Pál József Gergő

The early study compared six prediction models and five staking strategies across 539 matches, using 300,000 simulations. One pairing was profitable on the sample examined.

I extended the later research to 11 seasons and more than 50,000 matches, using seven different methods. Models could learn only from earlier data. On this dataset and under the tested conditions, none of the methods demonstrated a sustained edge exceeding the bookmaker margin. The methodology and results are publicly available.

Since childhood, what interested me in sport was understanding why a match unfolded the way it did. Studying computer science gave me the tools to explore that question with data and models. My BSc research led to a publication, followed by a study on a larger sample.

I created LabOfMetrics so others could use the tools developed during that research: explore probability estimates, evaluate their own strategies and understand the limits of the results. In both the platform and the course, I focus on interpreting data, using methods that can be checked and making responsible decisions.

All features are free during the open test period; I cover the costs of data, servers and AI analyses myself. Later, I plan to fund the full course and analytics toolkit through subscriptions, without advertising or commissions.

The research is freely available. Of the course’s 47 chapters, 19 are available without registration: all of Modules I and VIII, plus chapters on the mathematics of luck, cognitive biases, the tipster industry, responsible gambling and taxation. These are planned to remain free.