Why a 500-Card Deck Feels Impossible
Open a 500-card deck and the number does the damage before you answer a single question. The arithmetic says it is fine — five hundred cards, fifteen seconds each, a couple of hours — but it does not feel fine, and you close the tab.
The reason is not laziness, and it is not the total study time. An unstructured deck presents every card as an independent thing to learn: card 14 tells you nothing useful about card 15. You are not learning one subject with 500 facts in it — you are learning 500 separate things, each held on its own.
What the Research Actually Says About Chunk Limits
Miller (1956) is the famous one: the "magical number seven, plus or minus two," and one of the most misquoted results in psychology. Miller was careful — the limit applied to chunks, not raw information, and he was explicit that the recurring number might be coincidence rather than law. Cowan (2001) revisited it with tasks that stopped people silently rehearsing or grouping items on the fly; remove those tricks and capacity looks smaller, roughly four chunks. Four is the figure most working-memory researchers use today.
Either way, the number is not the interesting part. The limit is measured in chunks, not items, and chunk size is not fixed. Gobet et al. (2001) showed this with chess players: a strong player glancing at a real game position remembers it far better than a novice, but on a board with pieces scattered at random the advantage mostly disappears. The expert is not holding more pieces — just fewer, larger units built from years of exposure.
You cannot raise your working-memory capacity. You can change how much meaning is packed inside each unit occupying it. That is chunking — the only lever you actually control.
Grouping by Meaning, Not by Count
The common misreading is that chunking means "study in batches of seven." It does not. Seven unrelated cards are seven chunks; seven cards describing stages of one process are closer to one chunk, because each answer constrains the others — if the process runs A, B, C, forgetting B leaves a shaped hole rather than a blank. So the first question when splitting a deck is never "how many cards" but "which cards explain each other." Splitting alphabetically, or into ten blocks of fifty, does almost nothing: the pile is smaller, but no card is cheaper to hold.
How to Chunk a Deck
1. Find the structure that already exists
Most well-built decks are already organised by section, module, or tag — use it. A natural sequence (fundamentals, then mechanisms, then applications) gives you chunk boundaries ready-made. If there is no structure at all, skim the first fifty cards and invent four or five categories yourself. Rough categories beat no categories by a wide margin.
2. Cap the session at 20–30 new cards
Twenty to thirty is where most people land after a few weeks of honest self-observation. Below fifteen, progress feels pointless. Above thirty-five, the later cards stop landing — you are still turning them over, and you see the damage days later when they all come back at once.
On LearnCoachAssist the study page carries a daily study goal widget with an editable target. That number is your chunk-size dial: set it to 20 and the interface stops rewarding you for grinding past the point where cards stop sticking.
3. Master a chunk before adding the next
"Mastered" here is a deliberately low bar: you can answer every card in the chunk without hesitation, on a day when you did not just study it. Clear that bar and open the next chunk. Miss it and run the same chunk tomorrow, adding nothing new. This is the step people skip, and skipping it produces the review avalanche three weeks in.
4. Interleave old chunks once you have several
Blocked practice — one chunk, over and over — is right while a chunk is new, and wrong once it is solid. With three or four chunks behind you, mix them in the same session. Interleaving practice feels worse and performs better: it forces you to work out which category a card belongs to before you can answer it, which is exactly the discrimination a real exam demands.
Let Spaced Repetition Own the Schedule
Here is what makes this manageable: you only have to chunk the intake. Deciding which 20 new cards enter today is your job. Deciding when card 37 comes back for its fourth review is not — that is arithmetic, and a spaced repetition system does it better than you would. Every card carries its own due date, expanding when you get it right and collapsing when you get it wrong, so you never hold the schedule in your head — which matters, because a 500-card deck generates one nobody could track manually.
The split: you control new cards per day, the scheduler controls reviews per day. If the daily load is uncomfortable, the lever is always new cards. Never skip reviews to make room — that moves the debt forward with interest.
Worked Example: 500 Cards in Five Weeks
Take the AI Agents (500 Questions) deck: five hundred cards, one broad subject, plenty of internal structure — architectures, tool use, planning, memory, evaluation, failure modes. Set the daily goal to 20 new cards. That is 25 sessions; at five a week, weekdays only, the intake finishes in five weeks.
- Week 1. Short sessions: 20 new cards plus a few reviews, ten to fifteen minutes. Stay inside one topic area all week.
- Week 2. Week 1's reviews arrive in volume and session length roughly doubles, even though the new-card count has not changed. This is where most people quit. Do not raise the target.
- Week 3. Review load flattens as week 1 moves to longer intervals. Two hundred cards in — start interleaving weeks 1 and 2.
- Weeks 4–5. Same rhythm, last two hundred cards. Early material now appears weekly or less, freeing the session for new chunks.
- After week 5. No new cards. Reviews only, shrinking steadily — five to ten minutes a day to hold five hundred cards in place.
Five weeks, twenty-something minutes a day at the peak. The same 500 cards attacked in one heroic weekend produce a deck you cannot recall and will not revisit.
Three Mistakes That Break Chunking
Cramming an entire chunk in one sitting
Running the same 25 cards six times back-to-back produces a strong feeling of mastery and almost no durable memory: within one session the answer never leaves working memory, so you never really retrieve it. One pass today and one tomorrow beats six passes tonight.
Chunks that are too big
"I'll do a hundred a day and finish in a week." The first thirty land, the rest are processed rather than learned, and the whole hundred comes due on the same day — at which point the deck genuinely is unmanageable, and it feels like the method failed rather than the batch size.
Never revisiting an old chunk in a new context
A chunk that only appears alongside its own siblings gets learned in a way that depends on that context — you know the answer because you know which section you are in. Interleaving removes the crutch, and that has to happen before the exam.
FAQ
Isn't the real limit seven items, not four?
Both come from real work. Miller (1956) reported around seven; Cowan (2001) argued the honest figure is nearer four once rehearsal and covert grouping are prevented, and four is the more defensible modern estimate. For deck planning the difference barely matters: neither number is your session size, because 20 cards studied one at a time are not 20 simultaneous chunks.
What if the deck has no tags or sections at all?
Make your own. Skim the first fifty cards, write down four or five recurring themes, then work through the deck in whatever order groups those themes together. The categories need not be correct in any academic sense — they only need to make consecutive cards relate to each other.
How long should a chunk take before I move on?
Judge by recall, not by days: when you can answer every card cold, the chunk is done. That is usually two or three sessions for 20–30 cards. If it has taken more than four, the chunk is too big or the cards are badly written, and splitting it will do more good than repeating it.
Summary
A 500-card deck is unmanageable only as an undifferentiated pile. Group cards by meaning, take 20–30 new ones per session, clear a chunk before opening another, mix old chunks in once you have a few, and hand the schedule to spaced repetition. Working memory is small and you cannot expand it — but what fits inside one unit of it is up to how you organise the material.