CCAR-P One-Pager: Claude Certified Architect – Professional¶
Unofficial summary. Every exam fact on this page comes from
16-certification-tracks/registry.yaml(verified 2026-10-24 against the official exam guide v1.0, effective July 2026) and is summarised in exam-facts.md. The official guide on the Anthropic Partner Academy is the source of truth. Pearson VUE has announced updated guides for 2026-11-17 and updated exam content from 2026-12-07, so re-check before you book. Full track: 16 · CCAR-P.
Exam format¶
| Item | Value (from the registry) |
|---|---|
| Audience | Mid- to senior-level solution architects, AI/ML engineers, technical leads, senior software engineers |
| Recommended experience | 3+ years in systems architecture or platform engineering; 6+ months hands-on with LLM systems in production (not mandatory) |
| Items | 63, multiple-choice and multiple-response |
| Time | 120 minutes |
| Passing | Scaled score 720 on a 100–1,000 scale |
| Fee | USD 175 (list price) |
| Validity | 12 months |
| Delivery | Pearson VUE, online proctored or test centre |
| Eligibility | Organisations in the Claude Partner Network |
Domains, weights and what to know¶
D1 · Solution Design & Architecture — 17%¶
- Translate a business problem into a Claude solution with measurable outcomes (CCAR-P-1.1).
- End-to-end design: input, processing, output, feedback loops (CCAR-P-1.2).
- Pattern choice: single call → workflow (prompt chaining, routing, parallelisation, orchestrator–workers, evaluator–optimiser) → agent; pick the simplest that works (CCAR-P-1.3).
- Multi-agent orchestration: when sub-agents help (parallel, context isolation) and what they cost (CCAR-P-1.4).
- Decomposition and value alignment: efficiency, cost, SLAs (CCAR-P-1.5, 1.6).
- Study: 10 · Workflows vs agents, 13 · Architecture patterns.
D2 · Claude Models, Prompting & Context Engineering — 13%¶
- Model trade-offs: capability, latency, cost, context; tune effort before switching (CCAR-P-2.1).
- System prompts, templates and guardrails (CCAR-P-2.2).
- Zero-shot, few-shot and reasoning (adaptive thinking, effort) (CCAR-P-2.3).
- Context window and token optimisation: retrieval, compaction, context editing (CCAR-P-2.4).
- Reuse: prompt caching (prefix stability), modular prompts, Agent Skills (CCAR-P-2.5).
- Study: 04 · Context engineering, cheat sheet: model selection.
D3 · Integration — 19% (largest domain)¶
- Capability bloat: too many or over-powered tools; trim, scope and split (CCAR-P-3.1).
- AuthN/AuthZ gaps: user-bound permissions, audience-bound tokens, no shared admin keys (CCAR-P-3.2).
- Accuracy–latency trade-offs: model, effort, retrieval depth, streaming, caching (CCAR-P-3.3).
- Observability at scale: tracing tool calls, token usage, request IDs (CCAR-P-3.4).
- RAG design: chunking, embeddings, hybrid search, reranking, metadata filters (CCAR-P-3.5, 3.6).
- Study: 07 · Tool design, 08 · MCP, 11 · RAG.
D4 · Evaluation, Testing & Optimization — 16%¶
- Metrics: accuracy, latency, cost, safety, security (CCAR-P-4.1).
- Datasets and mixed methods: code-graded, model-graded, human-graded (CCAR-P-4.2).
- A/B testing and iterative improvement (CCAR-P-4.3).
- Diagnose prompt failures, hallucination and model mismatch (CCAR-P-4.4).
- Token, latency and cost optimisation: caching, batches, effort, routing (CCAR-P-4.5).
- Monitoring with logging and observability tooling (CCAR-P-4.6).
- Study: 14 · Evaluation and testing.
D5 · Governance, Safety & Risk Management — 14%¶
- Guardrails and safety controls in and around the model (CCAR-P-5.1).
- Failure modes: injection, hallucination, excessive agency, drift (CCAR-P-5.2).
- Human-in-the-loop design: where, who, with what information (CCAR-P-5.3).
- Regulatory compliance: data residency, retention, sector rules (CCAR-P-5.4).
- Bias, fairness and transparency (CCAR-P-5.5).
- Study: 12 · AI security, cheat sheet: AI security.
D6 · Stakeholder Communication & Lifecycle Management — 14%¶
- Discovery and requirements (CCAR-P-6.1).
- Communicate decisions and trade-offs; architecture decision records (CCAR-P-6.2, 6.4).
- Expectations and SLAs; feedback loops (CCAR-P-6.3).
- Lifecycle: discovery, design, handoff, monitoring, iteration (CCAR-P-6.5).
- Study: 13 · Stakeholder communication, 13 · Lifecycle management.
D7 · Developer Productivity & Operational Enablement — 7%¶
- Configure Claude Code for teams: shared settings, permissions,
CLAUDE.md, hooks, MCP (CCAR-P-7.1). - AI-assisted developer workflows: review, tests, CI (CCAR-P-7.2).
- Debugging and operational issue resolution (CCAR-P-7.3).
- Study: 09 · Team rollout, cheat sheet: Claude Code.
Architect's decision reflexes¶
| Situation | Reflex |
|---|---|
| Unclear if an agent is needed | Start with a workflow; add autonomy only where the path cannot be pre-defined |
| Repeated large prefix | Prompt caching with stable content first |
| Work that can wait | Message Batches |
| Machine-consumed output | Structured outputs or strict tools |
| Consequential action | Human approval plus deterministic checks |
| Many tools | Trim, namespace, or use tool search / Skills to load on demand |
| Quality question | Answer with an eval, not an opinion |
Practice: 17 · CCAR-P questions, 18 · Mock exams, 23 · Flashcards.