How SayLocal decides when to show you a word again

You meet a word, you get it, and a day later it is gone. That is not a failure of effort, it is how memory works. The question is not whether you forget. It is when the app should bring a word back so you do not. SayLocal answers that with a scheduler called FSRS. Here is what it is doing and why.

Forgetting is the default

Memory of a single study session decays quickly. Hermann Ebbinghaus mapped that curve in the 1880s, and the basic shape has held up ever since: without review, recall drops steeply at first and keeps sliding. Cramming all your practice into one sitting fights that curve in the worst possible way, because everything decays from the same moment.

The fix is spacing. In a synthesis of 184 articles and more than 300 experiments, Cepeda and colleagues found that spreading the same amount of study across time produces substantially better recall than massing it.^[1] A later study pinned down the timing: the best gap is not fixed, it grows as the delay before you need the material grows.^[4] Review too soon and you waste a rep on something you still know. Review too late and you have already forgotten it. The win is in the band between.

What FSRS actually schedules

SayLocal uses FSRS, an open scheduling algorithm built on a simple model of memory with three moving parts for every card: how hard the item is for you, how stable the memory currently is, and how likely you are to recall it right now. After each review the model updates those numbers and picks the next date so the word resurfaces just as your odds of recalling it dip toward a target, rather than long after you have lost it.

That target is a dial, not a guess. Asking for higher retention means shorter gaps and more reviews; accepting slightly lower retention means longer gaps and less daily work. FSRS lets the schedule sit wherever that trade-off makes sense, instead of using one fixed ladder for everyone.

The goal is to see a word again right as you are about to forget it, not a week too early and not a week too late.

Why the gaps grow

Each time you recall a word successfully, the memory gets more stable, so the next review can wait longer. A word might come back after a day, then three days, then a week, then a month. The intervals expand because the memory can carry the extra distance. The idea is old: Paul Pimsleur proposed a graduated schedule of ever-widening recall intervals back in 1967.^[6] A lapse does the reverse: miss a word and it drops back into shorter, tighter spacing until it firms up again.

Why we ask you to rate yourself honestly

FSRS updates from how the recall actually felt, which is why reviews ask you to grade yourself rather than just flip the card. That grade is doing real work: it tells the model whether to stretch the next interval or pull it back in. There is a temptation to rate everything as easy to clear the queue faster, but that quietly breaks the schedule and the words stop sticking.

It also helps that a review is a test, not a re-read. Roediger and Karpicke showed that retrieving information from memory produces far better long-term retention than studying it again the same number of times.^[2] A follow-up using foreign-language vocabulary found the same thing in our exact setting: once you can produce a word, repeated testing keeps it, while repeated studying does almost nothing.^[5] The small struggle of pulling a word out before you flip the card is not friction to remove; it is the part that works. That is the same desirable difficulty principle behind the i+1 design: effort that feels hard in the moment but builds memory that lasts.^[3]

What this means for you

You do not have to plan any of this. Show up, review what is due, and rate yourself honestly. The words you find easy will drift to the back and barely interrupt you; the ones you keep missing will come back often until they do not. The schedule is quietly bending the forgetting curve in your favour, one review at a time.