The seven CCAR-P domains and what each one tests

Here is where the questions actually are, and what each domain rewards.

Solution Design & Architecture · 17%

Choosing the shape of the system: workflow versus agent, where the boundaries sit, how many stages there are. The recurring lesson is that adding a stage is usually the wrong move. Credited answers tend to move or remove one instead. Complexity has to be earned by numbers in the scenario, not by ambition.

Claude Models, Prompting & Context Engineering · 13%

Model selection against real constraints, prompt structure, and how context is assembled and paid for. Know what platform primitives already solve before you reach for a custom workaround, and know what caching does and does not do — prefix matching means dynamic values belong after static content, and conversation history is a growing stable prefix.

Integration · 19%

The largest domain, and the one candidates most often underestimate. Retrieval design, tool and system boundaries, handoffs between components. Match small and ground big on retrieval. Validate contracts deterministically at machine-to-machine boundaries. When you need another team's judgment over data only they hold, request a verdict rather than importing their tools.

Evaluation, Testing & Optimization · 16%

How you know the thing works, and how you know a change improved it. Label quality is the recurring theme: an eval is only as trustworthy as the agreement between the people who labeled it. Watch for statistical traps too, particularly many-arm comparisons where at least one false winner is close to expected.

Governance, Safety & Risk Management · 14%

Guardrails, fail-closed behavior, and regulatory obligations. This is where the general answer-ranking heuristic gets suspended: legal questions want the assessment done before processing and the rights of the affected person, which is exactly the shape of answer you learn to downrank elsewhere. Expect fairness questions.

Stakeholder Communication & Lifecycle Management · 14%

Discovery, conflicting requirements, ROI defense, and knowing which lifecycle stage you are in. Conflicting stakeholder requirements are resolved by making the conflict visible and facilitating a decision, not by documenting both, picking a side, or inventing a compromise nobody agreed to. ROI means net unit economics against a measured baseline.

Developer Productivity & Operational Enablement · 7%

The smallest domain. Tooling and enablement for the teams who build and run the system. Worth a light pass rather than deep study — seven percent is roughly four questions, and time spent here is time not spent on Integration.

What the weights mean for a study plan

Integration alone is nearly a fifth of the exam, and Solution Design and Evaluation together are another third. Those three carry just over half the questions between them.

Developer Productivity is seven percent, which is about four questions. Studying it to the same depth as Integration is a straightforward misallocation, and it is the most common one.

Practice forms should carry the same proportions. If a domain is 19% of the exam it should be 19% of your practice, which is how ours are built.