Back to blog

How to organize flashcard decks (without building a filing system)

One deck per course, tags for topic and source. Why deep subdeck trees split your daily queue, when a subdeck earns its place, and the cost of reorganizing.

Short answer: one deck per course, and tags for everything else. Tag the topic, tag the source, tag the exam. Do not build a tree.

The reason is boring and it is the whole article. A deck is a scheduling unit. A tag is a finding unit. Decks decide what shows up in front of you today. Tags decide what you can pull back out three weeks from now. Most students invert this, because a deck list looks like a folder tree in a file browser, and folder trees are the one organizing skill everyone already has. So they build ECE216 > Week 4 > Op-amps > Inverting, end up with fourteen little decks, and never open eleven of them again.

I work on Nebulearn, so read my product opinions with the suspicion they deserve. The structure below is app-agnostic and I used it for two years before I built anything.

Decks schedule, tags retrieve

Before the structure, it is worth being precise about what each thing does, because most advice on this question treats them as two flavours of the same feature.

A deck is where scheduling settings live. In almost every spaced repetition app that lets you nest decks, the deck is what carries the daily new-card limit, the review order, and often the algorithm parameters. When you tap into a deck, the app builds a queue out of that deck's cards and stops there. That is the important part: the deck is a boundary on your queue.

A tag is a label with no scheduling power at all. It does not cap anything, it does not create a queue by itself, and a card can carry five of them. Tags exist so that in week nine you can type titration and see eleven cards from three different lectures.

Once you say it that way, the design question answers itself. How many separate daily queues do you actually want? For most people the honest answer is one per course, because a course is the unit that has one set of exams, one instructor, and one pace. Topic is not a scheduling unit. You do not want a Tuesday where op-amps are due but diodes are not because the app decided those were different worlds.

This is also why one idea per card and one deck per course are not in tension. Cards get small. Queues stay whole.

What the tree actually costs you

I want to be concrete about the failure, because "just use tags" sounds like tidiness advice and it is not. The subdeck tree has a real arithmetic cost.

Say you have four courses. You are a careful student, so each course gets subdecks by week, and by the end of term each course has ten or twelve of them. That is somewhere north of forty decks. Three things go wrong, in order.

First, the queue fragments. If the daily limit lives on the deck, then twelve week-decks each with their own cap means your course no longer has a daily number, it has twelve. Whichever decks you happen to open get studied. The others accumulate silently. A card in Week 3 that you last saw in September is not organized, it is lost, and it is lost inside a structure you built specifically so you would not lose it.

Second, every session now starts with a decision. Which of the forty do I open tonight. That sounds trivial and it is exactly the friction that ends streaks. The point of a due queue is that it removes the choice. A tree hands the choice back to you at the worst possible moment, which is when you are tired and looking for a reason to close the app.

Third, and this is the one that actually burns people: the tree encodes the wrong thing. You filed by where the card came from. Exams do not ask by lecture week. They ask by topic, and topics smear across weeks. My thermo final pulled entropy from an early lecture, a much later one, and a lab handout. A week tree is the one shape that guarantees those three never appear in the same review.

You can see the same reflex in what happens when a pile gets too big. People reorganize instead of cutting, and reorganizing does not change the card count. If your queue is genuinely unmanageable, the fix is a smaller deck or a saner new-to-due ratio, not a deeper hierarchy.

The structure I would set up on day one

So here is the concrete version, with my own course load as the example, because a rule you cannot copy is not a rule.

One deck per course. Four decks: ECE216, MAT292, AER210, MSE160. That is the entire deck list. Inside each one, every card gets tags from three families:

Tag familyExampleWhat it is for
Topicfourier, entropy, op-ampFinding cards by what they are about, across weeks
Sourcelec-07, tut-03, textbook-ch5Tracing a bad card back to where it came from
Exammidterm-2, finalBuilding a temporary cram filter, then dropping it

Topic tags are the ones you will use most and the ones people skip, because at card-creation time the lecture number is the thing on screen and the topic feels too obvious to write down. It will not feel obvious in November. Write it down.

Source tags earn their keep on the second pass. When you hit a card that is unanswerable, lec-07 tells you which slide deck to reopen, and more usefully, it tells you that the other nine cards tagged lec-07 are probably bad in the same way. That is how you fix a batch instead of a card.

