The usual bar
Looks like the real thing
Right subject, right format, right number of options. This is where most question banks stop.
- Matches the paper's layout
- Uses the right question stems
- Has a plausible answer key
Question standards
Accurate practice questions are made, not found.
Anyone can put a thousand questions behind a paywall. The hard part is making sure each one behaves like the real paper — that the right answer is right, the wrong answers are wrong for the reasons the exam cares about, and the difficulty is where it claims to be.
8
review stages before a question can publish
6
difficulty drivers, scored blind
2
independent checks on the answer and the explanation
0
questions that publish without a person signing them off
Why this page exists
A bad practice question is worse than no practice question.
It doesn't just waste the twenty minutes you spent on it. It teaches you a method that won't work on the day, and it moves your confidence in the wrong direction.
Our bar
Behaves like the real thing
The reasoning it demands is the reasoning the exam actually rewards. This is the part that takes eight stages.
- Every wrong option is wrong for a named reason
- No shortcut reaches the answer without the method
- Difficulty is rated blind, against a floor
The process
Eight stages, and a question can die at any of them.
Most of these stages exist to catch a specific failure we have actually seen. Nothing here is ceremony.
Content teamAutomated checksReview panelSubject specialist
- 01
Blueprint
Content teamNo question starts from a blank page. Each one is commissioned against a brief: the reasoning skill it has to test, the question pattern it follows, and the difficulty band it has to land in.
Those patterns were catalogued by hand from official past papers, question by question, before a single new question was written.
- 02
Authoring
Content teamThe question is written to that brief in full — stem, options, answer key, a worked explanation, and a separate rationale for every wrong answer.
If the writer cannot say why a wrong option is tempting, it is not a distractor. It is noise, and it comes out.
- 03
Pattern control
AutomatedCan rejectBefore anyone reads it for meaning, the question runs against a list of house rules that reject the tells — the small habits that let a candidate answer without reading properly.
Answer positions are reshuffled so the key never drifts towards A. Wording that quietly defuses the question, like telling you a figure can be ignored, is rejected outright.
- 04
Adversarial attack
Review panelCan rejectA reviewer then tries to break it. Not to solve it properly — to find the cheapest route to the right answer without doing the reasoning the question claims to test.
Find a shortcut and the question is rewritten to close it. If it survives into the final draft, the question loses its difficulty rating and drops a band.
- 05
Answer verification
Review panelCan rejectThe question is solved again, independently, with the key revealed. The marked answer has to survive a majority verdict from reviewers working separately.
Agreement is counted, not assumed. A split decision sends the question back rather than through.
- 06
Explanation verification
Review panelCan rejectThe explanation gets the same treatment as the answer, line by line, including the rationale attached to every single option.
A right answer with a wrong explanation teaches you a method that fails on the next question. That is worse than no question at all.
- 07
Difficulty calibration
Review panelCan rejectDifficulty is scored blind against six drivers by a rater who never sees what the author thought the question was worth.
It has to clear an absolute floor and land within one band of its target. Miss either and it goes back — and at the top of the range, two raters score it and the results are averaged.
- 08
Specialist sign-off
Subject specialistCan rejectA specialist in that exam reads the finished question end to end — stem, key, explanation, every rationale — and signs it off before it can go anywhere near a student.
Nothing publishes on its own. The sign-off is recorded against the reviewer and the question, and it cannot be edited afterwards.
What gets thrown away
The questions you never see.
Rejection notes are written in plain language and kept with the question forever. Here are four of the reasons that come up most.
The stem gave the game away
The wording told the candidate which figure to ignore. The question was supposed to test whether they could work that out themselves.
There was a shortcut
You could reach the key by comparing two numbers and skipping the method entirely. Every reasonable candidate would have taken it.
Right answer, wrong reason
The explanation arrived at the correct option by a route that happens to work here and fails everywhere else.
Too easy for where it sat
Labelled hard by its author, rated three points under the floor by a blind rater. It went back to be rebuilt rather than quietly relabelled.
Two strikes
A question that fails once goes back to be rewritten. A question that fails twice is not rewritten again — it is retired, and the reason stays on file.
Calibration
Difficulty is measured, not asserted.
Calling a question hard does not make it hard. Each one is scored against six drivers on an eighteen-point scale, blind, and has to clear a floor before its label means anything.
