Spaced repetition, and the reading that feeds it
Two old, unglamorous ideas. One is a schedule that asks you about a word just before you lose it. The other is a book that gave you the word in the first place. Most of what makes vocabulary stay put is the pair of them, and each one is weaker alone than people expect.
What the schedule is for
Memory does not fade at a steady rate. Hermann Ebbinghaus, testing himself on nonsense syllables in the 1880s, lost most of what he would lose in the first day; after that the curve flattened out and the remainder stayed with him a long time. His syllables were not German nouns and his numbers do not transfer to them, but the shape has kept turning up in the century of work since, and everyone who has crammed for an exam knows it from the inside.
The useful part is what happens when you interrupt it. Recall a word successfully and the memory comes back more durable than it went in: the word now survives days where it used to survive hours, and after the next successful recall, weeks. A review is not a top-up of something draining away. It is a step up a ladder, and the step only counts if you actually climbed it.
That last condition matters more than it sounds. Re-reading a word and its translation teaches you very little. Producing the translation from memory, before you see it, is the thing that moves the curve. It also feels worse, which is why so many people quietly avoid it: the app puts it plainly on the review screen — recalling a word is harder than recognising it, and that difficulty is the mechanism, not a flaw in it.
Why the gaps widen
If reviews worked best packed together, everybody would simply drill. They don't. The same number of repetitions spread over a longer period produces better long-term recall than the same repetitions crammed — one of the most replicated findings in the study of memory, and the reason every serious review system widens its intervals rather than repeating on a fixed drumbeat.
How far apart depends on how long you want the word to last, and the gap grows faster than instinct expects: keeping a word for a year does not take a slightly wider gap than keeping it for a month, it takes a several times wider one. The right distance is forgiving — close enough does the job — but it is nearly always further out than it feels like it should be, which is why a schedule that is working looks, from the outside, like neglect. With Folioki's default settings, a word you keep answering correctly comes back after about 2 days, then 11, then 46, then 163. Four short answers inside eight months — and then the word is not due again for another sixteen.
The corollary is the part people resist: seeing a word more often is usually wasted work. If you can still recall it easily, the review bought you almost nothing, and you paid a minute for it.
What a scheduler actually decides
A spaced repetition scheduler has one job: estimate the day your chance of recalling a word will have fallen to some chosen level, and ask you on that day. The chosen level is a setting, not a law of nature. In Folioki it runs from 80% to 95% and sits at 90% by default.
Ninety percent is a target rather than a promise: over a lot of reviews it means about one word in ten will be gone when it comes up. That is not the schedule failing; it is the schedule spending your time where it does the most good. Dragging the setting towards "remember better" is the counter-intuitive move: a few points of extra retention costs a great many more reviews, because you are asking to be tested while you still know everything. If you are going to move it at all, give the default a month first, and watch what the queue does before you decide which way to go.
What the scheduler cannot do is know whether you remembered. That information only exists in your head, and it reaches the algorithm through the four buttons you press — which is why how you answer a card decides more about your results than any setting on the screen.
FSRS, and what it replaced
The idea is older than any software: Sebastian Leitner's boxes, in 1972, moved a card one box along when you got it right and back to the front when you didn't. The first algorithm to put real numbers on it was SM-2, published with SuperMemo in 1987 and imitated by most flashcard software since. SM-2 multiplies the current interval by a per-card factor and nudges that factor up or down as you answer. It works. It also hands every learner the same ladder and adjusts it slowly.
FSRS — the Free Spaced Repetition Scheduler — models a card as three quantities instead: how hard it is for you, how stable the memory currently is, and how likely you are to recall it right now. The last of those falls away as the days pass, and the second decides how slowly — and both are fitted to the answers you have actually given rather than to an average learner. The algorithm is openly researched and openly implemented; Anki has shipped it since 2023, and Folioki schedules with it too. This is deliberately not a proprietary ingredient: a scheduler you cannot inspect is a scheduler you cannot trust with two years of your evenings.
Why a schedule alone stalls
Here is where most vocabulary apps stop, and where the results get disappointing. A schedule is a machine for keeping things you have learned. It does not decide what is worth learning, and it cannot supply the meaning.
Word lists teach one translation per word. Real texts do not cooperate. A word carries a register, a set of collocations, a shade of meaning that shifts with the sentence around it, and a list flattens all of that into a pair of words you will one day produce at exactly the wrong moment.
There is also an arithmetic problem. The figure reading research keeps landing on is that you need to know something like 98% of the running words on a page before it stops being work — about one unknown word in fifty. Getting there takes thousands of word families, and past the first few thousand, the words are precisely the ones no general list will cover: they belong to the book you happen to be reading. Drilling them out of context is the slowest way to acquire them. Meeting them in the novel that needed them is free.
Why reading alone also stalls
The mirror image is just as familiar. You meet a word, work it out from context, understand the sentence, read on — and by morning it has gone. Nothing was retrieved, because nothing asked you to retrieve it.
You might hope the book will bring the word back. For the commonest few thousand words it will. Beyond that, the distribution is brutal: most of the words you had to look up appear once or twice in the whole novel, and the second encounter, if it comes at all, comes hundreds of pages after you have forgotten the first. Reading hands you encounters at random. A schedule hands them to you at the moment they are worth something.
The two halves together
Put them in one place and each fixes the other's weakness. The book decides what you learn, and the decision is a good one: these are the words standing between you and the thing you actually wanted to read. The schedule decides when you see them again, and keeps what the reading would have lost.
The join between them is the sentence. A word saved out of the paragraph you met it in keeps a hook — a scene, a voice, a situation — and a card with a story behind it is answerable a month later in a way that der Anblick — sight never is. That is why every card Folioki writes carries the sentence you found the word in, without your having to copy it across.
In practice the loop is short. Read what you chose. Tap the sentence that stopped you, take the translation, ask for a breakdown if the grammar is the problem. Save the word worth keeping — the card arrives written, with the translation, an example, its forms and the sentence in place. Then answer a few minutes of reviews a day, and let the intervals stretch.
Defaults worth knowing: twenty new words a day, which is roughly what an hour of reading turns up; no cap at all on reviews, because a cap ends a session in the middle of a word and leaves the schedule lying to you; and a word that has defeated you eight times drops out of review with a flag, on the theory that eight failures means a bad card rather than a bad memory.
Where to go next
The other half of this is mechanical and takes five minutes to learn: how to answer a flashcard — what the four buttons mean, the two habits that quietly wreck a deck, and what makes a card you will still answer in a month.
Try it on a book you already own
Free to read with, on iPhone and iPad. Every word you save becomes a card that already carries the sentence you found it in.
