Open our scoreboard on any match day and you'll see two forecasts sitting side by side for the same game — ours and the market's — and they rarely agree exactly. Most people's instinct is to ask "so which one is right?" That's the wrong question, and asking a better one is the single biggest upgrade you can make to how you read any prediction. The disagreement isn't noise to resolve. It's the most informative thing on the screen.
Here's how to actually read a match forecast.
First: a forecast is a probability, not a prediction
"Spain to win" is a prediction. "Spain 55%" is a forecast. The difference matters enormously. A prediction is binary and unfalsifiable in the short run — when Spain wins, the predictor looks smart; when they don't, they mutter about a bad day. A probability commits to something testable: across everything you call at 55%, the outcome should happen about 55% of the time. You can't hide behind a probability. That's the whole point of using them.
So when you see a line like **Spain 55% / Draw 24% / Belgium 21%**, don't read it as "Spain wins." Read it as: *if this exact match were played many times, Spain takes it a little more than half of them, the draw shows up about a quarter of the time, and Belgium springs the upset one time in five.* One match tells you almost nothing about whether that split was right. The split itself is the claim.
The two forecasters, and how they think differently
On our board, each match carries two independent estimates, built in completely different ways.
**The market** is a price. Thousands of people put real money behind their opinion, and the price settles where the money balances. It's fast, it's liquid, and it's very hard to beat — money is a truth serum. Its blind spots are equally well documented: it drifts toward stories and favourites, and it tends to *underprice underdogs* (the famous favourite-longshot bias).
**Our model** is an engine. It takes each team's attacking and defensive strength — goals, expected goals, form, who's injured — turns them into an expected number of goals for each side, and runs the match thousands of times in probability space to get win/draw/loss. It has no idea who the crowd's favourite is. It only knows the numbers.
Two honest methods, two different lenses. When they diverge, each is telling you something the other is missing.
A live example: Spain vs Belgium
Take today's quarter-final. Here's what the two forecasters said before kickoff:
- **Our model:** Spain 55% / Draw 24% / Belgium 21% - **The market:** Spain 59% / Draw 24% / Belgium 17%
Look closely and the disagreement is precise. On the **draw**, they're identical — both land on 24%. Where they split is the **upset**: our model gives Belgium a 21% shot, the market only 17%. That four-point gap is the entire story of this forecast.
Why the gap? Spain arrive with the tournament's meanest defence — five straight clean sheets, not a goal conceded. The market, reasonably, rewards that with a big favourite's price. Our model respects it too — but it also sees that Belgium, even without the injured Amadou Onana, still field De Bruyne and Courtois and haven't stopped scoring. It refuses to round a live underdog down toward zero. That refusal is deliberate: rounding "unlikely" to "impossible" is exactly the mistake that favourite-longshot bias is made of, and correcting for it is where our model earns its keep on the nights it's right.
How to read the disagreement (not "who wins")
Once you stop asking "who's right?" a forecast becomes far more useful. Try these reads instead:
- **Where they agree, trust it more.** Both put the draw at 24%. When two independent methods land on the same number, that number is on firmer ground than either alone. - **Where they diverge, find the reason.** The gap is always about a specific thing — here, how much of a chance the underdog really has. Name the disagreement and you understand the match better than either number does alone. - **Notice the direction of the gap.** Our model sitting *above* the market on the underdog is a pattern, not a one-off — it's the calibrated correction to the crowd's favourite-longshot tilt. A model that were *below* the market on the underdog would be telling a different story worth questioning. - **Then wait for the score — and keep it.** The single match won't crown a winner between the two forecasts. Only the long run of scored calls does, which is why every one of ours is logged before kickoff and Brier-scored after, wins and losses alike.
The disagreement is the signal
A forecast isn't a tip and it isn't a promise. It's a measured claim about uncertainty, and the most interesting version of that claim is when two good methods look at the same match and see it slightly differently. The market's price and our model's engine are both trying to tell the truth about how likely things are. When they line up, you can lean on it. When they split, the gap points straight at what makes the match genuinely uncertain — and that's worth far more than a false sense of who's going to win.
Read both numbers. Find the gap. Then let the scoreboard settle it, over a hundred matches, not one.
*Educational content about forecasting and prediction markets — not betting or financial advice.*