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 | 401 | 0.639 | 61% |
| Base rate | 401 | 0.693 | 47% |
| Recent form | 401 | 0.654 | 60% |
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 Mar 2026 | 133 | 0.662 | 0.487 | −35.9% |
Fighters move between promotions, so every rating also counts their UFC and Bellator fights; only PFL fights are forecast and scored in the track record.
The PFL's season tournament (regular-season points, then play-offs) isn't simulated; the page gives each fight's odds and the fighter ratings.
A fight is a win or a loss for the rating, however it ended; a draw counts half each and a no contest not at all.
Ratings carry from one fight to the next with no pull back to average: fighters don't have seasons.
There is no weather here: cards are indoors.
Weight changes, layoffs, age and camp changes are not in the model.
Betting lines aren't in the source, so there is no market comparison.
Questions and answers
How accurate are PodiumOracle's PFL predictions?
On 401 games from held-back seasons (2024 to 2026), which the model never saw while it was tuned, the side it favoured won 61% of the time. Its log loss was 0.639, against 0.693 for the base rate and 0.654 for recent form. Against the betting markets on the same finished games, its log loss was 0.662 against the bookmakers' 0.487 over 133 games (lower is better).