Every shot gets a number: the share of shots like it that are scored. Like it means from that spot on the pitch, with that part of the body, from that kind of move: a cross, a through ball, a corner, a counter-attack, a rebound. A tap-in from six yards is worth about half a goal; a hopeful strike from thirty yards about one in forty. The numbers come from about 1.7 million shots in fifteen leagues over sixteen seasons, and they describe the chance before the ball was struck, never where it went or who struck it.
A team's expected goals in a match is the sum over its shots: the quality of the chances it made. Its opponent's is the quality of the chances it allowed. The expected-goals difference is one minus the other, and it is the site's measure of performance in a match. Over a season it is the season's total created minus the total allowed, divided by the matches played: how many more goals' worth of chances the team makes than it allows in a typical match. A team at +0.5 has, over 38 matches, made about nineteen goals' worth of chances more than its opponents.
Chances predict a team's future better than its goals do while its record is short, because goals are scarce and swing on a deflection or a save. Over a long record the two say nearly the same thing.
Take each side's shots and let every one score with its own probability. That gives the chance of each scoreline, and so the shares of a home win, a draw and an away win the match's chances implied. Deserved points are three times the win share plus the draw share. Over a season they add up to the points a team's chances would have earned on average. The gap between a team's points and its deserved points is stated as a fact. Part of it is chance; part of it is finishing, goalkeeping and things the shot model does not see, and the site does not pretend to know which.
Two numbers per team, in expected goals a match as the rating counts them, against an average opponent: what it creates and what it allows. Because the rating counts no chance for more than 0.25 of a goal, its numbers run about five per cent below the goals a team scores and concedes. The rating is a memory of the team's past matches, the recent ones counting most (a match a season ago counts about a third as much as yesterday's), mixing what the chances said with what the goals said once the record is long enough to trust the goals. A team new to the league carries part of its record from the division below. While a team's record is short, or its squad has changed, the rating leans on what the squad is worth. It is the site's estimate of how good a team is now; the season's expected-goals difference is what the team has done so far. Early in a season the rating is the steadier guide; by the second half of a season the two say the same thing.
The match pages show each side's expected goals three ways: in full, excluding penalties, and as the rating counts them. Excluding penalties is the full figure less the value of any penalties: what the side made from open play and set pieces. The rating does not take every chance at face value: a penalty counts for half its value, because winning penalties barely repeats from one season to the next; a chance that came from an opponent's mistake counts for half, for the same reason; and no chance counts for more than 0.25 of a goal, because a team's expected goals from its very best chances repeat from one season to the next about half as well as the rest of its expected goals. The full figure describes the match; the counted figure is what the match adds to the rating.
Every fixture still to play is given the chances of a home win, a draw and an away win from the two teams' ratings now. The projection behind those chances is fitted on the eight seasons before this one and on every match of this season played so far, refitted as each week's results come in, to how often sides of such ratings won, drew and lost, with this league's scoring level and home advantage. The rule from the ratings to the goals is a curve, not a straight line: a strong attack scores more than a straight line would say, and a weak defence concedes more, which is why the stronger side's chances are not held back. The home advantage is fitted with the recent matches counting most: a match six months older counts half as much. In the last eight matches of a season, a side whose season is settled at the table before the match, because it can no longer win or lose the title, a European place or its place in the division whatever happens, scores a little less than the ratings say and its opponent about a sixth more, and the projection adds that. A side that is settled at today's table keeps it for all its remaining fixtures; a side that becomes settled later counts from the day it does. The chances are then sharpened by one constant fitted on all fifteen leagues' matches. The level is not exact: on the four seasons 2019/20 to 2022/23 this form gave home wins 45 per cent on average where 44 per cent happened, draws 23 where 23 happened and away wins 32 where 33 happened; the bookmakers' forecasts gave 43, 24 and 33. So the home side is given about a point and a half too many, and an away side one and a half too few. The forecasts page shows the forecasts beside the market season by season. The rest of the season is then played out 10,000 times at random with those chances, and the final table counted each time. A team's projected points are the average over those plays; its range runs from the 10th to the 90th of them, so eight plays in ten finish inside it. The range covers the randomness of the matches still to play and the ratings' own error: before the fixtures are projected, each team's two ratings are moved by their own random factors, in a hundred draws, by a spread measured on the seasons 2019/20 to 2022/23 as the amount at which the range held eight final points in ten (with the ratings held fixed it held about two in three in the first ten matches). The range narrows as the season goes on. The front page shows what the latest round changed: the same projection as it stood before the round's matches, from the ratings the teams brought into them, against the projection now.
It prices no bets and recommends none. It shows what happened, valued by the chances, beside the score, and projects the rest of the season from the ratings, saying what that projection's range covers and what it does not.