Back to blog

Studying Two Subjects at Once: When to interleave and when not to

Interleaving helps when your problem is telling similar things apart, and hurts while you are still encoding. Here is how to tell which stage you are in.

Yes, study two subjects at once. That is the short answer, and it is also the least useful sentence in this post, because the mixing is not what decides whether the session works. What decides it is which stage each piece of material is in when it shows up in front of you.

Mixing helps when your problem is telling two similar things apart. It gets in the way while you are still getting something into your head for the first time. Most of the advice you find picks one of those jobs and turns it into a rule. "Interleave everything, blocks are a trap" is one version, "focus on one subject per session or you will confuse yourself" is the other, and both are describing something real. They are describing different weeks.

A little about me first: I am a third year engineering science student at the University of Toronto, specializing in machine intelligence. I also built Nebulearn, so a due queue is literally a thing I sell you. Keep that conflict of interest in mind for the whole post.

Interleaving is one of the better documented ideas in the learning literature, and I am not going to quote you an effect size, because I do not have a study open in front of me. What I can do is explain why the effect exists and tell you what it looks like in your own review log, which is the only data you can actually check.

Interleaving is a discrimination tool, not a variety tool

Start with what a blocked session does to you, because that is what people are defending when they say focus.

When you work through one topic in a block, you load the rule once and then apply it thirty times. The retrieval you practise is "apply this", and you get very good at it. The retrieval you never practise even once is "which of the four things I know is this one". The rule was handed to you by the chapter heading.

Then the exam arrives with no chapter headings. In thermo, nobody writes "this is a control volume question" above the problem, and picking control volume over control mass is half the marks. If you did every problem inside the control volume section, you have never made that pick under pressure. Same shape in Spanish: preterite and imperfect are each easy in a drill where the whole exercise is one tense, and they collapse the second a paragraph asks you to choose.

That is what interleaving is for. Not novelty, not beating boredom. Mixing forces the selection step to happen inside your practice instead of for the first time on the paper. The cost of the mix is exactly the benefit: every card arrives without a context telling you what it is about, so you have to identify before you can answer. My post on the context dependency problem is the same mechanism from the other side, where the deck is quietly feeding you the hint.

Now the counterweight, and it is the part the pro-interleaving crowd skips. All of that assumes you already have the pieces in memory and your failure is choosing between them. If you have never once produced the answer, mixing hands you a search problem stacked on top of a storage problem. You cannot find it, you cannot even tell whether you half know it or have never really seen it, and what the session teaches you is that you are bad at the subject. Encoding wants repetition, close together, with the thing in view, until you can say it once from nothing. That is a block.

The one cold pass test

So the real question is not "should I mix", it is "is this material past encoding yet", and you need something better than a feeling to answer that. Here is the test I use.

Close everything. No notes, no slides, no deck open. Write the answer to one question from the topic, in full, on paper. Then look at what came out. There are three outcomes, and each tells you what to do next.

Nothing comes, or you produce a vague gesture at the right area. You are encoding. It does not matter that you read the chapter twice or that the lecture made sense. Block this. One topic, one sitting, cards close together, until you can do a cold pass on each one. The blank sheet version works better for whole topics than for single cards.

You produce it and it is right. Stop blocking. This material has nothing left to gain from a session where every neighbour is its cousin. Put it in the mixed queue and let it come back on its own schedule.

You produce something confidently, in the right shape, and it belongs to the wrong item. You write the glycolysis answer for the Krebs question. You conjugate an -ir verb with the -er endings. You give a clean definition and it is actually the definition of the term next to it in your notes. This is the only failure that interleaving fixes, and it is the one people misread as "I need to go back and reread the chapter."

Those outcomes are worth naming as error types, because they look identical when you are tired and all of them just feel like getting it wrong.

What the error looks likeWhat it meansWhat to do
Blank. Nothing arrives.Not encodedBlock the topic for one sitting.
Right idea, half the steps missingEncoded but weakNormal review. Mixing is fine.
Confident, belongs to the neighbourDiscrimination failureInterleave that pair on purpose.
Right only after a related cardContext is answering for youInterleave, and check the card stands alone.

The bottom two rows are the ones mixing fixes. The top row gets worse.

What a mixed queue does to your review time

The reason this decision feels hard is that the correct choice feels worse while you are making it, and it also costs real minutes. Articles about interleaving never mention that bill, so here it is.

