Elite athletics has acquired an impressive quantity of measurement and comparatively little shared understanding of what to do with it. Force plates, timing systems, markerless video, wearables and athlete management platforms have all improved. The decision they exist to inform, which is whether this athlete should do this work today and why, is still made largely from memory, relationship and the coach’s own record-keeping. It is still very rarely written down in a form that survives a change of staff.
Key findings at a glance
- Athletics is a family of sports and the evidence is unevenly spread across it. Sprint and endurance research is comparatively rich. Throws, combined events and race walking are served far more thinly. A model trained on the whole of published sports science will be confidently wrong most often exactly where coaches most need help.
- Measurement has outrun interpretation. Markerless video now reports sagittal joint angles to within a few degrees of laboratory systems in gait, while transverse-plane rotations remain poor. Sprint force-velocity outputs are reproducible, but the slope most often used to individualise training is not.
- The dominant load metric does not survive its own statistics. A sustained methodological literature shows the acute to chronic workload ratio is mathematically coupled, that its apparent optimal zone can be an artefact of grouping, and that it produces the familiar association even when fed meaningless inputs.
- Prediction is the wrong ambition and continuity is the right one. Prospective work in track and field reports that athletes who lose fewer weeks to injury and illness are more likely to meet their performance objectives. That association is a better foundation for a product than any prediction claim the current evidence supports.
- Junior ranking tells a programme very little. Across Olympic sports, junior performance explains around two per cent of the reliable variance in senior performance. In athletics specifically, between six and 24 per cent of world-ranked juniors reach the equivalent senior ranking, depending on event and sex.
- Explainability is a performance requirement rather than a compliance chore. A recommendation a coach cannot interrogate will be ignored by the good ones and followed uncritically by the inexperienced, and both outcomes are bad.
Why track and field needs an architecture of its own
Athletics is routinely treated as one sport by the technology sold to it. It is not. A hammer thrower, a 10,000 metre runner and a 400 metre hurdler share a governing body and very little else in the way of performance model, training structure or injury profile.
That matters because the published evidence is not distributed evenly across those events. It clusters where research is easy and where funding follows, which means sprints and distance running. A general-purpose system trained on that literature will produce fluent, plausible advice for a shot putter, drawn from almost nothing. The fluency is the danger. It reads exactly like the advice that is well founded.
From the annual plan to Tuesday morning
The paper’s central argument is that the useful contribution of artificial intelligence in athletics is an explainable decision layer, aware of the event, honest about uncertainty, and designed to leave a named human being accountable.
The decision that matters is small and repeated. A coach stands on a cold track in February and decides whether the athlete in front of them does the session as written. That decision is made hundreds of times a season, it is where the annual plan either survives contact with reality or does not, and it is almost never recorded with its reasoning.
Three things the market sells that the sport should refuse
- Single-score readiness indices. A composite number that collapses sleep, load, mood and heart rate variability into one figure hides the disagreement between its inputs, which is the only genuinely useful information in it.
- Individual injury prediction. The published performance of these models does not support the decisions they are sold to make.
- General-purpose training generators that have learned the shape of coaching advice without the substance of any particular event.
What follows for a high-performance programme
The practical recommendation is narrower than the technology permits and more useful than it promises. Connect the records before buying any intelligence, because a system reasoning across four incompatible spreadsheets will reason confidently about the wrong athlete. Require every recommendation to name its sources, state its confidence and offer alternatives. Put a named person between any recommendation and any athlete, with that person’s decision recorded alongside it.
Pilots should measure availability, completed planned weeks and the quality of the decision record, rather than accuracy claims that cannot be audited. A decision record is the only asset in this field that appreciates.
Free download
Get the full paper
Read The Intelligent Track in full, with every claim carrying an evidence grade and the full reference list. Give your name and email and the PDF opens straight away.
Frequently asked questions
Is the acute to chronic workload ratio reliable for athletics?
Not as an individual injury predictor. A sustained methodological literature shows the ratio is mathematically coupled, that its apparent optimal zone can be an artefact of how athletes are grouped, and that it produces the familiar association even when fed meaningless inputs. Load records remain useful for describing what an athlete actually did. The ratio does not support the individual predictions made from it.
Can AI predict which junior athletes will succeed as seniors?
The underlying relationship is too weak to build a selection system on. Across Olympic sports, junior performance explains around two per cent of the reliable variance in senior performance. In athletics, between six and 24 per cent of world-ranked juniors reach the equivalent senior ranking, depending on event and sex. Selection systems built on early results discard people for no good reason, and a model trained on those past selections learns the programme’s history rather than the sport.
How accurate is markerless video for athletics biomechanics?
It depends entirely on the plane of movement. Markerless systems now report sagittal joint angles to within a few degrees of laboratory systems in gait. Transverse-plane rotations remain poor. A coaching decision that rests on a rotation measurement is resting on the weakest output the system produces, and most interfaces do not say so.
What should an athletics programme measure instead of a readiness score?
Availability and completed planned training weeks. Prospective work in track and field reports that athletes who lose fewer weeks to injury and illness are more likely to meet their performance objectives. Both measures are countable and auditable, which no proprietary composite score currently is.
Why does event specificity matter so much for AI in athletics?
Because the published evidence is unevenly distributed. Sprint and endurance research is comparatively rich, while throws, combined events and race walking are served far more thinly. A model trained across all of it will be most confident precisely where its foundation is thinnest, and a coach has no way of telling from the interface which is which.
About this series
This is the second of seven papers on the responsible use of artificial intelligence in performance sport. The governance framework it draws on is set out in The Responsible Performance System. The community athletics counterpart is Closing the Coaching Gap, and the series also covers talent identification in football, triathlon, elite football and golf.
Legal and regulatory notice. This article and the paper it summarises describe legislation, regulatory guidance and governing-body requirements. Nothing here constitutes legal advice, and the provision of legal advice sits outside the terms of any engagement with the author or with Athleet.AI. The material is presented to support discussion and further review by qualified advisers. Every athlete, squad and scenario used as an illustration is synthetic.