HOW WE CALCULATE
THE PERCENTAGES

Finished-match scores, league tables and season statistics for the European leagues come from the free football-data.org service.

For the 5 big European leagues, each match's expected goals (xG), the likelihood that shots become goals, come from Understat.

Süper Lig matches and the match archive of every league (score, goal minute, cards, corners) come from ESPN data we collect ourselves.

Match context (injuries, motivation, press reports) is gathered by AI scanning current news on the internet. This is for Pro members only.

For the Premier League, La Liga, Serie A, Bundesliga and Ligue 1 the calculation has 3 steps:

1. i.

   Attack and defence strength of each team

   The quality of the chances each team created and conceded in its last 38 league matches (expected goals, xG) is combined with the real goals scored and conceded: 75% chance quality, 25% goals, with newer matches weighted more heavily. The result is compared with the league average.

2. ii.

   Caution for teams with few matches

   A team with few matches played is pulled towards the average, as if it had played 10 matches at the league average. This avoids exaggerated figures early in the season or for newly promoted teams.

3. iii.

   The probability of every scoreline

   Home expected goals = the league's average home goals × the home side's attack strength × the away side's defensive weakness (the reverse for the away side). From these two numbers we work out the probability of every scoreline from 0-0 to 6-6 (in statistics, the Poisson distribution), with a small correction for low scores such as 0-0 and 1-1 (the Dixon-Coles correction).

The settings were tuned on matches from August 2025 to January 2026, then tested once on the last 8 months of matches: every match was predicted using only data from before its date. In that test, the match-result, 3+ goals and both-teams-to-score probabilities were all more accurate than the previous calculation. The current measurement is in the Track record (Turkish).

In the Champions League teams come from different leagues, so a cross-league strength rating (Elo) is used: a single rating built from results in the 5 big leagues, the Süper Lig, the Champions League and the Europa League (since 2014); leagues are linked to each other through European matches, and expected goals come from the rating difference between the two teams. It was measured on the main-tournament matches of the 2019/20–2025/26 seasons (970 matches, each season using only earlier seasons). If a rating is not ready, or a team cannot be matched, a simpler fallback (the average goals of recent matches) is used.

At the end of these steps we have the probability of each of the 49 scorelines from 0-0 to 6-6 (a “scoreline table”); 13 different probabilities (match result, total goals, both teams to score, handicap result, correct score, half-time / full-time…) come from this table.

Not every probability is measured separately. The weekly track record measures the accuracy of three main probabilities: match result, 3+ goals and both teams to score. The others come from the same scoreline table and are not measured separately. Pick lists are built only from measured probabilities: the match result and what derives from it (e.g. “doesn't lose”); 3+ goals and both teams to score only in leagues where we beat the simple league average.

Team strength (a strength rating in the Champions League) gives the expected goals; after that only small adjustments that were seen to improve predictions in testing are applied. The full technical list is below; values such as ρ are the size of the adjustments.

• TOP-5 LEAGUES

  Chance quality + goals give attack/defence strength → strength rating (Elo) difference (expected goals by at most ±8%) → expected-goal difference over the last 5 matches (at most ±5%) → a setting that widens the gap between strong and weak teams slightly → low-score correction (Dixon-Coles, ρ = −0.13).

• SÜPER LİG

  Attack/defence strength from goals only (no chance-quality data) → strength rating (Elo) difference (at most ±15%) → goal difference over the last 5 matches (at most ±10%) → low-score correction (Dixon-Coles, ρ = −0.15).

• CHAMPIONS LEAGUE

  Expected goals come directly from the cross-league strength rating (Elo) difference between the two teams. Without a rating, the simpler fallback (average goals of recent matches) applies.

• NOT USED IN THE CALCULATION

  Past meetings between the teams, form streaks (W/D/L), league position, referee, rest days, line-ups and injuries. They did not improve predictions in testing, so they stay out; the match page shows them as information and says so openly when they point to a different team than the calculation does.

Every setting is chosen without knowing the future: a season's settings are set using only the seasons before it, then tested on that season's matches. The 5 big leagues were measured on 10,734 matches from 2020/21 to 2025/26 (match-result prediction improved in all 5 of 5 leagues), the Süper Lig on 2,279 matches from 2019/20 to 2025/26, and the Champions League on 970 matches.

Current accuracy is re-measured every Monday in the Track record (Turkish) on the last 8 months of matches; each match is predicted using only data from before it.

• CHANCE QUALITY ONLY IN THE 5 BIG LEAGUES

  There is no free expected-goals (xG) source for the Süper Lig and the Champions League; in those competitions the calculation uses goals and results only.

• LINE-UPS ARE NOT IN THE CALCULATION

  Injuries to key players are mentioned in the AI summary but do not change the percentages.

• THE EDGE IN GOAL PROBABILITIES IS SMALL

  On 3+ goals and both-teams-to-score we are only slightly ahead of the simple league average, and behind it in some leagues. Which leagues is shown every week in the Track record (Turkish); in those leagues these probabilities stay out of the pick lists.