Question bank leaks: your exam is on a homework site, now what

Assume the leak, then measure it

Search for your own questions the way a student would: paste distinctive stems, in quotes, into a search engine and into the major homework and study sites directly. Distinctive is the key word. A generic question about supply and demand proves nothing. Your question with its invented company name, its odd numbers, and its fourth distractor intact is your bank, photographed. Log what you find with URLs, dates, and the share of the live exam the leaked set represents. That share, not the existence of a leak, decides the response.

A leak covering ten percent of a deep bank is a maintenance task. A leak covering most of a shallow bank, the same thirty questions recycled every term, is an exam that has been measuring memorisation of the leak for longer than you know. Grade distributions usually confess it first: scores clustering tightly at the top on the leaked items and scattering normally on the rest.

Takedowns work on copies, not on knowledge

Copyright takedown processes at the major platforms do remove posted banks, and institutions should file them with the same routine as any rights enforcement: identify, document, request, confirm. Understand what you bought. The copy comes down. The students who downloaded it, the group chats holding the file, and the answer keys already printed keep working. A takedown is hygiene, like closing a door after the corridor has copied the key.

That is not an argument against filing. It is an argument against filing only. Every takedown that is not paired with item replacement tells the cohort the institution believes the problem is solved, while the leak's value quietly transfers to whoever saved a copy, which is the worst distribution of advantage an exam can have: identical preparation available, priced by who thought to download it.

Bank design that survives being read

Depth is the defence. A bank where each learning objective carries many items, drawn randomly so two students rarely share a full paper, makes a leaked subset a study guide to a fraction of an exam. Parameterised items go further: same concept, different numbers, names, and scenarios per draw, so a memorised answer key fails on its face. Writing parameterised items costs more at the desk and pays back every term, in integrity and in item writing workload alike.

Retire and refresh on a schedule tied to exposure, not to suspicion. High stakes items that have been live for years should rotate whether or not you have found them posted, because your search is a sample, never a census. Track item statistics for the signature of memorisation: an item that got steadily easier across terms while its neighbours did not is an item the cohort has seen.

The demand side deserves one paragraph without scolding. Students buy and share banks where the exam is predictable, the stakes are brutal, and the teaching felt disconnected from the test. Institutions cannot fix all of that inside an integrity policy, but they can stop pretending a recycled bank under those conditions measures learning. It measures access. The fix list runs from item depth to exam design, and the integrity office's job includes saying so upward, in those words.

Students, incentives, and the honest conversation

When you find your material, tell the current cohort plainly that banks have been refreshed and leaked preparation will mislead as often as it helps. It is true once you have done the refresh, and it converts your silent maintenance into deterrence. Students make rational choices about preparation. Give the honest route the better expected value.

Frequently asked questions