Exam Prep: Claude Certified Architect Professional (CCAR-P)

Day 1 — Design, Models & Integration

Exam domains 1 (17%), 2 (13%), and 3 (19%) — 49% combined. Consolidate architectural judgment, defensible model/context decisions, and secure integration.

Module 1 · Solution Design & Architecture (Domain 1)

• Decomposition before architecture; who owns what across Claude, systems, and humans
• Pattern selection — augmented call vs. workflow vs. agent — via the predictability × autonomy map and the five-factor framework
• Multi-agent orchestration: fan-out, where failures can and cannot be caught, human-in-the-loop checkpoints
• Reference architectures and the retrieval-on-live-state anti-pattern; aligning to business-value pillars
• Drills: Match the Primitive to the Job · Critique the Orchestration Design

Module 2 · Models, Prompting & Context Engineering (Domain 2)

• Model selection: start with Sonnet, eval-gate every swap (the 7× Opus cost-overrun case)
• System-prompt architecture, templates, and diagnosing underspecification; technique selection by task complexity
• Context-window sizing and the working-memory cliff; progressive ↔ monolithic strategy; caching mechanics; libraries vs. Skills
• Drills: Author the Reusable Prompt Asset · Cost & Latency — find the load-bearing control

Module 3 · Integration (Domain 3)

• Entry point / build-time interface / delivery-route and connection-protocol selection (MCP, API/CLI, agent-to-agent)
• Identity and authorisation, least-privilege tool configuration, capability bloat; data handling and the PII-in-the-prompt case
• RAG pipeline: chunk by what the corpus is, index by how queries are phrased; retrieval matched to data shape and query
• Drills: Critique the Integration Diagram · Design the RAG Pipeline

Exam checkpoint: scenario quiz on pattern selection and orchestration recovery, model/context/caching trade-offs, and least-privilege + RAG integration.

Day 2 — Quality, Safety, Delivery & Mock Exam

Exam domains 4 (16%), 5 (14%), 6 (14%), and 7 (7%) — 51% combined — then a full mock exam.

Module 4 · Evaluation, Testing & Optimisation (Domain 4)

• Evals as acceptance criteria and as the gate on every change; model-based vs. code-based evals and the grading ladder
• Cost/latency modeling, reliability controls per call, failure modes by architecture (the demo-bill-vs-production-bill case)
• A/B testing without overclaiming, shadow testing, observability at scale, and diagnosing prompt failure vs. hallucination vs. model mismatch
• Drills: Sort the Eval Types · Place on the Experiment-Design Plane

Module 5 · Governance, Safety & Risk Management (Domain 5)

• The safety stack: training-time alignment vs. inference-time control, and who owns each layer
• Guardrail placement for safe degradation; chaining model-based and deterministic checks; injection via retrieved content/tool outputs; fail-open vs. fail-closed
• Fairness, decision logging, routing to humans by stakes not volume; obligation → control → evidence (GDPR, HIPAA, FedRAMP)
• Drills: Place the Controls on the Path · Justify the Control Choice

Module 6 · Stakeholder Communication & Lifecycle (Domain 6)

• Discovery as structured elicitation; surfacing the undocumented assumption; presenting a trade-off a stakeholder can act on
• Feedback loops and SLAs; triaging production signals; documentation that survives handoff; outcome documents that measure the right thing
• Drills: Recommend the Option and Name the Missing Element · Triage the Production Signals

Module 7 · Developer Productivity & Operational Enablement (Domain 7)

• Configuring Claude tooling and environments for teams and choosing a distribution strategy; verification checklists; supporting debugging and operational issue resolution (guided review + self-study handout)