How we write and check every question
Every question here is original, written from the public exam guides and official documentation, and checked before it is published. This page shows exactly how, including the numbers behind your readiness range.
Where questions come from
- Each question is written for one task statement in the official exam guide for its exam. The guide version is shown on each exam page.
- Every option is explained, and every question links to an official source for the right answer: a documentation page or, for some questions, the official exam guide.
- No question is a real or recalled exam question, and none is copied from another question bank. We never ask for, accept or use exam content.
- Questions are drafted with the help of AI tools and checked against the official documentation before they are published.
Automatic checks on every question bank
A validator script checks each bank before it goes live:
- every question has an exam, a domain, a task statement, an explanation and a source link;
- every question has 4 or 5 different options, and none uses "all of the above" or "none of the above";
- source links point to official Anthropic, Claude or Model Context Protocol pages;
- the correct option is the longest one in no more than 35% of single-answer questions, so picking the longest option is not a winning strategy;
- correct answers are spread across positions, with no position holding more than 40%;
- near-duplicate questions are flagged and rewritten.
How a mock exam is built
- A mock has the real exam's number of questions and is timed at the real exam's pace. If a bank is still smaller than the real exam, the mock is shorter and the time is scaled to match.
- For Architect – Foundations, 4 of its 6 scenarios are drawn at random, as on the real exam, with 15 questions from each. The official guide does not say how many questions each scenario has, so this even split is our own choice.
- For the other exams, questions are drawn in proportion to the official domain weights.
- You get no feedback until you submit. Options are shuffled for every mock, so you learn the idea, not the position of the answer (A, B, C…).
How scoring and readiness work
- A multiple-response question counts as correct only when you pick exactly the right options.
- Once you have 20 answers for an exam, from practice and mock exams together (a question answered twice counts twice, and a mock question left blank counts as wrong), you see a range rather than a single number: a 90% Wilson score interval around your accuracy. It narrows as you answer more.
- The target is 80%. It is set high on purpose: the real exam's pass mark is a scaled score of 720, and Anthropic does not publish how many correct answers that needs.
- A domain with at least 5 answers and accuracy below 60% counts as a weak spot.
- "Ready to book" means all of these: the low end of your range is at 75% or more, every domain has answers and none is weak, and your latest mock exam scored 80% or more.
Why there is no 100–1,000 score
The real exams report a scaled score with a pass mark of 720, and Anthropic does not publish how questions convert to that scale. Any scaled score we showed would be invented, so we show your accuracy against a clear target instead.
Keeping facts and questions current
Exam facts were last checked against the official exam guides, FAQ and policies on 2026-11-15. Each exam page links to its official exam guide, and each booking, retake and renewal answer links to its own source. When Anthropic publishes a new exam guide, the domains, task statements and questions are updated to match.
If you think a question or fact is wrong, tell us: corrections come first.
Report a wrong question or fact
What we do not do
- Use real or recalled exam questions.
- Show an invented scaled score or promise that you will pass.
- Show fake user counts or testimonials.
Last updated 15 November 2026.
More about CertificatePilot
About · Contact · Privacy · Terms · Accessibility