Sezi is an analysis tool that reads football through numbers, and reads those numbers back in plain English. It exists for people who want to know where a probability actually comes from before kickoff.
Not a model that ignores instinct, and not instinct that ignores the model. A practice where both lean on each other.
Sezi is the model's voice. Every morning it opens the fixture list and puts the model's output, each team's last ten matches, xG difference, injury news and the context that never fits on a stats screen on the same table.
It doesn't preach. It doesn't play oracle. It isn't a bookmaker and it doesn't write betting slips. It translates probability into plain language — why a match looks a certain way and how it could end up completely different, in 80–120 words.
Its tone sits between a quiet analyst and someone who's watched a lot of football: it doesn't speak in certainties, it likes to weigh things. When it's wrong, it says so; when it's right, it doesn't gloat. Every morning's note isn't a claim — it's a reading, offered.
A model is a map, the pitch is the terrain. A good map points a direction — but it's still a person walking the terrain.
The distance between 'likely' and 'certain' isn't the width of a sentence — it's the width of a season.
Data ahead of instinct goes sterile. Instinct ahead of data goes to noise. The two should stand side by side.
Brier score, hit rate, which match we got wrong — published every season, same as the rest.
The model is transparent. Which data feeds each layer, which assumption tunes it, and where it tends to fail — we publish it all in our seasonal reports.
The order below isn't random: each layer checks the one before it.
A time-weighted, Poisson-based baseline. Attack and defence parameters are fit per match, with a draw correction applied.
Team ratings update after every match. pi-ratings track home/away form separately; win weight scales with score margin.
A meta-learner. xG, head-to-head history, rest days, injuries/suspensions and weather join the Dixon-Coles + ELO outputs. One combined probability comes out the other end.
Numbers become plain language. 80–120 words, a tactical read, one preference. Zero betting language, zero guarantees, zero emotional selling.
Sezi's limits shape the product. Not doing something is often a bigger part of our identity than what we do. The six lines below aren't a slogan — they're product decisions.
We never use the words '100%', 'guaranteed' or 'certain' for any match. Every pick is a probability, and it's on record when that probability is wrong.
No operator links, no affiliate codes, no payment integrations, no odds comparisons — none of it. Our revenue is subscriptions.
Lower-division games, fixtures under match-fixing suspicion, markets showing fixing signals — none of it enters our analysis. It gets pulled from the list.
'Don't miss this', 'last chance', 'this week's steal' — none of that language exists on Sezi. Sezi speaks calmly, because it isn't in a hurry.
Nothing is stored beyond email and subscription status. Processed under applicable data-protection law, never sold to a third party.
Age confirmation is required at sign-up. Ad targeting is 18+. We don't publish content on social platforms that skips the age gate.
Note: apps promising 65–70% accuracy are making a mathematically impossible promise. The known ceiling of the 1X2 market is around 55%. Saying that plainly is a founding setting, not our marketing budget.
We try to keep Sezi as careful as a newsroom and as skeptical as a research lab. Four roles, none chasing a title, all there to remind the room the model's output can be wrong.
Picks the fixture list, decides which matches enter the model, holds the line on suspicious-match filtering.
Feeds the Dixon-Coles, ELO and XGBoost layers. Prepares the calibration report and runs mid-season retraining.
Reads every piece of Sezi's text by hand. The third eye that keeps betting language, hype and emotional selling out.
Every note that comes back from readers is a signal — a wrong read, missing context, a missing league — all tracked.
Today's fixture list, model probabilityfor every match, and Sezi's 80–120 word read — no sign-up required.