Methodology · ODDSXI

ODDSXI estimates the probability of football match outcomes, shows the signals behind each estimate and publishes a complete record of how the estimates performed. It is an information service. It does not take bets and does not tell anyone to place one.

The model

The current goal model (oddsxi-goal-model p6-v5) predicts each team's expected goals from its attacking and defensive record in the current and two previous league seasons. It uses goals and, with a smaller weight, shots and shots on target, with recent matches counting more than older ones. Match probabilities for 1X2, total goals, Asian handicap and both teams to score follow from the two expected-goals values. The model does not use market prices.

Frozen specification. Any change to the model is published as a new version; old forecasts keep the version that made them.
Point in time. A forecast only uses matches that had finished before it was generated.
Newcomers. A team without last-season data in the league, for example a newly promoted club, starts from the average first season of past newcomers (who are, on average, clearly weaker than established teams) and is updated with each of its matches. This rule was tested on past seasons before release. Matches with little data are marked Limited data. The previous version, p6-v1, which treated newcomers as league-average teams, is still published and scored as a baseline.
Three seasons of history. Since p6-v5 the model also learns from the season before last, with a small weight, which steadies early-season estimates. This was tested on past seasons before release. The previous version, p6-v2, trained on two seasons and is still scored as a baseline; matches forecast before p6-v5 keep their p6-v2 forecast.

Signal breakdown

For each team, expected goals equal the league scoring level multiplied by a small set of factors: home advantage, attack and opponent defence from goals, and attack and opponent defence from shots. The match page shows each factor's effect, and the product is exact. Recent results and rest days are shown as context only.

Model and market

Each forecast is shown next to the market: the consensus of a sharp betting market with the margin removed, as seen at the time of the forecast. The difference between the two is a research figure. A backtest of an earlier model on the 2025/26 season found no advantage over the closing market, and ODDSXI will not show an “edge” indicator until forward results demonstrate one.

Research selections

To measure whether the model's disagreements with the market mean anything, every pre-match case where the model probability is at least 3 percentage points above the margin-free market probability is stored before kickoff, together with the market probability seen at the forecast time, and can never be changed. After kickoff the Performance Lab shows how often these cases came true compared with what the market expected, grouped by the size of the disagreement (3, 5 and 8 points). The rules were registered before the first result. Large disagreements often point to a model weakness rather than a market mistake; the record exists to find out which.

Publication rules

Every forecast is published before kickoff.
Forecasts are stored immutably: the database rejects any change or deletion.
The Performance Lab scores every forecast, with rules registered before the first result was known.

Cards (beta)

The cards model predicts the total number of cards from each team's card history and its opponents'. It passed a preregistered historical validation and is compared in the Performance Lab with a league-average baseline. A corners model did not pass validation and is not published.

Responsible use

Probabilities are uncertain estimates, and the most likely outcome often does not happen. If you bet, only use money you can afford to lose. Gambling is for adults only.