Skip to content
Kioku

Spaced repetition & FSRS

Spaced repetition schedules each review for the moment just before you would forget, so you spend the least effort for the most retention. Kioku schedules Items with FSRS, the Free Spaced Repetition Scheduler.

This page explains the ideas behind that scheduling. It covers Items — flashcards graded with a Rating. Topics and Extracts are reading material and are paced separately, not by FSRS; see Incremental reading for how those work.

The forgetting curve

Section titled “The forgetting curve”

Memory decays. After you learn something, the probability that you can recall it falls over time — slowly at first, then faster. Review it and that probability snaps back up, and crucially the decay afterward is slower than before: each successful recall makes the memory more durable.

Spaced repetition exploits this. Instead of reviewing on a fixed timetable, it waits until your recall probability has dropped to a target level, then prompts you. Review too early and you waste a review on something you still know. Review too late and you have to relearn it. The art is in predicting that “just before you forget” moment for each Item independently, because some facts stick and others slip.

Stability, difficulty, retrievability

Section titled “Stability, difficulty, retrievability”

FSRS predicts that moment by modeling each Item’s memory with three quantities. Kioku tracks all three.

  • Stability is how long the memory lasts — the time it takes for your recall probability to decay to your retention target. Higher stability means longer intervals. Each successful review increases stability.
  • Difficulty is how hard the Item is to retain, intrinsic to the material. A harder Item gains stability more slowly, so its intervals grow more cautiously.
  • Retrievability is the probability you could recall the Item right now. It starts high after a review and decays until the next one. A brand-new Item is treated as fully retrievable; stability and difficulty only become meaningful after its first real review.

When an Item comes due, FSRS takes its current memory state, applies your rating, and works out the new stability, difficulty, and next due date. Topics and Extracts are paced by their own approach rather than by this model.

Ratings and what they mean

Section titled “Ratings and what they mean”

After you reveal an Item’s answer, you grade your recall with one of four Ratings. Each feeds FSRS and produces a different next interval:

  • 1 = Again — you failed to recall it. The Item lapses, its stability drops, and it comes back soon.
  • 2 = Hard — you recalled it, but with effort. The interval grows, but less than for Good.
  • 3 = Good — you recalled it cleanly. This is the expected rating and the baseline for interval growth.
  • 4 = Easy — it was effortless. The interval grows more aggressively, on the assumption the Item is more stable than Good would imply.

Before you answer, Kioku previews the interval each rating would produce, so you can see the consequence of your grade. Ratings of Good or Easy count as a hit; Again or Hard count as a miss in your session accuracy. For how grading fits into a session, see Review your queue and The review queue.

Choosing a retention target

Section titled “Choosing a retention target”

The Desired retention setting is the one dial that shifts FSRS’s behavior across every Item at once. It is the recall probability you target at the moment an Item comes due — the level retrievability is allowed to decay to before you see the Item again.

  • Higher retention (closer to 99%) schedules Items sooner. You remember more, but you review more often.
  • Lower retention (down to 70%) lengthens intervals. You review less, at the cost of more lapses.

The default is 90%, a balance most users never need to change. Changing it doesn’t rewrite the due dates already stored on Items you’ve learned. Instead, the new target takes effect the next time each of those Items comes due, when FSRS recomputes its interval under the updated setting. For how to change it and what to expect, see Tune FSRS retention.

Kioku can also tune FSRS to your own review history, fitting a Collection-specific set of parameters. It keeps the result only if it predicts your recall better than your current parameters — otherwise your parameters are left unchanged. Settings also show you how well-tuned the scheduler is by comparing predicted recall against what actually happened. Both live in settings and are covered in Tune FSRS retention.

Section titled “Related”