | 57% | 2582/4470 |
| 5 | Grok xAI | 58% | 55% (2007) | 59% | 2927/5076 |
| 6 | Qwen 2.5 Alibaba (local) | 57% | 56% (1672) | 53% | 2292/4007 |
| 7 | Claude Anthropic | 57% | 56% (2072) | 60% | 2894/5084 |
Overall = match winner + handicap + over/under combined. Models joined at different dates, so "Last 100" compares them on equal sample size.
Scorebase model by league
Match-winner hit rate
| League | All-time | Last 30d | Last 7d | Strong picks | Graded |
|---|---|---|---|---|---|
| Premier League | 46% | 45% | — | 100% (4) | 196/425 |
| LaLiga | 50% | 54% | — | 100% (12) | 195/389 |
| Bundesliga | 48% | 56% | — | 70% (10) | 142/294 |
| Serie A | 44% | 50% | — | 80% (5) | 166/374 |
| Ligue 1 | 46% | 61% | — | 100% (2) | 140/302 |
| Champions League | 52% | 73% | — | 100% (3) | 61/118 |
| MLS | 46% | 39% | 100% | 17% (6) | 219/477 |
| K League 1 | 40% | 29% | — | 0% (1) | 38/94 |
| NBA | 67% | — | — | 78% (464) | 877/1311 |
| NHL | 54% | 51% | 54% | 57% (649) | 772/1441 |
| MLB | 55% | 57% | 57% | 56% (1783) | 1992/3633 |
| KBO League | 53% | 60% | 76% | 53% (239) | 485/907 |
| NPB | 56% | 56% | 38% | 59% (427) | 459/823 |
Football draws count as a miss unless the model picked the draw. Strong picks are fixtures where the top probability clears a league-specific threshold.