How to read a PreviaXI preview
Start with the match analysis, then use the statistics to check it. This guide explains what to look for and uses a Premier League match to show how to revisit a forecast after full time.
The favourite can still be unlikely to win
Suppose the model gives the home team 45%, a draw 30% and the away team 25%. Home win is the largest single outcome, but draw and away win together account for 55%. “Most likely” does not mean certain, or even more likely than all alternatives combined.
A match result and an exact score answer different questions
The home-win probability adds every score in which the home team wins. A 2–1 score is only one of those possibilities. A score display is not a promise, and rounding expected goals is not the same calculation as finding the single most probable score. Read the score probabilities when available alongside the win, draw and loss figures.
Recent form needs a denominator
Three wins from five matches and six wins from ten both represent 60%, but they describe different samples. Check the dates, competition and home/away split before comparing teams. A strong record against weaker opponents does not automatically imply the same performance against a stronger opponent.
Observed statistics and model estimates are different
Season goals and results describe matches already played. Expected goals labelled “model” describe this model’s estimate for the upcoming fixture; they are not the team’s observed shot-based xG from a completed match. Missing data should remain missing, not be read as zero. Model coverage and limitations are explained in our methodology.
Look for the uncertainty as well as the explanation
Read the factors and uncertainty shown beside the model. Team news may be newer than the model snapshot: a listed absence does not prove it was included in the probability. Check timestamps and avoid treating a confident writing style as evidence of a confident forecast.
Judge a record across comparable matches
One correct score does not establish reliability, and one upset does not disprove a probability. Compare the same outcome type over many matches, with the sample size and dates visible. A 60% estimate describes roughly 60 successes out of 100 comparable cases if well calibrated; it does not guarantee the next match. Our record keeps model-backed results separate from unverified predictions and grades the 90-minute result.
A real example from the Premier League
Before Sunderland played Chelsea on 8 November 2026, PreviaXI gave Chelsea a 42% chance to win, Sunderland 31% and the draw 27%. Chelsea were the most likely single outcome, while a Sunderland win or draw together made up 58%.
Sunderland won 1–0 through a first-half penalty. Chelsea had 69% possession and led 4–2 in shots on target. Sunderland still finished ahead in expected goals from shots, 2.58 to 0.55. Possession and shots on target tell you who had more of the ball and tested the goalkeeper more often. Expected goals help you see how good the chances were.
Understand the analysis
Worked examples, uncertainty and the difference between a likely result and an exact score.