Pass the Claude Architect Professional exam
A primer on what this exam really is — a judgment test, not a recall test — then seven guides, one per exam domain, covering every objective of the official exam guide. The design decisions the scenarios hinge on, the numbers worth knowing, and the traps: wrong answers that sound sensible. About 81 minutes of reading, and the credential lasts a year.
The exam at a glance
Claude Certified Architect — Professional (CCAR-P) is the advanced tier of Anthropic’s architect track — and it tests something the foundation exams do not: organization-level design judgment. Model selection, RAG pipelines, evaluation, governance, stakeholders. Almost no flags, files or configuration trivia; the questions are scenarios, and the wrong answers are the decisions experience has already taught you not to make. We sat it with zero preparation and passed — not because it is easy, but because shipping AI systems is the preparation.
No prerequisites: you can sit Professional without holding Foundations, and the credential is awarded on exam performance alone — but eligibility runs through a Claude Partner Network organisation. If you don’t pass: wait 14 days before your second attempt, 30 before your third, 90 before your fourth — up to four attempts per rolling year. The exam-day logistics are in the prep post; these guides are the study material.
The seven domains
Study where the marks are. Integration (19%), Solution design and architecture (17%) and Evaluation, testing and optimization (16%) — the three highest-weighted domains — carry 52% of the exam between them, so read them first. Note the order below follows the official exam guide, not the weights: Integration is Domain 03 even though it is the heaviest. Developer productivity and operational enablement is the lightest at 7% — one focused sitting. Each guide page lists the objectives it covers, in exam-guide order.
What the Professional exam actually is (a judgment test), how it differs from Foundations, and the study order — the floor the seven guides build on. Five minutes; skip it if you have shipped AI past the pilot stage.
| Domain | Weight | Modules | Read | Study priority |
|---|---|---|---|---|
| 01Solution design and architecture | 17% | 6 | 12 min | 2nd of 7 |
| 02Claude models, prompting and context engineering | 13% | 5 | 12 min | 6th of 7 |
| 03Integration | 19% | 8 | 15 min | 1st of 7 |
| 04Evaluation, testing and optimization | 16% | 6 | 10 min | 3rd of 7 |
| 05Governance, safety and risk management | 14% | 5 | 10 min | 4th of 7 |
| 06Stakeholder communication and lifecycle management | 14% | 5 | 9 min | 5th of 7 |
| 07Developer productivity and operational enablement | 7% | 3 | 5 min | 7th of 7 |
The architect’s core move: turning a business ask into a system — choosing between augmented LLM, workflow and agent, designing the feedback loop, and aligning the design to value the business recognises.
Choosing the model the task deserves, writing prompts as engineered artifacts, budgeting the context window like the scarce resource it is, and making reuse pay: caching, modular prompts, Skills.
The heaviest domain: least-privilege tool design, the auth gaps agents create, accuracy–latency trade-offs, observability at scale, RAG pipelines end to end, retrieval strategy, and choosing between MCP, direct APIs and agent-to-agent.
How you prove a system works before it ships and keep proving it after: multi-metric definitions, real-traffic datasets graded by humans, code and calibrated LLM judges, A/B discipline, failure diagnosis, and the optimization levers.
Guardrails as a layered stack rather than a paragraph, the failure modes LLM systems actually exhibit, human review designed as a real control, compliance regimes mapped to design decisions, and the ethics questions that precede them.
The domain the developer track does not have: structured discovery, trade-offs communicated to three different audiences, honest SLAs for probabilistic systems, documentation that outlives the pilot, and the lifecycle phases that connect them.
The smallest domain: rolling Claude Code and agentic tooling out to a team as a managed capability — standard configuration, workflows where it pays first, and the support loop that keeps it healthy.
How to use this guide
Read the domains in weight order — that is what the study-order toggle above is for — because hours spent should follow marks available. Each page stands alone: the design concepts taught from the ground up, the decision tables the scenarios are written against, and a traps section collecting the wrong answers that sound sensible.
Prerequisites: shipped experience. This guide assumes you know what an API call looks like, what a system prompt is, and roughly what tokens cost. If the mechanics underneath are the gap, the CCAR-F guide covers them — flags, files, defaults — and its agentic-architecture, tools-and-MCP and context domains are the technical floor this one builds on. But CCAR-P needs none of that memorised: it tests what the mechanics are for.
How this was written and checked: by Fabs, who sat all three Anthropic certifications in 2026 — Developer, Architect Foundations, Architect Professional — and wrote the Professional up after passing it three hours after booking. The structure follows the official exam guide (version 1.0, effective July 2026) and covers every objective it lists; exam format, domains and policies were verified against the official pages in October 2026; the technical facts are anchored in Anthropic’s documentation. We are not affiliated with or endorsed by Anthropic.
Models, prices and policies move — if you are reading this in six months, re-check anything numeric against the official pages before you memorise it.
One honest caveat: a certificate proves you studied, not that you can ship. But this exam comes closest of the three to proving you have shipped — the domains that feel least like an exam, governance and stakeholders, are the ones that matter most in the job. That is the half of AI architecture nobody outsources to a model.
Where to verify us
Anthropic’s own material is free and it is the source of truth — the exam guide PDF above all, because every question is written against its objectives. This guide is the compression: same syllabus, about 81 minutes, the decision tables in one place, the traps marked. Already holding the developer or foundations certifications? The CCDV-F and CCAR-F guides cover those exams the same way.
Registration and the link to the exam guide — the authoritative source for domains, weights and policies.
Version 1.0: the blueprint with every objective, the sample questions, scoring and retake policies. Read it before scheduling.
Where the design-level facts live — retrieval, caching, evaluation, governance. Skim it for gaps, not coverage.
One of the few third-party CCAR-P resources — a study guide and practice sets for the professional tier.
A free unofficial GitHub repo covering all four Anthropic exams at a glance. Useful for orientation.