Sports Analytics Weekly by kubeia.io - 7/2026

futi is a next-gen live score app built to bring pro-quality analytics to fans in a familiar format. We’re rolling out model explainers and early data releases so you can preview the models behind the app. Follow us to keep up to date and be part of the future of fan-facing stats at the links below:

What I mean is: how good is this pass, quantitatively? Go ahead, try to put a number on it.

As far as most football stats are concerned, the number is 1. Messi attempted one pass. He completed one pass. Maybe you get fancy and label it a throughball or a final third entry or something, but it’s still just 1/1.

See the problem here? Counting stats treat different actions the same. They starve them of context. A pass is simply a pass whether it’s a slow roller to the goalkeeper or a heat-seeking missile between the opponent’s defensive lines. A final third entry could travel six inches or 60 yard

For years, football analytics tried to understand teams by adding up individual actions. More passes. More touches. More duels. More running.

But teams don’t play as collections of players.

They play as systems.

This paper makes a deceptively simple claim: if you want to understand how a team really plays, you shouldn’t start with players at all — you should start with relationships.

A review of how machine learning has been applied to analyze attacking performance, identify key indicators, and support tactical decision-making in professional football.

How do you put football data into tactical context?

Most of the time, people don’t. A pass is a pass and a tackle is a tackle no matter how they happen. Football stats typically treat each action as an isolated event, ignoring what’s going on in the game around it.

That’s because most raw football data — what’s known as “event data” — doesn’t contain information about what’s happening away from the ball. Somebody watches the game and logs every on-ball action, adding details about when and where it happened, who did it, whether it was successful, stuff like that. There’s a lot you can do with that data, but it doesn’t tell you how the teams are set up or how players are moving off the ball. As somebody once put it, event data is like listening to the game on the radio.

To help fill in the missing context, futi developed a model based on a common framework that coaches and analysts use to describe the game: phases of play.

Traditional player rating systems in football generally fall into two categories: subjective evaluations and objective statistics-based approaches. In subjective systems, experts or journalists assign ratings based on their perception of performance. In objective systems, ratings are derived from recorded match data such as passes, duels, or other events, often using statistical or machine learning techniques. Both approaches have limitations. Subjective ratings can be influenced by bias, while event-based models often require extensive tracking or event data that are not universally available.

The paper “A football player rating system” introduces a new approach: an adaptation of the Elo algorithm, originally developed for individual sports such as chess, to evaluate individual players in football. The key idea is to construct an objective and adaptive rating system that relies only on official match reports. These reports include the final score, lineups, substitutions, and minutes played.

In this post, we take a deep dive into how TFMs like TabPFN and TabICL are pre-trained to enable in-context learning (= single forward pass without weight updates). We’ll have a look into the pre-training procedure and how the pre-training data are generated (also called the prior). This post is a bit more general about pre-training TFMs, not a particular one, but I’ll reference TabPFN and TabICL mostly.