A Narrow Window for Feedback

Picture a kid learning to ride a bike with training wheels bolted on. They wobble off down the path, upright, grinning, apparently cycling. Take the training wheels off and they fall over in a metre.

That’s the problem with feedback at the wrong moment, and it’s the image Peter Brown reaches for in Make It Stick:

Immediate feedback is like the training wheels on a bicycle: the learner quickly comes to depend on the continued presence of the correction.

Hold that thought, because it sits in direct contradiction to almost every other thing I’ve read about how people get good at things.

A genuine contradiction

Open articles and books in the area of learning science and ask one simple question — when should feedback arrive? — and the books start arguing with each other.

Scott Young, in Ultralearning, lists immediacy as one of the defining features of how the best learners practise. Eva Keiffenheim’s learning newsletter bolds the instruction outright: get immediate feedback. And then Brown, above, says a brief delay beats immediate. And the evidence goes further still: in a separate essay on motor skills, Young relays a blunt verdict from the researchers Schmidt and Wrisberg — “concurrent visual feedback,” correction that arrives while you’re still performing, “is usually disastrous for learning.”

Immediate is essential. Immediate is disastrous. The same author — Young — appears to say both, depending on which learning problem he’s looking at. Either learning science is incoherent, or the word immediate is hiding something.

It’s hiding something.

One window, two ways to miss it

The trick is to stop thinking of feedback as a dial between “fast” and “slow.” There’s a window of opportunity you’re aiming for — and you can miss it in two directions.

Too early is feedback that arrives while you’re still performing — the training wheels. Correct someone mid-sentence, every time, and they learn to lean on the correction instead of building their own sense of when something’s off. They get fluent with you in the room and fall apart the moment you leave. Motor skills aren’t language — but the warning carries: when your correction replaces the learner’s own monitoring, you build dependence, not skill.

Too late is feedback that arrives weeks later, or never. This is the quiet killer in language learning especially: your conversation partner nods and smiles whether your grammar was right or not, so your mistakes harden into habits while you feel like you’re getting better. Confidence goes up; skill doesn’t. The decision scientist Robin Hogarth calls these “wicked” learning environments. The feedback is missing or misleading, so experience teaches you the wrong lesson.

The window itself is the stretch between them. It opens the moment you commit to an answer of your own, and closes once the mistake hardens into habit. Land your feedback inside it and it sticks. And there’s a lovely reason it works. Researchers call it the hypercorrection effect:

When students are absolutely positive they know something and it turns out to be incorrect, they’re much more likely to retrieve the correct answer in the future.

You have to commit first. The confident wrong guess is what makes the correction stick — there’s nothing for a fix to grab onto if you never reached for an answer of your own. (That effect comes from memory experiments — people recalling facts, not speaking a new language — so I’m leaning on it as a principle, not a proven law of grammar. It rhymes with what the classroom studies show.) Jump in too early and you rob the learner of the very mistake that would have taught them.

The contradiction dissolves

Now the books stop fighting. When Young praises immediacy, he means don’t let the loop stretch to months — don’t be the silent learner studying grammar for a year before speaking a word. He’s guarding the edge where the window closes. When Brown praises delay, he means don’t shrink the loop to zero — don’t be the training wheels. He’s guarding the edge where it opens.

They’re standing at different edges of the same window. It sounds like a disagreement — but it’s the same target, seen from opposite sides.

There’s a deeper reason they talk past each other: they aren’t measuring from the same starting line. The memory researchers’ immediate means feedback during the attempt, before you’ve committed. The language researchers’ immediate means correction just after you’ve produced the sentence — you’ve already committed; the ink is dry. Two clocks, one word. No wonder the books seemed to fight.

And the language-learning research backs the instinct on its own terms. It even has a name for the far edge of the window: an error proceduralises — sets into an automatic habit — and classroom studies find that correcting a learner mid-task, before that happens, beats waiting until later. Not borrowed from cycling or tennis. Measured on people doing the actual thing.

