Regular readers will know I have spent the last couple of months playing with AI agents to scrape and visualise publicly available sport data — first football muscle injuries, then a live dashboard for the Giro d’Italia. Those were fun, low-stakes experiments to learn what the tools could do. This post is the point where the experiment turned into something more serious: a full research study, now posted as a preprint, built almost entirely with the same family of AI tools.
I planned, conducted and executed this study — and developed the accompanying digital twin — using Claude Sonnet 5 and Claude Opus, together with Claude Cowork, Claude Code and Claude Design. Cowork orchestrated the data gathering and organisation across ten years of race results and weather records, Code built the statistical models and the digital twin engine, and Design shaped how the outputs are presented. I stayed firmly in the loop throughout as the domain expert: setting the research questions, checking the physiology and the methodology at every step, validating the data, and deciding what the numbers actually meant. I also modified some of the coding as things progressed to develop various analytics steps.
Why triathlon, why heat
Triathlon is an Olympic sport, and elite Olympic-distance racing is shaped by the interplay of swim-bike-run pacing, transition efficiency, the quality of the field, and — increasingly — the heat athletes race in. Major championships are more and more often held in hot conditions, which matters enormously for how athletes and support staff plan training, pacing, cooling and acclimatization. Despite there being a decade of publicly available race results and weather records out there, nobody had linked the two together at scale. That gap was the starting point.
The study had four aims: characterise the performance signature of a podium finish; reconstruct the thermal environment of championship venues over ten years; identify which athletes seem resilient to heat; and, building on all of that, prototype a digital twin that could support race planning.
Ten years of racing, in numbers
The dataset ended up covering 7,415 athlete-race records across 339 events and 3,060 unique athletes, with wet-bulb globe temperature (WBGT) at those races ranging from a mild 7.1°C to a much more challenging 28.0°C.
A few things stood out. Bike and run splits each contribute roughly equal unique variance to total race time, but the run leg is the real discriminator between the podium and the rest — it was the single most important feature in the podium-prediction models, at 45.5% importance for men and 48.9% for women. Somewhat counter-intuitively, the slope linking heat to performance did not reach statistical significance for either sex (men p=0.065, women p=0.104), which is a useful reminder not to over-interpret heat effects from headline temperature alone. DNF rates, on the other hand, told a clearer story, ranging from 12.7–16.3% for men and 11.7–18.8% for women across World Triathlon’s Green and Blue flag heat-risk categories.
Building the digital twin
The last aim — and the part I am most excited about — was turning ten years of descriptive analysis into something forward-looking. The digital twin prototype couples a performance model with a thermo-physiological heat-strain model, so it can be used to explore pacing, cooling and acclimatization decisions ahead of a race rather than just explaining results after the fact. On a temporal hold-out (training on earlier years, testing on later ones — the fairest test for something meant to inform future decisions), it achieved a mean absolute error of 0.490 z-units for men and 0.468 for women (r=0.383 and 0.295 respectively). Those are honest, prototype-grade numbers, not a finished predictive tool, and I say so explicitly in the paper.
The usual health warning
This is a preprint. It has not been peer reviewed, and I want to be upfront about that rather than let the AI-workflow angle overshadow it. The underlying data are scraped from publicly available sources, so they carry all the usual caveats about completeness and accuracy that come with that. What I can say is that the statistical approach, the modelling choices and the interpretation were all reviewed and directed by me at every stage — the AI tools accelerated the mechanics of gathering, structuring, analysing and building, but the scientific judgement was, and had to be, human. Critical thinking stays firmly the job of the person in the loop, whatever is doing the typing.
Why this matters to me
I hope this prototype is the beginning of something bigger rather than a one-off curiosity. There is an enormous amount of publicly available data across sport that nobody has the time to properly interrogate and organise — results archives, weather records, GPS feeds, injury registries. What changed for me this year is that a single person, using these AI tools well, can now plan, run and analyse a study of this scale in weeks rather than requiring a full research team and many months. I want to keep using that capability to ask more questions like this one across sports science and sports medicine, and to be transparent about the process as I develop more tools and research questions.
The Lactate Threshold app started as a simple tool to turn a step test into a clean set of threshold values. Over the past few development cycles it has grown into something closer to a complete profiling and prescription workspace. This post walks through the most significant additions — anaerobic speed reserve, maximum speed determination, switchable 3- and 5-zone models — and the smaller refinements that came with them.
Anaerobic Speed Reserve (ASR)
The headline feature is the ability to determine an athlete’s anaerobic speed reserve — the velocity range that sits between the speed at maximal oxygen uptake (vVO2max, or maximal aerobic speed) and maximal sprinting speed (MSS). Everything an athlete does above their aerobic ceiling happens inside this band, which is exactly why it matters for events decided by surges, kicks and repeated high-intensity efforts.
