Why it mixes topics up.

What you actually want.

To finish something. Take one notion, work it until it is solid, close it, move on. It is how books are organised, how courses are timetabled, and how almost everyone studies when left to their own devices — and it is not an unreasonable instinct, because it genuinely does make the practice go better.

What you run into.

Doing all the questions on one topic in a row means you never have to work out which topic you are in. It has already been decided for you, by the block you are sitting in. So the practice trains one half of the skill — carrying out a method — and quietly skips the other half: recognising, from the problem alone, which method this is.

That second half is most of what using the knowledge later consists of. Problems in the world do not arrive labelled with their chapter. In studies that compare the two, blocked practice reliably wins during practice and loses at a delayed test — the dissociation has been reproduced in labs, in classrooms and in randomised trials in real schools.

What we do instead.

Interleaving happens in three places in the pipeline, and the third one is the interesting one.

• When the points are chosen. A session's due points and its new points are alternated rather than served as one block of revision followed by one block of new material (compose, sessions/compose.rs).
• When the exercises are dealt — and only for points you have met before. The step that writes the exercises emits every slot of one point before moving to the next, so a session composed as an alternation could still reach you in blocks. A final pass fixes that, conditionally: a point you have answered before has its exercises dealt round-robin against the others, while a point you are meeting for the first time keeps its exercises together (deal_by_maturity, sessions/grid.rs). Mixing buys the ability to tell methods apart, and that is only worth its switching cost once you can carry the method out at all. Consolidation first, discrimination after.
• In the review queue, where it is yours to set. Reviews are reordered round-robin across the sections they come from, so consecutive items come from different parts of the course. It is a setting — Mix topics inside each review — on by default, and off is a supported answer.

One ordering decision sits alongside these and is not about interleaving at all: within a session, due material is served before new material. Whoever stops early loses whatever came last, and a missed review costs retention on something already learnt while a postponed first encounter only costs progress. The code marks that one as a calibration choice rather than a derived one, because it is.

What this does not buy.

This is the principle in the app with the most boundary conditions, and stating them is not throat-clearing — two of them are places where the effect is known to reverse.

• Mixing unrelated things is just task-switching. The discrimination benefit needs material that could plausibly be confused with each other. Ignia gets this by construction rather than by cleverness: a session is drawn from one skill, so what is being mixed is always the parts of one subject.
• The sign can flip, and we do not measure the thing that flips it. Interleaving wins when the hard part is telling similar categories apart. When the hard part is instead spotting what the members of one category have in common, blocking wins. That depends on the similarity structure of the material — which we do not compute for your course. Whether you have met a point before is a proxy for readiness, and a coarse one. It is not a measure of how confusable your notions are.
• The benefits are moderate and condition-dependent. The most sceptical serious review of this literature says the effects are real but neither large nor universal, and we would rather cite that here than leave it out. This is a well-supported default, not a guarantee about your next session.
• It costs you in the moment, and part of that cost lands in our data. Switching is slower and more error-prone while it is happening. An error caused by having just switched topics looks, in the record, exactly like an error caused by not knowing the answer — the app cannot separate them, so a little switching noise goes into your learner model. Known, unfixed, and small enough not to justify giving up the ordering.
• The mixing is not perfect, by construction. The deal is a plain rotation, so when one point holds more exercises than there are points to rotate through, its last few end up adjacent anyway. That residue is inherent to the method rather than a bug we have not got to.

Questions.

Sources.

What each one is doing here, so you can tell which claim rests on which work. Every one of these is a published paper you can look up.

• Rohrer, D., & Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35, 481–498. · Taylor, K., & Rohrer, D. (2010). The effects of interleaved practice. Applied Cognitive Psychology, 24, 837–848.

  The dissociation itself: blocked practice wins during practice and loses at a delayed test. Taylor & Rohrer is the cleanest single demonstration that practice-time performance misleads.

• Rohrer, D., Dedrick, R. F., Hartwig, M. K., & Cheung, C.-N. (2020). A randomized controlled trial of interleaved mathematics practice. Journal of Educational Psychology, 112(1), 40–52.

  The randomised trial in real schools referred to on this page. It is the strongest ecological evidence in this literature.

• Kornell, N., & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the "enemy of induction"? Psychological Science, 19(6), 585–592.

  The painters' styles study, including the part this page leans on: most participants believed the blocked condition had worked better, having just done both.

• Birnbaum, M. S., Kornell, N., Bjork, E. L., & Bjork, R. A. (2013). Why interleaving enhances inductive learning. Memory & Cognition, 41, 392–402.

  Separates the two things mixing buys — discrimination (needs related material) and spacing (needs only a gap). The app's design follows this split.

• Carvalho, P. F., & Goldstone, R. L. (2014). Putting category learning in order: Category structure and temporal arrangement affect the benefit of interleaved over blocked study. Memory & Cognition, 42(3), 481–495.

  The boundary condition where the effect reverses, and the reason this page says we do not measure the thing that flips it.

• Firth, J., Rivers, I., & Boyle, J. (2021). A systematic review of interleaving as a concept learning strategy. Review of Education, 9(2).

  The most sceptical serious source in this area: benefits real, but moderate and condition-dependent. It is here on purpose.

• Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140.

  The switching cost is real and measurable — in reaction time and in-the-moment accuracy. What it does not show is a cost to long-term retention.

You will find effect sizes in these papers. We deliberately do not print them on this page: an average across studies does not predict what any one person will get, and a number quoted without its conditions reads as more precise than it is. In the papers they come with those conditions.