So “how fast should feedback be?” turns out to be the wrong question. The right one is: what did the learner get to do first? Feedback should arrive after the learner has committed, and before the error settles into habit. Everything else — how often, how detailed, one correction or five — follows the same rule. Good feedback waits long enough for the learner to own the attempt — but not so long that the error becomes the lesson.

It’s worth being honest that when researchers went hunting for a single best delay — correct now? in five minutes? tomorrow? — they couldn’t pin one down; the answer kept shifting with what the learner was doing. Which is exactly what you’d expect if the target was never a fixed delay, but a window that opens and closes around the attempt.

As a teacher, that’s a genuinely uncomfortable discipline, because every instinct I have is to fix the mistake the instant I hear it. But the most useful thing I can do is usually to wait: let the learner finish the sentence, give them a beat to catch it themselves, and then offer one clear correction. Not five. Not a lecture. Just enough to help them hear the gap.

Building the window yourself

Take the teacher away and this gets harder, not easier. Studying a language alone is the most wicked environment there is: nobody corrects you, the native speaker nods and smiles, and your mistakes set like concrete while you feel fluent. So the teacher’s discipline doesn’t vanish — it becomes a protocol you run on yourself. You stop being the learner who gets corrected and become the architect of your own window. Four moves, and a tool for each.

Commit before you peek. The solo training wheels is the glance — checking the translation before you’ve tried to say it, flipping the Anki card before you’ve really retrieved. That collapses the window to zero. Produce first, reveal second. Anki and spaced-repetition decks are this move in a box — as long as you genuinely retrieve before you flip.

Get your delay from a recording. You can’t monitor your own pronunciation while you speak — that’s the concurrent-feedback trap. So split performing from reviewing: record a voice memo, then listen back a few minutes later, inside the window. It’s the one deliberate-practice move that survives with no coach in the room.

Engineer a check that hands back one thing. An LLM tutor can stand in for the coach — but prompt it to behave like a good one: make you attempt first, then give a single correction, not fifteen. Graded readers and dictation practice do this too — you predict the meaning or the spelling, then the next line or the transcript confirms.

Import an outside ear for what you can’t hear. The honest ceiling: a Thai tone you genuinely can’t hear, you can’t self-correct — the error is invisible to you. So book the occasional tutor or language-exchange session — not for constant correction, but to drag the invisible mistakes into view, where your own monitoring can finally work on them.

References

Agarwal, P. K., & Bain, P. M. (2019). Powerful Teaching: Unleash the Science of Learning. Jossey-Bass. (Source of the hypercorrection passage quoted above.)

Brown, P. C., Roediger, H. L., & McDaniel, M. A. (2014). Make It Stick: The Science of Successful Learning. Harvard University Press.

Butterfield, B., & Metcalfe, J. (2001). Errors committed with high confidence are hypercorrected. Journal of Experimental Psychology: Learning, Memory, & Cognition, 27(6), 1491–1494.

Fu, M., & Li, S. (2022). The effects of immediate and delayed corrective feedback on L2 development. Studies in Second Language Acquisition, 44(1), 2–34.

Hogarth, R. M., Lejarraga, T., & Soyer, E. (2015). The two settings of kind and wicked learning environments. Current Directions in Psychological Science, 24(5), 379–385.

Li, S. (2010). The effectiveness of corrective feedback in SLA: A meta-analysis. Language Learning, 60(2), 309–365.

Lyster, R., & Saito, K. (2010). Oral feedback in classroom SLA: A meta-analysis. Studies in Second Language Acquisition, 32(2), 265–302.

Schmidt, R. A., & Wrisberg, C. A. Motor Learning and Performance. Human Kinetics. (Source of the “concurrent visual feedback” finding, as cited in Young’s essay below.)

Xu, M., & Zeng, S. (2023). Optimal timing of treatment for errors in second language learning: A systematic review of corrective feedback timing. Frontiers in Psychology, 14, 1026174.

Young, S. (2019). Ultralearning. HarperBusiness — and his essay The Science of Learning Movement Skills.