How to Pass the Claude Certified Architect (CCA-F) Exam
A field-tested game plan: where the exam actually tests you, where to spend limited study time, and the habit that moves your score more than anything else.
The CCA-F is not a trivia test about API parameters. Every question is a scenario: given a set of production constraints, pick the architectural decision that best protects them. That single fact should change how you study.
The exam at a glance
- 60 questions, 120 minutes, Skilljar-proctored
- Scaled score 100–1,000; pass mark 720
- ~301-level — assumes roughly six months of hands-on work with the Claude API and Claude Code
- Domain-weighted: Agentic Architecture 27%, Claude Code Config 20%, Prompt Engineering 20%, Tool Design & MCP 18%, Context Management 15%
Where to spend your time
Domain weighting matters more than breadth. Agentic Architecture alone is worth more than Tool Design & MCP and Context Management combined. If you're time-constrained, over-invest in the top two domains — Agentic Architecture and Claude Code Configuration — and you'll clear the pass mark comfortably even with weaker coverage elsewhere.
The habit that matters most: interrogate every wrong answer
For every question you get wrong, don't just read the explanation and move on. Ask: what production concern did the right answer protect that mine didn't? The answer is almost always one of five things — latency, cost, observability, reliability, or human-in-the-loop safety. Once you can name which concern a scenario is testing, the "best" answer usually becomes obvious even among several defensible options.
Three-week structure
Week one: take the diagnostic to find your weakest domain, then work through that domain's glossary and quick reference while doing daily practice sets filtered to it. Week two: rotate through the remaining domains in weight order, re-attempting missed questions after 48 hours so the reasoning actually sticks — practice mode surfaces these automatically. Week three: full-length timed mock exams under real conditions, reviewing every explanation, even on questions you got right.
Common mistakes that cost points
- Treating stop_reason as optional. Any loop-control design that parses assistant text instead of branching on
stop_reasonis wrong, no matter how reasonable it sounds. - Ignoring the CLAUDE.md hierarchy. A specific-scope file adds to more general ones — it doesn't replace them. Options built on "override" logic are traps.
- Picking the more powerful option over the more appropriate one. The exam consistently rewards the narrowest tool/permission/pattern that solves the actual scenario, not the most capable one available.
Exam-day tactic
Read the scenario twice before looking at the options — the correct answer usually pivots on one specific constraint mentioned in passing. When two options both seem defensible, pick the more production-grade one: the one that degrades more gracefully under failure, not just the one that works on the happy path.
Realistic timeline: two to three weeks at about an hour a day, almost all of it spent on practice questions rather than reading. Start a full mock exam once you've cleared the diagnostic to see exactly where you stand.