The Round of 32 is done, the field is halved, and my forecasting model has re-run the whole tournament on the wreckage. Here’s where it lands, who it just lost to the cruelest lottery in sport, and the one call where I flatly refuse to agree with my own code.
Two heavyweights gone on penalties. Twelve group winners I called before a ball was kicked, all still standing. And France, the team the model keeps politely filing under second best, just handed a clearer road to the final without kicking a single different ball.
The first knockout round is complete, which means the Champion Predictor has been fed every result and rebuilt from the ground up. Before the Round of 16, here’s the state of play: what the model is sure of, what it got right when it actually counted, and the one place my eyes and my mathematics are no longer on speaking terms.
The state of the race
After conditioning on every result through the Round of 32, the title picture:
🇦🇷 Argentina, 21.8% 🇫🇷 France, 18.4% 🇪🇸 Spain, 15.6% 🇧🇷 Brazil, 10.3% 🏴 England, 7.0%
Argentina sits clear at the top. France is now alone in second, Spain a length back, and the top four hold roughly two thirds of the title probability between them. The single most-likely final the model spits out is still Spain vs Argentina.
The setup, for anyone new here
The engine is an ensemble of two models: a Dixon-Coles goals model and a World Football Elo rating, blended at the scoreline level and run through 10,001 Monte Carlo simulations of the tournament from here to the final. It does not get nervous. It does not have a favourite nephew. It does maths, and it occasionally embarrasses me by being right.
The part I’m actually proudest of
Anyone can look clever editing their predictions after the goals go in. The honest test is the version you lock the night before the opener and never touch again.
Graded against reality through the Round of 32, that frozen forecast has:
Group winners: 12 out of 12.
Teams in the Round of 16: 12 of 16, chosen blind.
Match-outcome accuracy: 61%.
Mean Ranked Probability Score: 0.156.
That last number is the one I trust most, so it’s worth a sentence. The Ranked Probability Score grades not just whether you were right but how confident and how close you were. Missing a result “by a draw” barely stings; calling a home win that turns into the opposite hurts a lot. A blind guess scores about 0.278. My model’s long-run form is 0.169. So far this tournament it’s running 0.156, comfortably better than its own average. Translation: it isn’t just picking winners, it’s well calibrated on the probabilities underneath them.
But the line I keep coming back to is the first one. Twelve group winners out of twelve, and twelve of the sixteen knockout survivors, all committed to before a single ball was kicked. I’ll be mentioning this at dinner parties until roughly 2030.
The fallen
Two pre-tournament somebodies are already at the airport, and both left the cruelest way. Germany and the Netherlands, each dispatched on penalties in the Round of 32. Going into that round the model still gave them a live title shot, 4.6% and 2.2%, contenders, not passengers. But the penalty spot is where forecasts go to be humbled. A shootout is a coin flip in expensive boots, and no model that respects the data pretends otherwise. It took two live contenders with it, and there was nothing in the mathematics that could have saved them.
What’s moved since the last update
Compared with the forecast I last published, Version 6 at the end of the group stage, the first knockout round did what good information should do. It sharpened the picture, and in one case rewrote a team’s odds without that team doing anything at all.
France got a promotion for doing nothing. With Germany and the Dutch cleared out of its half of the bracket, its number jumped from 14.5% to 18.4% without France having to play a single different match. This is how tournament probability actually works. Your odds are hostage to the draw as much as to your own form.
Spain holds third at 15.6%, still one of the strongest sides in the field. Brazil firmed to 10.3%. And Morocco keeps sidling up the board with genuine dark-horse menace. Best team on paper and last team standing are graded on very different rubrics, and the bracket is the rubric that pays out.
Where I overrule the machine
Here’s the honest part. The model’s pick is Argentina. I don’t think it’s right.
The machine watches football through a spreadsheet. It sees results, ratings, squad values, home advantage. What it can’t see is the thing I can’t stop seeing. From where I’ve been sitting, France aren’t just winning, they’re operating on a different frequency. Sharper, deeper, meaner from front to back. To my eye they’re at least a level above everyone else in this tournament, and the bracket has just tilted their way on top of it.
The model has them a close, climbing second at 18.4%. My eyes have them first, and it isn’t close. One of us is going to look silly when this is over, and I’m suspiciously comfortable with those odds. A forecast you only ever nod along with isn’t worth much. The value is in the disagreement, and in being honest about which side of it you’re standing on.
And the flip side, the team that let me down most: Portugal. A genuine pre-tournament dark horse who never got going, second in their group and out before the story ever started. For a side many of us fancied, a real disappointment.
Under the hood
For the curious: the two models are blended at the scoreline level and the remaining tournament is simulated 10,001 times, conditioned on every actual result so far. Two small acts of rigour, because details are the whole game in this work. FIFA quietly changed the 2026 tiebreaker rules so head-to-head now outranks overall goal difference, a break from every previous World Cup that materially changed who topped a group. And the Round-of-32 third-place allocation follows FIFA’s official deterministic table, not a convenient approximation. Both were caught and coded properly. A forecast is only as trustworthy as the plumbing beneath it.
What’s next
The Round of 16 is loading, the bracket is fixed, no redraws. The model gets re-run after every round, and I’ll keep publishing where it lands, and where, like now, I quietly disagree with it.
Model says Argentina. My perception and football awareness says France. We’re about to find out which one of us was right.
The Champion Predictor is a quantitative model built for research and curiosity, not betting advice. If you’d like the round-by-round updates through the knockouts, subscribe. The next one lands after the Round of 16.
from Risk Premium https://ift.tt/acK5bhL
via IFTTT
