- Model: xCup L2 = L1 Dixon-Coles + Elo correction (w=0.5)
- 100,000 sims, seed 0, pre-registered 2026-06-06 (git commit = timestamp)
- Validation: Elo correction is a significant out-of-sample improvement: pooled RPS skill +10.0%->+13.6% over 2010-2022, Diebold-Mariano p=0.002 (see scripts/backtest_elo_compare.py, docs/RESEARCH_LOG.md).
- Market gap Σ|model−market|: L1 raw 1.105 → L1+Elo 0.685
| # | Team | L1+Elo champ | L1 raw | Polymarket |
|---|---|---|---|---|
| 1 | Argentina | 20.3% | 24.0% | 8.8% |
| 2 | Spain | 13.0% | 4.7% | 16.0% |
| 3 | Brazil | 11.5% | 19.0% | 8.2% |
| 4 | Colombia | 8.3% | 11.6% | 1.8% |
| 5 | France | 6.2% | 2.4% | 16.1% |
| 6 | England | 5.5% | 3.4% | 11.2% |
| 7 | Ecuador | 4.5% | 5.4% | 0.9% |
| 8 | Uruguay | 4.0% | 6.9% | 1.1% |
| 9 | Portugal | 3.8% | 2.3% | 10.0% |
| 10 | Japan | 2.6% | 2.6% | 2.1% |
| 11 | Netherlands | 2.1% | 1.0% | 4.0% |
| 12 | Germany | 2.1% | 1.1% | 5.3% |
| 13 | Morocco | 1.8% | 3.1% | 1.6% |
| 14 | Mexico | 1.6% | 1.4% | 1.2% |
| 15 | Belgium | 1.5% | 0.8% | 1.9% |
| 16 | Paraguay | 1.4% | 2.0% | 0.2% |