Five chances from the match model, then expected goals
The feed delivers five probabilities for each priced fixture: home win, draw, away win, over 2.5 goals and both teams to score. They come from a match model maintained on the odds and scores platform that supplies our fixtures, built from each team’s scoring and conceding rates, home advantage and the current market.
Where the model has not rated a fixture, the average bookmaker prices stand in with the margin removed, and the match page says so in the pick card.
A pair of expected goal figures, one per team, is then fitted so that a Poisson score grid reproduces the five chances as closely as possible. The total is solved first from the over 2.5 chance, the home share from the result split, and both numbers are refined together.
The pair is printed in the pick card of every match page, and the Poisson predictor reproduces the page from them.
Every market from one grid
Double chance, draw no bet, every goal line from 0.5 to 4.5, both teams to score, clean sheets, win to nil, odd or even, correct score and half-time / full-time are all read from the same grid.
Half-time markets split each team’s expected goals 45% before the interval and 55% after, the long-run share in the top European leagues.
Because one distribution feeds every number, the markets on a page never contradict each other. A high over 2.5 chance always travels with a thin correct score list, and a high draw chance always travels with a strong under 2.5.
The sentences on each match page are generated from these figures, which is why they can be checked word by word against the tiles beside them.
The price, the edge and the grades
Each pick is matched to the best price among the bookmakers in the comparison. Edge is the chance times the price minus one. Positive means the price pays more than the chance deserves, negative means a likely result at a poor price, and the page says which.
Where no bookmaker prices a market, the fair price from the model is shown and labelled fair. Grades then rank picks within their market. Banker, Strong and Lean describe a pick’s chance against its own market’s scale, because an 85% double chance and a 14% correct score are both excellent for what they are.
The bars are fixed per market and listed on each market page. A grade never promises a result.
Freeze, grade and publish
When a match comes within 40 hours of kick-off, eight picks are frozen with their chance and price. Match result, double chance, over 1.5, over or under 2.5, both teams to score, correct score, half-time / full-time and first-half goals.
After full time each is graded won, lost or void from the score and published on the results pages with the hit rate and flat-stake return per market. The daily accumulator tickets are stored with their legs and settled the same way. Nothing is edited after the whistle and nothing is removed.
Days before the first published tip have no entries, because a record that starts earlier is a backtest.
What the model does not know, and how it is checked
How we predict football matches has one known blind spot. Team news, suspensions, weather and motivation reach the model only through the prices, which move on them. A late injury to a striker shows up as a shorter away price an hour before kick-off, and the probabilities follow at the next refresh.
For a match where you hold information the market has not absorbed, the probability tool prices your own view against the line.
Calibration is the final test of how we predict football matches. Picks rated 70% should land about 70% of the time over a large sample, and the results pages show that figure per market for the last 7 and 30 days.
If a market’s hit rate drifts well below the chances it was given, its grading bars are raised and the change is noted on the market page.