Flashcards that work: atomic cards, cloze, leech control

By Better Wise

You already have a review schedule with expanding intervals from the last lesson. That schedule only works if the cards inside it are worth reviewing: a deck full of bloated cards faithfully reminds you to fail the same way for weeks. This lesson is card surgery. You will split oversized cards into atomic ones, write cloze deletions that keep their meaning, and fix cards that keep lapsing instead of grinding them forever.

Three card skills you will build

• Rewrite a bloated card into three to five atomic question cards, each testing one fact.
• Write cloze deletions that leave enough context that only one answer fits.
• Diagnose a leech card and choose a fix: split it, recue it, or delete it.

Worked example: from one bloated card to five sharp ones

A learner building a 4-week nutrition project wrote this card:

Why it fails:

• It contains at least four facts. You can recall two and still feel like you "sort of" knew it.
• The prompt gives no cue about which comparison is being asked for.
• After a miss, you cannot tell which piece was missing, so nothing gets repaired.

The repair takes five steps.

1. Extract the facts. Macronutrients = energy, large amounts. Micronutrients = no energy, small amounts. Macro examples: carbohydrate, protein, fat. Micro examples: vitamins, minerals.
2. One fact per card. Each card gets a single answer slot.
3. Write a cue that points without giving away. "Which group supplies energy?" works; "What is the difference?" does not.
4. Keep answers to one line, and tag the context so the card is not an orphan: "In nutrition, ..." is enough.
5. Test cold after a delay. Answer with the back hidden, then check.

The result:

Five cards, five single-step recalls, each one diagnosable.

Quick check. Split this card into two atomic cards: Front: "The forgetting curve." Back: "Ebbinghaus showed that memory of new material drops steeply at first and then levels off, and that each review flattens the curve."

Model split for the forgetting curve card

One defensible split: (1) Front: "What shape does the forgetting curve have right after learning?" Back: "Steep initial drop, then a plateau." (2) Front: "What does a review do to the forgetting curve?" Back: "It flattens it, slowing later forgetting." If your version tested "who was Ebbinghaus", that is a third card, not part of these two.

Cloze deletions that keep their footing

A cloze deletion hides part of a sentence so the surrounding words do the cueing. It beats question cards when the fact lives inside a relationship: definitions, formulas, causes and effects. Three rules:

• Delete the thing you want to recall, not the grammar. Delete the key term, keep the sentence readable.
• One deletion per card, unless you are drilling a formula with distinct parts.
• Keep the context that makes one answer correct. "It increases retention" hides everything useful.

A card that tests exactly one fact is called an _____ card. A card that keeps lapsing after several reviews is called a _____. Before rewriting a leech, first check whether you actually _____ the underlying idea. Cloze deletion works best when the fact sits inside a _____ that supplies context.

Quick check. Repair these two clozes. (a) "_____ _____ _____ means better retention when study is spread out." (b) "_____ increases _____."

Repaired clozes

(a) The single term was shredded across three cards, and "The" tests nothing. Repair: "The _____ means better retention when study sessions are spread out over time." (b) "It" has no referent, so the card cannot be answered cold. Repair: "Spreading reviews over time improves long-term _____." If you wanted the mechanism, add a separate card: "Why does spacing beat massing?" with "More retrieval attempts under mild difficulty."

Leech control: fixing cards that will not stick

A leech is a card that has lapsed repeatedly, commonly four or more times. Most learners respond by reviewing it more, which is exactly the wrong move: the card is telling you the problem is in the card, not the effort.

flowchart TD
    A["🃏 Card failed again"] --> B{"Failed 4 or more times?"}
    B -->|No| C["Keep it on the normal schedule"]
    B -->|Yes| D{"Do you understand the idea?"}
    D -->|No| E["📖 Go back to the source, then rewrite"]
    D -->|Yes| F{"Does it test more than one fact?"}
    F -->|Yes| G["✂️ Split into atomic cards"]
    F -->|No| H["🔧 Add a cue, context, or image"]
    G --> I{"Still failing after two more reviews?"}
    H --> I
    I -->|Yes| J["🗄️ Suspend or delete"]
    I -->|No| C

Lists are a common leech source. "What are the four stages of the cell cycle?" is a set card, and set cards feel like progress while teaching nothing reliable. Split the list into ordered links instead: "Which stage follows G1?" and "Which stage follows S?".

Quick check. You have failed a set card five times and can list two of four items each time. What is your plan?

Leech plan for the set card

Two moves together. First, split into one card per link in the sequence (G1 to S, S to G2, G2 to M). Second, if the labels themselves are the hard part, convert to cloze: "During _____ the cell grows and copies organelles." Do not add a fourth "memorize the list" card, and do not delete the topic: it is understood, just badly cued.

Why cards fail, and the fix for each

Card triage rules of thumb

• Write cards after you understand the material, never as a substitute for understanding.
• If a card takes more than a few seconds to parse, the prompt is the problem.
• One fact, one cue, one line of answer. If the back needs a paragraph, make another card.
• Fix a leech once, deliberately. If it fails again after the fix, suspend it; you can always relearn it from the source.
• Prune monthly. A 300-card deck you actually review beats a 1,200-card deck you dread.
• Test production, not recognition. "Could I write this answer cold?" is the only honest check.

What to carry into your deck

You can now cut a bloated card into atomic questions, write clozes that keep their context, and triage leeches instead of grinding them. Next, Interleaving and varied practice takes these cards and asks a scheduling question: should a single session mix topics, and how does mixing change what your deck has to cover?

Front: What is an atomic flashcard? Back: A card that tests exactly one fact or idea per review.

Front: Why does a bloated card fail even when you review it often? Back: You cannot tell which part you missed, so the error never gets isolated and repaired.

Front: What should a card's cue do? Back: Point to the answer without giving it away, while supplying the context the answer needs.

Front: What makes a card an orphan card? Back: It has no context tag, so the prompt is ambiguous and only one vague answer seems possible.

Front: When is a cloze deletion a better fit than a question card? Back: When the fact lives inside a sentence, such as a definition, formula, or cause and effect.

Front: What should you delete in a cloze deletion? Back: The key term you want to recall, not filler words like articles or pronouns.

Front: What is a leech card? Back: A card that has lapsed repeatedly, commonly four or more times, and is not sticking.

Front: What is the first question to ask about a leech? Back: Do I actually understand the underlying idea? If not, return to the source before rewriting.

Front: Two fixes for a leech you already understand? Back: Split it into atomic cards, or add a cue, context, mnemonic, or image.

Front: When should a leech be suspended or deleted? Back: When it keeps failing after a deliberate rewrite and no longer earns its review time.

Front: Why are set cards, such as list all four stages, risky? Back: They test several facts at once, so partial recall feels like knowing and errors stay hidden.

Front: What is the honest test of whether a card works? Back: Answer it cold after a delay, with the back hidden, and produce one specific fact.