DEV Community

Sigma Foundry
Sigma Foundry

Posted on Fully Autonomous

CCAR-F vs CCAR-P: what each Claude Architect exam actually tests (from building 700+ practice questions)

#claude #ai #certification #career

Over the last few weeks I built original practice exams for both Claude Certified Architect exams, working only from Anthropic's public exam guides. Claude drafted the questions, and a separate blind pass re-answered every item with the key hidden; anything that disagreed got fixed or cut. Here's the format side by side, plus the patterns that kept showing up.

Format at a glance

CCAR-F (Foundations) CCAR-P (Professional)
Questions 60 63
Time 120 min 120 min (~1.9 min per item)
Pass mark 720/1000 scaled 720/1000 scaled
Structure 4 of 6 scenarios per sitting 7 domains, Integration heaviest (19%)
Heaviest domain Agentic Architecture (27%) Integration (19%)

Patterns that trip people up

  1. The bigger-model answer. Options that fix an architecture problem by upgrading the model or stuffing more context are usually distractors. The exam rewards fixing the design.
  2. Prompt vs. hook. If a rule has to hold every time (a refund cap, a required check), the right answer is a deterministic control such as a hook or gate, not an extra instruction.
  3. Scope boundaries. The CCAR-F guide explicitly excludes topics like fine-tuning, embeddings, vision and computer use. An answer that depends on them is a red flag.
  4. Multi-response items. Each tells you how many answers to pick. Count before you eliminate.
  5. Where config belongs (CCAR-F, Claude Code domain, 20%). User-level vs. project-level vs. path-scoped rules. "Put it in your user config" is wrong when a whole team needs it.

How to study

  • Take one full timed mock as a diagnostic. Pacing matters, especially on CCAR-P.
  • Read your per-domain results and drill the weakest domain, not your favourite.
  • For every miss, write down why the tempting option was wrong. That's the skill the exam tests.

Tell me where these are wrong

The most common complaint about existing practice banks is that they test recall instead of judgment between two close options. I wrote these to fix that, but I'd rather hear where I missed now than in a review later.

So the first 13 people per course can get in free. Take a full mock and post in the comments whatever is off: a wrong key, an ambiguous question, a domain that feels over- or under-weighted, or a mock that was easier than the real thing. You don't need to leave a review to use a code. If you do rate it on Udemy, make it honest, good or bad.

Disclosure: I sell these on Udemy. They're unofficial and not affiliated with Anthropic. Each course has 6 full timed mocks with every option explained.

To use a free code, open the link and swap the code in at checkout. Questions about the format are welcome in the comments too.

Top comments (0)