Eighteen Matches, One Certainty
Published 4 September 2026 · figures read from the model the same day
The Champions League league phase starts on Tuesday, and we have published a prediction for all eighteen opening matches. Before a ball is kicked, we want to point out something the numbers make obvious the moment you line them up next to each other.
Across all eighteen fixtures, our model gives one team — one — a probability above 70%.
That team is Manchester United, at home to Sabah FA, at 72%. Everything else falls short of it, and most of it falls a long way short.
The shape of the board
Sort the eighteen matches by how confident the model is in its favourite, and the distribution is striking. Six have a side it makes more likely than not to win: Manchester United at 72%, Paris Saint Germain at 63% against Slovan Bratislava, Bayern München at 60% against Bodo/Glimt, Barcelona at 59% against Feyenoord, PSV Eindhoven at 54% against Shakhtar Donetsk, and VfB Stuttgart at 53% against Viking.
The other twelve have no favourite the model considers more likely than not to win. The median favourite across the whole round sits at 45%, which is another way of saying that in a typical match this week, the team we lean towards is more likely to fail to win than to win.
At the bottom of the confidence range sit the fixtures the model simply cannot separate. Real Madrid host Inter at 37% / 28% / 35%. FC Porto host Manchester City at 39% / 28% / 33%. Napoli host Arsenal at 32% / 31% / 37%, which is about as close as three numbers can reasonably get. In each of those, the gap between the two sides is smaller than the honest error bar on any model, ours included.
This is not hedging
It would be easy to read a board full of 40-somethings as a model refusing to commit. It is the opposite. The model commits precisely: it says these matches are close, because they are close, and it declines to manufacture a confidence the data does not support.
The temptation in this business is to round up. A 45% favourite becomes “the pick.” A 39% becomes “value.” The language gets louder as the evidence gets thinner, because loud reads better. We would rather publish the number and let it be as unremarkable as it actually is.
There is a straightforward test of whether that discipline is worth anything, and we run it publicly. On our track record page, every settled prediction is scored against the real result and grouped into calibration bands. The question is not whether we got a match right. It is whether our 40% teams win about forty times in a hundred and our 60% teams about sixty. A model that says 40% and means it is more useful than one that says 80% and is right slightly more often than chance.
Our headline hit rate is 47% correct results across 1,680 logged predictions. That number is not going to impress anyone at a glance, and we are not going to dress it up. Three-way football outcomes are hard, and a large share of the matches we cover look exactly like this Champions League round: close enough that the correct answer is a probability, not a name.
Where the model disagrees with the reputation
A few of this week’s calls will read oddly if you go by badge rather than by form and expected goals.
FC Porto are favoured over Manchester City, 39% to 33%. Como are favoured over RB Leipzig, 41% to 32%. Club Brugge KV are favoured over Aston Villa, 47% to 27%. Slavia Praha are favoured over Lens, 39% to 33%.
Home advantage carries a good deal of that, and recent scoring and conceding rates carry the rest. The model does not know which of these clubs has a European history and which does not. Sometimes that blindness is exactly what you want from it. Sometimes it is a weakness, and the league phase is where that weakness is most exposed: teams from very different leagues meet each other rarely, so there is less directly comparable evidence than in a domestic season. We say so on our methodology page, and we would rather you knew it going in.
The draw is worth a note too. In the closest fixtures on the board, our draw probability sits in the high twenties and low thirties — 31% in Napoli against Arsenal, 28% in both Real Madrid against Inter and FC Porto against Manchester City. In a genuinely even match, the draw is often the single most likely outcome even when it is not the outcome anyone talks about.
The eighteen calls
Here is the full round: the model’s three-way split, its favourite, and the most likely scoreline. Kickoff times are UTC. Every match links to its own page.
| Match | Home / Draw / Away | Favourite | Likely |
|---|---|---|---|
| Tuesday 8 September | |||
| Club Brugge KV vs Aston Villa | 47 / 26 / 27 | Club Brugge KV 47% | 2-1 |
| AEK Athens FC vs Lask Linz | 49 / 27 / 24 | AEK Athens FC 49% | 1-0 |
| Lille vs Real Betis | 39 / 30 / 31 | Lille 39% | 1-0 |
| Borussia Dortmund vs Villarreal | 41 / 29 / 30 | Borussia Dortmund 41% | 2-1 |
| FC Porto vs Manchester City | 39 / 28 / 33 | FC Porto 39% | 2-1 |
| Real Madrid vs Inter | 37 / 28 / 35 | Real Madrid 37% | 2-1 |
| Wednesday 9 September | |||
| VfB Stuttgart vs Viking | 53 / 26 / 21 | VfB Stuttgart 53% | 2-0 |
| Barcelona vs Feyenoord | 59 / 23 / 18 | Barcelona 59% | 2-0 |
| Liverpool vs Atletico Madrid | 46 / 27 / 27 | Liverpool 46% | 2-1 |
| Paris Saint Germain vs Slovan Bratislava | 63 / 22 / 15 | Paris Saint Germain 63% | 2-0 |
| Sporting CP vs Galatasaray | 44 / 29 / 27 | Sporting CP 44% | 1-0 |
| Napoli vs Arsenal | 32 / 31 / 37 | Arsenal 37% | 0-1 |
| Thursday 10 September | |||
| PSV Eindhoven vs Shakhtar Donetsk | 54 / 26 / 20 | PSV Eindhoven 54% | 2-0 |
| Fenerbahçe vs AS Roma | 30 / 27 / 43 | AS Roma 43% | 1-2 |
| Manchester United vs Sabah FA | 72 / 18 / 10 | Manchester United 72% | 3-0 |
| Bayern München vs Bodo/Glimt | 60 / 23 / 17 | Bayern München 60% | 2-0 |
| Slavia Praha vs Lens | 39 / 28 / 33 | Slavia Praha 39% | 2-1 |
| Como vs RB Leipzig | 41 / 27 / 32 | Como 41% | 2-1 |
Each fixture has its own page with the expected goals behind the split, current form, head-to-head record, the full scoreline distribution, and fair odds. During the matches themselves you can follow live scores alongside live probabilities, which is the quickest education available in how much a single goal actually changes.
Check us on Friday
These eighteen predictions are logged, timestamped and public. On Friday we will publish the review: what happened, how many we called correctly, and — more usefully — whether the round as a whole came out roughly where the probabilities said it should.
If we get twelve right, that will not mean the model is brilliant. If we get six right, it will not mean the model is broken. Eighteen matches is a small sample, and a small sample of coin flips tells you very little. The honest verdict lives in the full track record, accumulated over hundreds of matches, and it is available to you at any time without asking.
We do not take bookmaker sponsorship or gambling-affiliate advertising, and none of this is betting advice. It is a statistical estimate published for its own sake, because we think the numbers are interesting and because we think somebody should be publishing them without an angle.
Enjoy the football. Tell us where we were wrong at [email protected].
These probabilities were read from the model on 4 September and the board rebuilds every day, so the numbers on each match page are the current ones. Where they disagree with this article, the match page is right. Read more about why we publish any of this.