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 | 360 | 0.974 | 54% |
| Base rate | 360 | 1.031 | 51% |
| Recent form | 360 | 1.010 | 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) | 19 Feb 2026 | 83 | 0.961 | 0.893 | −7.6% |
| Polymarket | 23 Oct 2025 | 157 | 0.965 | 0.893 | −8.1% |
| Kalshi | 6 Nov 2025 | 160 | 0.987 | 0.920 | −7.3% |
Clubs are rated on their domestic league games as well as all three UEFA competitions, so a Norwegian champion and a mid-table Premier League side sit on one scale, linked by European meetings. Only Conference League matches are scored in the track record.
Knockout ties are two legs; the simulation plays each as one game on neutral ground.
Clubs from leagues football-data.co.uk doesn't carry (Czechia, Croatia, Serbia, Ukraine, Israel, Cyprus and others) are rated only on their European matches.
The Conference League has only a few seasons, so it uses the settings fitted for the Europa League, which rates the same clubs on the same scale.
Questions and answers
How accurate are PodiumOracle's Conference League predictions?
On 360 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 54% of the time. Its log loss was 0.974, against 1.031 for the base rate and 1.010 for recent form. Against the betting markets on the same finished games, its log loss was 0.961 against the bookmakers' 0.893 over 83 games, 0.965 against Polymarket's 0.893 over 157 games and 0.987 against Kalshi's 0.920 over 160 games (lower is better).