Skip to the main content

Exam guide

Claude Certified Architect (CCAR-F) exam guide

What the Claude Certified Architect — Foundations (CCAR-F) exam looks like: its length, how it is scored, and what each of its five domains asks of you.

SpacedPublished 4 min read

Nearly every question on the Claude Certified Architect — Foundations exam (CCAR-F) describes a system built on Claude that is misbehaving (an agent that stops early, a tool the model keeps picking wrongly, a rule that is ignored one time in a hundred) and asks what you would do about it. The exam tests judgement under constraints.

The format

The examDetail
Questions60
Time120 minutes, about two a question
Options4 a question; each says how many answers to select
Pass mark720 on a scale that runs to 1000
Wrong answersNo penalty stated in the guide
DeliveryPearson VUE, online proctored or at a test centre
Help allowedNo AI assistance and no documentation
Rule · remember thisAnswer every questionNothing in the exam guide penalises a wrong answer over a blank one, so answer everything. If a question is eating your time, cross out what you can, choose between what is left, and move on.

Some questions ask for two answers, and each question says how many to select. Spaced marks them all or nothing, naming one of the two right options scores the same as naming neither, so treat the pair as a single piece of knowledge.

How it is scored

Your result is a scaled score, so aim to clear 720 of 1000 with room to spare. Near the line a couple of questions either way decide the result.

The domains and their weights

The exam guide splits the content into five domains, and weight maps almost directly to question count: at 60 questions, one percentage point of weight is a little over half a question.

DomainWeightQuestions
1. Agentic Architecture & Orchestration27%about 16
2. Tool Design & MCP Integration18%about 11
3. Claude Code Configuration & Workflows20%about 12
4. Prompt Engineering & Structured Output20%about 12
5. Context Management & Reliability15%about 9
  • 1. Agentic Architecture & OrchestrationWeight: 27%Questionsabout 16
  • 2. Tool Design & MCP IntegrationWeight: 18%Questionsabout 11
  • 3. Claude Code Configuration & WorkflowsWeight: 20%Questionsabout 12
  • 4. Prompt Engineering & Structured OutputWeight: 20%Questionsabout 12
  • 5. Context Management & ReliabilityWeight: 15%Questionsabout 9

Domain 1 is the heaviest and Domain 5 the lightest. The lightest domain still carries about 9 questions, enough to decide a result near the pass mark.

What each domain asks

  • Agentic Architecture & Orchestration (27%): how an agent loop decides whether to continue, how a coordinator hands work to subagents and what context they start with, and where a hook rather than a prompt has to enforce a rule.
  • Tool Design & MCP Integration (18%): how a model chooses a tool (by its description, not its name), how errors should come back, what tool_choice does and does not guarantee, and how MCP hosts, clients and servers connect.
  • Claude Code Configuration & Workflows (20%): how CLAUDE.md files combine, the difference between a command and a skill, rules scoped to paths, when plan mode is the right call, and running Claude Code non-interactively in CI.
  • Prompt Engineering & Structured Output (20%): naming criteria explicitly, few-shot examples, schemas that do not force the model to invent, retries that carry the real error, and when the Batch API is worth its latency.
  • Context Management & Reliability (15%): what to carry forward through a long context, how to escalate with the context intact, why a silent failure is the worst kind, and how to measure whether output can be trusted.

How a question is built

Each question sets out a production scenario and a symptom, then offers 4 remedies. Read the last sentence first. It tells you whether the exam wants the root cause, the best first step, the most effective solution, the correct approach or one choice for each of three cases. On the same scenario those have different right answers. How to answer CCAR-F questions walks through the method.

Trap · looks right, isn’tA bigger number is the tempting answerMore retries, a larger context window, a higher iteration cap. These sound responsible, and they answer a problem of size when the question describes a different one. The wrong-answer patterns covers the rest.

Before you book

  • Registration needs a work email from an organisation in the Claude Partner Network. Membership starts free at its entry level.
  • A failed attempt starts a waiting period before you can try again, and the wait grows with each further failure. Attempts are also capped within a rolling year, so it pays to prepare for a first-attempt pass.
  • Exam facts here follow the CCAR-F Exam Guide v1.0 (July 2026). Check the official guide for fees and current policy before you book.

How to prepare

The exam rewards telling similar mechanisms apart under time pressure. Producing a rule yourself, before you check it, is what makes it stick. Space that practice out, so each topic comes back just as you were about to lose it.

Spaced is built on that. Every module of the exam is taught, then brought back on a plan that runs itself around your evenings, with every mistake filed under the principle behind it. Domain 1 is free in full.

All posts

Ready to learn? Let’s get started.

Three questions and a short baseline quiz, and you have a plan. Whether you are booked weeks out or panicking days out, you can start tonight, with or without an exam date.

Start learning

Domain 1 is free. No card needed.