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 | 531 | 0.983 | 52% |
| Base rate | 531 | 1.064 | 45% |
| Recent form | 531 | 1.001 | 51% |
| Betting market | 531 | 0.970 | 52% |
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) | 20 Jan 2026 | 172 | 0.984 | 0.971 | −1.4% |
| Polymarket | 22 Nov 2025 | 228 | 0.964 | 0.953 | −1.2% |
| Kalshi | 29 Nov 2025 | 223 | 0.976 | 0.965 | −1.2% |
A draw is its own outcome: the model has a band around an even contest in which a draw is the likeliest single result, fit on the tuning seasons.
The market comparison uses the bookmakers' closing odds with their margin taken out.
The five post-split games are only scheduled once the split is known, so until then the simulation plays the 33 scheduled rounds and reports the table after them; after the split it plays the real fixtures.
Team news, injuries and rotation for European games are not in the model.
Expected goals are not in the rating yet; it learns only from results.
Questions and answers
How accurate are PodiumOracle's Scottish Premiership predictions?
On 531 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 52% of the time. Its log loss was 0.983, against 1.064 for the base rate and 1.001 for recent form, and 0.970 for the closing betting odds. Against the betting markets on the same finished games, its log loss was 0.984 against the bookmakers' 0.971 over 172 games, 0.964 against Polymarket's 0.953 over 228 games and 0.976 against Kalshi's 0.965 over 223 games (lower is better).