Home › Articles › The six CCAR-F exam scenarios explained, and what each one tests

28 September 2026 · 5 min read · Anton Pavelko

The six CCAR-F exam scenarios explained, and what each one tests

CCAR-F is the only Claude exam built from scenarios. All six are published in the exam guide. Here is what each one describes, which domains it draws on, and the one idea to have ready for each.

The Claude Certified Architect - Foundations exam does something none of the other three Claude exams do: every question sits inside a scenario, a short description of a production system, and the questions ask what you would do in it. There are six scenarios. On the day you get four, chosen at random, with about fifteen questions each.

The guide prints all six. Reading them before the exam is the single best preparation step, because a question makes sense only against its scenario, and because the scenarios tell you which domains to expect. Here they are, with the domains each one draws on and the one idea that most questions in it come back to.

1. Customer Support Resolution Agent

An agent built with the Claude Agent SDK handles returns, billing disputes and account issues. It reaches backend systems through custom MCP tools: get_customer, lookup_order, process_refund, escalate_to_human. The target is 80% first-contact resolution while knowing when to escalate.

Domains: Agentic Architecture, Tool Design and MCP, Context Management and Reliability.

The idea to have ready: guarantees versus good intentions. Verification before refunds is a prerequisite gate in code, not a sentence in the prompt. Refund limits are hooks. Escalation is triggered by explicit requests and policy gaps, not by sentiment. The refund amount lives in a case-facts block, not in a summary.

2. Code Generation with Claude Code

A team uses Claude Code for generation, refactoring, debugging and documentation, with custom slash commands and CLAUDE.md configuration, and needs to know when to use plan mode rather than direct execution.

Domains: Claude Code Configuration, Context Management and Reliability.

The idea to have ready: shared things in the project, personal things at home. Commands in .claude/commands/, standards in the project CLAUDE.md, path-scoped rules in .claude/rules/. Plan mode for architectural, multi-file work; direct execution for a one-file fix with a stack trace.

3. Multi-Agent Research System

A coordinator delegates to specialised subagents: web search, document analysis, synthesis, report generation. The output is a comprehensive, cited report.

Domains: Agentic Architecture, Tool Design and MCP, Context Management and Reliability.

The idea to have ready: the coordinator carries everything. Subagents start with empty context and get their findings in the prompt; they are spawned with the Task tool, in parallel when the calls go out in one response; every message routes through the hub; errors come back structured with partial results; citations survive only as structured claim-source mappings.

4. Developer Productivity with Claude

An Agent SDK agent helps engineers explore unfamiliar codebases, understand legacy systems, generate boilerplate and automate repetitive work, using the built-in tools (Read, Write, Bash, Grep, Glob) and MCP servers.

Domains: Tool Design and MCP, Claude Code Configuration, Agentic Architecture.

The idea to have ready: explore incrementally. Grep for entry points, Read and follow imports, Glob for files by name, Edit with unique anchors and fall back to Read plus Write. Scratchpad files and subagents keep a long exploration from degrading; sessions resume by name and fork from a shared baseline.

5. Claude Code for Continuous Integration

Claude Code runs inside a CI/CD pipeline: automated code review, test generation, feedback on pull requests, with actionable findings and few false positives.

Domains: Claude Code Configuration, Prompt Engineering and Structured Output.

The idea to have ready: precision comes from criteria and structure. -p to run without a terminal, --output-format json with --json-schema for findings that can be posted as comments, explicit categorical criteria rather than "be conservative", per-file passes plus an integration pass for big pull requests, and a separate session for review so the author is not judging its own work.

6. Structured Data Extraction

A system extracts information from unstructured documents, validates it against JSON schemas, handles edge cases and feeds downstream systems.

Domains: Prompt Engineering and Structured Output, Context Management and Reliability.

The idea to have ready: schemas guarantee syntax, loops guarantee meaning. Tool use with a JSON schema for valid output, optional fields so absent information becomes null rather than invention, retry with the specific validation error attached, and human review routed by calibrated confidence and checked by stratified sampling.

How to use the six

Count the domains above and you will see the pattern: Agentic Architecture appears in three scenarios, Context and Reliability in four, Claude Code in three. That is why the weights are what they are, and why Domain 1 is the place to start.

Since you get four of six, a third of what you prepare will not be asked. You cannot know which third. Prepare all six, and do not generalise from a colleague's report of an easy or brutal exam; their four may not be yours. The CCAR-F study guide covers every task statement behind these scenarios, and the cheat sheet puts the exact names on one page.

Written by Anton Pavelko, who holds the Claude Certified Architect, Associate and Developer credentials. An independent resource, not affiliated with Anthropic. Spotted an error? Email a correction.