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 | 450 | 0.857 | 61% |
| Base rate | 450 | 0.977 | 53% |
| Recent form | 450 | 0.928 | 56% |
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 | 137 | 0.826 | 0.827 | 0.1% |
| Polymarket | 18 Feb 2025 | 237 | 0.878 | 0.862 | −1.8% |
| Kalshi | 29 Apr 2025 | 253 | 0.870 | 0.858 | −1.3% |
Clubs are rated on their domestic league games as well as all three UEFA competitions, so a side that dominates the Eredivisie and one that finishes fourth in England sit on one scale, linked by their European meetings. Only Champions League matches are scored in the track record.
The knockouts are two-legged; the simulation plays each tie as a single game on neutral ground, which is close for well-matched sides and a little generous to underdogs.
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.
Countries are linked only through the UEFA competitions, and ESPN's feeds leave out the qualifying rounds.
Questions and answers
How accurate are PodiumOracle's Champions League predictions?
On 450 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 61% of the time. Its log loss was 0.857, against 0.977 for the base rate and 0.928 for recent form. Against the betting markets on the same finished games, its log loss was 0.826 against the bookmakers' 0.827 over 137 games, 0.878 against Polymarket's 0.862 over 237 games and 0.870 against Kalshi's 0.858 over 253 games (lower is better).