How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Held-back seasons
| Forecast | Forecasts | Log loss | Favourite won |
|---|---|---|---|
| The model | 622 | 1.003 | 51% |
| Base rate | 622 | 1.052 | 45% |
| Recent form | 622 | 1.032 | 50% |
Against the betting markets
Each market's price before the start, scored on the same finished games as the model. A negative figure means the market was more accurate.
| Market | Since | Priced | Model log loss | Market log loss | Model vs market |
|---|---|---|---|---|---|
| Bookmakers (Pinnacle, DraftKings, bet365) | 14 Mar 2026 | 243 | 0.991 | 0.955 | −3.8% |
| Polymarket | 28 Mar 2026 | 195 | 0.995 | 0.951 | −4.6% |
| Kalshi | 14 Mar 2026 | 228 | 0.997 | 0.961 | −3.7% |
Summer heat counts as rough weather here, as well as wind and rain.
The NWSL Shield is the 'Finish first' column.
International call-ups and injuries are not in the model.
No historical betting lines are in the source, so there is no market comparison.
Questions and answers
How accurate are PodiumOracle's NWSL predictions?
On 622 games from held-back seasons (2024 to 2026), which the model never saw while it was tuned, the side it favoured won 51% of the time. Its log loss was 1.003, against 1.052 for the base rate and 1.032 for recent form. Against the betting markets on the same finished games, its log loss was 0.991 against the bookmakers' 0.955 over 243 games, 0.995 against Polymarket's 0.951 over 195 games and 0.997 against Kalshi's 0.961 over 228 games (lower is better).