1X2, Over/Under, both teams to score and the other terms explained simply
The most classic market: the 1 sign means a home win, X a draw, 2 an away win. It refers to the result after 90 minutes (regulation time plus added time), without extra time or penalties.
The AI Analyses published with their most likely 1X2 outcome are collected on the today's football predictions page; how the probability is estimated is explained in the guide to probabilities.
Covers two of the three possible outcomes at once: 1X means "the home team does not lose" (win or draw), X2 "the away team does not lose", 12 "the match does not end in a draw". It is the outcome with the highest probability of happening, and by nature the most cautious.
It concerns the total number of goals scored in the match by both teams. Over 1.5 means at least 2 goals in total; Under 1.5 at most 1 goal. Over 2.5 means at least 3 goals in total; Under 2.5 at most 2. The ".5" avoids ambiguity: there are no half goals, so the outcome is always settled.
GoalCast AI's analyses use Over/Under 2.5 as the main market, the same one as the accuracy page; older analyses reported Over/Under 1.5, a lower line where the Over is almost always more likely.
Today's predictions on this market, with verified outcomes and calibration, are on the Over/Under 2.5 predictions page.
Also known as BTTS: GG (Goal) means both teams score at least once; NG (No Goal) means at least one of the two fails to score. A 3-1 is GG; a 0-0 and a 2-0 are NG.
Today's predictions on this market, with verified outcomes and calibration, are on the both teams to score predictions page.
A combo joins two outcomes of the same match, for example «1 + Over 2.5» or «1X + BTTS»: it comes true only if both do. Its probability is always lower than that of its most uncertain leg, because two conditions must hold together; the two outcomes, however, influence each other (a favourite that wins often scores a lot), and the model estimates the joint probability taking this correlation into account. In the odds, instead, accumulators also multiply the bookmaker's margins.
AI Analyses suggest up to four combos per match, each with its estimated probability, verified one by one after the final whistle: you will find them on the today's suggested combos page.
Probability expresses how expected an outcome is, as a percentage. The probabilities of the three 1X2 outcomes always add up to 100%. A 60% probability is not a certainty: it means the opposite outcome still has a 40% chance of happening, that is, on average, 4 times out of 10.
To understand why a 60% is not a certainty and how to compare it with odds, read how to read probabilities; the method the model uses to estimate them is described in the methodology.
In AI Analyses every prediction reports the model's confidence: the probability the artificial intelligence assigns to the indicated outcome. 70% confidence = very likely but not a given; 45-50% confidence = uncertain match, in which the prediction should be read with particular caution.
How well the declared confidence matches real results can be checked on the accuracy page, updated automatically after every finished match.
The list of players the model considers most likely to score, with their probability. It takes into account position, scoring form, expected minutes on the pitch and the quality of the opposing defence.
The estimated market value of many of these players is in the value profiles.
The sequence of a team's recent results (typically the last 5 matches). A team "in form" wins and scores consistently; form weighs heavily in the model's assessments, together with the quality of the opponents faced.
The history of matches played between the two teams. Useful for spotting recurring patterns (fixtures always full of goals, historical dominance of one side), but it must be weighed: teams and coaches change, and very old meetings say little.
The statistical advantage of playing at home: friendly crowd, no travel, familiarity with the pitch. Historically it is worth a few extra percentage points of probability for the host team, with significant differences between leagues and teams.
Said of a team that finishes the match without conceding. Clean-sheet frequency is a good indicator of defensive solidity and weighs in the Over/Under and both-teams-to-score markets.
A decimal price (for example 2.00) is the multiplier of a winning stake and hides a probability: the implied probability is 1 divided by the odds, so 1 / 2.00 = 50%, 1 / 1.50 = 66.7%, 1 / 4.00 = 25%. On match pages you find the average market odds next to the AI Analysis probabilities: comparing them tells you whether the model is more or less optimistic than the market about an outcome.
The implied probabilities of the three 1X2 outcomes always add up to more than 100%: the difference is the bookmaker's margin, explained in the next entry. A full example is in the guide to probabilities.
If the odds on a match are 2.10 / 3.40 / 3.60, the implied probabilities are 47.6% + 29.4% + 27.8% = 104.8%: that extra 4.8% is the margin (overround) the bookmaker keeps. The higher the margin, the less the odds reflect the real probability. The probabilities of an AI Analysis, instead, add up to 100% because they describe the possible outcomes, not a commercial offer.
Expected goals measure the quality of chances created: every shot is assigned the probability of becoming a goal based on position, angle, type of move and body part, and the sum of the values in a match is its xG. A team that scores 1 goal from 2.3 xG has wasted chances; one that wins 1-0 with 0.4 xG was lucky or clinical. In AI Analyses the xG of recent matches is one of the form factors: it says more than the plain number of goals.
A model is calibrated when its probabilities match real frequencies: among all predictions given at 60%, about 60% must come true. So it is not enough to count how many predictions hit: it also matters that the declared percentages were honest. It is the criterion with which we read the numbers on the accuracy page, and the reason a 55% prediction that fails is not necessarily a model error.
The calibration table by confidence band, updated at every verification, is on the accuracy page.