A question bank rots when its cards stop requiring retrieval — the card's wording becomes the cue, and you answer from "the first four words" instead of from memory. "When was the Peace of Westphalia?" triggers "1648" before you finish reading, because the card's shape is the retrieval path, not the fact. The card isn't testing you anymore; it's reciting itself.

The fix is maintenance, not volume. Each week: delete two cards that are too easy (you've memorized them, not the material), rewrite two that you answer from the wording (invert them — given the answer, produce the condition), and add one transfer (the mechanism in a new story). The bank's health isn't how many cards it holds; it's how many still make you think. A bank you prune is a tool; a bank you just add to is a museum.

Why the growing bank rots

Adding 80 new cards and never pruning is the bank's version of massed practice — volume masquerading as coverage. The deck grows, the reviews pile up, and each card becomes a recognition test: you see the first words, the answer surfaces, and the "retrieval" was really the card reading itself to you. The bank is full; the learning is stale.

Spacing only helps if the item still requires retrieval — the card has to make you produce something, not recognize its own wording. A burned-in stem is rereading in disguise: the answer comes from the card's shape, not the concept. The pruning restores the demand — delete the ones that answer themselves, rewrite the ones the wording gives away, and the bank goes back to testing instead of reciting.

The pruning workflow

Delete the too-easy two. Each week, two cards you can answer without thinking — the fact is memorized, the card is redundant. Deleting isn't losing the knowledge; it's removing a card that's stopped testing. The bank isn't a record of everything you know; it's a set of things you're still practicing.

Rewrite the wording-answerable two. The card you answer from the first four words — "When was the Peace of..." → "1648" — gets inverted or recontextualized: "name two consequences of 1648 the lecture contrasted with earlier treaties." The rewrite forces the retrieval the wording was giving away — the fact now has to be produced, not cued.

Invert the stale card. Given the answer, produce the condition: not "when was Westphalia" but "what treaty made state sovereignty the default." The inversion is the transfer — the same fact, different direction, which is the exam's angle. A card that only runs one way is half-tested.

Add one transfer, not five more facts. The weekly addition isn't more volume — it's one stem that takes the mechanism to a new story. "1648 → sovereignty" becomes "what earlier treaty failed to do this" — the contrast, not another date. The bank grows by depth, not by duplication.

The bank is a living set. Cards enter (new material), get rewritten (stale wording), get deleted (mastered), and transfer (new angle). A museum preserves; a bank practices. The pruning is the difference — the deck that changes is the deck that tests.

A 40-minute session on the clock

The weak answer mistakes speed for knowledge — the fast answer means the card is memorized, which means it's stopped testing. The strong answer kills or inverts: the date is burned in, so the card becomes "what did 1648 change" — the consequence, the contrast, the exam's actual verb. The rewrite restores the retrieval the wording was bypassing; the card goes back to work instead of reciting itself.

Why the museum isn't a bank

A card you can answer from its wording is a card that's stopped working — the retrieval path is the card's shape, not the concept. Singing "1648" at "When was the Peace of..." is recognition of the card, not retrieval of the fact — the card has become a cue for itself. The museum preserves these dead cards; the bank prunes them.

The exam doesn't have the card's wording — it has a different stem, a different angle, a different verb. A card that only works in its own phrasing fails the exam's rephrasing; a card that's been inverted, recontextualized, and transferred works in any phrasing because the retrieval is the concept, not the cue. The museum's cards are dead text; the bank's cards are living tests — and only the living ones survive the exam's new wording.