Spaced repetition is one idea, and it is smaller than the jargon around it suggests: you remember something for far longer if you are made to recall it just as you are about to forget it.
Everything else — the algorithms, the intervals, the ease factors — is machinery for working out when that moment is for each individual thing you are trying to remember.
Forgetting is predictable, which is the useful part
Learn a new word today and you will not lose it evenly. You lose most of what you are going to lose quickly, and then the decline flattens. Hermann Ebbinghaus mapped this in the 1880s by memorising nonsense syllables on himself, and the shape has held up remarkably well for well over a century.
The important consequence is not that you forget. It is that forgetting has a shape, so it can be anticipated.
Ebbinghaus' retention curve, R = e^(−t/S), plotted from the formula. The lower line is a word met once. The upper one is the same word recalled on days 1, 3, 8 and 18 — each recall returns it to full and makes the next fall slower. Illustrative: the shape has replicated for over a century, the exact percentages depend on the person and the material.
The counterintuitive bit: difficulty is the point
Reviewing a word ten minutes after learning it feels productive. It is almost worthless. The word is still sitting in your head; retrieving it takes no effort; nothing is strengthened.
Reviewing it three days later, when you have to genuinely dig — pause, almost lose it, and then have it arrive — does far more. The struggle is not an unfortunate side effect. Cognitive scientists call this a desirable difficulty, and the ease of the recall is roughly inversely proportional to how much it strengthens the memory.
Leave it too long, though, and you have not reviewed anything — you are learning it again from scratch, at full cost. So there is a window, and it moves.
Where flashcards go wrong
Paper flashcards, and most apps that imitate them, have three problems that spacing alone does not fix.
They are reviewed on your schedule, not the word's
A box of cards gets flipped through when you have time. That means easy words get reviewed as often as hard ones, which wastes most of the session. The entire value of the technique is in reviewing each item at its own interval.
The cue does half the work for you
A card reading el turno — the shift with the answer immediately below is barely a retrieval task. If your eye reaches the answer before your memory does, you have practised reading, not recall.
An isolated word is a word you cannot use
Knowing that turno means shift does not tell you how to say you are on nights, or ask someone to swap. Words live in phrases, and a card carrying a sentence you would actually say teaches you the word and a structure at the same cost.
What a good implementation does
- Schedules each item separately, based on how you specifically have performed on it.
- Asks you to produce, not recognise. Prompt in English, answer out loud in the target language — not four options.
- Shortens the interval after a miss, rather than resetting to zero. You have not lost everything.
- Puts the word in a sentence you would use. Context is free and it roughly doubles what the card teaches.
Modern schedulers — SM-2 in older software, FSRS in current — differ in how they model your forgetting, but they are all solving that same problem: predict the last day this will still stick, and ask then.
The failure mode nobody warns you about
It is possible to build a deck you cannot maintain. Add fifty words a day for a month and the reviews compound into an hour-long daily obligation that eventually collapses — taking the habit with it.
Ten to fifteen genuinely new words a day is sustainable almost indefinitely and gets you past three thousand words in a year, which is more than B1 needs. The instinct to add more is the single most common way people quit.
What those word counts mean in levels and months.
How Arbor uses it
Review comes first in the daily plan, before anything new, because a word reviewed on the right day is cheap and the same word reviewed a week late has to be learned again.
The cards are not typed by you. Words you reach for during a conversation — including the ones you reached for in English and could not find — are added automatically, in a sentence from that conversation. And the counting is done in word families rather than forms, so jestem and być are one word rather than two, and the number you see is not quietly inflated.
Arbor builds this into the app.
Every lesson, conversation and review session is made from your own job, your own week and the people you are learning for. Spanish and Polish are ready today.