How spaced repetition works
You looked the word up on Monday, used it once, and by Thursday it is gone. That is not laziness and not a bad memory. Forgetting has that shape for everyone, and a schedule cut to fit the shape is still the cheapest way to get ahead of it.
Why memory lets go of what you learned
A word you have just met is not stored, it is held for a while. Nothing is wrong with it on the first day: it comes up instantly, you can even drop it into a sentence. What changes is not whether the word is there but what it costs to reach it. Every day without the word the reach takes a little longer, until one day the hand comes back empty.
The temptation is to run through the list again, just in case. It feels like work and it does almost nothing. You reread what you already know perfectly, you spend attention and you do not move the memory: there was no effort, and effort is the part that holds. Meanwhile the twenty words that were genuinely at the edge got the same three seconds as the two hundred you already knew.
Hence the familiar feeling that the list gets heavier every week while the vocabulary does not. The work grows with the length of the list, not with what is actually about to go.
The whole trick is the day, not the number of passes
Spaced repetition does exactly one thing: it picks the day a word comes back. Everything else, from the cards to the dictation and the examples, is packaging around that decision.
Bring the word back too early and you get the useless review from above: it is still fresh, there is nothing to retrieve, nothing gets stronger. Bring it back too late and it is gone, so you pay the full price, the same as for a new word. Between those two misses there is a window where the word can still be pulled out, but only just, with a real pause and a small effort. One review that lands in that window is worth ten lazy ones.
So the algorithm keeps a single number per word: the day that word will be at the edge. It is not an average over all words; it is recomputed from your own answers about that one word.
Every interval is longer than the one before it
The first success buys little, a few days. The second buys more, the third more again. Answer that you know it a few times running and the same word stops coming back tomorrow: first in four days, then ten, then twenty five, then in a couple of months.
That growth is what makes the whole thing affordable. Words leave the daily queue without leaving your head: they are still on the schedule, just far apart. A deck of five hundred words does not mean five hundred reviews a day; most of it sits in the future and today brings the handful that is genuinely at risk. Get one wrong and it drops back to short intervals, which is right: your honest "no idea" is the only signal the schedule has.
Where this comes from
The forgetting curve is not a marketing picture. Hermann Ebbinghaus drew it in 1885 by testing himself for months on lists of nonsense syllables, picked so that neither meaning nor prior knowledge could help him. He measured how much survived after an hour, a day, a week, and got a curve that falls steeply first and then flattens out.
His numbers do not carry over to real vocabulary: a syllable with no meaning is the worst case, and a word you can picture holds far better. What does carry over is the shape of the curve and his second result, the one that matters here: every relearning made the next drop slower, and repeats spread out in time beat cramming them together. In more than a century that effect has been found again so many times that it is one of the most solid results in the psychology of memory.
Turning the curve into a schedule is arithmetic, and in most apps the arithmetic is still the same one: SM-2, written by Piotr Wozniak for SuperMemo in the late 1980s. It keeps an interval and an ease factor per word and multiplies them after every answer. Verba runs on it too. It is a heuristic from 1987, not a law of nature: newer models such as FSRS fit real review logs better, and the data schema here lets the algorithm be replaced without touching your history. What is proven is the principle, not any one constant in the formula.
What spaced repetition does not give you
It trains one skill: pulling a specific item out of memory on demand. For vocabulary that is the bottleneck, but a language does not reduce to it.
It will not make you fluent in speech. Speaking happens under time pressure with no hint on the screen, and the only practice for it is speaking. It will not teach you grammar either, at least not the part that matters: a rule is a system you apply to new sentences, not a list of items to recite, and endings drilled as cards produce someone who knows the table and still builds the sentence wrong. Listening is a third skill: a word you recognise on paper can go straight past you in fast speech.
There is honest small print too. A card installs recognition first, and from there to free use the road is long: knowing that "apskaita" means accounting is not the same as reaching for it in the middle of your own sentence. The schedule assumes you show up; skip three weeks and the queue that comes back is a wall, because every interval expired at once. And no algorithm decides which words are worth your time. Choose badly and you will remember perfectly, forever, words you will never need.
Where Verba fits in
Verba is the vocabulary notebook that runs this schedule for you. You drop in the words you actually met, yourself or through your AI bot, and every morning you get the short list of the ones that are due: ten words, three buttons, a couple of minutes. The translation, the example, the pronunciation and the story of the day come with them, so the word meets you inside a sentence and not only in a queue.