A CCAR-F study plan: two weeks, four weeks, or a weekend
Three preparation schedules for the Claude Certified Architect – Foundations exam, weighted to the blueprint, with what to cut when time runs short.
The exam guide suggests 15–20 hours for someone with hands-on Claude experience and 30–40 for someone newer. Here is what to do with those hours, depending on how many you have.
Every plan below follows the same principle: time follows the blueprint weights, not your interest. Domain 1 is 27% of the exam and Domain 5 is 15%. If your study time does not roughly reflect that, you are optimising for the wrong thing.
| Domain | Weight | Share of study time |
|---|---|---|
| 1 · Agentic Architecture & Orchestration | 27% | ~7 lessons, most of your first week |
| 3 · Claude Code Configuration & Workflows | 20% | 6 lessons |
| 4 · Prompt Engineering & Structured Output | 20% | 6 lessons |
| 2 · Tool Design & MCP Integration | 18% | 5 lessons |
| 5 · Context Management & Reliability | 15% | 6 lessons |
Note that Domain 5 has six lessons for the lightest weight. Read them; do not linger.
Before any plan: two hours you should not skip
Sit a mock exam cold. Before reading a single lesson. It will feel unpleasant and it is the highest-value two hours in any of these schedules, because everything after it is targeted rather than general. The per-domain breakdown tells you which plan you actually need.
Read the six scenarios. Fifteen minutes. The exam frames every question inside one of them, all six are published, and they turn the abstract domain descriptions into recognisable situations.
The four-week plan (~30 hours)
For anyone newer to Claude, or anyone who wants the credential to be worth having rather than just held.
Week 1 — Domain 1, all seven lessons. Two per sitting, with the build exercises. By the end you should have written one complete agentic loop: tool calls handled, stop_reason driving control flow, a real termination condition, and session resumption that survives being killed mid-run. This single artefact covers more of the exam than anything else you will build.
Week 2 — Domains 3 and 4. Configure a real repository properly: a project CLAUDE.md with imports, a directory-scoped rule under .claude/rules/ with a glob, and one custom command with allowed-tools and argument-hint frontmatter. Then build an extraction pipeline that enforces a schema through tool use and retries on validation failure. Re-test Domains 1, 3 and 4 at the end of the week.
Week 3 — Domains 2 and 5. Write an MCP server with two deliberately similar tools and watch the model pick the wrong one until you fix the descriptions — that lesson does not land any other way. Then read Domain 5 properly; it is the lightest domain but its questions punish guessing hardest.
Week 4 — consolidation. Full mock exam at the start. Study only what it says is weak. Second full mock 48 hours before. Cheat sheets and the divergence log the night before, nothing new.
The two-week plan (~18 hours)
For someone already building with the Agent SDK, Claude Code and MCP who needs to close the gap between working knowledge and exam answers.
Days 1–4 — Domain 1. Same as above, and do not compress it. Read all seven lessons, do the quizzes, and build the loop even if you think you already know it. The exam's version of "agentic loop" is more precise than most working implementations.
Days 5–7 — Domains 3 and 4. Read the lessons, do every quiz, skim the build exercises and actually do the two you are least confident about.
Days 8–10 — Domains 2 and 5. Lessons and quizzes. The MCP tool-description exercise is the one to do if you only do one.
Days 11–12 — mock exams. Two full timed attempts with a day between. Study the gaps between them, not everything.
Day 13 — the divergence log and the cheat sheets. This is where the "answer from the guide, not from this week's docs" habit gets installed.
Day 14 — nothing. Cramming the day before a judgement test does not work.
The weekend plan (~10 hours)
Not recommended, but sometimes the exam is booked and the weekend is what is left. This is triage, and it only works if you already build with these tools daily.
Saturday morning: mock exam, cold. Two hours.
Saturday afternoon: the two weakest domains from the score report, quizzes only. Skip the build exercises — there is no time for the way they teach.
Saturday evening: all five cheat sheets, then the six scenarios.
Sunday morning: Domain 1 lessons regardless of your score. It is 27% and it feeds the rest.
Sunday afternoon: second mock exam. Then the divergence log and the FAQ.
What you are giving up is real: the build exercises are where judgement comes from, and this plan skips them. If you can move the booking, move it.
What to cut when time runs short
In order, the first things to drop:
- Re-reading lessons you have already done. It feels productive and teaches nothing. Do a quiz instead.
- Out-of-scope material. The guide publishes the list — fine-tuning, model internals, embeddings and vector databases, computer use, vision, streaming, rate limits and pricing, OAuth, cloud provider specifics, MCP server hosting. If you are reading about any of it, stop.
- Domain 5 build exercises before Domain 1 build exercises. Weight order applies to what you cut as much as to what you do.
- The glossary, as a thing to read front to back. It is a reference; use it when a term stops you.
The last thing to cut, always, is the mock exam. Without the per-domain breakdown you are guessing at what to study, and guessing is how people spend nine hours on the domain worth 15%.