A six-week CCAR-P study plan

Six weeks. This exam rewards experience more than reading, so most weeks pair study with something you build, measure or write down.

Week 1 — Close the Foundations gap, then move on

If you have not sat CCAR-F, spend this week on its material: the agent loop and stop_reason, subagents and context isolation, tool and MCP design, structured output, caching and batching. Professional assumes it rather than testing it directly. If you have sat it, skim and go straight to week 2.

Week 2 — Integration, the heaviest domain (19%)

Build a retrieval pipeline end to end: chunk a real corpus, add contextual enrichment, run hybrid embedding and lexical search, rerank, and generate with citations enabled. Measure retrieval failure at each stage so the published improvement figures mean something to you. Then work the authorisation problem: make the agent inherit a caller's permissions rather than using a service account, and see what that costs you architecturally.

Week 3 — Evaluation and optimisation (16%)

Build an eval suite of at least a hundred cases for something you already run. Grade it three ways — code, LLM judge, and a human sample — and validate the judge against the human labels. Establish a baseline, then change exactly one variable and measure the delta in quality, latency and cost together. This week teaches the habit that most of the domain's questions test.

Week 4 — Governance and risk (14%)

Take a system you know and write its compliance map: what data flows where, under what basis, retained how long, reviewed by whom, evidenced how. Learn the failure-mode table and the guardrail layering order. Read Anthropic's certification and data-handling position so you can distinguish what the vendor provides from what remains yours.

Week 5 — Stakeholder communication and lifecycle (14%)

The domain you cannot cram. Write three artefacts for a real system: a decision record with the rejected alternatives and why, a one-page capability and limitation statement for a business audience, and a draft SLA that commits only to things you can actually measure. Then have someone non-technical read the middle one. Add the 7% enablement domain as an afternoon: standardise a repository's Claude Code configuration and make a CI run reproducible with --bare.

Week 6 — Drill

Practice questions filtered to your weakest domains, then the anti-pattern list. Because the items are standalone rather than scenario-anchored, practise the specific skill of reading a fresh stem and naming the deciding constraint before looking at the options — regulatory, measurable target, or organisational reality. That habit is worth more marks than any single fact.

← Previous: Facts to memorise  |  Next: Official resources →