Claude Certified Architect – Professional: what changes from Foundations
The Claude Certified Architect – Professional exam (CCAR-P) is 63 questions, and more than a quarter of them are about governance, stakeholders and the life of a system rather than code. How it differs from Architect – Foundations, what its sample questions reward, and how to prepare.
Tecyfy · Published · 6 min read
The Claude Certified Architect – Professional exam (CCAR-P) tests whether you can own a Claude system from the first meeting to the day it is handed over. That is more than building it. Of its 63 questions, 28% sit in 2 domains with almost no code: governance, safety and risk (14%), and stakeholder communication and lifecycle (14%). Integration is the largest domain, at 19%.
The exam runs 120 minutes and costs $175. The pass is a scaled 720 out of 1,000, and each question says how many answers to choose. The guide recommends 3 or more years in systems architecture or platform engineering and 6 months with Claude or a similar system in production. It names solution architects, AI/ML engineers, tech leads and senior engineers as the audience, and says it is not for entry-level developers. Only staff of Claude Partner Network organisations can register.
How it differs from Architect – Foundations
You do not need the Foundations exam first, and the two test different jobs. If you are choosing between them, this is the difference:
| Architect – Foundations | Architect – Professional | |
|---|---|---|
| Questions | 60, inside 4 of 6 production scenarios | 63, across 7 domains |
| Fee | $125 | $175 |
| Claude Code | Its own domain, 20% | Part of developer enablement, 7% in all |
| Agents | Agent architecture and orchestration, 27% | Within solution design, 17% |
| Named in its guide only | Agent SDK loops, hooks, structured output, the 6 scenarios | Compliance, evaluation design, A/B tests, observability, discovery, documentation, SLAs |
| For | People who build agent systems | Architects who own enterprise systems end to end |
Foundations mostly asks how to make a given agent work. Professional also asks whether this should be an agent at all, how you will know it works, who signs off, and what happens after launch. The Foundations study guide covers the other exam.
The 7 domains
| Domain | Weight | What it adds over a builder's view |
|---|---|---|
| Integration | 19% | Capability bloat, authentication gaps, RAG chunking and indexing, MCP versus API or CLI versus agent-to-agent |
| Solution design and architecture | 17% | Workflow, agent or augmented LLM; multi-agent orchestration; tying the design to cost, productivity and SLAs |
| Evaluation, testing and optimization | 16% | Metrics for accuracy, latency, cost, safety and security; evaluation datasets; A/B tests; observability |
| Governance, safety and risk | 14% | Guardrails, human review, GDPR, HIPAA and FedRAMP, bias and transparency |
| Stakeholder communication and lifecycle | 14% | Discovery, explaining trade-offs, managing expectations and SLAs, documentation, handover |
| Models, prompting and context | 13% | Model trade-offs, system prompts, context and token use, reuse through caching, modular prompts and Skills |
| Developer productivity and enablement | 7% | Setting up Claude Code and other tools for a team, and helping with operational issues |
What the sample questions reward
The guide's 3 samples share one instinct: fix the cause, not the symptom.
Remove what the role does not need. A support agent can read tickets, draft replies, issue refunds and delete accounts, but staff only read and draft. The answer is to take the refund and delete tools away. Logging them or adding a confirmation step are the distractors, because they watch a risk that need not exist. This is least privilege, and it is also the "capability bloat" objective in the integration domain.
Put what repeats first, then cache it. Every request sends the same 8,000-token system prompt and policy, then a short user message. Put the static part first and turn on prompt caching, which reuses a repeated prefix and cuts both cost and time. Truncating the policy loses content you need.
Look at retrieval before the model. After a document refresh, a RAG system gives confident wrong answers while the model and latency are unchanged. Check the retrieval and indexing step first. Only something tied to the refresh explains a change that started with it.
Preparing for the part that isn't code
Most candidates are strong on the build and weak on the other 28%. For governance, know which controls prevent a problem and which only detect it, and when a person must review an output (human in the loop). For the stakeholder domain, practise writing a one-page decision record for a design you know: the options, the trade-off you chose, the SLA it promises and how you will measure it. For team enablement, know that Claude Code settings can be managed for a whole organisation as well as per project or per person.
For the technical domains, the AI engineering course covers RAG, evals and production in order. The free Architect – Professional practice questions are written to the guide's 7 domains. Our comparison of the 4 Claude certifications covers eligibility and retake rules.
Questions
- How do I prepare for the Claude Certified Architect – Professional exam?
- Prepare for an architect's whole job, not only the build. Integration is the largest domain at 19%, then solution design at 17% and evaluation at 16%, but governance and stakeholder communication are 14% each. Practise writing down architecture decisions and their trade-offs, designing an evaluation set, and choosing between removing a risk and monitoring it.
- How many questions are on the CCAR-P exam?
- 63 multiple-choice and multiple-response items in 120 minutes. Each item says how many answers to pick. The pass mark is a scaled 720 on a scale of 100 to 1,000.
- Do I need Architect – Foundations before Architect – Professional?
- No. The guide has no prerequisites. It recommends 3 or more years in systems architecture or platform engineering and 6 months with Claude or a similar LLM system in production.
- What is the difference between Claude Architect Foundations and Professional?
- Foundations sets its 60 questions inside 6 production scenarios and puts 20% on Claude Code. Professional has 63 questions across 7 domains, puts Claude Code inside a 7% domain on developer enablement, and adds governance, compliance, evaluation design and stakeholder communication.
- How much does the Claude Architect Professional exam cost?
- $175 per attempt, the most of the 4 Claude certifications. The credential lasts 12 months; renewing on time is a free assessment.
- Who can sit the exam?
- Only people at organisations in the Claude Partner Network, registering with a company email on a domain in their organisation's partner record.