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 | 650 | 0.617 | 68% |
| Base rate | 650 | 0.734 | 57% |
| Recent form | 650 | 0.648 | 66% |
| Betting market | 650 | 0.600 | 69% |
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) | 5 Mar 2026 | 177 | 0.561 | 0.529 | −6.1% |
| Polymarket | 8 Aug 2026 | 32 | 0.533 | 0.531 | −0.4% |
| Kalshi | 5 Mar 2026 | 204 | 0.561 | 0.520 | −7.9% |
The finals are simulated with the real system: since 2026 a wildcard round (7th v 10th, 8th v 9th) and then the top-eight finals (qualifying and elimination finals, semi-finals, preliminary finals and the grand final at the MCG). Finals already played keep their real result; seasons before 2026 are replayed in the top-eight system they used.
Home advantage is learned from every game; interstate travel is where most of it comes from, and the model does not yet separate a Melbourne derby at the MCG from a trip to Perth.
The market comparison uses the implied probabilities Squiggle records from the bookmakers (its 'Punters' source).
Selection, injuries and suspensions are not in the model.
Games under the Docklands roof are treated as indoors for the weather.
Questions and answers
How accurate are PodiumOracle's AFL predictions?
On 650 games from held-back seasons (2024 to 2026), which the model never saw while it was tuned, the side it favoured won 68% of the time. Its log loss was 0.617, against 0.734 for the base rate and 0.648 for recent form, and 0.600 for the closing betting odds. Against the betting markets on the same finished games, its log loss was 0.561 against the bookmakers' 0.529 over 177 games, 0.533 against Polymarket's 0.531 over 32 games and 0.561 against Kalshi's 0.520 over 204 games (lower is better).