The 5 Competency Domains of the CCAR-F Exam Explained (With Domain Weights)

Understanding the CCAR-F Evaluation Architecture

Anthropic's Claude Certified Architect – Foundations (CCAR-F) exam is designed to validate advanced technical judgment in designing, deploying, and governing production AI systems. Unlike entry-level certifications that focus on surface-level terminology, the CCAR-F evaluates how architects solve complex, scenario-based engineering challenges across five core competency domains.

Understanding the domain weights is critical for structuring your exam preparation and allocating your study time effectively. Because the test evaluates real-world system architecture, candidates frequently rely on comprehensive preparation platforms like ccaftraining.com to explore scenario drills, practice exams, and architectural trade-offs that mirror the official exam standards.

The Five Competency Domains Breakdown

The CCAR-F exam distributes its 60 multiple-choice and multiple-response scenario items across five distinct technical areas. Here is what each domain covers and why it carries its specific weight:

Domain 1: Agentic Architecture & Orchestration (27%)

As the single heaviest domain on the exam, Agentic Architecture represents more than a quarter of your total score. This domain focuses on building autonomous systems where Claude can reason, execute actions, observe results, and self-correct without continuous human intervention.

• Control Flow Management: Focuses on handling system execution boundaries and interpreting stop reason signals to manage transitions between reasoning loops and tool execution.

• Multi-Agent Orchestration: Covers designing coordinator-subagent architectures to delegate specialized tasks across multiple models while maintaining centralized state management.

• State Persistence and Task Decomposition: Evaluates breaking complex enterprise goals into sequential execution gates and preserving conversation history across session forks and resumes.

Domain 2: Claude Code Configuration & Workflows (20%)

This domain focuses on deploying and governing AI coding assistants within enterprise software engineering teams. It requires hands-on mastery of how Claude Code integrates into collaborative development environments.

• Configuration Hierarchy: Evaluates structuring rules across project repositories, from global user settings down to directory-specific instructions.

• Execution Modes: Covers configuring autonomous execution versus interactive plan-and-review modes to balance developer velocity with architectural oversight.

• CI/CD Pipeline Integration: Tests embedding automated code reviews, non-interactive execution flags, and vulnerability scanning directly into automated build pipelines.

Domain 3: Prompt Engineering & Structured Output (20%)

At an architectural level, prompt engineering is treated as a deterministic software practice. This domain tests your ability to force models to deliver predictable, machine-readable data structures required by backend engineering systems.

• JSON Schema Enforcement: Focuses on extracting unstructured enterprise documents into validated JSON schemas that eliminate syntax failures.

• Schema Optimization: Covers designing schemas with nullable fields and explicit boundary constraints to prevent forced model hallucinations during data extraction.

• Advanced Few-Shot Patterns: Evaluates constructing contextually robust examples within system prompts to guide model behavior through complex reasoning workflows.

Domain 4: Tool Design & MCP Integration (18%)

An AI agent relies entirely on the quality of its external tools to interact with enterprise databases, APIs, and software environments. This domain focuses on Anthropic's Model Context Protocol (MCP).

• MCP Implementation: Tests configuring, securing, and scaling Model Context Protocol servers to standardize data connections across enterprise systems.

• Interface Clarity: Covers writing unambiguous tool descriptions and parameter schemas so the model can accurately route functions without misfiring.

• Structured Error Handling: Evaluates configuring error metadata flags, such as retryable states and error categories, enabling agents to self-correct when external integrations fail.

Domain 5: Context Management & Reliability (15%)

While it carries the smallest weighting, context management is essential for long-running enterprise sessions where accumulating data can degrade reasoning accuracy over time. Combined with Domain 1, architecture and context optimization represent 42% of the entire exam.

• Progressive Summarization: Focuses on condensing lengthy conversation histories and session states into lean memory checkpoints.

• State Consolidation: Covers moving heavy background reference datasets out of the active context window and into externalized vector storage or state objects.

• Human-in-the-Loop Escalation: Evaluates designing defensive guardrails that automatically trigger human intervention when a system reaches cognitive thresholds or logical ambiguity.

Strategic Exam Preparation

Because four of the six published enterprise scenarios appear randomly on any given exam sitting, candidates must achieve proficiency across all five domains rather than focusing solely on the highest weights.

To build a well-rounded skillset, successful candidates often structure their review sequentially—starting with Agentic Architecture and ending with Context Management. Utilizing specialized training resources and mock scenario labs at ccaftraining.com allows you to test your architectural judgment under timed conditions, ensuring you are fully prepared to navigate every domain on test day.

To see how these five domains translate into real-world exam questions and architectural trade-offs, check out the Claude Certified Architect Exam Study Guide. This video provides a detailed breakdown of all five domains and reviews the six production deployment scenarios you will encounter on test day.