Every match on Touchline Atlas comes with three numbers: the chance of a home win, a draw and an away win over 90 minutes. They look simple, and that is the trap. This guide explains what those numbers mean, why a favourite can lose without the forecast being “wrong”, and how to read the rest of a match page.
A probability describes many matches, not one
Say a match shows 48% home win, 27% draw, 25% away win. That does not mean the home side will win. It means that if this exact match, with the same teams, the same form and the same conditions, could be played 100 times, we would expect the home side to win about 48 of them, draw about 27 and lose about 25.
Football is only played once, so you only ever see one of those outcomes. When the away side wins, it is one of the 25 results the forecast expected, not proof that the forecast was broken. A forecast can only be judged fairly over many matches, which is why we publish every one of them on the results archive.
48% is a probability, not a promise.
Why the favourite still loses so often
The side with the highest probability is the favourite, but “favourite” and “likely to win” are not the same thing. At 48%, the home side fails to win more often than it wins: 52 times in 100 the result is a draw or an away win.
Even strong favourites miss regularly. A 70% favourite still fails to win about 3 matches in 10. Over a season of a few hundred matches, that is dozens of “upsets” that a well-calibrated model fully expected. If a forecaster never seems to miss, it is usually because the misses are not being shown.
This also explains why the most likely result is right only around half the time across football. With three outcomes, and draws that are hard to call, a model that picks the right result in a little over half of matches is doing well. Ours has been right in 65% of 72 graded matches so far; the track record shows how that compares with simple benchmarks.
The three numbers always add up to 100
Home, draw and away cover every possible 90-minute result, so the three numbers add up to 100%. We round them so that the displayed numbers still add up to exactly 100, which is why you might occasionally see 33 / 33 / 34 rather than three equal numbers.
The draw deserves its own attention. In most leagues roughly a quarter of matches end level, yet a draw is rarely the single most likely result. That is because the draw has to compete with two separate win outcomes. A forecast of 38 / 30 / 32 is a genuinely uncertain match, even though the home side is technically the favourite.
Most likely result and most likely score are different questions
Our model does not start from the three numbers. It first estimates how many goals each side is likely to score, then builds a grid with the probability of every exact score: 0–0, 1–0, 1–1, 2–1 and so on. The home, draw and away probabilities are the sums of the right parts of that grid.
That is why a match page can show a home win as the most likely result and 1–1 as the most likely score at the same time. A home win can happen through 1–0, 2–0, 2–1, 3–1 and many more scores, each with a modest share. A draw is concentrated in a few scores, mostly 1–1. So 1–1 can be the single biggest square in the grid while the home win is still the biggest total.
Exact scores are very hard to predict. Even the most likely score usually has a probability of only about 10–14%, so it misses most of the time. We show it, but we grade the site on the full probabilities, not on exact scores. Our correct score page shows the top scores for upcoming matches.
What the confidence labels mean
Each forecast has a label based on its biggest probability:
- High: the most likely result is 65% or more.
- Medium: 50% to 64%.
- Low: under 50%, a genuinely open match.
The label describes the probability, not a level of certainty about the outcome. In testing, High picks have been correct 70–83% of the time, which also means they miss roughly one time in four or five. We publish the hit rate for each label so you can check whether the labels mean what they say.
Comparing forecasts across matches
A few habits make the numbers more useful:
- Compare probabilities, not picks. A 52% home win and an 80% home win are both “home win” picks, but they are very different forecasts.
- Be careful when combining. Three separate 60% results all happening is much less likely than any one of them: 0.6 × 0.6 × 0.6 is about 22%, roughly one time in five.
- Look at the spread. A forecast like 40 / 30 / 30 says the match is open; 75 / 17 / 8 says it is one-sided. Both are informative.
- Odds are a different number. Bookmaker odds can be turned into probabilities, but they include a margin and are set by a market, not a model. We do not use odds in version 1 of our model.
Read the rest of the match page
The probability is only the headline. Before you rely on it, check:
- The lock time. Every forecast is locked before kickoff, normally 24 hours ahead; the match page shows the exact lock time. Team news after that is not in the numbers.
- The data status. “Limited context” means the numbers are published but we could not verify team news from a dated source in time.
- Form and head-to-head. Recent results and past meetings explain much of what the model sees.
- The lock fingerprint. The SHA-256 hash shows the forecast was not changed after the lock; you can check it in the ledger.
Key takeaways
- A probability tells you how often a result would happen over many similar matches, not what will happen in this one.
- Favourites lose often, and that is built into the numbers.
- The most likely score and the most likely result answer different questions.
- Judge any forecaster over hundreds of matches, misses included.
Our forecasts are for information and entertainment, not betting advice. If you bet, read our responsible gambling page first. 18+.
