17 · Practice Questions¶

Unofficial. Every question here was written independently for this repository. None was taken from, adapted from or reconstructed from an Anthropic exam, an exam dump or any other question bank, and none should be assumed to resemble live exam content. Anthropic's exam policy treats all real exam content as confidential.

What's here¶

Track	Folder	Files
CCAO-F — Claude Certified Associate – Foundations	`ccao-f/`	`d1.json` … `d7.json`, one per exam domain
CCAR-F — Claude Certified Architect – Foundations	`ccar-f/`	`d1.json` … `d5.json`, one per exam domain; each question is tagged with one of the six exam scenarios
CCAR-P — Claude Certified Architect – Professional	`ccar-p/`	`d1.json` … `d7.json`, one per exam domain

Question counts per domain roughly follow the official blueprint weights in registry.yaml, and every official objective has several questions. The generated CCAO-F and CCAR-P coverage matrices show which questions test each objective.

How to practise¶

Interactively: use the mock-exam engine, in timed or learning mode, by domain, or by retrying questions you missed.
By reading: the JSON is human-readable. Cover the answer field and work through a domain file.

Question format¶

Each domain file validates against schema/question-bank.schema.json:

{
  "schema_version": 1,
  "track": "CCAR-P",
  "domain": 3,
  "unofficial": true,
  "notice": "Independently authored practice questions. Not official Anthropic exam content.",
  "questions": [
    {
      "id": "CCAR-P-D3-001",
      "objective": "CCAR-P-3.1",
      "type": "single",
      "difficulty": "intermediate",
      "category": "security",
      "scenario": "A fictional organisation and situation…",
      "question": "What should the architect do first?",
      "options": {"A": "…", "B": "…", "C": "…", "D": "…"},
      "answer": ["C"],
      "explanation": "Why C is right, and the principle behind it.",
      "why_incorrect": {"A": "…", "B": "…", "D": "…"},
      "module": "07-tool-use-and-function-calling/06-tool-security.md",
      "references": ["https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview"]
    }
  ]
}

Field	Rule
`id`	`TRACK-D<domain>-<nnn>`, unique across the repository
`objective`	An objective ID from the registry, belonging to the file's domain
`type` / `select`	`single` has exactly one key; `multiple` has 2+ keys and `select` equal to the number of keys
`difficulty`	`foundation`, `intermediate` or `advanced`
`category`	`concept`, `scenario`, `architecture`, `troubleshooting`, `security` or `cost-performance`
`why_incorrect`	One rationale for every distractor, and only for distractors
`module`	A learning page that teaches the concept (must exist)
`references`	Official `https://` documentation URLs
`scenario_set`	Scenario-based exams only (CCAR-F): one of the scenario IDs in the registry, such as `support-agent` or `data-extraction`

Quality rules¶

Test reasoning and judgment in a realistic scenario, not recall of trivia.
Use fictional organisations and synthetic data.
Make every distractor plausible, and explain why it is wrong.
Make every question distinct. The validator rejects stems that are more than 85% similar.
Base answers only on facts verified against official documentation, as of the date in the registry.
No test-taking cues. Write distractors as competent but inferior choices of similar length to the key. The validator fails a bank when the key is the longest option in more than 40% of items, or when one answer position is used far more or less often than chance.

After adding or editing questions, run python tools/shuffle_options.py --ids <the IDs you changed>. It gives every question a fair, reproducible answer order (seeded by the question ID) and remaps the key and rationales, so never refer to options by letter in question text.

Validation¶

python tools/validate_questions.py

The validator checks the schema, unique IDs, objective/domain consistency, that answer keys are valid options, select counts, distractor rationales, explanations that claim a different letter is correct, module paths and near-duplicates. CI runs it on every pull request.

Contributing questions¶

Read CONTRIBUTING.md and docs/STYLE.md. Add a question to the right domain file with the next free ID, run the validator, then regenerate the coverage matrix with python tools/build_coverage.py.