The standard math study method is redoing homework in order — problem 1, then 2, then 3, in the sequence the professor assigned. It feels thorough and it fails a specific way: the exam doesn't present problems in homework order, and it doesn't label them with the section's method name. The exam presents a naked problem and asks which method — a decision homework order makes for you.

The fix is generating your own problems: take the lecture's worked example, change the numbers, and add one trap the original hid. A domain issue, an extra term, a lookalike method that almost applies. Then solve it closed-book and name the method in one sentence. That's the practice the exam requires — selection, not sequence.

The worked example is a template. The exam is a variant. Practicing only the template is memorizing an instance; generating the variant is learning the method.

Why homework order fails the exam

Redoing homework in order is the math version of rereading — fluent, sequential, and structurally wrong for the test. Homework comes pre-sorted: section 4.2 is the chain rule section, so problem 7 in section 4.2 is a chain rule problem. The method selection is done before you start. The exam hands you an unlabeled problem where the method is the question — and the student who only practiced pre-sorted problems has never made that decision.

Interleaving lookalike problem types improves method selection — the desirable difficulty of mixing makes you choose the approach rather than inherit it from context. And generating a variant is retrieval of the procedure, not recognition of the homework number: when you invent the problem, you have to know what makes it a related-rates problem versus a plain optimization, which is exactly what the exam asks.

The variant-generation workflow

Scope to one lecture. Today's related-rates example, not the whole derivatives unit. The methods your professor demonstrated are what the exam draws from — practicing on a random textbook's problems is preparing for a different test.

Clone the method, change the story. Same underlying technique — relate variables, differentiate, plug the instant — but a different container: a conical tank instead of a spherical balloon, a sliding ladder instead of a rising rocket. The method is what's invariant; the story is what changes.

Add one trap the original hid. A radius that depends on height (can't treat it as constant), a quantity that goes negative (domain issue), a lookalike that tempts the wrong method. The trap is what makes the variant a test — without it, you've just renumbered the homework.

Name the method in one sentence. After solving, say what it was: "related rates via similar triangles, differentiate with respect to t." The naming is the retrieval — if you can't name the method, you executed a walkthrough; if you can, you selected an approach.

Retest with a new trap. Same method, different hidden condition. The 1/3/7-day spacing holds; the new trap prevents memorizing the variant instead of learning the method's boundary conditions.

A 40-minute session on the clock

The weak answer is the label — "also related rates, differentiate both sides" — which is what you'd say if you recognized the category without seeing the difference. The strong answer produces the method's steps and names the trap: in a cone, radius isn't constant, it's proportional to height, and missing that turns the problem into mush. The trap is what the exam adds to the template.

Why inventing beats redoing

When you invent a variant, you learn what makes the problem that problem — the constraint that requires related rates instead of plain differentiation, the relationship that forces the chain rule. You see the method's necessary conditions because you're the one setting them. Redoing the homework shows you the method applied; inventing the variant shows you when to apply it.

This is why one good cousin beats five clones. Five variants that only change digits are the same problem five times — you memorize the sequence of operations without learning what triggers them. One variant with a real trap forces the decision the exam makes: is this related rates or optimization? Is the radius constant or dependent? That decision is the skill; the arithmetic is the easy part.