Walk-forward test on 15 slates (127 games) from 2026-10-30 to 2026-11-14.
Each slate was run through the same engine as the live page with every strategy, then settled from the box scores. Leg chances use the same mix
of our models and Soccerbet's fair price as live (50% models), chosen on earlier games only.
Prices are Soccerbet singles multiplied. The ± is one standard error: a result within two of zero is noise, so no edge is proven.
All engine tickets
606 tickets: predicted 20.3%, won 19.5%; Brier score 0.1440 (always guessing the average: 0.1568); return -10.0% ± 10 per ticket. Single legs: predicted 56.8%, won 56.9% (1,920 legs).
Before the rework, the per-game builder made 22 tickets in 206 test games (predicted 12.1%, won 4.5%).
By strategy
| Strategy | Slates with a ticket | Tickets | Predicted | Won | Avg odds | Return per ticket |
|---|---|---|---|---|---|---|
| Collective odds 3 | 15/15 | 60 | 35.4% | 33.3% | 2.65 | -11.2% ± 16 |
| Collective odds 5 | 15/15 | 60 | 20.9% | 21.7% | 4.44 | -2.8% ± 24 |
| Collective odds 8 | 15/15 | 60 | 13.0% | 13.3% | 7.20 | -5.2% ± 31 |
| Collective odds 12 | 15/15 | 59 | 9.8% | 6.8% | 10.33 | -30.8% ± 34 |
| Collective odds 20 | 15/15 | 56 | 5.6% | 5.4% | 17.49 | -7.8% ± 52 |
| Best over points | 15/15 | 45 | 28.5% | 22.2% | 3.41 | -24.9% ± 21 |
| Best under points | 15/15 | 45 | 30.4% | 20.0% | 3.37 | -31.6% ± 21 |
| Best player props | 15/15 | 56 | 20.5% | 19.6% | 4.70 | -10.9% ± 25 |
| Best game ticket | 0/15 | no ticket in the test: it needs team-market prices, which the test does not have | ||||
| Conservative | 12/15 | 46 | 38.6% | 45.6% | 2.23 | +1.9% ± 17 |
| Balanced | 15/15 | 60 | 22.1% | 26.7% | 4.11 | +2.3% ± 22 |
| High upside | 15/15 | 59 | 6.0% | 5.1% | 25.33 | +4.9% ± 62 |
Tickets per strategy include its distinct alternatives (highest chance, lowest correlation, other market mix). They share legs, so they are not independent bets.
By option
| Tickets | W-L | Predicted | Won | Brier | Avg odds | Return per ticket | |
|---|---|---|---|---|---|---|---|
| Best for the strategy | 162 | 31-131 | 20.8% | 19.1% | 0.1388 | 8.10 | -7.7% ± 23 |
| Lowest correlation | 154 | 34-120 | 20.6% | 22.1% | 0.1593 | 8.06 | -1.9% ± 19 |
| Different market mix | 128 | 17-111 | 18.2% | 13.3% | 0.1160 | 8.83 | -19.9% ± 24 |
| Highest chance | 162 | 36-126 | 21.0% | 22.2% | 0.1569 | 7.25 | -12.1% ± 16 |
By number of legs
| Tickets | W-L | Predicted | Won | Brier | Avg odds | Return per ticket | |
|---|---|---|---|---|---|---|---|
| 2 | 196 | 60-136 | 33.4% | 30.6% | 0.2068 | 2.89 | -15.9% ± 9 |
| 3 | 185 | 41-144 | 20.9% | 22.2% | 0.1696 | 4.54 | -4.5% ± 13 |
| 4 | 161 | 14-147 | 9.4% | 8.7% | 0.0778 | 14.82 | -5.3% ± 28 |
| 5 | 55 | 3-52 | 6.1% | 5.5% | 0.0517 | 16.50 | -6.1% ± 53 |
| 6 | 9 | 0-9 | 4.7% | 0.0% | 0.0022 | 17.52 | -100.0% ± 0 |
By combined odds
| Tickets | W-L | Predicted | Won | Brier | Avg odds | Return per ticket | |
|---|---|---|---|---|---|---|---|
| 16+ | 102 | 5-97 | 5.5% | 4.9% | 0.0463 | 22.45 | -1.3% ± 44 |
| 2.5-4.5 | 265 | 64-201 | 27.6% | 24.1% | 0.1820 | 3.52 | -18.9% ± 9 |
| 4.5-9 | 121 | 23-98 | 16.3% | 19.0% | 0.1497 | 6.11 | +6.8% ± 21 |
| 9-16 | 72 | 5-67 | 9.5% | 6.9% | 0.0646 | 11.02 | -25.5% ± 33 |
| under 2.5 | 46 | 21-25 | 38.6% | 45.6% | 0.2518 | 2.23 | +1.9% ± 17 |
By league
| Tickets | W-L | Predicted | Won | Brier | Avg odds | Return per ticket | |
|---|---|---|---|---|---|---|---|
| EuroLeague | 40 | 17-23 | 19.2% | 42.5% | 0.2978 | 8.44 | +221.5% ± 96 |
| Mixed | 56 | 6-50 | 16.1% | 10.7% | 0.0794 | 10.99 | -66.1% ± 14 |
| NBA | 510 | 95-415 | 20.8% | 18.6% | 0.1391 | 7.66 | -22.0% ± 9 |
Live record published tickets, settled from the box scores
No published engine ticket has been settled yet.
What the test cannot see
- player props only: Game Analysis team-market prices before tip-off are not joined in this test
- no historical injury status, minutes ranges or games-played counts: tiers use expected minutes only
- main lines only (alternative and milestone lines are not stored historically)
- prices = Soccerbet singles multiplied; same-game prices are not stored
- alternatives share legs with the best ticket of the same preset, so tickets are not independent bets
- about 3 weeks of test games: ROI standard errors are large; no edge is claimed