Plain-English definitions of the analytics terms behind our predictions.
Separate scores measuring how effectively a team creates goals and how well it prevents them. Splitting strength into these two parts gives a sharper picture than a single overall rating.
A method of updating beliefs as new evidence arrives, blending prior knowledge with fresh data. In soccer, it lets the model refine team ratings game by game rather than starting from scratch each time.
In tournaments where several groups each send their top finishers forward, this rule ranks the third-place teams across all groups so the strongest of them can also advance.
A score that grades how accurate probability forecasts are, where lower is better; a perfect prediction scores 0. Our model's out-of-sample Brier score is about 0.31 on a walk-forward evaluation, meaning its match predictions are well-calibrated against real results.
How well a model's stated probabilities match real-world frequencies. If events the model calls 70 percent likely actually happen about 70 percent of the time, the model is well-calibrated and its numbers can be trusted.
A visual indicator showing how certain the model is about a given prediction. Higher confidence reflects clearer evidence in the data, while lower confidence flags matchups that are genuine toss-ups.
A range that shows where a quantity, like a team's expected goals, most plausibly falls. A 90 percent credible interval means there is a 90 percent chance the true value lies within that range.
A widely used soccer forecasting method that predicts match scorelines from team attack and defense strengths. It improves on basic models by better handling low-scoring results like 0-0, 1-0, and 1-1.
A rating system, borrowed from chess, that assigns every team a single number for overall strength. Teams gain points for wins and lose them for losses, with the size of the swing scaled by how surprising the result was — beating a far stronger side moves your rating much more than beating a weaker one. It is the starting view of relative quality our model builds on.
A measure of how many goals a team or player should have scored based on the quality of their chances, where each shot is assigned a value between 0 and 1 reflecting its likelihood of being scored.
Two added 15-minute periods, played in full, when a knockout match is level after 90 minutes. If the score is still tied at the end of extra time, the tie goes to a penalty shootout.
The measurable edge teams tend to gain playing at their own stadium, driven by familiarity, travel, and crowd support. Models account for it by giving the home side a small boost in expected goals.
A method that plays out a match thousands of times using random chance to map every possible outcome. We run 50,000 simulations per game to estimate win, draw, and loss probabilities and the spread of likely scorelines.
A stadium belonging to neither team, often used in finals and tournaments. Playing on neutral ground removes home advantage, so the model treats both sides evenly.
Data the model never saw during training, used to test how it performs on genuinely new matches. Strong out-of-sample results, like our Brier score of about 0.31, show the model generalizes rather than just memorizing the past.
The tiebreaker that decides a knockout match still level after extra time. Each side takes five penalties, alternating, and whoever scores more advances. Football records it separately from the scoreline, so the match itself still counts as a draw.
A statistical pattern that describes how often rare, independent events occur in a fixed span, like goals in a soccer match. It is the mathematical backbone for estimating how likely each possible score is.
The starting assumption a model holds before seeing new data, such as an early estimate of a team's strength. As real results come in, the prior is updated into a more informed, evidence-based view.
The tendency for extreme performances to be followed by more average ones. A team on a hot or cold streak will usually drift back toward its true level over time, so recent outliers rarely last.
A technique that pulls extreme estimates toward a sensible average to prevent overreacting to small samples. It keeps a team's rating from swinging wildly after just one fluky win or loss.
The full set of probabilities for every possible final score, from 0-0 to 3-2 and beyond. Instead of one prediction, it shows how likely each exact result is across the range of outcomes.
A knockout format where losing a single match ends a team's run. With no second chances, every game is decisive and upsets carry outsized weight.
A measure of how much better a model predicts than a naive baseline, such as always guessing the average outcome. A positive skill score means the model adds real forecasting value beyond simple guesswork.
The phase of a penalty shootout that begins once both teams are level after their first five kicks. The sides trade one penalty each until, in a single round, one scores and the other misses.
A rating system that scores each team's overall strength, rising after wins and falling after losses, with bigger shifts from surprising results. It offers a quick, single-number snapshot of relative quality.
The natural game-to-game randomness that makes soccer unpredictable, where the better team does not always win. Understanding variance explains why upsets happen and why single results can mislead.
The model's estimated chances of the three possible match outcomes: a home win (1), a draw (X), or an away win (2). These three probabilities always add up to 100 percent.
The flip side of Expected Goals, measuring the quality of chances a team allows its opponents to take. Lower xG Against suggests a strong defense that limits dangerous opportunities.