The Claude Architect Professional exam changes on 7 December - Here is what's different
DEV Community

The Claude Architect Professional exam changes on 7 December - Here is what's different

This is part 5 about ClaudeArena, the free practice site I built for Anthropic's Claude Certified Architect exams (CCAR-F and CCAR-P). Last time it was about what AI agents score on my practice exam. This time the exam itself moved.

On 17 November Anthropic published version 1.1 of both exam guides. For Foundations almost nothing changes. For Professional it's a different exam from 7 December. I read both guides in full and compared them with v1.0 line by line. Here is what I found, and what I did about it.

Which guide applies to you

  • You sit Professional before 7 December: you're tested on v1.0. Keep studying what you're studying
  • You sit Professional on or after 7 December: you're tested on v1.1. Read on
  • You sit Foundations: nothing that is tested has changed. See the end of this post

What stays the same in Professional

120 minutes, a pass mark of 720 on a scale of 100 to 1,000, a $175 fee, and a credential that is valid for 12 months. The seven domains keep their names.

What changes

More questions, and not all of them count. v1.0 had 63 questions. v1.1 has 70, of which 60 are scored and 10 are not. You aren't told which 10, so you answer all 70 as if they count, with less time for each: about 1 minute 43 seconds a question, down from 1 minute 54.

Every question is now a scenario. v1.0 described its questions as multiple-choice and multiple-response. v1.1 calls them scenario-based multiple-choice. The three new sample questions show what that means: each one is a long situation to read before you get to the options.

The weights moved, and one domain was cut in half:

  • Domain 1, Solution Design and Architecture: 17% -> 18%
  • Domain 2, Claude Models, Prompting, and Context Engineering: 13% -> 13%
  • Domain 3, Integration: 19% -> 21%
  • Domain 4, Evaluation, Testing, and Optimization: 16% -> 17%
  • Domain 5, Governance, Safety, and Risk Management: 14% -> 16%
  • Domain 6, Stakeholder Communication and Lifecycle Management: 14% -> 7%
  • Domain 7, Developer Productivity and Operational Enablement: 7% -> 8%

Domain 6 lost half its weight and is now the smallest domain. If you spent a lot of time on discovery and lifecycle phases, most of that weight now sits in Integration and Governance.

The objectives were rewritten. v1.0 listed short one-line tasks. v1.1 has 39 named skills, each with a description, written as things to evaluate, analyse, identify, select or diagnose. That's a change in what is asked of you: less "do you know what this is", more "here is a design or a failure, what is wrong with it".

New ground in v1.1

These are named in the v1.1 objectives and weren't objectives of their own in v1.0:

  • Choosing a deployment surface and an agent runtime, including Claude Managed Agents
  • Deciding what goes to Claude, what stays with people, and what they do together
  • Cache breakpoints, reasoning effort and how a context is managed over its life
  • Reliability controls for integrations in production
  • Contextual retrieval
  • Bias in an LLM used as a judge, and how to design an experiment
  • Moving a system to a new model version
  • Where prompt injection can get in, and ranking risks against each other
  • Mapping compliance duties to controls in the architecture
  • Plugins, marketplaces and managed settings for Claude Code, and its permission modes
  • Who is accountable for code an AI wrote

How I would prepare for v1.1

The guide's own preparation section changed too, and it's worth taking at its word:

  • Know Foundations first. v1.1 says the scope of the Foundations certification is assumed knowledge. It isn't a prerequisite, but it is treated as known
  • Practise reading other people's work. The guide says exam scenarios present artifacts such as architecture diagrams, configuration files, logs, traces, evaluation results and code, and expect you to reason from them
  • Study by weight. Integration, Solution Design, Evaluation and Governance are 72% of the scored questions
  • Train your reading speed. Longer scenarios and less time per question is the part most people will feel

What I built

ClaudeArena has one rule above all others: the exam guide decides every question and every answer. So a new guide meant a new question bank, not a patch on the old one.

  • 140 new questions written from the v1.1 guide, spread over the seven domains by the new weights and covering all 39 skills
  • Both versions side by side until 7 December. On Professional you choose "the exam until 6 December" or "the exam from 7 December". The v1.0 questions weren't touched
  • A full v1.1 mock exam: 70 questions in 120 minutes. All 70 are scored on the site, because I can't know which 10 the real exam leaves out
  • Its own leaderboard and its own medals, so a score always says which version it was earned on
  • A v1.1 section in the study guide with the new weights and all 39 skills

The new questions are at Normal difficulty. Hard and Expert for v1.1 are marked "Coming soon" on the site, and I'd rather show that than fill the gap with questions that haven't been properly reviewed and verified.

A tell I had to write a check for

Every batch of new questions was reviewed independently before it went in. One finding had nothing to do with the content. In a number of drafts the right answer was the only option built as "do this and that", while the three wrong ones were single actions. You could pick it without reading the scenario.

Reviewing by eye catches that sometimes. A build check catches it every time, so there is one now: a right answer can't be the only option of its shape. It stopped 15 drafts. That brings the checks that run before every deploy to 83.

And Foundations?

The v1.1 Foundations guide changes registration, identification and support. The domains, weights, task statements and sample questions are the same as in v1.0, so the Foundations questions already follow it.

One practical change, in both guides: you can cancel or reschedule an online exam up to 1 hour before it starts, and an exam at a test center up to 48 hours before. In v1.0 it was 24 hours for both.

Try it

Still free and unofficial, with no account and no email needed. The v1.1 practice is live now, next to v1.0.

→ https://claudearena.dev

A caveat I owe you: I haven't seen the real v1.1 exam and nobody outside Anthropic has. The questions follow the published guide. Whether they feel like the real thing is something only people who sit it after 7 December can say.

One question for you: are you sitting Professional before or after 7 December, and did the new guide change your plan?

Read on DEV Community ↗ ← Back to News

Comments

No comments yet. Start the discussion.