Ligue 1 · Sunday, 20 September 2026. Final score 2–1 (half-time 1–0). Before kick-off we rated it 54% under 2.5 goals — the match finished with 3, so that call missed. That rating was frozen into a version-controlled ledger before the game and has not been touched since.
Open this match in the app →| Market | Yes | No | Our call |
|---|---|---|---|
| Over 2.5 goals | 49% | 51% | Under |
| Over 3.5 goals | 31% | 69% | no pick |
| Both teams to score | 51% | 49% | Yes |
The result model puts Nice at 20%, the draw at 26% and Lille at 54%, from expected goals of 0.86 and 1.6.
The league sets the starting point. Ligue 1 has produced over 2.5 goals in 53% of its matches, and that is where a team with no record of its own would sit.
Each side's own record. Nice have 1 in 4 this season (0 of 2 at home), Lille 1 in 4 (0 of 2 away). Those records are pulled toward the league rate with the weight of about 24 matches — enough that two or three unusual games cannot swing a rating — giving 49% and 49%.
Last season still counts, briefly. With 4 and 4 games played this season, last season's rates are still blended in and fade out entirely by game 40 — the point at which our testing showed they stop adding accuracy. A single season is part luck, so each club's record is first pulled toward the league average by 34 games' weight. That moves the two figures to 51% and 46%.
The mix. Weighted 40% home side, 40% away side and 20% league — the split that beat every alternative across 30,582 matches of testing — that gives 50%.
Head to head. These sides have met 5 times in three years, producing over 2.5 goals in 60%. That nudges the number 5% of the way toward the head-to-head pattern — deliberately small, because fixtures change more than reputations do — landing on 50%.
Goal expectation. Separately a Poisson model built from attack and defence strength — using real expected-goals data where the league has it — puts this match at 45%, from expected goals of 0.86 for Nice and 1.6 for Lille. Folding in 30% of that view gives 48%.
Calibration. Tested against 97,729 past matches, our ratings below 55% have run consistently too pessimistic — a figure in the high 30s, for instance, has come in about four points higher. So ratings in that range are lifted part-way back, giving 49%.
This breakdown is generated from the same inputs the rating uses, so it cannot describe a calculation we did not perform. The full model is set out in the methodology.
More: Nice season stats · Lille season stats · Ligue 1 table and form. Ratings are calibrated probabilities, not tips — a 60% call is expected to miss two times in five.