New: The State of Code Review 2026 — 77,949 PRs analyzed. Read the report →

FAQ

Who takes the quiz?

Whoever reviews the pull request. Reviewsaur posts a link on the PR; a reviewer opens it, takes the quiz, and their result counts toward the required approvals.

What if a reviewer fails?

They get a fresh quiz — Reviewsaur generates a new set of questions about the same PR for them to try again, rather than re-serving the one they just failed. Each attempt is graded instantly with per-question feedback so they know what they missed, and they can keep trying until they pass.

The result page waits for the next quiz and links straight to it — usually in under a minute — so nobody has to go back to the pull request to find it. It is posted as a new comment on the PR too.

Do reviewers need a Reviewsaur account?

Yes. To take a quiz, a reviewer must be signed in to Reviewsaur (via GitHub) and be a member of the team that owns the repository. Opening the quiz link while signed out sends them to sign in first, and someone outside the team can't take it. This is how each result is reliably attributed to a specific person.

The link itself contains a unique token, but that only identifies which quiz it points to — it doesn't grant access on its own.

Can an admin override a quiz?

Yes. Admins can reset a quiz, or manually mark it as passed or failed, from the quiz's responses view. Use this for edge cases — a trivial change, an outage, or a reviewer who demonstrated understanding another way.

What does the quiz cover?

Questions are generated from the PR diff and focus on the parts of the change that matter most; cosmetic changes (styling, copy, renames) get none. Each question has a category: Understanding, Risk, Architecture, Trade-off or Verification. How deep they go is set per repository by the difficulty setting: Easy questions can be answered from a single hunk, Hard ones need several parts of the change put together. On Pro an admin can give one person a different difficulty. Files matching your excluded paths are left out, and if every changed file is excluded, no quiz is generated and the check passes automatically. Very small changes — a typo, a comment, a version bump — get a shorter quiz, and on Pro they can skip the quiz altogether.

What if I click "I don't know"?

If your admin turned the option on, every question has an I don't know button.

The question counts as wrong — there is no partial credit — but the correct answer and its explanation appear right away, and the question locks so it cannot be answered afterwards. The result page marks these apart from wrong guesses, so an honest gap reads differently from a bad answer.

Is my code used to train AI models?

No. We don't use your code or quiz data to train models, and our AI provider's commercial API terms exclude submitted data from model training. The diff is sent only to generate or grade a specific quiz and is discarded afterwards — see Where does your code go? for the full picture.

What exactly is sent to the AI provider?

The PR title and description, the diff (with your excluded paths stripped out first), the changed file paths, and recent commit messages. Your repository's name is not included in the prompt, and PRs over 50 files or 300,000 diff characters are rejected before any AI call. Grading a written answer sends only the question, its rubric, and the answer — not the diff.

How do I delete my data?

The raw diff is never stored — it's discarded once the quiz is generated. Quiz questions and results are kept while the repository is connected; removing a repo from Reviewsaur deletes its quizzes, responses, and settings with it. For account or team deletion, email [email protected].

Does it block merging?

Only once you make Reviewsaur Quiz a required status check in branch protection. See GitHub status check. Until then, quizzes are generated and linked but don't gate the merge.

What are the limits?

Free allows one active repository and 50 lifetime tests; Pro is unlimited. Full breakdown on Plans & limits.

Still stuck?

Email [email protected] and we'll help.