How FSRS spaced repetition works in Lira

What happens when you tap a word while reading?

You're reading a text in Lira, in English, German, or another target language, and you hit a word you don't know. You tap it. Its translation shows up right there, in the context of the sentence it came from, without you having to leave your reading to open an external dictionary.

That single tap triggers a second, invisible action: the word gets added to your review queue. Lira stores the word, its translation, the original sentence, and schedules its first review. You don't have to do anything else, no list to maintain, no card to build by hand.

This mechanism changes what "learning a word" actually means. You don't learn it by seeing it translated once. You learn it by seeing it again several times, at intervals calculated to match the moment your memory is just starting to let it slip. That's where FSRS comes in.

How does a review session work?

When you open a review session, Lira shows you the words whose recall interval is due that day, not every word you've ever looked up. Some words come back daily, others once a month. That's not random. It's the direct result of how you rated them last time.

For each word, you first see the word alone, sometimes with its original sentence. You try to recall its meaning before revealing the translation. Once the answer shows, you rate yourself on four levels:

Again, you didn't remember the word at all. Hard, you remembered it but with effort or hesitation. Good, you remembered it without much effort. Easy, the answer came to you instantly, no doubt at all.

That rating isn't cosmetic. It's the single input FSRS uses to recalculate the next interval for that specific word. If you rate "again," the word comes back soon, often the very next day. If you rate "easy," the interval stretches out noticeably, sometimes by several weeks at once.

This is why some words in your vocabulary come back almost daily while others only show up once a month. It's not that Lira treats them differently by default. It's that your own memory, measured through your successive ratings, treats them differently.

What is FSRS, exactly?

FSRS stands for Free Spaced Repetition Scheduler. It's an open source algorithm built for one job: predicting, for a given word and a given user, the best moment to schedule the next review.

Before FSRS, most spaced repetition apps, including Anki for a long time, ran on an older algorithm called SM-2, built in the 1980s. SM-2 works with fairly simple fixed rules: a correct answer multiplies the interval by a set factor, a wrong answer shrinks it. That works, but it applies roughly the same logic to everyone and every word.

FSRS does something different. It models memory more precisely, factoring in how difficult a specific word is for you, the full history of your past reviews on that exact word, and how much time actually passed since you last saw it. Two people learning the same word can end up with different intervals, because their review history on that word is different.

The practical result, documented in the comparative benchmarks published on the public FSRS repository, is better accuracy: fewer wasted reviews on words you already know well, and better-timed reminders on fragile words, the ones you're about to forget without noticing.

You don't need to configure any of this yourself. Lira runs FSRS in the background for every word in your queue, silently adjusting each schedule as your ratings come in. There's no setting to tweak, no deck to organize by hand. The only input the algorithm needs from you is an honest rating at each review, which is exactly the part that's fully in your control.

The idea of memory stability, in plain terms

FSRS is built around one central idea: the stability of a word in your memory. Without getting into formulas, you can think of stability as a measure of how solid a word feels. A word with low stability is fragile, you're likely to forget it fast if nobody reminds you. A word with high stability is solid, you can wait a long time before reviewing it again without much risk of forgetting.

Every time you review a word successfully, its stability goes up. Every miss brings it back down. And the more stable a word is, the longer the interval before the next review stretches, because the short-term risk of forgetting it is low.

This mechanism echoes an old observation in memory psychology known as the forgetting curve, described by psychologist Hermann Ebbinghaus in the late 19th century. Ebbinghaus showed that forgetting follows a predictable slope after learning, fast at first, then slower and slower. FSRS builds on that same principle, but applies it word by word, person by person, using real data instead of one theoretical average curve.

You don't need to understand the math to benefit from it. What matters is that your rating at every review directly feeds that stability calculation, word by word.

Why honest ratings change everything

This is the easiest thing to overlook and the most important for the system to work. If you rate "easy" out of laziness, to get through your session faster, or because you vaguely recognized the word without really recalling it, you're feeding FSRS false information about that word's real stability in your memory.

The practical result: the interval before the next review stretches out while the word is still fragile. When it comes back weeks later, you've forgotten it. You have to relearn it almost from scratch, which actually takes longer than if you'd honestly rated it "hard" or "again" in the first place.

This is overconfidence bias applied to memory: you think you remember a word because you vaguely recognize it, but passive recognition and active recall are two different skills. You can recognize a word inside a sentence without being able to produce it yourself in a conversation or in writing.

Rating "hard" when it genuinely was hard isn't a failure. It's exactly the information FSRS needs to bring that word back at the right moment, before you fully forget it rather than after. On the flip side, under-rating out of excess caution, marking "again" by default when you actually knew the answer, floods your daily session with words you've already mastered and wastes time on unnecessary reviews.

The simple rule to keep in mind: rate what actually happened in your head while the translation was still hidden, not what you wish you'd known.

Where the crossword fits into memorization

Spaced repetition in Lira mostly trains recognition and recall: you see a word, you try to recall its meaning. The crossword generated from your vocabulary adds a different layer, active recall through production.

In a crossword, you don't see the word, you have to reconstruct it letter by letter from a clue. That's a more demanding exercise than simple recognition, and that's exactly what makes it complementary. Memory research has long distinguished passive recognition, recognizing a word you're shown, from active recall, producing a word from nothing. The second reinforces memorization more durably than the first.

Since the grid is generated from words you've already met while reading and put into review, every game stays inside your real vocabulary range, neither too easy nor beside the point. It's an occasional complement to your daily review sessions, not a replacement for them. Playing a crossword doesn't change a word's FSRS schedule directly, but it strengthens the underlying memory that schedule is trying to track, which tends to show up as fewer "again" ratings the next time that word comes up for review.

FAQ

What happens if I don't review for several days?

Nothing gets lost right away. Words whose interval had already come due simply pile up in your review queue, waiting for you to come back. That said, the longer you wait, the higher the chance those words are genuinely forgotten, and you'll likely rate them "again" more often when you return. FSRS then recalculates the following intervals accounting for that gap, it doesn't start from zero.

Can I undo a rating if I tapped the wrong button?

Yes. If you tapped the wrong rating by mistake, you can go back within the current session and correct it. It's worth doing if the mistake actually changes your answer, an incorrect rating skews the stability calculation for that specific word and can push its next review interval off in an unjustified way.

Can I end up with too many words to review at once?

It can happen if you import a lot of new texts at once and tap on many unfamiliar words in a short time. In that case, it's better to spread your sessions across several days rather than clearing everything in one sitting. Attention drops during long sessions and ratings become less reliable, which hurts FSRS's accuracy over time.

To learn more about the reading method behind Lira, check our article on comprehensible input. For any other question about the app, see our FAQ or contact us directly.