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 | 239 | 0.649 | 59% |
| Base rate | 239 | 0.694 | 54% |
| Recent form | 239 | 0.674 | 56% |
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) | 21 Jan 2026 | 113 | 0.634 | 0.639 | 0.8% |
| Polymarket | 9 Jan 2026 | 121 | 0.632 | 0.643 | 1.6% |
| Kalshi | 4 Dec 2025 | 165 | 0.640 | 0.646 | 1.0% |
The finals are simulated with the current format: third plays fourth and fifth plays sixth, the 3-4 winner takes the third seed and the 3-4 loser meets the 5-6 winner for the fourth; then best-of-three semi-finals (1 v 4, 2 v 3) and a best-of-five grand final series.
ESPN carries the NBL from the 2022–23 season, so the model has only a few seasons to learn from; its parameters are fit on two and scored on one.
Imports and injuries swing NBL teams more than most leagues; the rating only sees results.
The NBL Cup and games against touring NBA sides are left out.
Questions and answers
How accurate are PodiumOracle's NBL predictions?
On 239 games from held-back seasons (2025–26 to 2026–27), which the model never saw while it was tuned, the side it favoured won 59% of the time. Its log loss was 0.649, against 0.694 for the base rate and 0.674 for recent form. Against the betting markets on the same finished games, its log loss was 0.634 against the bookmakers' 0.639 over 113 games, 0.632 against Polymarket's 0.643 over 121 games and 0.640 against Kalshi's 0.646 over 165 games (lower is better).