Blocked homework is a trap that feels like mastery — all of type A, then all of type B, each problem solved because you already knew which method to use. The block told you the method; the exam won't. When the exam shuffles the types, the student who only practiced blocked sets can't choose — they never had to. The method was given; the skill of selecting it was never trained.

The fix is to mix the lookalikes deliberately. Take three confusable types from the unit — separable, linear first-order, exact equations — and shuffle them in one set. Before you compute, write the method in four words: "exact: check My=Nx" or "linear: y'+Py=Q." Then compute. The naming is the exam's skill — the discrimination between lookalikes that blocked practice let you skip. The computation is the follow-through; the choice is the test.

Why blocked practice fails

Doing all of type A then all of type B trains execution without selection — each problem comes pre-sorted, so the method is given by the set's position, not chosen by you. The fluency feels like mastery because every problem is solvable; it's false because the solving never required the discrimination. The block answered the question the exam will ask: "which method applies here?"

Interleaving is the desirable difficulty that trains the discrimination — the lookalikes shuffled together force the choice each problem, which is the skill blocked practice skipped. The mixed set feels worse because the decision is hard; that difficulty is the mechanism working. Blocked fluency is a false signal — the 90% on Tuesday's type-A set says nothing about whether you can tell type A from type B, which is what the shuffled exam will ask.

The interleaving workflow

Pick two or three confusable types. The lookalikes — separable vs linear vs exact, integration by parts vs substitution, Type I vs Type II error — the pairs the exam will put side by side. Two or three is enough to force the choice; five is chaos. Start with the pair you confuse most.

Shuffle them in one set. Not "A, B, A, B" in a pattern — shuffled, so the type is unpredictable. The unpredictability is the point: the exam doesn't announce the type, and the set shouldn't either. A pattern gives the method away; a shuffle makes you choose it.

Name the method before you compute. Four words: "exact: check My=Nx," "linear: integrating factor," "separable: split and integrate." The naming is the discrimination — the decision the exam scores. Writing it first means the choice exists before the algebra can substitute for it.

Then compute — and check the choice. If the method stalls, the name was wrong: not exact, so try linear; not linear, so separable. The stall is the feedback — the chosen method didn't fit, which means the discrimination was wrong. The re-choice is the second rep, the one that teaches the boundary.

Misses are discrimination errors, not computation errors. A wrong method on a lookalike is a different miss than a wrong integral — the first is "can't tell the types apart," the second is "can't execute the chosen method." Both matter; the discrimination miss is the one interleaving is for, and it's the one blocked practice never caught.

A 40-minute session on the clock

The weak answer skips the discrimination — "recognize as I go" means the algebra will reveal the type, which means the method was never chosen, just discovered. The strong answer names first: the exact-check, the linear-form, the separable-fallback — a decision tree for the lookalikes. The naming is the exam's operation; the computation is what follows a choice the blocked homework made for you.

Why the mixed set feels worse

The mixed set is harder because it adds a step blocked practice skipped — the method choice. Blocked homework starts at "compute"; the mixed set starts at "decide." The extra step is the skill, which is why the mix feels slower and more error-prone: you're doing the part the block removed. The discomfort is the discrimination being trained.

This is why students avoid interleaving — it feels like regression after the fluency of blocked sets. The blocked fluency was real but narrow: execution speed on a pre-sorted type. The mix's difficulty is the selection skill appearing — the same problems, now with the decision the exam will require. The worse feeling is the skill being built; the better feeling was the skill being skipped.