The Problem

Anthropic’s Claude Certified Architect exams (CCAR-F & CCAR-P) test real scenario judgment, not documentation recall. Official resources give only sample questions, so most engineers have no clear signal of readiness.

Existing practice sites have three common problems:

1. Aggressive gating — email, account, or credit card before you see a single scenario

2. Coarse feedback — domain-level scores without telling you which exact skill is weak

3. Documentation drift — exam blueprints freeze while SDKs, MCP, and APIs keep changing

What I Built

ClaudeArena is a free, privacy-first practice engine for AI engineers and solutions architects preparing for the CCAR exams.

• 950+ original scenario questions focused on agentic workflows, subagent isolation, and tool-choice guardrails
• Objective-level gap analysis that surfaces precise weaknesses (e.g. “Configure subagent termination conditions”) instead of vague domain scores
• Server-side answer validation — no client-side answer leakage
• Every question cites the exact Anthropic docs page it rests on
• Practice mode needs no account, no email, and sets zero tracking cookies

Architectural choices

• Edge APIs for scoring so the question bank can’t be scraped via DevTools
• Heavy real-time audit effort on the Foundations exam (the one getting the next content update) while keeping Professional on the current baseline
• Deep page-level citations + automated link health checks to fight doc rot

Feedback I’m looking for

1. How do you monitor upstream documentation when URLs restructure?

2. Is sub-task / objective drilling more useful in your study flow than full timed mocks?

Try a free drill (no signup): https://claudearena.dev/app#practice