The CCAR-P exam is a breadth exam for architects who make full-lifecycle decisions on production Claude deployments. It covers seven domains, and no single domain carries more than 19% of the weight, so you can't pass on one strength. This guide covers the format, what each domain tests in practice, and how to study for it.
The format facts and domain weights below come from the published exam guide (v1.0). The domain summaries and study angles are Sigma Foundry's own guidance. Check the guide itself before you book, because details can change between versions.
The format at a glance
- Items and time: 63 items in 120 minutes, or about 1.9 minutes per item.
- Item types: multiple-choice and multiple-response.
- Scoring: scaled, with a pass mark of 720 on a 1,000-point scale.
The seven domains and what they test
- Integration: 19%
- Solution Design & Architecture: 17%
- Evaluation, Testing & Optimization: 16%
- Governance, Safety & Risk: 14%
- Stakeholder Communication & Lifecycle: 14%
- Claude Models, Prompting & Context: 13%
- Developer Productivity & Operations: 7%
Our advice is to prepare for applied judgment calls more than definitions. Most of what follows is about choosing, diagnosing and justifying, not reciting.
Solution Design & Architecture (17%)
Translating business needs into Claude solutions. In practice this means choosing between an augmented LLM, a workflow and an agent based on how predictable the steps are, not on how advanced the option sounds. It also covers multi-agent orchestration, task decomposition and tying each design choice to a named business-value pillar.
Claude Models, Prompting & Context (13%)
Model selection trade-offs, system prompts, prompt engineering, context optimization and token strategy. Expect questions where a bigger context window is offered as the fix. The durable fix is usually active state management, such as a facts block or checkpointing.
Integration (19%)
The heaviest domain. It spans tool configuration, retrieval design, MCP and observability, all under accuracy, latency and cost trade-offs. Practice fixing a bloated tool registry, matching chunking and retrieval to the data, and picking the right monitoring layer for a stated need.
Evaluation, Testing & Optimization (16%)
Metrics, evaluation datasets, A/B testing, debugging failures and production monitoring. You'll need to separate prompt failure from hallucination from model mismatch before choosing a fix. You'll also need to spot a broken experiment design, such as an A/B test that changes the model and the prompt at the same time.
Governance, Safety & Risk (14%)
Guardrails, failure modes, human-in-the-loop validation, regulatory compliance and ethical AI. The recurring theme is that a well-written system prompt alone is not a guardrail. Rules that must always hold belong in deterministic layers, such as a hook that runs before the tool call.
Stakeholder Communication & Lifecycle (14%)
Discovery, communicating decisions, SLA alignment, documentation and lifecycle management. Prepare to present trade-offs in business terms rather than benchmark scores, and to treat deployment as one phase in a continuing loop.
Developer Productivity & Operations (7%)
Configuring Claude tools for teams, developer workflows and debugging support. It's the smallest domain, about 4 of 63 items if weights map to item counts. Know it well enough to collect those points, but don't over-invest.
How to study
Learn the anti-patterns
Wrong options on an architecture exam tend to be plausible shortcuts. Learn these by name:
- Enforcing a hard business rule, like a refund cap, with prompt wording instead of code or a hook.
- Fixing misrouting or context exhaustion with a bigger model or a bigger window.
- Using an agent where a fixed workflow is cheaper, faster and easier to debug.
- Checking a policy after the tool has already run.
- Treating a 200 OK response as success without checking
stop_reason. - Changing two variables at once in an A/B test.
- Upgrading to a new model version without re-running the eval suite.
- Trusting an aggregate metric that hides a failing subgroup.
- Having the session that generated the output also review it.
When you miss a question, write down which anti-pattern the wrong option used. The pattern repeats across domains.
Weight your time, then cover everything
Integration, Solution Design and Evaluation together make up 52% of the exam. Put your first study hours there, but remember that with seven domains and a flat spread, a gap anywhere costs points.
Use timed mocks to train pacing
At about 1.9 minutes per item, pacing is a real constraint. Take full mocks under exam conditions: 63 items, 120 minutes, no notes. Mark long items, answer the quick ones first, and come back to the marked ones at the end. Record your score by domain after every mock.
Drill weak domains, then retest
Use the domain breakdown to pick one weak area. Re-study that domain's objectives, then re-drill only the questions you missed. Retest with a fresh full mock, because repeating the same mock measures memory, not readiness.
Common traps
- "Agent" as the advanced answer. A workflow is often the correct, cheaper choice when the steps are predictable.
- Prompt-only safety. If the rule must hold every time, look for the deterministic layer.
- Bigger is better. A larger model or window rarely fixes a design flaw.
- Done at launch. Lifecycle questions expect monitoring and iteration after deployment.
- Multiple-response miscounts. Read how many answers the item wants before you eliminate.
Practice with Ready Room
Ready Room for CCAR-P gives you full mock exams with a missed-question re-drill and a domain heat map, so you can see which of the seven domains need work. Sigma Foundry makes this app. It's an unofficial practice resource, not affiliated with Anthropic.
Prefer Udemy? Our CCAR-P mocks are on Udemy in two volumes of six timed exams each, every option explained: Vol. 1 and Vol. 2. Each is $12.99 at launch with those links (list $19.99) through December 5.
FAQ
How many questions are on the CCAR-P exam?
63 items, multiple-choice and multiple-response, in 120 minutes.
What score do I need to pass?
A scaled score of 720 out of 1,000, per the published exam guide (v1.0).
Which domain is weighted most?
Integration, at 19%. Developer Productivity & Operations is the lightest, at 7%.