Spaced Repetition, Explained: Why Your Words Come Back at 3, 11 and 35 Days

You learn a word on Monday. On Tuesday you still know it. By the weekend it is a vague shape. That is not a failing of your memory. That is how memory works, and once you accept it you can use it.

Spaced repetition is the technique of showing you a word again just before you would have forgotten it. Done right, each review makes the memory last longer than the one before, so the reviews get rarer while the word gets more permanent. This article explains how it works, how the scheduler in Lexer decides when a word comes back, and why “3 days, 11 days, 35 days” is not an arbitrary sequence.

The forgetting curve

In the 1880s the psychologist Hermann Ebbinghaus memorised lists of nonsense syllables and tested himself at intervals. He found that memory decays steeply at first and then flattens: you lose most of a new memory within days, and what survives the first week survives much longer.

He also found the remedy. Reviewing a memory resets the curve, and each reset makes the next decline slower. Review at the right moment and a memory that would have lasted days now lasts weeks; review again and it lasts months.

Two things follow:

• Too early is wasted. Reviewing a word you still know well teaches you almost nothing. The memory was not in danger.
• Too late is wasted differently. If the word is already gone, the review is really a first encounter again. You pay the full price twice.

The sweet spot is the moment the memory is about to fail but has not yet. A good scheduler is a machine for finding that moment.

From a fixed schedule to a model of your memory

The earliest spaced-repetition software (the SM-2 algorithm from the late 1980s) used a simple rule: each successful review multiplied the interval by a fixed factor. It worked, and it still powers a lot of flashcard apps. But it treats every card the same way at the start and adjusts slowly.

Lexer uses FSRS (Free Spaced Repetition Scheduler), a newer, open algorithm that models two things for every word:

• Stability: how long the memory is expected to last before it drops below a target recall probability. Measured in days.
• Difficulty: how hard this particular word is for you, which affects how much each review helps.

After every answer the model updates both numbers and schedules the next review for the day your probability of recalling the word is predicted to fall to about 90%. Easy words race ahead; hard words come back sooner. Nothing is on a calendar; everything is on your memory.

3 days, 11 days, 35 days

Here is what the schedule looks like for a word you get right every time, which is what the real scheduler in Lexer computes for a card with no history.

Review	When	What happened
New word	Day 0	You meet it in its news sentence
1st review	Day 3	Correct. Stability grows.
2nd review	Day 14 (+11)	Correct. Stability grows faster.
3rd review	Day 49 (+35)	Correct. The word is now counted as learned.

Notice that the gaps are not doubling. 3 → 11 → 35 is roughly tripling, because each successful recall of a word that was about to fade is a strong signal that the memory is consolidating. The scheduler trusts you more each time.

If you get a word wrong, the opposite happens. The interval collapses, the word comes back within minutes in the same session, and the model notes that this word is harder than it looked. Difficulty goes up, future intervals grow more cautiously, and the word gets the extra attention it needs.

What “learned” means

Many apps call a word learned the moment you tap “I know this”. That is a feeling, not a measurement.

In Lexer a word counts as learned after three correct reviews on separate days. By then it has survived three forgetting curves of increasing length, which is strong evidence it is in long-term memory. It leaves your active review deck, and your progress screen counts it. A word you mark as “I already know this” skips the deck entirely, because reviewing what you already know is the waste we are trying to avoid.

How reviews are asked

Recognition is easier than recall, and recall is what you need when you speak or write. So Lexer mixes the question types:

• Which word fits this definition? Four options, one definition. Fast, good for early reviews.
• What does this word mean? The reverse direction.
• Which word fills the blank? The word’s own news sentence with the word removed. This is the hardest and the most useful: you retrieve the word from meaning and context at once.
• Self-graded recall. You see the word, think of the meaning, flip, and grade yourself: Couldn’t recall · Hard · Good · Easy. Each button shows the interval it would produce, so you can see the schedule you are choosing.

A correct multiple-choice answer counts as a Good grade. Honest self-grading matters more than the grade itself: telling the scheduler “Easy” when it was hard just moves the word out of reach until it is forgotten.

Making it a habit

Spaced repetition only works if the reviews happen when they are due. Three things help.

Keep sessions short. Thirty-seven words due is about twelve minutes. A review session is a coffee, not a class.

Let the due count guide you, not the calendar. Some days nothing is due. Some days a lot is. Both are the system working.

Use the reminder, not the guilt. Lexer sends at most one notification a day, and only on days you have not studied yet. It asks about a word (“What did resilient mean?”) rather than nagging about a streak, because recalling one word is the review.

Spaced repetition is the second half of learning a word

The first half is meeting the word somewhere real, in a sentence that gives it meaning. That is why we built Lexer around words from real news. The second half is making sure you still have it next month. That is what the scheduler is for.

You can try both halves for free on iPhone and iPad: five words a day, reviewed exactly when they need to be.