Understanding Win Probabilities

When we show a prediction as "58% home win, 24% draw, 18% away win," it's easy to read that as "the home team will win." That's not quite what it says — and the difference matters if you want to actually understand what a prediction is telling you.

A probability is about many matches, not one

A 58% win probability means: if this exact match were played over and over under the same conditions, the home team would win roughly 58 times out of 100. It doesn't mean this specific match is 58% "decided." Football has one result per match, and even a strong favourite doesn't win every time — that's exactly why upsets exist without breaking the model that predicted against them.

Why a 60% favourite still loses a lot

A team predicted at 60% to win will, over a long run of similar matches, lose or draw roughly 4 times out of 10. That's not the model being wrong — it's what a 60% probability means. A model would only be doing its job badly if 60%-favourites won far more or far less than 60% of the time across many matches, which is exactly what our accuracy page tracks.

Three numbers, not one

Every match has three probabilities — home win, draw, and away win — and they always add up to 100%. A draw probability sitting around 24-28% is normal in football; it's a genuinely common result, not a hedge. When two probabilities are close together (say, 42% vs 38%), that's the model telling you the match is genuinely close, not a sign it couldn't make up its mind.

Confidence is a separate number from probability

Don't confuse the win probability with the confidence score shown alongside it. Probability describes the match itself; confidence describes how much recent data the model had to work with. A 65% favourite backed by a full season of results and a 65% favourite based on three games look identical in the headline number, but the confidence score tells them apart.

What this means in practice

Treat the percentages as exactly what they are: an estimate of how often an outcome happens, not a guarantee of what will happen today. That's also why we publish how the model works and our real, ongoing accuracy — a probability is only useful if you can check whether it's been honest over time.