Exam tags are deliberately disposable. Before a midterm I filter to midterm-2 and do an extra pass on that subset without touching my normal reviews. Afterwards the tag just sits there, harmless. That is the difference from a subdeck: a stale tag costs nothing, a stale deck quietly holds cards hostage.

Two naming details that save real annoyance. Pad your numbers, so lec-07 sorts next to lec-08 instead of after lec-70. And pick one separator and never switch, because op-amp and op_amp are two different tags and you will not notice for a month.

When a subdeck genuinely earns its place

I am not going to pretend subdecks are always wrong, because there is one test they pass cleanly, and it falls straight out of the first section.

A subdeck earns its place when the cards inside it need a different schedule. Not different content. Different schedule. In practice that means one thing most of the time: a separate exam with its own date.

The real case is something like a certification with a fixed date landing in the middle of a term. It had a fixed date, it needed a much higher daily new-card count than any of my courses, and after the exam I wanted it gone from my mornings entirely. Different pace, different end date, different decision about when to stop. That is a deck.

Two more qualify by the same test. A deck someone else wrote that you are running alongside your own cards, because you want to suspend the whole thing in one action when it turns out to be bloated. And a language deck, which usually wants a steady daily trickle forever while a course deck wants a burst in September and nothing in December.

What does not qualify: chapters, weeks, lecture numbers, easy versus hard, theory versus problems, and anything you split because the deck list looked untidy. Those are all tags. If you cannot name a date or a pace that makes the subdeck different, you are filing, not scheduling.

Reorganizing a deck you already study

Most people reading this already have a mess, which is a different problem from setting one up, and the costs are not symmetric. This is the part I would have wanted to know first.

Retagging is cheap. Adding, renaming, and bulk-applying tags does not touch a card's history or its next due date in any app I have used. You can spend twenty minutes adding topic tags to a deck you have been reviewing for two months and nothing about tomorrow changes. Do this freely.

Moving cards between decks is usually cheap too, but read the warning. The card keeps its own history, and what changes is which deck's settings now govern it. Collapse twelve week-subdecks into one course deck and your daily limits become one number instead of twelve, which is the point, but tomorrow's session will look bigger than yesterday's. That is not a bug. That is the backlog becoming visible. Expect a rough week and do not respond by splitting it again.

The genuinely expensive move is rebuilding. Deleting cards to recreate them properly, or exporting and reimporting to clean up structure, throws away review history. That history is what the scheduler uses to pick intervals, so a rebuilt card is a brand new card, and a thousand rebuilt cards is a brand new thousand-card deck. It is the same reason history does not come along when you move between apps: the text moves, the schedule does not. Reorganize in place, or do not reorganize.

The cost nobody counts is the afternoon itself. A reorganization day is a day you did not review, and the queue does not pause for tidying. If the cleanup will take more than half an hour, split it: reviews first, tidying afterwards, stop when the timer goes.

When this breaks anyway

Two failure modes show up after people adopt this, and both look like organization problems while being something else.

The first is a course deck that becomes genuinely unpleasant to open. One deck per course only works while a course's card count is sane. If a single course is producing hundreds of cards a month, splitting it does not help, because the cards still come due either way. That is a card-count problem, and it gets solved by deleting the leeches and by auditing the deck before the exam window, not by moving things around.

The second is tags you never search. If you have been tagging for a month and have never once filtered by a tag, your tags are decoration. Keep the topic family, keep the exam tag, drop whatever you have never typed into a search box. Organizing you do not use is just a slower way to make cards.

Credit where it is due: Anki is better than we are at the case where you genuinely want deep hierarchy. Its per-deck options presets make a real subdeck tree behave, and if you are running something like the AnKing collection, that hierarchy is the product and you should keep it. Quizlet is fine too when a unit really is one small set you will cram on Friday and never revisit, because then a set per unit costs you nothing.

For everyone else the split in this article is the whole design: one queue per course so nothing can hide, tags so you can still find anything. That is why FSRS is on by default in Nebulearn, with no per-deck scheduling to get wrong, and why the thing you organize with is a tag rather than another folder. A course is a deck. Everything else is a label. If your Sunday goes on next week's cards instead of on your deck list, the structure is working.

FlashcardsStudy WorkflowSpaced RepetitionOrganization