FSRS, explained.

What FSRS does

You forget a new word a little more every day, until you meet it again. Spaced repetition brings each flashcard back right before you’d forget it, and each review makes the word last longer before the next one. FSRS is the part that decides when “right before you’d forget it” is, separately for every flashcard.

To do that, it keeps three numbers for each flashcard and updates them every time you review it.

• D

  Difficulty

  How hard this flashcard is for you, from 1 to 10. It goes up each time you miss it.

• S

  Stability

  How many days it takes for your chance of remembering to fall to 90%. Each review you get right raises it.

• R

  Retrievability

  Your chance of remembering it right now. Close to 100% just after a review, then falling day by day.

The forgetting curve

After a review, you’re all but certain to remember a word. Then the chance falls, quickly at first and then more and more slowly. FSRS draws that curve for each flashcard from its stability: the higher the stability, the flatter the curve.

The review goes where the curve reaches your desired retention. At the usual 90%, a flashcard with a stability of 12 days comes back after about 12 days; ask for 95% and it comes back after about 5. Get it right, and its stability grows, so the next gap is longer.

What a review changes

That second point on the left is why timing matters. A review just before you’d forget does more for a word than one the day after you learned it, so FSRS doesn’t waste your time with reviews you don’t need yet.

Desired retention

Desired retention is the chance of remembering that FSRS aims for when a flashcard comes back. Most people use 90%: about nine reviews in ten go right, and the tenth tells FSRS the word needs more time.

• Higher, like 95%: you forget less, but every word comes back much more often, so reviews take far longer.

• Lower, like 80%: fewer reviews, but you’ll fail more of them, and failed words start over on short gaps.

Misu keeps it at 90% for everyone. You don’t need to tune it.

Why it beats older schedulers

Older schedulers, like SM-2 from the 1980s, stretch each flashcard’s gap by a multiplier that shifts a little with each answer. They don’t estimate your chance of remembering, so they can’t aim for one. FSRS does, and it shows: in the open-source FSRS benchmark, built on the review histories of about 10,000 learners, FSRS-6 predicted recall better than SM-2 for 99.5% of them.

Better predictions mean better timing. Simulations by FSRS’s makers put the saving at 20 to 30% fewer reviews for the same retention.

Tuned to your memory

FSRS-6, the version Misu runs, shapes its curves with 21 numbers, its parameters: how fast stability grows, how much a missed review costs, how steep the forgetting curve is. Everyone starts with the same defaults. Your own review history shows how your memory actually behaves, and fitting the parameters to it, called optimising, makes your schedule more accurate.

In Misu that happens on its own, once a month. Misu fits FSRS-6 to all your reviews, counting recent ones more, and keeps the new parameters only if they predict your reviews better than the ones you have. Your next reviews stay on the days they were planned; the new parameters shape the gaps after them.

How Misu uses FSRS

• Every flashcard, every language. Ready-made courses and the flashcards you make yourself are scheduled the same way.

• Two answers. Swipe right if you knew it, left if you didn’t. That’s the whole review.

• Desired retention at 90%. The usual choice, kept for you.

• Fitted to you every month. With nothing to set up, and only when it makes your schedule more accurate.