Claude Cert Practice

A free, community-built practice-exam platform for the Claude certifications:

Exam	Track	Level	Questions
CCAO-F	Associate	Foundation	316
CCAR-F	Architect	Foundation	456
CCAR-P	Architect	Professional	309
CCDV-F	Developer	Foundation	678

Built with Next.js (App Router) + Tailwind CSS. No database, no accounts — progress is stored in the visitor's browser. The quiz needs no AI at runtime; it just serves pre-generated question JSON, so hosting is cheap.

Features

• Practice mode — answer one at a time, get instant right/wrong feedback and a full explanation.
• Exam mode — timed simulation with a question navigator, flag-for-review, score, pass/fail, and a per-domain breakdown.
• Filter by domain, choose 10 / 25 / 50 / all questions, shuffled each attempt.
• Light/dark aware, responsive, keyboard-friendly.

Run locally

npm install
npm run dev
# open http://localhost:3000

Build for production:

npm run build && npm start

Project structure

app/                 # routes (home, /about, /exam/[exam])
components/           # ExamRunner (setup/run/results), QuestionView
lib/                 # types, exam metadata, quiz logic, localStorage
content/*.json       # the question banks the app serves  (committed)
data/source/*.json   # private raw source banks           (gitignored)
scripts/rewrite.mjs  # rewrite/expand pipeline (see below)

Question format (content/<EXAM>.json)

{
  "id": "CCDV-F-001",
  "exam": "CCDV-F",
  "type": "single",           // or "multi"
  "domain": "Agent SDK",
  "question": "…",
  "options": [{ "id": "A", "text": "…" }, { "id": "B", "text": "…" }],
  "correct": ["A"],
  "explanation": "…",
  "rewritten": true
}

You can hand-edit these files directly — add, fix, or remove items. The domain filter is derived automatically from whatever domain values appear in the data.

Generate original questions from the official objectives (recommended for a public site)

The cleanest way to build a publishable bank: generate fresh questions from the official exam objectives (data/objectives.json, drawn from Anthropic's score-report blueprints) rather than adapting a third-party bank. Nothing is derived from any vendor's questions.

# PowerShell
$env:AI_GATEWAY_API_KEY = "your_gateway_key"
npm run generate                       # all exams, ~3 questions per objective (~300 total)
npm run generate -- --exam CCAO-F --per 5
npm run generate -- --exam CCDV-F --limit 2 --per 3   # cheap trial (first 2 objectives)
npm run generate -- --dry              # no API calls; just prints the plan
npm run generate -- --append           # add to the existing bank instead of replacing

• Writes content/*.json (what the app serves). Replaces that exam's bank unless you pass --append. Your private PassQuestion-derived bank stays in data/source/*.json and is untouched.
• Model via GENERATE_MODEL (or REWRITE_MODEL); default anthropic/claude-sonnet-4.5. Use a smaller model (e.g. anthropic/claude-haiku-4.5) to cut cost.
• Edit data/objectives.json to add/adjust objectives or change coverage.

Rewrite / expand pipeline (alternative: adapt the source PDFs)

The banks in content/ were extracted from source PDFs and cleaned. To turn them into polished, original items (freshly worded stems, expanded explanations, tidier domain tags) — recommended before publishing publicly — run the rewrite pipeline. It uses a Claude model through the Vercel AI Gateway.

# PowerShell
$env:AI_GATEWAY_API_KEY = "your_gateway_key"
npm run rewrite                      # all four exams
npm run rewrite -- --exam CCDV-F     # one exam
npm run rewrite -- --exam CCAO-F --limit 20   # first 20 pending (a cheap trial run)
npm run rewrite -- --force           # re-rewrite everything

• The pipeline reads data/source/*.json, rewrites each question, and writes content/*.json. It resumes — items already marked "rewritten": true are skipped unless you pass --force.
• The correct answer id(s) are taken from the source and never changed by the model.
• Pick the model with REWRITE_MODEL (default anthropic/claude-sonnet-4.5); use a smaller model to cut cost, e.g. REWRITE_MODEL=anthropic/claude-haiku-4.5.
• Prefer a direct provider instead of the Gateway? Set ANTHROPIC_API_KEY and change the model string in scripts/rewrite.mjs to an @ai-sdk/anthropic model.

Deploy to Vercel

1. Push this folder to a Git repo (GitHub/GitLab/Bitbucket).
2. In the Vercel dashboard, import the repo. Framework preset Next.js is detected automatically; no environment variables are needed for the site itself.
3. Deploy. That's it.

Or with the CLI:

npm i -g vercel
vercel          # preview
vercel --prod   # production

Content & licensing note

The site is live at https://claudearchitects.org.

The bank currently committed in content/ is lightly-cleaned PassQuestion text — a paraphrase of a commercial question bank, not original work. It is being served publicly today. Replacing it with generated originals is the outstanding task on this project; until that happens, the published items remain a derivative of someone else's paid product and the project carries that risk.

Two content paths produce the bank:

• npm run generate — original questions authored from the official exam objectives in data/objectives.json. Safe to publish, and the path this project should be on.
• npm run rewrite — a close paraphrase of the source PassQuestion PDFs. Lower risk than verbatim, but still a derivative of a commercial bank; suitable for private study only.

To move the public site onto original content, run npm run generate (it replaces each exam's bank unless you pass --append), review the output, and deploy. The private PassQuestion-derived source stays in data/source/ and is gitignored.

This project is not affiliated with, endorsed by, or sponsored by Anthropic. "Claude" and the certification names are trademarks of their respective owner and are used only to describe what the practice content covers.