Minimum Information Principle: Why One Card = One Idea
Most people's flashcard problems aren't with the app or the schedule — they're with the cards themselves. Cards that hold too much information are the single biggest reason spaced repetition fails. Here's the principle that fixes it, why it works cognitively, and 8 before-and-after rewrites across real subjects.
Why dense cards silently break spaced repetition
Broken grading
You remember fact A on the card but forget fact B. Any rating you give is a lie. The algorithm schedules based on the lie — and both facts suffer for it.
Wrong intervals
A mixed "I remember some of it" rating produces a middle-ground interval. Fact A gets reviewed too soon (wasted time). Fact B gets reviewed too late (forgotten again).
Exhausting reviews
Reading a long, multi-part card before you can even attempt recall takes 10× longer per card. Sessions feel draining and you stop doing them.
False confidence
You recognise that something was on the card — but recognition isn't recall. Dense cards produce familiarity, not retrieval strength. The knowledge isn't actually there when you need it.
"The more you put on a single flashcard, the less you will remember from it."
Piotr Wozniak — creator of SuperMemo, the first spaced repetition software — listed the minimum information principle as Rule #1 of his 20 rules for formulating knowledge. Not rule #7 or #12: rule #1. Before everything else — before images, before mnemonics, before context — you must minimise the information per card.
The principle is simple: each card should test exactly one piece of knowledge. One question. One answer. One fact that can succeed or fail independently of everything else in your deck.
The atomicity test
A card is atomic when it cannot be meaningfully split further. The Greek root atomos means indivisible — and that's the target. An atomic card contains the smallest testable unit of the knowledge you're trying to build.
"Atomic" does not mean trivially simple. A card that asks "What causes the spacing effect at the synaptic level?" is atomic — it has one correct answer. A card that asks "Explain the spacing effect" is not atomic — it has no defined correct answer and could be answered with one sentence or twenty.
The practical test: could two people disagree on whether the back of this card correctly answers the front? If yes, the card is not atomic. A well-formed question has exactly one right answer that any competent reviewer would agree is correct.
Why small cards work better — what's happening in the brain
Working memory has a hard limit
George Miller's classic research established that working memory holds roughly 7 ± 2 items. A card with 4 facts simultaneously occupies most of working memory just for reading — leaving nothing for actual retrieval. An atomic card uses one slot for the question and one for the answer.
Retrieval requires a single cue
Memory retrieval works through cues — one piece of information activates a specific memory trace. A vague, multi-part question provides a vague cue, which activates a vague memory. A precise single-fact question provides a precise cue, which activates a precise memory trace — and reinforces it strongly.
Accurate signal = accurate scheduling
The SM-2 algorithm (and its successors) schedule your next review based on your rating. An accurate rating from an atomic card produces an accurate interval. An inaccurate rating from a mixed card produces a wrong interval — and errors compound over time.
Speed reinforces the habit
An atomic card takes 3–5 seconds per review. A dense card takes 20–40 seconds. A 50-card session takes 5 minutes with atomic cards — or 30 minutes with dense ones. The faster session is the one you'll actually complete every day.
- Read the front. Did it take more than 2 seconds? If yes — the question is too long or too complex. Simplify.
- Try to recall the answer. Is there exactly one correct answer? If you could give two or three reasonable answers, the question is too vague. Sharpen it.
- Check the back. Is there more than one sentence (excluding a usage example)? If yes — the back contains multiple facts. Split the card.
- Grade it. Could you give this card a clean 0 or 5? Or is it somewhere in between because "it depends"? If grading is ambiguous, the card is not atomic.
One dense card → multiple atomic cards
Example: Cellular respiration (Biology)
Front: What is cellular respiration?
Back: Process converting glucose + O₂ → ATP + CO₂ + H₂O. Three stages: glycolysis (cytoplasm), Krebs cycle (mitochondrial matrix), electron transport chain (inner mitochondrial membrane). Net yield: ~36–38 ATP per glucose.
Example: World War I (History)
Front: World War I
Back: 1914–1918. Started with assassination of Archduke Franz Ferdinand. Key alliances: Triple Entente vs Triple Alliance. Ended with Treaty of Versailles (1919). ~20 million dead.
Before and after: 8 subjects, 8 fixes
8 symptoms, 1 card. You'll remember 3 and feel like you "knew it."
Group the three polys (they share structure), then one card each for the remaining symptoms.
6 forms on one card. You'll know "je prends" and fail the rest.
One card per irregular form. Regular forms (je prends, tu prends, il prend) can share a card — they follow the same pattern.
10 methods on one card. You'll recognise all of them and produce none.
my_list.append(4) # [1, 2, 3, 4]
One card per method, with a concrete usage example on the back.
6 elements. A high-stakes exam card that will give you false confidence.
Card A tests the count and names. Cards B–G each test the definition of one element individually.
Five separate factual questions — each independently testable, independently schedulable.
Then a separate card for the relative minor, one for the interval pattern, one for the note sequence.
Card A names all five (count + overview). Cards B–F each ask: "Which of Porter's Five Forces concerns X?" — testing recall of each force's name and meaning individually.
One card per contribution. Each can be reviewed and retained on its own schedule.
Legitimate exceptions — and how to handle them
Pairs with inherent structure
If two facts always go together and testing one in isolation is meaningless — keep them. Example: latitude and longitude of a capital city are best tested together. The prefix "anti-" always means "against" — testing prefix and meaning together is atomic.
Chunked sequences
Experts chunk related information into single cognitive units. A medical student who has seen "polyuria, polydipsia, polyphagia" a hundred times can card all three "polys" together — they fire as one memory trace. Beginners should split them until the chunk becomes natural.
Mnemonic lists
A card that tests a full acronym (e.g., "HOMES — name the Great Lakes") is technically multi-part but is justified if the acronym itself is the learning target. The rule: the acronym is the atomic fact; the individual items are separate cards.
Example sentences
An example sentence on the back of a vocabulary card is not a violation — it's context, not a second fact. The tested fact is the word's meaning; the sentence illustrates it. Keep examples short (one sentence) and they're fine.
FAQ: minimum information principle
What is the minimum information principle?
A flashcard design rule from Piotr Wozniak (SuperMemo): each card should contain the smallest possible unit of testable knowledge — one question, one answer. Cards with multiple facts produce inaccurate ratings, wrong review intervals, and false confidence. Splitting them into atomic units fixes all three problems.
Why do multi-fact cards break spaced repetition?
When you partly know a card, you can't give it an accurate grade. A compromised grade produces a compromised interval — the known facts are over-reviewed (wasted time), and the unknown facts are under-reviewed (forgotten). Atomic cards produce clean, binary grades: you knew it or you didn't. The algorithm uses those accurate signals to schedule each fact optimally.
Does one-card-one-fact mean I'll need hundreds of cards per topic?
Yes — and that's better. 60 atomic cards covering a chapter give you 60 precise retrieval events. 10 dense cards covering the same material give you 10 confused retrievals that mix known and unknown facts. The total review time is similar, but the outcome is very different. More cards, more retention.
How do I know when a card is small enough?
The 5-second test: read the front in under 2 seconds, recall and state the answer, check the back in under 5 seconds total. If any step takes longer, the card is too complex. If grading is ambiguous because you partially knew it — split the card.
Is the minimum information principle the same as atomicity?
Yes — different names for the same concept. "Atomicity" (from Greek: indivisible), "one card one fact," "minimum information" all describe the same design target: the smallest testable unit. Wozniak called it Rule #1 of 20 in his formulation of effective knowledge. Others call it the atomicity principle or the one-fact rule.