Sezi isn't a black box. We study the world's best statistics platforms and prediction systems one by one, work out their SWOT and tier standing, then combine the best of each in our own engine. On top of that we add our own layers: live player-form, injury and suspension feeds; other AI models' predictions; pundit commentary; public consensus. Sezi filters those signals, combines them — one output: model probability.
Below is the full system: three thinking filters, four modeling layers, one calibration benchmark.
Every day Sezi runs the fixture list through three different filters: high confidence, balanced and surprise value. The same match can land in a different tier under each filter — that's how you read exactly where the model stands, and which signal tips it which way.
The line where the model shows its highest calibration — a clear favorite, a wide gap between the three outcomes.
The line where all three outcomes sit in a tight band — low risk, no strong lean either way.
The line where the model diverges from market consensus — where it thinks value is leaking through.
Our model is transparent. We publish which data feeds each layer and which assumption tunes it, every season. No black box — four layers, one combined output.
openfootball, SofaScore, Wikimedia, API-Football metadata. 10 sources, 3.2M events, daily ingestion.
xG aggregation, ELO + pi-ratings, a Dixon-Coles baseline, Bayesian calibration.
Pundit commentary, other AI predictions, public consensus, form, injuries, suspensions.
One probability plus a confidence band. An 80–120 word English read. A reading, not a certainty.
Sezi writes the model's output in the tone of a modest, sports-literate observer — not a lecturing expert. Every note carries context, a preference, and a reason.
How the model works, where the data comes from, who Sezi is — clear answers to what people ask most.
Other question? /en/contact