The concept owes much to the work of Dr Gareth Sandford and colleagues, who showed that ASR and maximal sprint speed are “untapped tools” for differentiating the world’s best middle-distance runners and for understanding the complexity of athlete profiles that traditional aerobic categories miss. Two athletes with an identical vVO2max can have very different reserves above it — and therefore very different tolerances to supramaximal work — information that is invisible if you only look at threshold and VO2max.
Dr. Martin Buchheit’s research extends this directly into programming. Prescribing high-intensity work as a percentage of maximal aerobic speed alone ignores the differing mechanical ceilings between individuals, so the same session can impose very different relative stress on two athletes. Anchoring supramaximal efforts to a percentage of the ASR instead normalises that stress, and the evidence shows it reduces the inter-individual variability of physiological adaptation. The app now makes that calculation a single step rather than a spreadsheet exercise.
Maximum Speed Determination
Because ASR depends on having a reliable upper anchor, the app now supports maximum sprint speed (MSS) determination as a first-class input. Enter the result of a short maximal sprint and the app uses it as the top of the reserve, pairing it with the aerobic anchor derived from the step test. This closes the loop: from a single profiling session you get the threshold values, the aerobic speed, the sprint ceiling, and the reserve that connects them.
Flexible Training Zones: 3 or 5
Training-zone prescription is now configurable. You can choose between a 3-zone model — the classic below-LT1, between-thresholds, above-LT2 structure favoured in polarised approaches — and a more granular 5-zone model for coaches who want finer resolution across the intensity spectrum. Zones are generated directly from the athlete’s own threshold and speed anchors rather than from generic percentages, so the prescription reflects the individual profile the test produced.
Switching between the two models takes a tap, which makes it easy to align the output with whichever periodisation philosophy a given athlete or training block calls for.
Other Improvements
Alongside the marquee features, this round of work brought a number of refinements: cleaner presentation of the threshold detection results, a more consistent workflow from data entry through to zone output, and better handling of the speed-based inputs that the ASR and MSS features rely on. The aim throughout has been to keep the app fast to use rink-side or track-side while quietly adding depth for those who want it.
Development is ongoing, and I’ll keep posting updates here as new capabilities land. If you’re using the app and have feedback or feature requests, I’d be glad to hear them. If you use it for any purposes make sure you reference it:
⚠️ UPDATE 1st June 2026 ⚠️: I have now added the option for Cycling Tests.
A note for team-sport coaches: if you are specifically after a tool to plan HIIT sessions with change-of-direction (COD) prescriptions, I’d recommend Dr Martin Buchheit’s dedicated COD shuttle prescription app, available here. It is purpose-built for that use case and complements the profiling work the Lactate Threshold app is designed for. A screenshot is below.
Key References
Sandford GN, Allen SV, Kilding AE, Ross A, Laursen PB. Maximal Sprint Speed and the Anaerobic Speed Reserve Domain: The Untapped Tools that Differentiate the World’s Best Male 800 m Runners. Sports Medicine, 2019.
Sandford GN, Laursen PB, Buchheit M. Anaerobic Speed/Power Reserve and Sport Performance: Scientific Basis, Current Applications and Future Directions. Sports Medicine, 2021.
Buchheit M, Laursen PB. High-Intensity Interval Training, Solutions to the Programming Puzzle. Part II: Anaerobic Energy, Neuromuscular Load and Practical Applications. Sports Medicine, 2013.
For most of my career, the richest data in elite football sat behind closed doors. Tracking systems, multi-camera optical feeds, possession context, off-the-ball movement — the kind of information that actually explains why a match unfolds the way it does — was the private property of a handful of federations and clubs wealthy enough to own it. Everyone else worked with shots, corners, fouls and possession percentage, and squinted to fill in the rest.
That is changing, and the FIFA World Cup has become the most visible stage for the shift. Over the last two tournaments FIFA has done something I think is genuinely important for our field: it has taken some of the most modern performance data ever produced in football and made a meaningful slice of it public. As someone who has spent years arguing that data only becomes knowledge when people are allowed to interrogate it, I find this exciting — so I wanted to write down where this data lives, who is doing interesting things with it, and why the act of sharing matters as much as the numbers themselves.
FIFA’s initiative: Enhanced Football Intelligence
The centrepiece is Enhanced Football Intelligence (EFI) — the set of metrics that first appeared as those small graphics in the corner of the screen during the 2022 World Cup. EFI was built by FIFA’s Football Performance Analysis & Insights team to move us beyond traditional counting stats and toward metrics that describe how a team plays.
