Interleave related skills, not random subjects

Mix things you might confuse with each other — not unrelated subjects thrown together.

Why it works

Interleaving works largely by training discrimination: noticing the subtle cues that say "this is a type-A problem, not type-B." That benefit only exists when the items are similar enough to be confused. Mixing genuinely unrelated material adds task-switching cost without the discrimination payoff.

How to do it

1. Pick categories that learners commonly mix up (e.g. different formula types, similar concepts).
2. Interleave within that confusable family rather than across totally separate domains.
3. Notice which pairs you keep misclassifying and weight those more heavily.

Evidence

Studies on inductive learning (e.g. classifying painters by style) and on math problem types show interleaving helps most when categories are similar and must be distinguished, supporting the discrimination account of the effect. (rct)

For wholly unrelated tasks the main mechanism (discrimination) does not apply, so benefits are weaker and switching costs can dominate.

Common mistake

Interleaving wildly different subjects (French verbs, then calculus, then history dates) and concluding "interleaving does not work" when the real issue is that nothing needed discriminating.