Guide

What a question bank is, and when you actually need one

A question bank is not a folder of past papers. It is a library where each question exists once, can be reused across papers, and carries enough history that a report written last year still means something.

The problem it solves

Left alone, question sets rot in a specific way. The same question ends up in four papers with slightly different wording. Nobody knows which version a cohort actually answered. A question is quietly reused with a batch that has already seen it. When someone challenges an answer key, the paper that was sat no longer exists in a recoverable form.

A bank fixes those by making the question — not the paper — the unit that is stored.

What separates a bank from a spreadsheet

De-duplication: Exact duplicates should be refused outright. Near-duplicates — the same question reworded or repunctuated — need fuzzy matching, because those are the ones that actually accumulate.
Versioning: Editing a question should append a version rather than overwrite it, so an attempt graded six months ago still resolves to the text that was on screen.
Usage tracking: Which papers a question has appeared in, so you can avoid showing a batch something it has already answered.
Translations as one question: A bilingual paper should draw one question and deliver it in the candidate’s language, not maintain two parallel banks that drift.
Structured metadata: Topic, difficulty, tags and status — the fields you filter on when assembling a paper to a blueprint.

When you need one

One paper a term, thirty questions, one author: a spreadsheet is fine, and a bank is overhead.

The threshold is usually crossed by one of three things — several authors adding questions, papers that need to be assembled to a blueprint rather than written start to finish, or an answer key being challenged and nobody able to reproduce what was asked. Any one of those is the signal.

Migrating what you already have

Export existing papers to a spreadsheet with one question per row and options in their own columns.
Import in batches — one paper at a time — so a validation failure is easy to locate.
Read the duplicate report. Overlap between old papers is the point of the exercise, not a problem with the import.
Tag as you go. Retrofitting topic and difficulty onto a thousand untagged questions is the part everyone underestimates.
Only then assemble a new paper from the bank, and check it against your blueprint.

Common questions

How is a question bank different from keeping past papers?

A past-paper folder stores documents; a bank stores questions. That lets one question be reused across papers, de-duplicated against near-identical copies, versioned so old reports stay reproducible, and tracked so you know which cohorts have already seen it.

Do I need a question bank for a single small test?

Probably not. It starts paying off with multiple authors, papers assembled to a blueprint, or a need to reproduce exactly what a candidate was asked months later.