Per card, a mixed queue is slower. In a blocked session the context stays loaded and you answer from something already sitting in working memory, so cards go by fast. In a mixed queue every single card starts with a beat of "what is this even asking", and that beat is the retrieval you came for, but it costs seconds each time and they add up across a session.

Accuracy in the session drops too, and that has a second order effect people do not expect. On a real scheduler, grading more cards Again or Hard shortens their intervals, so they come back sooner and tomorrow's queue is bigger than it would have been after a comfortable blocked session. Mixing is not only slower today. For a week or two it inflates the pile, and if you are already fighting a queue that is too big, that is a bad moment to add the mix on top.

That cost also tells you when to spend it. Those extra Again grades are information about which pairs you cannot separate, and you are either buying that information now at the price of a worse Tuesday, or on the exam at the price of marks. Mixing needs runway, and it is not a cram tactic. Two days out from a test on one chapter, block, and accept that you are cramming rather than doing spaced repetition. A month out with two courses running, mix, and expect the sessions to feel mediocre for a while.

The thing not to do is shrink the mix by subject to make the session comfortable again. Cap the clock instead, because session length is the lever that protects your evening.

What one week looks like in practice

The stages only make sense on a calendar, so here is the whole thing on one week of mine.

Monday, thermo lecture on entropy, all new. I build maybe twenty cards off the slides and review only those, back to back, that night. That session is blocked on purpose and from the outside it looks nothing like good practice. Same card twice in ten minutes, until I can do a cold pass on each one.

Tuesday, the entropy cards join everything else. My queue that night has week two definitions, week four cycles, a pile of Spanish, and some cards from last month I barely recognize, in whatever order the scheduler picked. I do not sort it by course. This is where the two subjects at once question answers itself: they are in one queue, and the mixing happens because of per-card scheduling, not because I built a clever rotation.

Wednesday, the confusable pairs get attention. Reversible and irreversible, closed system and open system, preterite and imperfect. Those I deliberately keep together in one mixed pass, because those specific cards are only hard next to each other, and this is the interleaving that pays. If a pair keeps failing I stop blaming my brain and go read the two cards, because usually they are two vague cards that could each be answered by the other, which is an encoding problem wearing an interleaving costume.

The only rule running through that week: new material stays blocked until it passes one cold pass, everything else goes in the mixed pile, and confusable siblings get shoved together.

When blocking is still the right call

I have been describing blocking as the earlier stage, so I should be direct about the cases where it stays correct even for material you already know. "Always interleave" will hurt you in at least four situations.

A procedure you cannot execute yet. If you cannot balance a redox equation at all, mixing reaction types is not practice, it is failing in variety. Block until the procedure runs, then mix the types.

Forty eight hours out, one chapter. The exam covers one unit, there is no selection step to rehearse, and what you need is coverage. Block.

Two subjects that are not confusable. Mixing Spanish vocab with fluid mechanics buys a context reload and nothing else, because no part of you was ever going to answer a fluids card with a Spanish word. That is fine, and it is what a single queue does anyway. Just do not congratulate yourself for interleaving. The real work is inside a subject.

A card that keeps dying. If one card fails in the mixed queue over and over, blocking will not save it. The card is the problem. This is where ghost words in language decks come from, and the fix is rewriting or deleting, not rescheduling.

Credit where it belongs: Anki's filtered decks are the best version of the block step I have used. You pick a tag, you get a temporary deck of exactly that material, you grind it, and it folds back into your collection when you are done. If you need to pull a single chapter out of one long running deck for a night, that is the tool, and nothing I have built beats it for that job. Quizlet earns the same slot honestly: a shared set for one chapter on a Friday is blocking, and blocking is correct for one chapter on a Friday.

Mixing two subjects is the easy version of this question

Which brings it back to the query, because "should I study two subjects at once" turns out to be the milder version of the real question.

If both courses live in one queue on a real per-card scheduler, you are already interleaving them and it took no decision on your part. What you control is whether brand new material gets held out of that pile for one sitting. That is the entire practice: block what has not passed a cold pass, mix everything else, and put the confusable siblings next to each other on purpose.

So the mixed queue is the default and blocking is the exception you buy deliberately, for one night, for a reason you can name. That split is the thing I tried to make Nebulearn do without configuration: FSRS is on by default, so one queue holds both courses and each card is scheduled on its own instead of by whichever deck it sits in. The part you still do by hand is the first night, while new cards are new, and I am not going to pretend the app makes that stage judgement for you. You make it with one cold pass on paper.

Study TipsInterleavingSpaced Repetition