How our questions are made
A wrong answer key is the worst thing a question bank can do to a candidate: it does not just cost a mark, it teaches a wrong fact that gets applied on exam day. This page describes exactly what stands between a draft question and your screen, including where the coverage stops.
1. Every question is written against a Learning Outcome Statement
Questions are not written to a topic, they are written to a specific LOS, and the code stays attached to the question for its whole life, and you can see it on every question on this site, including the free samples. That is what makes it possible to say a subtopic is genuinely covered rather than merely represented.
The format is fixed to the exam's: three options, one correct, no "all of the above". The wrong answers are written to be the results of specific, common mistakes. The value of a practice question is almost entirely in whether its distractors are the errors you would actually make.
2. A deterministic validator runs before anything is imported
Every question passes an automated check with no AI involved: schema and format, a valid topic and subtopic against the live list, a LOS code, a real explanation, and duplicate detection against the entire existing bank so the same question cannot enter twice under different wording.
Anything that trips a rule does not get published. Depending on severity it is held for human review or blocked outright, and it stays visible in a review queue. Nothing is silently dropped, because a question that vanishes is a question nobody fixes.
3. A second model re-solves the question blind
For every published question, an independent AI model is given the question without the answer key and asked to solve it. Its answer is then compared with ours. Agreement is weak evidence; disagreement is a flag, and those go to a stronger model and then to a person.
This has been run on 1,452 of 1,452 published questions (100%). It is a gate, not a survey: a question the audit cannot confirm does not stay on the quiz. It is pulled back into the review queue for a person, whether it is being imported or has already been live for months.
4. Real answers keep checking us after publication
Once a question is live, how candidates answer it becomes evidence. A question that nearly everybody gets wrong is usually not a hard question. It is usually a wrong answer key, or an ambiguous stem. A question that nearly everybody gets right is not teaching anything.
Both patterns are surfaced automatically and pulled back for review. Any user can also flag a question from inside the app, and those go to the same queue.
What we do not claim
- We are not affiliated with, endorsed by, or connected to CFA Institute. CFA® and Chartered Financial Analyst® are trademarks owned by CFA Institute.
- Our questions are not past exam questions. Nobody's are: CFA Institute does not release them.
- Our difficulty levels are our own judgement. No third-party provider knows the real exam's difficulty distribution, and a score here does not predict a result there. Use CFA Institute's mocks as the reference. See how to judge readiness.
- Questions are drafted with the help of large language models and then validated, audited and reviewed as described above. We would rather say so than let you discover it.
Figures on this page are generated from the live question bank, last updated .