wordledger · vocabulary notes

Spacing, and why cramming feels better than it works

Practice that feels smooth is usually practice that is not doing very much. The uncomfortable kind is the kind that lasts.

The observation is old. In the 1880s Hermann Ebbinghaus, memorising nonsense syllables on himself with some patience and no assistants, described how retention falls away steeply at first and then flattens — and, more usefully, that each successful review flattens the curve that follows it. Everything below is a consequence of that second half.

recall time with review without review review review
Each successful recall does two things: it restores retention, and it reduces the slope of what follows. The second effect is the one worth designing around — it is why the gaps can keep growing.

Spacing

Given a fixed number of repetitions, spreading them out beats bunching them together. Ten minutes on a word today and ten minutes next week will leave more behind than twenty minutes this afternoon, and the gap widens the longer you wait before measuring.

The catch is that cramming feels better while you do it. Massed practice produces fluent, confident performance in the session, and that fluency is easily mistaken for learning. Spaced practice feels worse in the moment — you have half-forgotten things, you hesitate, you get some wrong — and that discomfort is roughly where the benefit comes from. Robert Bjork's phrase for this is desirable difficulty, and it is worth keeping in mind on the days when the session feels unproductive.

Recall, not recognition

The second effect is at least as large and much easier to get wrong in practice. Retrieving an item from memory strengthens it substantially more than seeing it again does. Rereading a list produces a strong sense of familiarity and very little durable memory; being asked and having to dig produces the opposite.

In practice this means the prompt has to make retrieval genuinely necessary. A card that shows the word and its translation together tests nothing. A card that shows a sentence with the word removed forces the work. The moment of blankness before the answer arrives is not a sign that the card is too hard — it is the entire mechanism.

If you can answer without pausing, the card is no longer teaching you anything. Push it further out rather than enjoying it.

A schedule that survives contact with a normal week

Any expanding sequence works; the exact numbers matter far less than the expansion. What I settled on, after some fiddling:

ReviewGapWhat it is doing
1stsame dayConfirms the word was encoded at all, not just seen.
2ndnext dayCatches the ones that looked learned and were not.
3rd3 daysFirst genuinely effortful recall for most items.
4th1 weekWhere about a third of my cards used to fail.
5th3 weeksSurvivors here are mostly permanent.
6th2 monthsMaintenance.
later6 months+Nearly free. Keep them; they cost almost nothing.

On a failure, drop back — not to the very beginning, but two steps. Restarting from zero every time a word slips is how decks turn into treadmills.

Two ways a queue quietly goes wrong

The backlog spiral

Miss four days and the queue does not grow by four days of work; it grows by four days of work plus everything those days pushed into the following ones. The queue then looks punishing, which makes it easier to skip again. The only reliable fix I found is a hard daily cap on new items — mine is ten — set low enough that a bad week is recoverable rather than a reason to quit.

Leeches

A small number of items fail over and over and consume a wildly disproportionate share of the session. It is tempting to treat these as a discipline problem. They are almost always a design problem: the prompt is ambiguous, the word has two meanings sharing one card, or it was learned alongside a near-synonym and the two now retrieve each other. After roughly eight failures, the card gets deleted and rewritten from a fresh sentence, or dropped entirely. Rewriting fixes most of them; deleting the rest costs less than you would think.

What I stopped doing