Past exams are the best exam-prep resource — and the most misused. The standard approach is doing the same PDF three times until the answers feel automatic, which produces fluency with that exam, not skill for the new one. The 2024 midterm's numbers won't be on the 2025 midterm; the operations it asks for will.

A past exam is a verb list and a trap list, not a script. For each item, name the operation — compare, compute, justify, apply — and the trap the item sets. Then write a new story that needs the same operation. The old wording won't reappear; the old moves will.

The distinction is between memorizing an instance and extracting a pattern. "The answer was consumer surplus of $40" is an instance — useless next year. "The operation is calculate-surplus, the trap is the tax-vs-subsidy wedge direction" is a pattern — that's what transfers.

Why memorizing past exams fails

Doing the same past paper repeatedly is the exam-prep version of rereading — each pass feels more fluent because you're recognizing the answers, not producing them. By the third run, you're not solving; you're reciting. And recitation fails the moment the exam changes the story: same operation, different numbers, and the memorized answer is useless.

Memorizing stems trains recognition of last year's story — the fluency problem applied to a fixed document. Retrieving the operation in a new story is transfer: you take the verb (calculate surplus) and apply it to a different setting (subsidy instead of tax), which is the move the exam makes. The past paper tells you what operations to practice; the new stems are where you practice them.

The mine-and-regenerate workflow

Scope to one past paper. Last year's midterm, not a folder of every exam ever given. One paper is enough to harvest the verb list; more papers add story variety, not new operations.

Name each item's operation and trap. For every question: what does it ask you to do (the verb), and what does it hide (the trap)? "Calculate consumer surplus after a tax" is calculate-surplus plus a wedge-direction trap. The operation is the transferable skill; the trap is what makes it non-trivial.

Write a new story for the same operation. Same verb, different content — a subsidy instead of a tax, a different market, a new scenario. The new stem forces the operation without the memorized numbers; you can't paste last year's answer because the story is different.

Name the fake-progress version. For each item, say what memorizing would have missed: "I can recite the old surplus answer but can't do it with a subsidy." The contrast makes the extraction visible — recognition of the old story isn't skill on the new one.

Retest with a new story. Same operation, different setting — the 1/3/7-day spacing holds, and the new story prevents memorizing the variant instead of learning the operation.

A 40-minute session on the clock

The weak answer is the memorizer's reflex — "same graph the other way, surplus is surplus" — which is what you'd say if you'd done the old paper three times and recognized the category. The strong answer extracts the operation (find the wedge, integrate) and names what changes: the wedge flips, the incidence story shifts from tax burden to subsidy capture. That's the verb-and-trap extraction the new exam tests.

Why the operation is what transfers

The past exam's content is dead — the numbers, the specific scenario, the exact wording won't come back. What lives is the operation: the move the item requires. Calculate-surplus appears every year because the course tests it every year; the surface changes (tax vs subsidy vs price floor) are the costume, not the skill.

This is why two papers are enough. The verb list is short — a course has maybe ten operations it tests — and two papers reveal them all. The fifth paper adds nothing but more instances to memorize. Harvest the verbs, write the new stems, practice the moves; the papers have done their job once you have the list.