Correction, 13 October 2026: This article originally overstated the maturity of athlete digital twins and described TCS’s heart model as a full-athlete example. It also used market forecasts and a pooled AI accuracy figure to imply practical validation. Those claims have been removed or narrowed. Unsupported statements about Caribbean data collection and SportsBrain deployments have also been removed. The original publication date and byline are retained.
A coach considering a digital twin needs a clear answer to a modest question: what decision will this model help me make? A moving image of a runner and a dashboard of numbers can look impressive. Neither tells the coach whether the next recommendation deserves to influence training.
For Caribbean clubs weighing the cost of a new system, that question belongs before the purchase. Ask the supplier to name the body system or performance task it models, the measurements it needs, and the circumstances in which its estimates should be set aside.
Start with what the model actually represents
In this article, an athlete digital twin means a model linked to measurements from a particular athlete, or a defined part of that athlete’s body, and updated as relevant measurements change. Its scope might be quite narrow. Calling it a twin does not establish that it represents the whole person or predicts every useful outcome.
A wearable, a video-analysis tool and a simulation can all contribute information. They perform different jobs. A club should ask how each measurement reaches the model, how old the latest input is, and what happens when a sensor fails or a session is missing.
Des Linden’s example is a heart model
Tata Consultancy Services describes its work with runner Des Linden as a digital model of her heart. The company’s project page discusses a full-person twin as a possible extension. This is a useful example of organ-level modelling, with a clearly bounded subject. Read TCS’s project description.
The page is the developer’s account of its own project. It does not, on its own, establish that a whole-body model can choose a safe training load for a Jamaican sprinter. A club considering a different product needs evidence for that product, athlete population and decision.
Ask what was tested
A large market forecast tells a coach little about whether a particular recommendation is reliable. The same problem arises when an impressive accuracy score arrives without an explanation of what the model was asked to classify or predict.
Suppose, in a deliberately fictional example, a system correctly labels the type of drill in a video. That result would not show that it can predict the runner’s next injury. The input, outcome and cost of a mistake are different. Nor would an accurate risk estimate, by itself, prove that acting on it improves health.
Digital-twin researchers Guodong Shao, Joe Hightower and William Schindel describe verification, validation and uncertainty assessment as part of establishing model credibility. Their paper concerns manufacturing. We borrow its evaluation questions here as an engineering approach, without treating it as clinical evidence for athletes. See the NIST publication record.
- Verification: Was the model implemented as intended?
- Validation: Does it perform adequately for the particular use being proposed?
- Uncertainty: How much confidence belongs in the result, and when should the system decline to advise?
For a sports programme, a useful supplier discussion would also cover the number of people tested, the kinds of athletes included, how unseen data were kept separate, and the errors that occurred. Ask to see the comparison with the club’s current approach. A graph that looks plausible is a starting point for questions.
A small, private evaluation before an athlete trial
The following is an editorial suggestion for examining a proposed workflow. It is not a training prescription or a report of a SportsBrain deployment.
- Write one question. For example: can staff tell which fictional session records are incomplete?
- Use invented adult records. Make up session dates and missing-data flags. Do not copy an athlete’s identity, injury history, medical file or a child’s data into the exercise.
- Check the result by hand. Record where the system is correct, mistaken or unable to answer. Decide in advance what would count as a useful result.
- Test the handover. Can another staff member understand the assumptions and retrieve the records without relying on the person who set it up?
This exercise tests a basic record-handling workflow. It cannot validate physiological predictions. Any later proposal involving real athletes needs a separate plan for consent, access, retention, independent evaluation and professional responsibility. Health-related decisions belong with appropriately qualified clinicians and sports-science staff, with the athlete involved.
What would justify the next step?
A useful proposal should make the costs visible: staff time, data preparation, equipment, ongoing support and the ability to export records if the club leaves. Ask who can use the data and whether the supplier intends to train other models with it. Those details deserve an answer before real records enter the system.
Caribbean programmes should be able to assess a tool on the decisions it supports and the evidence behind it. We have not established in this article that SportsBrain has deployed or independently validated an athlete digital twin. A future claim of that kind should identify the model, intended use, evaluation and limitations. For now, the practical work is to define the question and demand a test that answers it.
Frequently asked questions
What is an athlete digital twin?
A model linked to measurements from a particular athlete or part of their body. Its developer should specify what it represents, how it is updated and which questions it has been tested to answer.
Did TCS build a full-body twin of Des Linden?
The TCS project page describes a digital model of her heart. It discusses a full-person twin as a possible extension. That distinction matters when evaluating what the project demonstrates.
Does a high AI accuracy score prove that a twin can prevent injury?
No. The task, test population, comparison method and errors all matter. A score from a different prediction task does not establish that changing training on the model’s advice improves an athlete’s health.
What should a club try first?
Start by testing the proposed workflow with fictional adult data and a written evaluation plan. Keep real medical records and children’s data out of public AI tools. Involve qualified professionals before any health-related use.
Sources and method: The TCS developer description establishes the scope of its heart project. The NIST publication supplies an explicitly adapted engineering evaluation framework. The fictional workflow and purchasing questions are editorial suggestions; no interviews or athlete trials are claimed.