Sports Analytics Weekly by kubeia.io - 29/2026

Using ScoutGPT, a value-aware generative Transformer trained on structured match-event sequences, to simulate hypothetical player transfers and estimate tactical fit

Evaluating counterattacks, rest defense, and counter-pressing through dual machine learning models with EPV and xBG.

How scoring in the Men's World Cup compares to domestic leagues

The paper argues that football metrics should incorporate context before being used in statistical or machine learning models. Raw defensive statistics like blocks, tackles, and interceptions often reflect how much a team had to defend rather than how well it defended. By normalizing these metrics based on time spent out of possession, the authors create measures that better capture defensive efficiency. Their results show that combining both raw and normalized metrics produces more accurate models than using either alone, highlighting that thoughtful feature engineering can improve predictive performance while making models easier to interpret.

At the 2026 FIFA World Cup, offside decisions are supported by an upgraded version of Semi-Automated Offside Technology (SAOT). The system combines calibrated stadium cameras, skeletal tracking, a 500Hz sensor package inside the adidas Trionda match ball, player-specific 3D models, and a centralized video review workflow.

That sensor package is not a vague “AI chip.” Public FIFA material describes a 500Hz motion sensor in the ball, while technical reporting on connected-ball systems describes inertial sensors such as an accelerometer and gyroscope, commonly packaged together as an Inertial Measurement Unit (IMU).

An accelerometer measures linear acceleration.

A gyroscope measures angular velocity, or how fast the ball is rotating around each axis.

Together, those signals help identify the exact instant of contact when a player kicks, heads, or deflects the ball.

For the first time since 2018, OpenCV has a new major version. OpenCV 5.0 arrived in June 2026, and the single biggest change for computer vision engineers is a completely rewritten DNN module with a new inference engine built for modern neural networks. Indeed, that makes it the perfect moment to answer a very practical question: can we run a state of the art object detector, fast, on nothing but a CPU?

To keep it grounded, we point the detector at something happening right now. The 2026 FIFA World Cup is underway across the United States, Canada, and Mexico. Therefore, we pull live clips from the tournament and ask a YOLO26 model, running entirely through the OpenCV 5 DNN module, to find every player and the ball.

Crucially, there is no GPU here and no PyTorch at inference time, just an ONNX file and OpenCV. This is Part 1, so it covers the full Python workflow. In turn, Part 2 will take the exact same pipeline into C++, and later parts will explore more of what OpenCV 5 can do, from pose estimation to segmentation.

It’s not a secret that most ML models for quantile regression tend to be too biased towards the mean. You want a 10% quantile? The model’s quantile estimate will tend to be too large. And for large quantiles, the estimate is often too low. For example, I used two xgboost models to estimate 10% and 90% intervals, but ended up with an average coverage way below 80% (more like 65%).

So, what is the best model (class) to use?

The short answer: Second-generation tabular foundation models like TabICL v2 and TabPFN-3.0.

These pre-trained models are capable of outputting the full predictive distribution. Meaning you get the quantiles “for free”. Ignoring that TFM inference is kind of expensive, there is at least no additional cost.

TabFM follows in the footsteps of especially TabPFN and TabICL, which the authors note in their research announcement post. It re-uses architecture elements of both:

For example, TabFM uses alternating row and column attention, as e.g. TabPFN 2 did.

TabFM uses row compression and performs ICL over the compressed rows, as TabICL does.

Also, TabFM is pre-trained on hundreds of millions of synthetic datasets generated with structural causal models, just like TabICL and TabPFN. However, not much is known about pretraining and prior, because the published code is just the inference code, and there is no white paper or paper published, just an inference code repo, the research announcement post, and the Hugging Face model (weight) release.

Being a tabular foundation model, TabFM inherits the “standard” pros and cons of modern foundation models: No tuning needed; highly performant; slow inference; need to provide training data at inference time; and so on.

So, TabFM is not the first tabular foundation model, nor the last. What is all the hype about?

One of the reasons for the hype: TabFM climbs to the top of TabArena, a benchmark for (primarily) tabular foundation models. If you want to learn more about TabArena, I've got you covered:

This snapshot supports the view that tabular foundation models are state-of-the-art models that stand side-by-side with boosted trees and tuned neural networks. However, the development of TFMs still has a lot of momentum. Go back to December 2025, and the recommendation would have been something like: use tabular foundation models only for very small datasets. Since then, they have been eating into the boosted tree territory improvement by improvement. The question is: how far will the TFM territory expand? I’m very curious to see how the TFM field will further develop, and, of course, will share with you what I learn along the way.

How quants separate genuine edge from market noise: The episode explains why financial markets are difficult to diagnose and how researchers distinguish persistent information from randomness, temporary anomalies, and misleading patterns.

How data problems create false strategies: It examines survivorship bias, timestamp errors, incorrect corporate-action adjustments, data leakage, and excessive data cleaning—problems that can make an impossible strategy appear profitable.

How strong trading hypotheses are created: Listeners will learn how quantitative ideas emerge from economics, behavioral finance, market microstructure, and alternative data, as well as why every hypothesis must be specific, measurable, and falsifiable.

Why quants try to disprove their own ideas: The discussion explores confirmation bias, narrative fallacy, curve fitting, moving the goalposts, and the importance of defining the methodology and success criteria before examining the results.

How strategies are tested under realistic conditions: The episode covers transaction costs, spreads, slippage, financing, liquidity, market impact, position sizing, parameter robustness, and performance across different market regimes.

How professional validation methods reduce overfitting: It introduces out-of-sample testing, walk-forward analysis, purged cross-validation, bootstrap methods, multiple-testing corrections, and the Deflated Sharpe Ratio.

What makes a quantitative strategy credible: The central lesson is that a strategy becomes credible only after surviving changing regimes, imperfect data, realistic execution costs, statistical scrutiny, and repeated attempts to prove it wrong.