The problem
Two people study for the CCAR-F. The first memorizes that there are five domains, memorizes the weights (27%, 20%, 20%, 18%, 15%), and knows by heart what every acronym stands for: MCP, ADR, CLAUDE.md. The second studies the same five domains, but for each one asks: "what decision does an architect make here, under what constraint, and what happens if they get it wrong?"
On exam day, both face the same scenario: a support team wants an agent that resolves tickets, decides between delegating to a specialized subagent or processing everything in a single session, and must choose where to place the escalation policy. The first person recognizes the technical terms in every option and cannot differentiate between them. The second recognizes which option solves the real problem without violating an implicit constraint — and usually, only one remains.
That is the difference between studying vocabulary and studying decisions. The CCAR-F is a scenario-based exam: each bundle features a scenario and several questions about it. It doesn't test whether you know what MCP is. It tests whether, given a misconfigured MCP, you can pinpoint what broke and why.
The blueprint is a working model, not a list of topics
Each domain of the CCAR-F represents a slice of real-world work for a Claude systems architect. The weight does not reflect difficulty — it reflects frequency. Agentic Architecture & Orchestration accounts for 27% because most architectural work revolves around deciding how to orchestrate agents, not because questions in that domain are inherently harder.
Use this distinction to build your study plan: allocate time proportionally to the weight, but within each domain, spend most of your time on Doing and Deciding, not on Knowing.
- Knowing: facts and vocabulary you recall ("
stop_reason: tool_usemeans the model wants to call a tool"). - Doing: a procedure you execute ("implementing the loop that reads
tool_use, executes it, and returns the result"). - Deciding: a choice under constraints that you resolve ("does this loop need a separate subagent, or does a single tool call suffice?").
Most weak study plans pile time into Knowing. The exam concentrates its weight on Doing and Deciding.
Stable facts vs. moving targets
Some principles remain constant across product versions: sensitive data requires an approved handling path before leaving the system; an irreversible action warrants more review than a reversible draft; a larger model is wasteful when a smaller one already meets the measured requirement. These are the principles underpinning most correct answers in scenarios.
Other facts change between when you study and exam day: model names and pricing, context limits, feature availability across tiers. Treat this second group as time-sensitive references rather than permanent memorization — check the official guide before scheduling your exam.
How to read a multi-question scenario
The CCAR-F format groups multiple questions around the same scenario bundle (four bundles selected out of six possible). This shifts the reading strategy: your first pass through the scenario should extract four things, in this exact order:
- The required outcome — what must exist in the end, not how.
- The explicit constraints — budget, latency, data policies, approval workflows.
- Risk and reversibility — what happens if the output is wrong.
- The minimum viable capability — the smallest solution that satisfies the three points above.
Distractors in scenarios are rarely absurd. They are usually technically correct, but overlook a key constraint — like using the most powerful model without measured justification, or automating an irreversible action without human-in-the-loop review. When reading options, ask which constraint each one ignores, not just whether it "would work."
Put it into practice
This lesson's lab asks you to take a domain blueprint with weights and turn it into a study plan: a script that distributes available hours proportionally to each domain's weight and flags when the actual allocation is unbalanced. It is the exact same logic you will apply mentally, under time pressure, during the exam.