Do two candidates really get different papers?

Yes, when the bank has more questions than the rules draw. Six drawn from nine gives a large number of distinct combinations, and shuffling a section reorders what is drawn. If the bank holds exactly as many questions as the blueprint asks for, everybody gets the same paper — which is why the coverage table is worth reading first.

Can I still hand-pick specific questions?

Yes. A section can name question ids directly instead of drawing. Mixing the two is normal: a compulsory case study everybody must answer, plus objective questions drawn at random.

What happens to a question I edit after the paper has been sat?

The sitting keeps what it served. A candidate is marked against the question as it was when they answered it — editing the bank does not rewrite anybody's completed paper.

How many questions should a bank hold?

Enough that the draw is not deterministic. As a rule of thumb, three to four times what a paper draws per rule: a rule asking for two easy questions on a topic wants six to eight there. Below twice, most candidates will see most of the bank.

Can two teachers contribute to one bank?

Yes — the bank belongs to the course, not to a person. Which is the point: it outlives whoever set this year's paper.

Question Bank & Randomised Exam Paper: How to Build One

Ninety seconds in the running product: building the bank, drawing a paper from it, and what is recorded while somebody sits it.

Why a bank and not a document

The usual way to set a paper is to write twenty questions into a document and give every candidate the same twenty. That has two costs. The first is that the questions are used once: next term somebody writes twenty more, and the good ones from last year are in a file nobody opens. The second is that identical papers make copying worth the effort — one answer key, passed around, works for everybody.

A bank fixes both. Questions are written once and drawn from for years, and because every candidate is dealt a different selection, an answer that reaches the wrong person is worth almost nothing.

Getting questions in

An empty bank is the normal starting point, and it offers the shortest way out of itself: adopt a quiz you have already written. Every question in it moves into the bank and becomes reusable, so a department with three years of quizzes has three years of bank without typing anything.

The empty Question bank screen, explaining that questions outlive the paper they were written for and that every candidate's paper is drawn from here, with an 'Adopt from a quiz' button offered as the way to fill it. — An empty bank names the migration path rather than leaving you to find it.

Tag as you go, because the blueprint draws on tags

Two fields do the work: topic and difficulty. They are not decoration — a draw rule asks for “two easy questions on data structures”, and a question with no topic is invisible to that rule. A bank tagged carelessly produces papers that are technically valid and pedagogically wrong: all four objective questions from one unit, none from the rest of the syllabus.

Difficulty must be easy, medium or hard. Anything else is refused with a message rather than accepted and quietly dropped — which is what used to happen. A department that tagged its bank “moderate” ended up with every question untagged and blueprints that could not be filled for no visible reason.

The question bank with nine questions, showing a Coverage table of topic against difficulty — Algorithms, Data structures, Databases and Trees across Easy, Medium and Hard — and a list of questions with their type, topic, marks, how many times each has been served, and its facility. — The coverage table is the one to read before setting a paper. An empty cell is a draw rule that cannot be filled.

Two more columns matter over time. “Served” counts how many candidates have been dealt this question, so an over-used question is visible before it becomes the one everybody has seen. “Facility” is the proportion who got it right once it has been sat — a question at 5% is probably broken rather than hard, and one at 98% is not discriminating between candidates.

Describing the paper

A blueprint is a list of sections, and each section is a list of draw rules. A rule says: draw this many, from this topic, at this difficulty. Leave difficulty blank and it draws from any. Sections carry their own instructions, and can be marked “answer any N” for the written half where a candidate chooses between essays.

The paper builder showing a blueprint with two sections: Section A objective with three draw rules — two easy Data structures, one medium Algorithms, one medium Trees — and Section B written with a rule drawing two Databases questions of which the candidate answers any one. The footer reads six questions per paper. — Two sections, six questions per candidate, drawn from nine in the bank. Change the bank and the paper changes with it — there is no sheet to reissue.

When the bank cannot fill it

This is the failure the software is most useful for catching, because it is invisible on paper: a rule that asks for twenty hard questions on trees when the bank holds one. Left alone it does not produce an error on exam day — it produces a short paper, or the same handful of questions for everybody, and nobody notices until the marks come back strange.

The paper builder showing a warning that the bank cannot fill this blueprint, naming the section and that it asks for one but only zero match, with a 'Save anyway' button beside it. — It names the section and the shortfall, and still lets you save. A blueprint written before the bank is full is a normal way to work.

“Save anyway” exists deliberately. Writing the blueprint first and filling the bank against it is a reasonable order of work, and software that refused to store an unfinished plan would just be pushing the plan back into a spreadsheet.

Do two candidates really get different papers?: Yes, when the bank has more questions than the rules draw. Six drawn from nine gives a large number of distinct combinations, and shuffling a section reorders what is drawn. If the bank holds exactly as many questions as the blueprint asks for, everybody gets the same paper — which is why the coverage table is worth reading first.
Can I still hand-pick specific questions?: Yes. A section can name question ids directly instead of drawing. Mixing the two is normal: a compulsory case study everybody must answer, plus objective questions drawn at random.
What happens to a question I edit after the paper has been sat?: The sitting keeps what it served. A candidate is marked against the question as it was when they answered it — editing the bank does not rewrite anybody's completed paper.
How many questions should a bank hold?: Enough that the draw is not deterministic. As a rule of thumb, three to four times what a paper draws per rule: a rule asking for two easy questions on a topic wants six to eight there. Below twice, most candidates will see most of the bank.
Can two teachers contribute to one bank?: Yes — the bank belongs to the course, not to a person. Which is the point: it outlives whoever set this year's paper.