Somewhere in a drawer, or in a box in your parents' garage, there's proof that you once knew things you no longer know. Maybe it's a report card with an A in French. Maybe it's a notebook full of chemistry you could once do in your sleep. And there's a small, private embarrassment that goes with it. You spent years on that language and can't order a coffee in it now. You sat through all that calculus and couldn't take a derivative today if someone paid you. It feels like a verdict on you: on your discipline, maybe your intelligence, maybe the person you were trying to become.

But not everything went. You probably still remember a handful of words from that language with complete certainty. You can still recognize faces from a school photo you haven't looked at in fifteen years. Some knowledge disappeared and some of it settled in as if it had been carved into stone. Memory researchers have a name for that stone: permastore. Why some learning ends up there and the rest drains away is one of the most useful things you can know about your own mind.

The man who tested memory over fifty years

Most memory experiments last an afternoon, or a few weeks at most. In the early 1980s the psychologist Harry Bahrick did something much more ambitious. He tested hundreds of people on the Spanish they had learned in high school and college, and the time since they'd studied ranged from a few months to roughly fifty years.

He couldn't follow one group of people for half a century, so he did the next best thing. He compared people at different distances from their last class and statistically accounted for how much they had studied, how well they had done, and whether they had used Spanish since. His paper, published in 1984, had a title that sounds like science fiction: Semantic memory content in permastore: Fifty years of memory for Spanish learned in school.

The shape of the forgetting was what surprised people. Retention fell steeply over the first several years after the last course, roughly three to six. Then it mostly stopped falling. Whatever survived that early drop stayed remarkably stable for decades, often about thirty years, and only declined again late in life. This was true even for people who said they had barely used Spanish since school.

So the forgetting curve doesn't slope down forever. For a portion of what you learn, it flattens into a plateau. Bahrick called that plateau permastore.

What decides what survives

The question that matters is what predicted how much knowledge made it onto the plateau.

It wasn't how recently people had studied, because after the first few years recency barely mattered. It wasn't mainly how much they had used Spanish since, either. The strongest predictor was how thoroughly the material was learned in the first place. Bahrick measured that through things like the number of courses taken and the grades earned. People who took several courses over several years kept far more than people who took one.

This is where the shame from the first paragraph should start to ease. Taking more courses doesn't just mean more hours. It means meeting the same material again and again across a long stretch of time. A second-year course reuses first-year vocabulary. A third-year course reuses both. Each time you come back, you have partly forgotten, and each time you have to work to retrieve it. That pattern of learning, drifting away, and relearning, spread over years, is what seems to move knowledge into permastore.

The one-semester student didn't fail because they were less capable. Their knowledge just never got enough returns, spaced far enough apart, to leave the steep part of the curve.

The family experiment

Bahrick later tested this directly, with a study that had a handful of participants who were deeply invested. He, his wife, and their two adult daughters (all psychologists) spent years learning sets of foreign vocabulary under different schedules. Some word sets were reviewed every two weeks, some every month, and some every two months. Some got more total sessions and some got fewer.

The results, published in 1993, matched the permastore pattern. The longest gaps between sessions produced the most durable memory years later. Reviewing at long intervals was efficient enough that fewer widely spaced sessions could hold up about as well as many sessions packed closer together. The spacing between returns did more work than the raw number of returns.

This cuts against intuition. When we want to really know something, our instinct is to see it more often. Over a span of years, it pays more to see it less often but at the right moments, after enough time has passed that retrieving it takes effort.

Why the plateau exists at all

Nobody fully understands how permastore works at the level of neurons. The behavioral science, though, fits a well-supported idea: memories differ in how well they're woven into everything else you know, and in how many times they've been successfully rebuilt from partial forgetting.

A fact you've met once hangs by a single thread. A fact you've retrieved across many years has been rebuilt many times, connected to other knowledge, and stabilized again after each retrieval. It stops being an isolated item and becomes part of how you think. That's why you can still produce gracias or the Pythagorean theorem without effort while the subjunctive mood is gone. Some things got enough returns over enough time, and some didn't.

There's a second encouraging point. Even what looks lost usually isn't entirely gone. Since Ebbinghaus in the 1880s, researchers have documented savings: relearning something you once knew is faster than learning it from scratch, even when you can't recall any of it on a test. The French you think is gone is mostly inaccessible, not erased. That changes how you should feel about the drawer full of old report cards. It's an asset that has been sitting idle.

The real lesson: design for the plateau

Put together, the research reframes what learning is for. Most of us study for the next test, the next trip, the next meeting. We tune for how well we'll know it next week, and the cost shows up years later, when it's gone and we privately conclude we were never good at it.

Permastore suggests a different goal. Learning something isn't finished when you can recall it. It's finished when you've returned to it enough times, across enough time, that it stops leaving. That takes less total effort than you'd expect, as long as the effort is spread across months and years instead of crammed into a week.

The things you remember from childhood aren't random. They're the things life kept bringing back: your best friend's phone number, the lyrics you sang every summer, the faces you saw every school day for years. You never scheduled those reviews, but you got them. With the things that matter to you now, you can do it on purpose.

Your next moves

  • Run a cold test on one lapsed skill tonight. Choose a language, an instrument's theory, or a subject you once studied seriously. Without looking anything up, write down everything you still know in ten minutes. What you produce is your permastore, and it's usually more than you feared.
  • Make a small "return deck" from the gaps. Take 15–20 items you almost remembered, the ones on the tip of your tongue, and write each as a question on a card. Those items are where savings make relearning fastest.
  • Space your reviews with widening gaps, not a fixed routine. Review the deck tomorrow, then in about three days, then a week, then two weeks, then a month. Once something stays solid, let the gap grow. The family study suggests long gaps are where durability comes from.
  • Pick one thing you're learning now and set a three-year horizon. Write down what you want to still know in 2029. Then put a recurring calendar note to revisit it every couple of months, even after it feels mastered, because that's when most people stop.
  • Retrieve, don't reread. At every return, try to recall first and check afterward. Effortful retrieval after partial forgetting is the step that seems to move knowledge toward the plateau.

Where Recall fits in

Scheduling widening gaps by hand works until you have a few hundred items and a busy life. Recall was built for that problem. Its FSRS scheduler models when each card is about to slip and brings it back right before that point, so the spacing that Bahrick's family planned by hand happens automatically. You can import the old Anki or Quizlet decks you abandoned years ago, take advantage of savings, and review offline wherever you happen to be. If there's something you want to still know a decade from now, you can start building that at recall.lumenlabs.works.