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 | 629 | 0.693 | 63% |
| Base rate | 629 | 0.705 | 56% |
| Recent form | 629 | 0.699 | 58% |
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) | 1 Mar 2026 | 138 | 0.688 | 0.672 | −2.2% |
| Polymarket | 31 Jul 2026 | 37 | 0.761 | 0.698 | −9.0% |
| Kalshi | 1 Mar 2026 | 159 | 0.684 | 0.679 | −0.6% |
The finals are simulated with the real top-eight system: qualifying and elimination finals, semi-finals, preliminary finals and the grand final. Games already played in the finals keep their real result.
A draw after golden point is rare, so the draw band the model learns is narrow.
The ladder is ESPN's official one, with two points for each bye.
The rating does not know who is injured, suspended or rested. Origin weeks are not treated differently.
Questions and answers
How accurate are PodiumOracle's NRL predictions?
On 629 games from held-back seasons (2024 to 2026), which the model never saw while it was tuned, the side it favoured won 63% of the time. Its log loss was 0.693, against 0.705 for the base rate and 0.699 for recent form. Against the betting markets on the same finished games, its log loss was 0.688 against the bookmakers' 0.672 over 138 games, 0.761 against Polymarket's 0.698 over 37 games and 0.684 against Kalshi's 0.679 over 159 games (lower is better).