The Connected Endurance Athlete: integrating training, recovery and race execution in triathlon with AI

Triathlon does not have a shortage of measurement. It has a shortage of coordination. Three disciplines, a gym, recovery, fuelling, equipment, environment and the rest of an athlete’s life have to be reconciled into one week, and no product on the market holds all of it in a form a coach can reason about.

The athlete this paper has in mind is not a professional. She is 38, has two children and a job, trains about nine hours a week and has been trying to go faster for four years. On her phone are six applications. One holds the swim, one the bike, one the run, one the strength work, one a sleep score she has learned to distrust and one a plan she bought in January and stopped following in March. None of the six knows about the other five.

Key findings at a glance

• The interference effect is narrower than the folklore. Across 43 studies, concurrent training cost nothing measurable in maximal strength or muscle size. The cost fell on explosive strength, and it appeared only when both modes were done in the same session.
• Strength work changes economy, not the engine. Concurrent training improved running economy by about 3.9 per cent, and explosive work by 4.8 per cent, while a recent meta-analysis found no improvement in maximal oxygen uptake or the maximum metabolic steady state.
• No commercial readiness score has been validated. A search of the peer-reviewed literature found no validation of any proprietary composite recovery or readiness score against a physiological or performance criterion. The evidence that exists is qualitative.
• The instruments are worse than the interfaces suggest. Wrist heart rate carries limits of agreement from minus 24 to plus 18 beats per minute. Sleep-stage specificity against laboratory measurement ranged from 0.18 to 0.54, and was worst on disturbed nights.
• How the bike is ridden changes the run. At identical average power, an hour of variable cycling made the following 9.3 kilometre run 42 seconds slower, and 35 of those seconds appeared in the first half.
• The intensity debate is softer than it sounds. A network meta-analysis of individual participant data found no difference between polarised and pyramidal distributions, and the same training classified two defensible ways differs by a factor of three.

Reconciliation and explanation rather than prescription

The honest opportunity in endurance sport is not a better plan generator. It is a system that holds the whole week in one record, names what it cannot see, states the error on the numbers it reports, and helps a coach and an athlete make a trade-off they can both defend.

Most of what limits the age-group triathlete is not physiological. It is that the plan does not know she rode hard on Tuesday because a meeting moved, the sleep score does not know she was awake with a child, and the strength app has never been opened by anybody but her. When she asks her coach why she is slower than last year, the honest answer is that nobody has the whole picture.

Why the single score has to go

A composite readiness number is attractive precisely because it hides disagreement between its inputs, and that disagreement is the useful signal. No proprietary composite score has been validated against a physiological or performance criterion in the peer-reviewed literature.

The instruments feeding those scores are also weaker than their presentation implies. Wrist heart rate carries limits of agreement from minus 24 to plus 18 beats per minute. Sleep-stage specificity against laboratory measurement ranged from 0.18 to 0.54 and was worst on exactly the disturbed nights an athlete most wants to understand.

Four things to refuse outright

1. No composite readiness score presented as a fact.
2. No automatic rewriting of a plan without a person approving it.
3. No inference about energy availability, which the governing consensus reserves for a physician and for which no single validated diagnostic test exists.
4. No claim to have measured what the device only estimated.

What follows for a coach or an athlete

Hold one record, not six. Reconcile the disciplines against one athlete identity and one clock, and treat the gym and the rest of life as part of the record rather than as context somebody remembers. Report error alongside every derived number. Measure completed planned weeks, which is countable, auditable and the measure most closely linked to hitting a performance goal in the published prospective work.

None of that requires a better model, which is the part that tends to disappoint. The changes that would help most are administrative rather than analytical, and the largest of them costs one decision about which source wins.

Frequently asked questions

Are wearable recovery and readiness scores accurate?

No proprietary composite recovery or readiness score has been validated against a physiological or performance criterion in the peer-reviewed literature. The evidence that does exist is qualitative. The underlying instruments are also weaker than the interfaces suggest, with wrist heart rate carrying limits of agreement from minus 24 to plus 18 beats per minute and sleep-stage specificity ranging from 0.18 to 0.54 against laboratory measurement.

Does strength training interfere with endurance training?

Far less than the folklore claims. Across 43 studies, concurrent training cost nothing measurable in maximal strength or muscle size. The cost fell on explosive strength, and only when both modes were performed in the same session. Strength work improves running economy, by about 3.9 per cent for concurrent training and 4.8 per cent for explosive work, without improving maximal oxygen uptake or the maximum metabolic steady state.

Does how you ride the bike affect the triathlon run?

Yes, and average power conceals it. At identical average power, an hour of variable cycling made the following 9.3 kilometre run 42 seconds slower, with 35 of those seconds appearing in the first half. A pacing plan built on average power alone is missing the variable that most affects the discipline it feeds into.

Is polarised training better than pyramidal?

The evidence does not separate them. A network meta-analysis of individual participant data found no difference between polarised and pyramidal distributions. The debate is further softened by classification, since the same training classified in two defensible ways can differ by a factor of three in how much time it appears to place in each zone.

What should a triathlete measure instead of a readiness score?

Completed planned training weeks. It is countable, auditable, and it is the measure most closely linked to hitting a performance goal in the published prospective work. It also has the advantage of being understood identically by the athlete and the coach.

About this series

This is the fifth of seven papers on the responsible use of artificial intelligence in performance sport. The governance framework it draws on is The Responsible Performance System. The series also covers elite athletics, grassroots athletics, talent identification in football, elite football and golf.

Legal, medical and regulatory notice. This article and the paper it summarises describe physiological, nutritional and environmental evidence and touch on health and competition regulation. Nothing here constitutes medical or legal advice, and nothing in it should be used to diagnose, treat or screen for any condition. The provision of such advice sits outside the terms of any engagement with the author or with Athleet.AI, and the material is presented to support discussion and further review by qualified practitioners and advisers. Every athlete and scenario used as an illustration is synthetic.