How the MalluSports Model Actually Works, Step by Step

The model starts by estimating each team's attacking and defensive rates from recent matches, adjusted for home advantage and the quality of the opposition faced. Those rates become the parameters of two Poisson processes, one per team.

From there it computes the probability of every plausible scoreline and aggregates those into probabilities for each market: match result, both teams to score, over and under each goal line, and the rest. That produces a full probability distribution rather than a single prediction.

The final step is the one that rejects most fixtures. Live bookmaker odds are converted into implied probability with the margin stripped out, and the model's number is compared against the market's. Where the gap is not large enough to survive uncertainty in our own estimate, nothing is published. Most matches produce no pick at all, which is the point.

Key points

Attacking and defensive rates estimated from recent matches
Two Poisson processes produce every scoreline probability
Bookmaker odds converted to true probability, margin removed
Most fixtures produce no pick; that is the filter working

Frequently asked

Why do most matches have no pick?

Because the model and the market usually agree. A pick is only published where the disagreement is large enough to be worth acting on.

Is the model always right?

No, and the full record including every loss is published at our open data page so you can see exactly how often it is wrong.

Can I see the underlying data?

Yes. Every settled pick is downloadable as CSV and JSON under a CC BY 4.0 licence.

Put it to use

How the MalluSports Model Actually Works, Step by Step

Key points

Frequently asked

Put it to use

Read next