CCAR-P: Architect – Professional Exam

Format

• Questions: 63.
• Length: 120 minutes.
• Fee: $175 (the most expensive Anthropic exam).
• Passing: 720 on the 100–1,000 scaled score.
• Delivery: Pearson VUE (OnVUE or test center).
• Target profile: senior architects with 3+ years systems architecture experience and 6+ months running LLM systems in production. The only Professional-tier exam in the 2026 lineup; the intended step up from CCAR-F (cca-f-exam-overview).

The seven domains

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

The signature of Professional tier: 28% of the exam (D5 + D6) is non-technical — governance and stakeholder work. CCAR-F asks "which pattern?"; CCAR-P asks "which pattern, under this compliance regime, defended to this steering committee."

Objectives by domain

• Solution Design & Architecture (17%) — business-problem translation (turning fuzzy requirements into Claude-shaped solutions), designing end-to-end architectures, workflow vs agentic pattern selection (predefined pipelines vs model-directed loops — agentic-patterns), and multi-agent orchestration (orchestrator-workers, when to split vs keep monolithic).
• Models, Prompting & Context (13%) — model selection tradeoffs at the architecture level (fleet-of-Haikus vs single-Opus economics), guardrails, chain-of-thought deployment, context optimization, and prompt reuse — the three mechanisms being prompt caching (prompt-caching), modular prompts (composable tested fragments), and Skills (agent-skills).
• Integration (19%) — the heavyweight:
  • Capability bloat evaluation — pruning tool/connector surface area; every capability an agent carries is context cost and attack surface.
  • Authn/authz gap analysis — finding where identity and permission checks are missing across model, tools, and downstream systems.
  • Observability — tracing, logging, and metrics for agentic systems.
  • RAG pipeline design — chunking and indexing strategy decisions (rag-patterns).
  • Integration surface selection — MCP vs direct API/CLI vs agent-to-agent communication.
  • Progressive discovery vs monolithic context — loading capability/knowledge on demand as the agent needs it, versus front-loading everything into the prompt (context-management).
• Evaluation, Testing & Optimization (16%) — choosing eval metrics and building eval datasets, A/B testing prompt and architecture changes, and systematic diagnosis of production quality regressions.
• Governance, Safety & Risk (14%) — regulatory regimes: GDPR (EU data protection), HIPAA (US health data), FedRAMP (US federal cloud authorization); plus ethics — appropriate-use boundaries and transparency obligations (org-scale anthropic-safety).
• Stakeholder Communication & Lifecycle (14%) — structured discovery (requirements elicitation that surfaces constraints early), defining SLAs for LLM-backed services (latency/availability targets under nondeterminism), and writing architecture docs that non-ML stakeholders can approve.
• Developer Productivity & Enablement (7%) — team enablement: rolling out Claude Code, shared skills/plugins, internal standards and golden paths so other teams build safely without the architect in the room.

How it differs from CCAR-F

	CCAR-F	CCAR-P
Question style	60 Q across 4-of-6 fixed scenarios	63 standalone senior-scenario questions
Center of gravity	Implementation patterns	Tradeoffs + governance + communication
Compliance content	Minimal	GDPR/HIPAA/FedRAMP explicitly in scope
Evaluation content	Light	Full 16% domain: metrics, datasets, A/B
Assumed role	Builder-architect	Accountable owner of a production system

Preparation pointers

1. Pass-level command of everything in cca-f-exam-overview — CCAR-P assumes it.
2. Drill the Integration domain: practice arguing MCP vs API vs agent-to-agent, and progressive discovery vs monolithic context, out loud with justifications — that's the question format.
3. Build one real eval: pick metrics, assemble a dataset, run an A/B. The 16% D4 weight punishes candidates who have only read about evals.
4. Skim primary summaries of GDPR/HIPAA/FedRAMP — the exam tests which regime applies and its architectural consequence, not legal minutiae.

See also

• certifications-overview — logistics, retake, and renewal rules
• cca-f-exam-overview — the Foundations tier this builds on
• agentic-patterns · rag-patterns · context-management — core technical inputs