What the exam is

CCA-F tests whether you can make sound design decisions when building agents on Claude. It is the foundations tier of Anthropic's architect certification. The questions are scenarios, not vocabulary checks: you are shown a situation an agent system runs into and asked which approach handles it best. You are rewarded for judgment about trade-offs, failure handling, and clean interfaces between components, not for memorizing definitions.

Format at a glance

• 60 scenario multiple-choice questions in one session.
• 120 minutes, which works out to about two minutes per question.
• Scored on a 0 to 1,000 scale. The pass mark is 720 out of 1,000.
• Judgment-based: most questions have a clearly best answer and tempting near-misses.

Our practice mock mirrors this exactly: 60 questions, a 120-minute timer, the same 720 pass line, and an explanation for every option so you learn why the near-misses fall short.

The four domains

Our mock splits its 60 questions evenly across four domains, 15 each. Each domain maps to a kind of system an agent architect is expected to reason about.

The full domain pages go deeper on what each one covers and why it matters. Browse the domains.

A study plan that uses this mock

1. Take one full timed run first, cold. A baseline under the real clock tells you where you actually stand and gets you used to the two-minute-per-question pace.
2. Review every explanation, including the ones you answered correctly. The reasoning is the point. Note the trap pattern behind each wrong option so you recognize it next time.
3. Drill your weakest domain. After a run you see a per-domain breakdown, so target the 15 questions where you lost the most points and work them untimed until the reasoning clicks.
4. Use the flashcards for quick recall between full attempts. They are good for short, repeated sessions when you do not have a free two hours.
5. Re-sit the full timed mock a few days later. Aim to clear 720 comfortably, not by one point, before you book the real exam.

Exam-day tips

• Read the whole scenario before the options. The constraint that decides the answer is often in the last sentence.
• Pick the answer that handles failure honestly. Options that hide errors or skip escalation usually lose.
• Watch the clock loosely. Two minutes per question is plenty if you flag the hard ones and come back rather than stalling.
• Prefer the simplest design that meets the requirement. Over-engineered answers are common traps.
• Answer everything. There is no benefit to leaving a question blank, so make your best call and move on.