What Is Spaced Repetition? A Plain-English Guide

The forgetting curve

In the 1880s Hermann Ebbinghaus memorised thousands of nonsense syllables and tested himself at intervals to find out how fast he lost them. The shape he found has held up ever since: memory decays steeply at first and then levels off. Without review, most of what you learn today is gone within a week.

The useful half of the finding is what happens when you review. Each review resets the curve and makes the next decay slower. Review a word once and it might last three days. Review it again at the right moment and it lasts two weeks, then two months, then a year. The intervals grow roughly geometrically, which is why a mature vocabulary costs so little to maintain.

Why cramming feels better and works worse

If you study fifty words tonight, you will know them tonight. That feeling of fluency is real and it is also almost entirely short-term memory, which is why the same fifty words are largely gone by Friday.

Spaced repetition deliberately makes studying feel worse. It shows you a word at the moment you are about to forget it, so recall is effortful and you get some wrong. That difficulty is not a bug — the strain of retrieving something you nearly lost is precisely what makes the memory durable. Researchers call these "desirable difficulties," and the name is a fair warning about how they feel.

How a scheduler decides

Every spaced repetition system answers one question: given everything I know about how you have handled this card, when should I show it next?

The old answer, SM-2, dates from 1987 and is still what most people picture. Each card carries an "ease factor." Answer correctly and the interval multiplies by it; answer wrong and the card resets to the beginning. It is simple, it works, and it has two well-known failure modes — a few lapses can drive a card into "ease hell" where it never escapes short intervals, and a full reset throws away memory you demonstrably still had.

FSRS, the Free Spaced Repetition Scheduler, replaces that with an actual memory model. It tracks three things per card: retrievability (how likely you are to recall it right now), stability (how slowly it decays), and difficulty (how hard this particular card is for you). It then schedules the review for the day your predicted recall probability crosses a target — 90% by default. Because it models memory rather than applying a multiplier, a lapse adjusts the estimate instead of erasing it, and you get roughly the same retention for meaningfully fewer reviews.

What the grades actually mean

Most systems ask you to grade each answer on four levels, and most people use them wrong. What they are asking about is the effort of retrieval, not whether you got it right.

• Again — you did not recall it, or you produced the wrong word. Not a failure; it is information the scheduler needs.
• Hard — you got there, but it took real effort or several seconds of digging.
• Good — you recalled it with normal effort. This should be most of your answers.
• Easy — it was instant and effortless. Use it sparingly; over-using Easy pushes intervals out faster than your memory can support.

If almost everything is Good, the schedule is calibrated. If you are pressing Again constantly, you are taking on new cards faster than you can absorb them.

The three mistakes that break it

Spaced repetition fails for very few reasons, and they are all self-inflicted.

• Grading dishonestly. Pressing Good on a card you actually fumbled tells the scheduler your memory is stronger than it is. The interval goes out too far, and you lose the card a month later — by which point the cause is invisible.
• Skipping days and then bingeing. The whole model assumes reviews land near their due date. A 300-card backlog is not just unpleasant; every card in it has drifted past the point the schedule chose.
• Adding too much at once. Every new card generates a review chain that lasts months. Twenty new words a day feels ambitious in week one and becomes an hour-long daily obligation by week six. Ten is plenty.

What it will not do

Spaced repetition is a memory tool, not a language course. It will make sure you still know a word in a year. It will not teach you to hear fast speech, to order food without freezing, or to feel when the subjunctive is required.

Treat it as the maintenance layer underneath everything else you do — the thing that means the vocabulary from the podcast you listened to in March is still there in September. Paired with real input and real conversation it is close to unreasonably effective. On its own it produces someone who knows a great many words and cannot use them.