Critics of the constraints-led approach (CLA) have always made one point that sounds hard to refute: it tells coaches to manipulate the task constraints, but never tells them which one, by how much, or how they’d know it worked. It’s actually an easy criticism to answer. It simply reveals a lack of understanding of the principles of Nonlinear Pedagogy, which the CLA is part of, and implies a default back to a strictly linear view of how skill develops.

In Karl Newell’s influential constraints model, constraints are categorised as related to the individual, the task and the environment. Constraints are continuously interacting and don’t sit in isolation. They interact inside a nonlinear, dynamic system, so changing one constraint on a movement system is almost impossible and rarely produces the single, predictable effect a linear model would expect. That’s not a conceptual gap in the CLA — it’s the whole point of it. But it does leave coaches with a real daily decision that they need to approach with an open mind: which specific constraint do you change, and at what level — and once you’ve made the change, how do you know it landed? Most of the time, it’s an informed estimate, based on experience, insight and evidence, followed by “wait-and-see.”

An example: you’re three weeks into pre-season and your defence keeps leaking goals. Do you run more video analysis? More reps? More drills on shape? Coaches make this call constantly, with no real way to know which lever is likely to move the needle on performance.

But what if there was a way? Ben Teune’s PhD programme of research didn’t just notice the gap — he built tools to open it, peer around, and close it, using analytics and machine learning to measure the effects of constraints, quantify how they interact, and place a value on something coaches have only ever eyeballed: how representative a drill actually is.

Teune is a sports scientist and analytics lead at the WTA (Women’s Tennis Association), and spent four years embedded with the Western Bulldogs, a Sydney AFL team, while completing his PhD at Victoria University under Prof Sam Robertson — work that turned a string of practical coaching frustrations into a genuine analytics toolkit for skill acquisition.

Teune started out the way many sports scientists do: working on strength and conditioning, GPS units, load monitoring. The pivot came on the training paddock, watching his AFL club get smashed week after week on the weekend, while he ran a kicking contest drill. He realised a 12-week sprint program might buy a player a tenth of a second in a running speed test. Reading the play earlier, and positioning better, would buy far more than that, for free, in the dynamic performance landscape of a team game.

That single observation sent him toward a PhD in skill acquisition — and toward the question his whole body of work now answers: if perceptual and positional skill matters more than marginal physical gains, how do you actually train and develop the systems that drive these skills in action?

Building on Farrow and Robertson’s specificity index (Farrow & Robertson, 2017), Teune used a machine-learning technique called rule association to measure not just individual performance constraints, but how often they emerge together to shape performance in competition versus training.

The output is a discrepancy score between training and competition — not a verdict on what’s wrong, but a number a coach can act on. Sometimes the gap should stay. Often it shouldn’t. Either way, it’s a decision to be made by the coaching staff instead of a guess.

A second strand of the PhD work tackles functional variability: how much should a drill vary, and when? Teune’s team used clustering algorithms to sort practice tasks by the specific performance behaviours that athletes actually produced inside these constraints — not by what the drill looked like on a whiteboard. The result regularly surprised the coaches who designed the sessions.

That’s the practical payoff: a coach can build a session with genuine inter-task variability, rather than three drills that look different and train the same thing twice.

The clearest finding from Teune’s PhD research concerns the outnumber: a numbers-up (overload) scenario in a small-sided drill where one team is given an extra player — say, four attackers against three defenders — to see how they use that advantage. Across his data, teams settled into one of two distinct solutions.

In a match, that numerical advantage rarely lasts — the defender recovers, the gap closes. Which means that seeking the “safe” solution, patiently probing for the right pass, can quietly train a team to lose the very advantage it was trying to exploit. Teune’s point isn’t that one solution is always right. It’s that a coach who can see both, and knows their practice design is producing one or the other, has an actual decision to make — and a constraint or more to manipulate, like a time limit, to bias the outcome.

The phrase coaches hear most from the analytics world is “data-driven solutions.” Teune thinks that’s backwards.

Teune and Keith Davids both point toward a “Department of Methodology”: specialists who work together to do the analysis that supports the work of coaches on the track, in the gym, and with the players in front of them.

As Teune puts it, citing Ian Graham’s How to Win the Premier League, the best decisions get made when the data and the coaching team’s insights and interpretations align.