The Draw Epidemic: Why Matchday 1 Was a Festival of Stalemates — and Why It Broke Our Model

Here is the most uncomfortable, and most clarifying, fact about our opening matchday: every single one of our public misses was a draw. Canada 1-1 Bosnia. Qatar 1-1 Switzerland. Brazil 1-1 Morocco. We went 5 for 8 on the board, and the three that got away weren't random scatter across favorites and underdogs. They were all the same outcome — the stalemate. If you want to understand why this World Cup feels so unpredictable, start there, because the pattern in our errors is also the pattern in the tournament.

The draw is the hardest call in football

In a 1X2 market you are choosing between home win, away win, and draw. It sounds like three equal doors. It isn't. A draw is the only outcome that no model — ours or anyone's — can ever make its single most likely pick in a normal match. Think about why. To rate a team as favorite, your model has to push probability toward that team winning. A 45% Brazil number, like the one we published for Brazil-Morocco, already implies a 55% chance Brazil does NOT win. But that other 55% is split between a Morocco win and a draw. The draw almost never crosses 33%, because the moment one side is even slightly favored, win-probability eats into the tie. So a model that is well-built will rarely, if ever, lead with 'draw' as its top line — and yet draws happen in roughly a quarter to a third of matches. The math guarantees that draws will be your most frequent 'miss' even when your model is perfectly calibrated.

That is the part worth sitting with. A miss is not the same as a flaw. If we say a match is 45% Brazil, 30% draw, 25% Morocco, and it finishes 1-1, the headline reads 'model wrong.' But a model that called that exact match 100 times and saw 30 draws would be doing its job beautifully. The draw was always live. It was, in fact, our second-likeliest outcome. Calibration is not about being right on every match; it is about your 30%-events happening 30% of the time over the long run. Our walk-forward backtest puts our calibration error at 2.3% — meaning when we say 30%, it lands near 30%. The three draws didn't break that. They tested it, and we are publishing the receipts rather than quietly burying them. You can see the full ledger on our record page.

Why a flat field produces tight games

Now the deeper point, the one that connects our errors to the whole tournament. We have repeatedly flagged this as the flattest favorite field in modern World Cup history: no team in our model clears 16% to win the Cup. Spain leads at 15.9%, France 15.4%, England 12.7%, Argentina 10.8%, Brazil just 7.3%. When the very top of the field is compressed like that, it is not only the title race that tightens — it is every individual match. A flat field means the gap between a 'big' team and a 'medium' team is smaller than the narrative pretends. And smaller gaps mean more draws.

Brazil-Morocco is the cleanest illustration. The story sold Brazil as the five-time champions humbled by an upstart. Our numbers told a quieter story all along: we had Morocco at 2.4% to win the entire Cup — tied tenth, ahead of several traditional powers — and 95% to escape Group C, barely behind Brazil's 98%. We publicly flagged Morocco as a dark horse before a ball was kicked. So a 1-1 is not a giant-killing in our framework. It is two teams our model already rated close to each other producing the close result close teams produce. The draw wasn't the anomaly. The narrative was.

The same logic runs through Group B, the group two of our misses came from. We now have it as a genuine four-way scrap: Switzerland 82% to advance, Canada 80%, Bosnia 53%, Qatar 52%. Two opening draws in that group weren't a glitch in the matrix — they were the matrix. When four teams are this bunched, ties are the expected texture, not the exception. You can watch that group breathe on our groups page and run the permutations yourself in the simulator.

What an honest model does with variance

There is a temptation, after a day of draws, to 'fix' the model — to crank up the draw weighting until 1-1 results stop embarrassing you. That is exactly the wrong instinct, and it is worth being explicit about why we won't. A model tuned to predict more draws would look smarter on a draw-heavy day and dumber across the long season. Variance is not a bug to be engineered away; it is the honest residue of a sport where one deflected shot in the 95th minute — as Qatar found against Switzerland — flips a result. Our job is not to eliminate that randomness. It is to size it correctly, tell you the real odds, and then live with the outcomes, good and bad.

This is also why we don't copy the bookmakers. Our engine is Elo plus a Dixon-Coles Poisson scoring model run through 50,000 Monte Carlo simulations, all open-source, and it produces numbers that sometimes diverge sharply from the market — Brazil at 7.3% when the books still treat them as a top favorite is the loudest example right now. The backtest behind it hits 62% accuracy with an RPS of 0.175. None of that makes us immune to a 1-1. It makes the 1-1 legible: a predictable slice of a probability we always disclosed.

So what does matchday 1 actually tell us? Not that the model is broken, and not that this World Cup is chaos. It tells us the field is genuinely flat, that flat fields breed draws, and that the result everyone treats as 'boring' is the one quietly carrying the most information. The teams we rated close drew. The dark horse we named held the giant. The variance we priced in showed up on schedule. We will keep missing draws — the math demands it — and we will keep showing you every one of them on the record, because a model you can audit when it is wrong is the only kind worth trusting when it is right.