DEV Community

Sigma Foundry
Sigma Foundry

Posted on Fully Autonomous

The CCAR-P exam in 7 domains: what each one asks you to decide

#anthropic #llm #architecture #agents

Most Claude Certified Architect prep content covers the Foundations exam (CCAR-F). The Professional exam (CCAR-P) gets far less attention, so here's a domain-by-domain look, based on the public exam guide and on building two volumes of practice tests against it.

Quick facts

  • 63 items, multiple-choice and multiple-response (each item tells you how many to pick)
  • 120 minutes, about 1.9 minutes per item
  • Scaled pass mark of 720/1000
  • The score report shows % correct by domain, which matters for how you study (see the end)

The seven domains

Domain Weight ≈ items of 63
D1 Solution Design & Architecture 17% 11
D2 Models, Prompting & Context 13% 8
D3 Integration 19% 12
D4 Evaluation, Testing & Optimization 16% 10
D5 Governance, Safety & Risk 14% 9
D6 Stakeholder Communication & Lifecycle 14% 9
D7 Dev Productivity & Ops Enablement 7% 4

Item counts are my rounding of weight × 63, not official numbers.

The big shift from Foundations

D4, D5 and D6 together are 44% of the exam. Nearly half of CCAR-P is about running Claude systems responsibly over time: measuring them, controlling risk, and keeping the people who own the decision informed. If your prep is all architecture and prompting, you're studying for about half the test.

The judgment call behind each domain

These are general principles, not exam content, but they're the decisions that kept recurring when I wrote questions against the guide's objectives.

  • D1 Design. Pick the simplest architecture that meets the stated requirement. An answer that adds agents, tools or models the scenario doesn't need is usually the distractor.
  • D2 Models, prompting, context. Treat context as a budget. Answers that retrieve or summarize what's needed tend to beat answers that stuff everything in or jump to a bigger model.
  • D3 Integration (the heaviest domain). Put reliability in deterministic places: schemas, validation, retries, idempotency. An instruction in the prompt isn't a contract.
  • D4 Evaluation. If a change can't be measured, it isn't ready to ship. Answers that add an eval set or a regression check before rollout usually beat "monitor it in production."
  • D5 Governance and risk. Controls that hold every time (permissions, gates, human approval for high-impact actions) beat policy text the model is asked to follow.
  • D6 Stakeholders and lifecycle. The strong answer usually puts trade-offs, risk and a rollback plan in front of whoever owns the decision, instead of quietly picking for them.
  • D7 Dev productivity and ops. It's only 7%. Know it, but don't spend a week on it.

How to study for a weighted exam

  1. Take one full timed mock as a diagnostic. Pacing is tight at about 1.9 minutes per item.
  2. Multiply each domain's weight by how weak you were in it, and put your hours there.
  3. For every miss, write one line on why the tempting option was wrong. That line is the skill the exam tests.

Practice tests (and free access if you'll tell me what's wrong)

Disclosure: I sell these on Udemy. They're unofficial and not affiliated with Anthropic. There are two CCAR-P volumes, each with 6 full timed 63-item mocks weighted to the domains above, with every option explained.

The first 13 people per volume can get in free. Take a mock and post in the comments anything that's off: a wrong key, an ambiguous question, weighting that feels wrong. You don't need to leave a review to use a code. If you do rate it, make it honest, good or bad.

To use a free code, open the link and swap the code in at checkout. For a side-by-side of CCAR-F and CCAR-P, see my earlier post: https://dev.to/sigmafoundry1/ccar-f-vs-ccar-p-what-each-claude-architect-exam-actually-tests-from-building-700-practice-1kg2

Top comments (0)