allowing more errors before advancing created learning resilience
In a field study of seventy-one novice soccer players, researchers found that practice governed by adaptive algorithms—ones that adjusted difficulty based on each learner's actual error rate—produced stronger skill retention and transfer than traditional blocked or random methods. The work, conducted over two days of instep kick training, suggests that matching challenge to individual performance may be a more faithful way to cultivate lasting motor learning than any fixed sequence. Where conventional wisdom prescribed a universal path, the data pointed toward a more personal one.
- Traditional practice methods—blocked repetition and random sequencing—failed to produce the retention gains that algorithmically adjusted practice achieved, exposing a gap between established coaching wisdom and emerging evidence.
- The sharpest divergence appeared not during practice itself, but in transfer tests two days later, when players faced new distances and conditions they had never explicitly trained for.
- Counterintuitively, the group allowed to make more errors before advancing outperformed even the stricter algorithmic group, suggesting that a wider margin for mistake may build deeper, more flexible learning.
- When researchers introduced financial incentives, live audiences, and social comparison to simulate competitive pressure, heart rates spiked—and the algorithmic advantage disappeared entirely, complicating the picture.
- The study marks the first field application of algorithmic practice on a real sport skill, opening questions about whether error-threshold tuning can scale across disciplines, experience levels, and the pressures of actual competition.
Seventy-one college-age soccer novices arrived to learn the instep kick—a fundamental strike requiring precise foot-to-ball contact—and were divided into five groups practicing over two days, each completing 324 kicks from varying distances. Three groups followed established motor learning methods: blocked repetition, random sequencing, and a control condition. Two groups followed something newer: algorithmic practice, where difficulty advanced only after a player reached a defined accuracy threshold. One group progressed at 67% accuracy, the other at 56%.
During the practice days themselves, no group clearly outpaced the others. But the stricter algorithmic group showed measurable day-to-day improvement that the others did not. When retention tests arrived two days later, both algorithmic groups outperformed the random and control groups—they had learned something durable.
The transfer test revealed the sharpest divide. Facing new distances and conditions they had never explicitly practiced, the group that advanced at the lower accuracy bar—the one allowed more errors before progressing—outperformed everyone, including its stricter counterpart. Allowing more mistakes, within reason, appeared to build a kind of learning resilience.
Researchers then introduced psychological pressure through financial incentives, audiences, and social comparison. Heart rates confirmed the stress was real. Under those conditions, the algorithmic advantage disappeared, suggesting that competitive pressure may flatten the differences that structured practice creates—a complication the researchers acknowledge without dismissing.
The study frames its findings within the challenge point framework, which holds that learning accelerates when task difficulty aligns with a learner's current ability. Algorithmic practice may hit that alignment more reliably than any preset schedule. The researchers are careful to note they observed outcomes, not mechanisms. What remains open is whether error-threshold tuning scales beyond soccer, whether it holds for experienced athletes, and what the brain is actually doing when a system adapts to the individual rather than the other way around.
Seventy-one college-age soccer players, none of them skilled at the sport, showed up to learn the instep kick—that fundamental strike where the inside of the foot meets the ball with precision. Researchers divided them into five groups and sent them to practice over two days, each completing 324 kicks from distances of five, six, and seven meters. What separated the groups was not the number of repetitions but the logic governing how practice evolved.
Three of the groups followed established wisdom about motor learning. One practiced in blocked order—the same distance, the same motion, repeated until muscle memory set in. Another practiced in random order, mixing distances unpredictably to force the brain to adapt. A third served as a control. But two groups followed something newer: algorithmic practice, a method that adjusted difficulty based on performance. The first algorithmic group moved to harder variations once players reached 67% accuracy. The second group waited until 56% accuracy before progressing. Both began with blocked practice, then shifted structure based on how well they were actually performing.
During the two days of acquisition, the numbers looked flat. No group clearly outpaced the others. But when researchers looked at day-to-day improvement, the algorithmic 33% group—the one with the stricter accuracy threshold—showed measurable gains from day one to day two. The others did not. Two days after practice ended, retention tests revealed the real divergence. Both algorithmic groups performed better than the random group and the control. They had learned something that stuck.
The transfer test, where players faced new distances and conditions they had not explicitly practiced, showed the sharpest separation. The algorithmic 44% group—the one that progressed at a lower accuracy bar—outperformed everyone else, including its stricter counterpart. This suggested that allowing more errors before advancing, within reason, created a kind of learning resilience. The brain had solved the problem in more ways.
Researchers then introduced psychological pressure. They offered financial incentives, placed players in front of an audience, and used social comparison to spike stress. Heart rates climbed, confirming the pressure was real. Yet under this duress, the algorithmic advantage vanished. All groups performed similarly, suggesting that stress flattens learning differences—a finding that complicates the picture but does not erase it.
The study frames its results within the challenge point framework, a theory suggesting that learning accelerates when task difficulty matches the learner's current ability. Algorithmic practice, by adjusting difficulty based on actual performance rather than a preset schedule, may hit that sweet spot more consistently. But the researchers acknowledge they did not directly measure the mechanisms at work—they saw the outcome, not the machinery beneath it.
This was the first time algorithmic practice had been tested on a real field skill rather than in a lab. The results suggest that error-threshold tuning—the idea that you can dial in how many mistakes a learner should make before advancing—might reshape how coaches design training. Not all learners need the same path. Not all errors are equally useful. The next questions are whether this scales beyond soccer, whether it works with experienced athletes, and what happens inside the brain when a system adapts to individual performance rather than forcing everyone through the same sequence.
Notable Quotes
error-threshold tuning may represent a promising avenue for future motor learning research— Study authors
The Hearth Conversation Another angle on the story
So the algorithmic groups started the same way as the blocked group—same distance, same motion—but then something changed?
Right. Once they hit a certain accuracy threshold, the algorithm shifted them into more varied practice. The 33% group had to be more accurate before moving on; the 44% group could progress with more mistakes still in their work.
And that mattered more in the transfer test than in retention?
Yes. In retention—testing the same skill two days later—both algorithmic groups were ahead. But when they faced new distances and conditions, the 44% group pulled ahead of everyone, even the stricter algorithmic group.
Why would allowing more errors lead to better transfer?
The theory is that you're solving the problem in more ways. If you progress too quickly, you might have only one narrow solution. More errors mean more attempts, more variation in how you get it right.
What happened when they added the stress—the money, the audience?
The advantage disappeared. All groups performed the same under pressure. Stress seems to erase the learning differences that showed up in calmer conditions.
So the algorithmic approach works in practice but not under game conditions?
Not quite. The stress test was artificial—a lab simulation. Real match pressure might be different. But it does suggest that however algorithmic practice builds skill, it doesn't automatically transfer to high-stakes performance. That's a separate problem to solve.