The correction that works is the fourth one

The first three corrections feel like progress

You write a sentence, something comes back marked, and you read the explanation. It makes sense. It might even feel obvious in hindsight. You accept it, you feel slightly better informed, and you move on. That whole sequence is pleasant and almost entirely useless on its own.

Two weeks later the same error is back. Nobody involved knows it's the same error. Not you, because the first correction was one sentence a long way back, and not whoever corrected you, because they're looking at today's sentence. So the explanation gets delivered again, from scratch, as if it were new information. It feels like being taught. It's closer to being reset.

Why the fourth one is different

By the fourth time, something has changed, and it isn't the explanation. It's the frame. Instead of a fact about a sentence, you're looking at a fact about yourself: this is a thing you do. That reframing is what makes the correction stick, because it moves the error from the category of accidents to the category of habits, and habits are the only thing you can actually set out to change.

It also changes what you do next. A one-off correction has no obvious follow-up. A pattern does: you now know a place to slow down, a construction to check before hitting send, a shape to watch for. Attention is the scarce resource in language learning, and a pattern tells you where to spend it.

The number four isn't sacred. The point is that a correction's value depends on the count attached to it, and the count is exactly what nobody keeps.

Why recurrence usually goes unnoticed

Repeated errors are hard to see from the inside for a straightforward reason: at the moment of writing, the wrong form is the one that feels right. That's what makes it a habit rather than a slip. You can't catch it by trying harder, because the trying happens after the choice.

They're also hard to see from the outside, because the outside is usually a different person each time, working from a single sentence with no history. Even a patient human tutor who sees you weekly is reconstructing your patterns from memory, and memory across months of sentences is not what human memory is good at.

So the recurrence lives in a blind spot that belongs to nobody. Both halves of the exchange are behaving reasonably, and the information still gets lost.

What has to be kept

Not the corrections themselves. A pile of corrected sentences is an archive, and nobody reads their own archive. What has to be kept is the grouping: which error this was, how many times it has happened, when it last happened, and what the last few instances looked like.

That's a small amount of information and it changes the exchange completely. A correction can arrive already knowing it's a repeat. Practice can point at the pattern with the highest count instead of whatever came up today. And the thing you have been waiting for, the moment where an error is genuinely gone, becomes visible: the count stops moving, and after enough writing without a recurrence, the pattern is finished.

That last part is worth more than it sounds. Learners rarely get to see a weakness close. Progress in a language is mostly felt as an absence, a mistake you don't make any more but never noticed stopping. Keeping count turns the absence into something you can look at, which is motivating in a way that no encouragement is.

The uncomfortable implication is that correction quality matters less than correction memory. A precise explanation delivered once and forgotten loses to a plain explanation attached to a tally. So if you're choosing where to spend effort on your own learning, spend it on making the count exist.