Blind difficulty rating
Worked example
15/18
Rated hard
ST
Reasoning steps
How many linked steps stand between the stimulus and the answer.
3/3
CC
Concepts combined
Whether it tests one idea or forces two together.
2/3
DH
Data handling
How much material has to be read, filtered and held at once.
3/3
CB
Calculation burden
The arithmetic load once the method is clear.
2/3
DS
Distractor subtlety
How closely the wrong options shadow real candidate errors.
3/3
IN
Insight demand
Whether the route is visible, or has to be found.
2/3
Automatic rejection, whatever the total
- STA question with no reasoning chain is rejected, however long it takes to answer.
- DSA question whose wrong options nobody would pick is rejected, however hard the maths.
The author scores the question too. If the two scores disagree badly, that is treated as a fault in the question rather than a difference of opinion.
The paper trail
Every check leaves a permanent record.
Each stage writes a row against the question: what was checked, who or what checked it, the verdict, and the reason if it failed. Those rows cannot be edited afterwards.
It means a question is never just approved. It is approved by someone, on a date, against a standard, and that is still true a year later.
Verification recordCannot be edited
- Pattern controlAutomatedPass
- Adversarial attackshortcut found — rewrittenReview panelFail
- Adversarial attackReview panelPass
- Answer verificationReview panelPass
- Explanation verificationReview panelPass
- Difficulty calibration14/18 — hardBlind raterPass
- Specialist sign-offSubject specialistPass
The last reviewer
You are the ninth stage.
Eight stages catch most things. They do not catch everything, and a bank that pretends otherwise is a bank that never improves.
So every question in the platform carries a report button. It sits in practice and in the mocks, all it takes one tap.
Report this question
- The marked answer looks wrong
- Inaccurate data or calculation
- Missing information
- Unclear or ambiguous wording
- Broken display or formatting
- Something else
Add a note (optional)
What happens next
- 1
You flag it
One tap, mid-question, with an optional note. Your session carries on where it was.
- 2
It reaches a specialist
Reports land in a review queue with the question, your note and the category attached.
- 3
A person decides
Reports are not actioned automatically. A specialist re-checks the question and either fixes it, retires it, or explains why it stands.
- 4
The record is permanent
Every report is kept, resolved or not. Repeat reports on one question are the strongest signal we get.
Where the standard comes from
We publish the real papers too.
Our questions are modelled on official past papers, so the honest test of whether we got it right is whether you can sit a real one and recognise the shape of it.
The past papers we hold are free and totally ungated.
Browse the past papersQuestions about the questions
The things people ask before they trust a question bank.
Are GhostPrep practice questions written by real people?
Yes. Every question is commissioned against a written brief, authored in full with a worked explanation and a rationale for each option, and reviewed by a subject specialist before it can publish. Automated checks sit in the middle of that process to catch mechanical faults — answer-position drift, wording that defuses the question, formatting breaks — because those are caught more reliably by a rule than by a tired reader.
How do you know a practice question matches the real exam?
Every question is built to a pattern taken from official past papers, then rated blind against a rubric that was itself calibrated on real questions from those papers. The check that matters most is the one you can run yourself: our past papers are free and ungated, so you can sit a real one and see whether our questions feel like it.
How is question difficulty decided?
Difficulty is scored against six drivers on an eighteen-point scale — reasoning steps, concepts combined, data handling, calculation burden, distractor subtlety and insight demand. The rating is done blind, by someone who has not seen what the author thought the question was worth, and it has to clear an absolute floor as well as land in the right band. Questions at the top of the range are rated twice and the scores averaged.
What happens if I find a mistake in a question?
There is a report button on every question, in practice and in mocks. It takes one tap, it does not interrupt your timer, and you can add a note. A specialist re-checks the question and either corrects it, retires it, or records why it stands. Reports are kept permanently and cannot be deleted, so repeat reports on the same question are visible.
Do you use official past paper questions in the question bank?
No. The question bank is original material written to the same standard. Official past papers are published separately and clearly labelled as what they are, free to download, so you always know whether you are sitting a real question or one of ours.
How many questions do you throw away?
A meaningful share never reaches the bank. The most common reasons are a shortcut that reaches the answer without the method, an explanation that is right by accident, and a difficulty rating that comes back well under what the question was written to be. Questions that fail twice are retired rather than rewritten again.
Pick your test
Sit one and see. No card required.
Free sample questions for every exam, with the full worked explanation on each one.