What makes EFI different is its source. Rather than relying only on on-the-ball event data, it combines event data with live tracking data from every player on the pitch, captured by a multi-camera optical system. When something happens, you know where all twenty-two players were in relation to it. That positional context is what unlocks metrics you simply cannot derive from a traditional stats sheet, including:
Line breaks — how often a pass cuts through an entire defensive unit, and whether it went through, around or over. FIFA’s own analysis shows that the more line breaks a team concedes in midfield, the more games it tends to lose.
Ball recovery time — how long, on average, it takes a team to win the ball back after losing it.
“In contest” possession — the honest third category that sits between “our ball” and “their ball,” capturing the messy phases when nobody is truly in control.
Receptions behind the midfield and defensive lines — where and how players make themselves available between the opponent’s units.
Pressure on the ball and forced turnovers — whether a side is genuinely disrupting the opponent or merely looking busy.
Crucially, FIFA makes this available to the public. The metrics are explained — with video and multilingual PDFs — on the FIFA Training Centre, and snippets of match data are published after games. For the 2026 tournament FIFA has gone further still, layering on two new initiatives: the FIFA Power Rankings, an objective player-rating system scoring every outfield player 0–10 for attacking, creativity and defending using EFI algorithms; and FIFA AI Pro, which gives all 48 teams the same generative-AI tools to explore match data and rebuild moments in 3D — explicitly framed as democratising analytics that used to belong only to the biggest budgets.
Where to find the data and the people working with it
Here are the sites I’d point any coach, analyst or curious fan toward. I’ve grouped them deliberately, because one distinction matters a great deal and is easy to miss: some of these work directly from data officially published by FIFA, while others produce excellent World Cup analysis using third-party providers such as StatsBomb. All are worth your time — but knowing which is which keeps you honest about where a number actually came from.
This is the well from which everything else is drawn. Free, open, and aimed at coaches of every level, it hosts the EFI metric explainers, the “Football Language” glossary, video breakdowns, the FIFA Insight interviews with the people who built EFI, and — the part I’d flag for 2026 — the Match Report Hub, a live index of post-match summary reports for every World Cup match, organised by group and added as the tournament unfolds. If you only bookmark one link from this article, make it this one.
The sharpest example I’ve seen of someone treating FIFA’s openness as a system rather than a curiosity. Tactics Journal points out that FIFA is publishing a roughly 52-page post-match summary for every group-stage game — formations, pressing phases, line-break tables player-by-player, defensive pressure maps, sprint-zone physical data — all in a consistent structure. Their argument is that because every report follows the same layout, you can parse all of them into a structured, auditable evidence layer for scouting and tournament-wide tactical questions. They call it “tactical infrastructure,” and I think that’s exactly the right way to think about it: the PDF is content; the structured layer you build from it is infrastructure.
An independent blog that does something quietly useful: it organises and explains the EFI metrics as a reference, pulling together the definitions of phases of play, line heights, team lengths, receptions and the rest into one navigable place. A patient companion to the official material that helps a newcomer go from “what is a line break?” to actually reading a match through that lens.
My favourite example of why open methods matter as much as open data. Parlak built an open-source implementation of FIFA’s EFI metrics — data, concept and visualisation layers — with the explicit goal of reproducing FIFA’s match reports and testing whether the published concepts are specified well enough to be rebuilt by an outsider. That is science in the best sense: take the published method, try to recreate it, and flag the ambiguities. For analysts and students it doubles as a practical toolkit for generating EFI-style visualisations. You can read his Master’s thesis on this project here.
The wider analytics and data-journalism ecosystem
These don’t all run on FIFA’s own feed, but they show what a culture of shared football data makes possible — and they’re some of the most engaging World Cup analysis being published right now.
A blog “powered by data science and written by journalists,” produced by Northeastern’s Network Science Institute (the NetSI Sport group led by Brennan Klein). It’s a masterclass in turning event data into narrative: passing networks and passing-cluster maps that fingerprint a team’s style, xG shot maps, dribble-and-carry graphics, and genuinely novel angles like whether the 2026 hydration breaks are changing scoring patterns. Worth knowing that their underlying data comes from Hudl StatsBomb(over 3,400 events per match), not FIFA’s EFI feed — a good illustration of how the official and commercial data worlds sit side by side.
Not a football site at all, but a weekly roundup of the best data visualisations from newsrooms around the world, and during the tournament it’s been a reliable showcase of World Cup charts — qualification journeys, the evolution of the match ball, player and game analyses from the likes of Reuters, The New York Times and El País. The best place to see how professional data journalists choose to present this kind of information. Some details are also here.
