Why frequency decks get abandoned
A word you've never seen in context is an abstract pairing — a foreign string and an English string, with nothing between them. It's maximally arbitrary and therefore maximally hard to retain, and every review is a grind.
A word you met *yesterday, in a sentence, in something you cared about*, already has a hook. You remember where you saw it. Reviewing it is a completion, not an imposition, and the difference in dropout rate is enormous.
Sentence mining
- Read or watch something slightly too hard for you. Slightly. If you're looking up every third word, go easier.
- When you hit an unknown word, card the whole sentence, not the word. Context is what makes it stick, and it teaches you the grammar for free.
- One unknown per card. A sentence with three new words tests nothing cleanly — you'll fail it and not know which part you failed. This is the atomic card rule.
- Add audio if you can. A word you can read and not hear is a word you don't have.
Then let the scheduler do its job
Vocabulary is the purest spaced repetition problem that exists — no mechanism, no understanding to derive it from, just arbitrary pairings that must be retained. This is exactly what FSRS is for, and it will handle thousands of them without you thinking about scheduling once.