Flashcard Design
Flashcard design is the craft of deciding what goes on a card and how it is phrased — distinct from, and upstream of, the scheduler that decides when to show it (see fsrs). A spaced-repetition system only optimizes the timing of retrieval; it cannot rescue a badly formulated card. The canonical statement of the craft is Piotr Wozniak’s “Twenty rules of formulating knowledge” (SuperMemo, 1999), written from long experience of what makes items retainable versus chronically forgotten.
The load-bearing rules
Wozniak lists twenty rules in order of importance; the early ones carry most of the weight and are the most often violated. 1
- Do not learn what you do not understand. Cramming material you cannot yet place is astronomically slow and produces inert knowledge. Comprehension precedes memorization.
- Learn before you memorize. Build the overall picture first, then break it into items. Isolated facts with no coherent structure are the hardest to retain.
- Minimum information principle. Formulate items as simply as possible. Simple material is processed the same way on every repetition (one consistent path through memory); complex items force a different path each time and the traces interfere. If a complex item keeps being forgotten, split it into atomic sub-items — more items, but each scheduled at its own pace, which is cheaper than re-forgetting the composite.
- Answers should be short. The goal is the minimum information to retrieve in one repetition. Questions short, answers shorter.
- Avoid sets, and avoid enumerations where possible. Unordered sets are nearly un-memorizable past ~5 members; convert them to ordered enumerations, and enumerations to overlapping cloze deletions.
- Combat interference. The single greatest cause of forgetting in a mature collection is items that are too similar to other items. Make items unambiguous and kill interference the moment it appears.
Recognition vs. production
For paired material (vocabulary, translations, definitions), Wozniak’s rule 17 notes that active recall of the foreign term does not guarantee passive recognition of it, and vice versa — so both directions can be worth a card each without violating the minimum information principle. This is the basis of the common recognition card (prompt in the target form, answer in the familiar one) versus production card (prompt in the familiar form, answer in the target one) split. Production is harder and lower-volume; recognition is the high-frequency workhorse. 2
Ambiguous fronts and sense markers
When two genuinely different lexemes share a surface form, the deck acquires an interference pair that no scheduling can fix. A Japanese vocabulary deck hit the textbook case: 本·ほん as a noun (“book”) and 本·ほん as the counter for long thin things (see japanese-counter-words) produced identical recognition fronts, and one entry was nearly merged away as a data-entry duplicate before the collision was recognized as the language’s own homography. Both senses are real words, so the repair is prompt-side: a sense marker on the front (# 本 for the noun, # 🔢 本 for the counter) makes each question unambiguous while leaving the tested content unchanged. Marking the whole class — every counter card carries the marker — beats marking only the colliding pairs, because a partial convention leaves the learner remembering which words were marked. This is the combat-interference rule applied to ambiguity the deck author did not create: when the language supplies the collision, disambiguate the prompt rather than delete a sense. 3
Cue leakage
A card’s prompt must not contain the surface form of its own answer. If the front of a conjugation-drill card literally shows the target suffix, or a definition front embeds an example that names the thing being defined, the retrieval is cued rather than recalled — the grading becomes meaningless and the scheduler’s signal degrades. Prompts should name what is being asked (an abstract label, a direction) while requiring the learner to produce the form. This is the card-design counterpart of the testing effect: the card only trains memory if the answer is genuinely generated, not read off the prompt. The inverse failure — a prompt that never states what is being asked — and the scoring rules that keep a grade honest live in assessment-integrity.
What flashcards are for
Wozniak’s rules converge on a boundary that practitioners repeatedly rediscover: flashcards train recognition and recall of atomic units — a word, a fact, a conjugation, a set phrase. They are poor at production-and-composition skills (constructing a novel sentence, solving a multi-step problem, writing), because those require holding and combining several pieces at once, which is a different cognitive task from the single-path retrieval the minimum information principle optimizes. Sentence-level and generative skills belong in live practice; the cards feed it by keeping the atomic units available.
Relation to the scheduler
Card design and scheduling are separable concerns that meet at the review event. A well-formed card gives the scheduler (see fsrs) a clean recall/lapse signal; a leaky or compound card gives it noise. Wozniak assumes a spaced-repetition system underneath the rules — the rules are about making each item worth scheduling. The distributed-practice evidence for why scheduling helps at all is in spaced-repetition.
Related pages
- fsrs — the scheduling algorithm that times the reviews these cards feed
- spaced-repetition — the distributed-practice and retrieval-practice evidence underlying the whole approach
- assessment-integrity — the item-and-scoring side beyond the card: question concealment, copy-vs-recall grading, assisted-pass labeling
- semantic-interference-in-lexical-selection — same-category competition during word production; a live-production phenomenon distinct from ambiguous card prompts
- japanese-transitive-intransitive-verbs — contrast complete case frames when related verb forms collapse in English translation
- japanese-location-particles — test location phrases with their verbs rather than isolated one-word glosses
Sources
- Wozniak, P. (1999). “Effective learning: Twenty rules of formulating knowledge.” SuperMemo. Piotr Wozniak 1999 — Effective learning: Twenty rules of formulating knowledge
- Wikipedia contributors. “Japanese counter word.” Wikipedia contributors — Japanese counter word