A reminder that you don’t need a newsroom to do this. This is a community-built interactive Power BI dashboard — one example aimed at World Cup Fantasy players, with a “Stat View” toggle to compare a player’s club versus international form before making transfer decisions — sitting within Fabric’s wider Data Stories Gallery of user-made World Cup dashboards. A nice window into the grassroots, build-it-yourself end of the spectrum.
(Two more worth a look in the same spirit: Tactical Football Analysis for written post-match tactical breakdowns, and Flourish, whose football chart templates are a quick way to build your own tournament visualisations.Also, have a look at this substack on how to build team-shape visualisations.)
What the analysis actually looks like
The reason this data is worth sharing is that it produces visuals that change how you see a match. A few of the workhorse formats:
Pass networks map who connects to whom and where, turning a team’s structure into a readable shape — you can see at a glance whether a side is building through its full-backs, overloading one flank, or bypassing midfield entirely.
Tracking heatmaps show where players and teams actually spend their time, exposing the difference between nominal position and real behaviour, and revealing how compact or stretched a side is in and out of possession.
xG (expected goals) shot maps put a probability on every chance so you can judge whether a team created genuine danger or simply accumulated low-value shots — useful, as long as you remember it’s a guide, not a verdict.
The principle underneath all of them is the same one I keep coming back to: the best statistics don’t make football more complicated, they make it easier to understand. A line-break count, a compactness number or a clean sprint map isn’t there to replace the coach’s eye — it’s there to sharpen the question the coach is already asking as well as providing additional information for support staff to improve how players are prepared.
The other half of the story: injuries and player health
It would be a strange omission, for me especially, to write about World Cup data and talk only about what happens when players are on the pitch. The same tournament that generates all this performance data also generates a great deal of information about the cost of playing it — and FIFA itself frames its data mission broadly, as unlocking the potential of video and data to drive technical development and education, not just tactical analysis.
This World Cup has put player load firmly in the spotlight. The expansion to 48 teams and 104 matches, layered on top of an already congested club calendar, has prompted warnings from sports-medicine specialists that fatigue and tight scheduling are pushing injury rates up — with the knee particularly exposed to the constant cutting, pivoting and rapid changes of direction the modern game demands. FIFA’s own venue medical staff have pointed to the injuries they see most often in elite players: ankle sprains, and hamstring and calf strains. None of that is new in kind, but the volume and the schedule are. At the end of each tournament there is always a published paper on the injury surveillance activities conducted by the FIFA medical team. If you want to read more about our experience at the FIFA 2022 World Cup you can read the following papers:
For the public, the most visible “injury data” is the running tracker — outlets such as ESPN, The Independent and others maintain live lists of who is ruled out or racing to be fit, and this tournament has tested several squads hard, with sides like Brazil and the Netherlands losing first-choice players before a ball was kicked. These are journalism rather than open datasets, but they perform a real function: a structured, continuously updated public record of availability.
FIFA has a long history of medical research around its tournaments, and the combination of tracking data (sprint loads, high-intensity distances, accelerations) with injury records is exactly the kind of linkage that could move us from counting injuries to understanding and preventing them. The performance data and the medical data are two halves of the same picture; sharing both is how we protect the players who generate it.
Why sharing the data matters
I want to close on the part I care about most, because it’s easy to treat “FIFA released some metrics” as a minor technical footnote. It isn’t.
It democratises insight. When tracking-derived data was private, the gap between the richest and poorest programmes was partly a data gap. Publishing EFI, and giving every World Cup team the same AI tools, narrows that gap. Insight stops being a function of budget alone.
It creates a shared language. When a coach in one country and an analyst in another can both point to the same definition of a line break or “in contest” possession, conversations get more precise and more productive. Common metrics are the grammar of a common discussion.
It invites scrutiny and reproducibility. The moment a method is public, people like Doğan Parlak can try to rebuild it, stress-test it, and improve it. That is exactly how a field matures — not by guarding methods, but by exposing them to challenge.
It stimulates new ways of analysing the game. This is the part that excites me as a scientist. Hand the same dataset to a hundred curious people and you will get analyses no single organisation would ever have commissioned. Open data is generative: it produces questions, tools and visualisations that wouldn’t otherwise exist, and the elite game gets smarter as a result.
Football has always been understood through stories and through the eye. What FIFA’s data initiative does is add a third lens — a transparent, shareable, contestable one — and then, remarkably, hand it to everyone. The numbers are interesting. But the decision to share them is what will change how we understand the game at the very highest level. So, well done to my FIFA colleagues for this initiative.
At Aspetar we have produced a special issue of our Journal dedicated to Football and the World Cup. You can access it by clicking on the cover page.
You must be logged in to post a comment.