← All certifications Claude Certified Architect – Professional · CCAR-P

Design it. Defend it. Operate it.

Eleven design-studio labs and a full practice exam for architects who own production Claude systems end to end. Written against Anthropic's official CCAR-P exam guide, where the right answer is usually the one that removes something.

Start Lab 1 → Take the Practice Exam

The Exam

What you're preparing for

The Professional credential validates that you can design, build, and deliver production-grade AI solutions on Claude: selecting models and architectures, engineering context, integrating into enterprise systems, and carrying evaluation, security, compliance, and governance through the whole lifecycle.

Exam code

CCAR-P

Questions

63 items

Time limit

120 minutes

Passing score

720 / 1000

Exam fee

$175 USD

Validity

12 months

This exam tests judgment, not syntax. The guide expects three or more years in systems architecture and at least six months running Claude or comparable systems in production. It asks whether the system you designed should exist in that shape, at that cost, under that regulation, and whether you can defend the choice to a stakeholder. Code appears where a decision is expressed in code, never as an end in itself.

Two domains have no API surface at all. Stakeholder Communication and Lifecycle Management is 14% (more than Models and Prompting) and Developer Productivity is another 7%. Engineers systematically under-study both. If you are coming from Architect – Foundations, expect RAG pipeline design, evaluation frameworks, observability at scale, GDPR/HIPAA/FedRAMP, discovery, ADRs, and SLAs: none of which Foundations tests.

Design Studio Labs

Eleven labs, weighted to the blueprint

Domains 1, 3, and 4 are 52% of the exam between them, so each gets two labs. The 7% domain gets one, and its lab opens by telling you not to over-invest. Each lab is a scenario, a decision framework, a trade-off table, and the sentence that kills the alternatives, then exam-style self-checks.

Solution Design & Architecture

From business problem to architecture. The pattern ladder (augmented LLM, workflow, agentic) and the four criteria an agent must pass. The five planes, and why the feedback plane is the one everyone omits.

Domain 1 · 17%

Multi-Agent Systems & Orchestration

When multi-agent is wrong. Orchestration topologies, context isolation as the real justification, explicit delegation briefs, and context editing versus compaction on a long-running agent.

Domain 1 · 17%

Models, Prompting & Context Engineering

Model trade-offs and routing. Adaptive thinking and effort as an architectural dial. Why prompt guardrails are advisory. Prompt caching that actually hits, and the ordering mistake that guarantees it never does.

Domain 2 · 13%

RAG Pipeline Design

When not to build RAG at all. Chunking on document structure, retrieval matched to query pattern, hybrid plus rerank, ACL filtering at retrieval time, and why changing the embedding model invalidates the whole index.

Domain 3 · 19%

Integration Protocols, Auth & Observability

Capability bloat and least privilege. The confused deputy. MCP versus API/CLI versus agent-to-agent. Accuracy-latency levers, and the three questions observability must answer.

Domain 3 · 19% — heaviest domain

Evaluation Frameworks & Datasets

The five metric families and what each hides. Programmatic gates, LLM-as-judge, and human calibration. Building a dataset from production failures. A/B testing with a guardrail metric.

Domain 4 · 16%

Diagnosis & Optimization

Differential diagnosis: symptom to plane. Hallucination is a symptom, not a diagnosis. The ablation ladder. Cost per request versus cost per resolved task, and why the first one falling can be a loss.

Domain 4 · 16%

Governance, Safety & Risk Management

Layered guardrails, and why anything enforced by asking the model nicely is not a control. Injection defense that works. Human-in-the-loop by blast radius. GDPR, HIPAA, FedRAMP as architectural obligations. Fairness that an aggregate metric conceals.

Domain 5 · 14%

Stakeholder Communication & Lifecycle

Structured discovery, and the two questions that surface a doomed project early. A worked ADR. SLA versus SLO versus SLI. The curated-demo trap. The runbook that decides whether the system survives you.

Domain 6 · 14% — the one engineers skip

Developer Productivity & Ops Enablement

Claude Code for teams: the CLAUDE.md hierarchy, settings precedence, permissions, and hooks that enforce what prompts merely request. Tool Runner versus Agent SDK versus Managed Agents. Worth four items, so budget your time accordingly.

Domain 7 · 7% — smallest domain

Capstone: End-to-End Architecture Review

A global bank's regulatory-change assistant, proposed with one defect seeded in every domain. Audit it, find all seven, and repair the design. Then price it and write the handoff package.

All seven domains

Blueprint

Where the points are

Domain	Weight	Approx. items	Lab
1. Solution Design & Architecture	17%	11	Lab 1, Lab 2
2. Claude Models, Prompting & Context Engineering	13%	8	Lab 3
3. Integration	19%	12	Lab 4, Lab 5
4. Evaluation, Testing & Optimization	16%	10	Lab 6, Lab 7
5. Governance, Safety & Risk Management	14%	9	Lab 8
6. Stakeholder Communication & Lifecycle Management	14%	9	Lab 9
7. Developer Productivity & Operational Enablement	7%	4	Lab 10

Integration alone is worth twelve questions. Developer Productivity is worth four. Domain 6 has no code in it and is worth nine, more than Models and Prompting. Study accordingly.

Assess Yourself

Then test what stuck

✓

Architect Professional Practice Exam

190 scenario questions across all seven domains. Run a full 63-question exam simulation drawn at the real blueprint weights, or drill one domain. Multiple-response items scored all-or-nothing, the real 720 cut line, and a per-domain breakdown.

Quick Reference

One-page cram sheet: the seven domains and weights, the pattern ladder, retrieval by query shape, the control-strength hierarchy, caching and batch rules, the symptom-to-plane differential, and SLA versus SLO.

Open a design doc

The guide's own advice: build and operate at least one end-to-end Claude solution, including RAG, evaluation, and observability. Then practise architectural decision-making. That's Lab 11. Everything before it gets you there.

Start Lab 1 →