# CCAR-P: the exam that asks you to justify, not to configure
The only Professional-level exam of the four, and the most expensive. Its objectives use verbs like evaluate and justify, and two of seven domains are about talking to people.
**Time:** 8 minutes

**What you'll have when you're done:** The exam facts, the seven domains and their unusually flat weights, what the objective verbs tell you about question style, and a preparation checklist.
---Claude Certified Architect, Professional (exam code CCAR-P) is the only
Professional-level exam in the program and the priciest of the four. It tests judgment
rather than implementation.

**It is not a step up from Architect Foundations.** CCAR-F is not required, holding it
does not upgrade into this, and the two credentials are separate. Anthropic frames
Foundations as proof an architect can build with Claude, and Professional as proof they
can design and govern Claude solutions at organizational scale.

Eligibility is the program-wide rule: Partner Academy registration, work email at a
Claude Partner Network company. See
[the program overview](/guides/claude-certification/claude-certification-what-it-is/).

## Exam at a glance

| | |
|---|---|
| Exam code | CCAR-P |
| Questions | 63, across seven domains |
| Time limit | 120 minutes. Budget about 135 in the chair. |
| Passing score | 720 on a scale of 100 to 1,000 |
| Valid for | 12 months |
| Retakes | 14 days, then 30, then 90. Four attempts per rolling 12 months. |
| Delivery | Proctored. Online with a webcam, or at a Pearson VUE test center. |
| Prerequisites | None. CCAR-F is not required. |
| Language | English only. No browser translation. |

This is reportedly the only exam of the four to use three question types: standard
multiple choice, multiple response (commonly two correct from five), and a
scenario-matching style with dropdowns.

## Who it is for

The guide targets mid-level to senior architects: solution architects, AI and ML
engineers, technical leads and senior engineers who turn business problems into systems
that scale. It suggests three or more years in systems architecture or platform
engineering, six months or more hands-on with Claude in production, and experience
carrying a system from first discovery through to running it live.

Note what that profile includes. Stakeholder conversations, client advice, leading design
decisions, and handling security, legal and executive discussions. **Two of the seven
domains, 21% of the exam, are about communication and lifecycle rather than technology.**
That is the portion engineers most reliably skip and most reliably lose marks on.

It rules out entry-level developers, casual users, and anyone who has never designed a
whole system.

## What it covers, and what each part is worth

| Domain | Weight | Roughly |
|---|---|---|
| Integration | 19% | 12 questions |
| Solution Design and Architecture | 17% | 11 questions |
| Evaluation, Testing and Optimization | 16% | 10 questions |
| Governance, Safety and Risk Management | 14% | 9 questions |
| Stakeholder Communication and Lifecycle Management | 14% | 9 questions |
| Claude Models, Prompting and Context Engineering | 13% | 8 questions |
| Developer Productivity and Operational Enablement | 7% | 4 questions |

This is the flattest weighting in the program. Nothing above 19%, nothing below 7%. No
domain can be safely dropped, and none rewards obsession.

Integration, the heaviest, is where the concrete engineering lives: retrieval pipelines,
chunking and indexing strategy matched to the shape of the data and the kind of question
being asked, and the choice between MCP, plain API or CLI calls, and agent-to-agent
protocols. Then three things that are easy to skip and clearly deliberate inclusions:
spotting capability bloat (an agent holding tools its role never needs), finding gaps in
authentication and authorization, and choosing how to monitor a system at scale.

## The verbs tell you the question style

The most useful signal in this guide is grammatical. Where CCAR-F says implement and
configure, this one says **evaluate, analyze, justify, select and diagnose**.

That changes what preparation looks like. Knowing that a retrieval strategy exists is not
enough. You need a defensible reason for choosing it over the alternative, in a specific
situation, with the trade-off named. Governance names GDPR, HIPAA and FedRAMP as design
constraints, plus human-in-the-loop validation and bias, fairness and transparency.
Lifecycle asks you to run structured discovery, explain trade-offs, manage expectations
including service levels, and document an architecture.

One honest warning: there is no official practice exam, and the guide holds three sample
questions for a 63-question exam across seven domains. That is the thinnest official
signal about question style anywhere in the program, and it is worth factoring into how
confident you feel before booking the most expensive sitting of the four.

## Does it count for your employer?

Yes. Architect Professional is one of the three that count toward Claude Partner Network
eligibility, alongside Architect Foundations and Developer Foundations. The recognized
work domain rule applies, or the pass is not credited to the company.

## Preparation checklist

**Before you study anything**

- Partner Academy account on a recognized work domain
- Read the exam guide end to end, and accept that three sample questions is all the
  official style guidance you get
- Open the official prep course, five lessons, several hours each
- Do not treat CCAR-F as a prerequisite. It is not one.

**Study**

Work all seven domains. The flat weighting means the usual strategy of front-loading the
biggest domain buys you very little here. If you are going to over-invest anywhere,
over-invest in the two people-facing domains, because they are 21% of the exam and the
part most technical candidates have never revised for in their lives.

**Hands-on, aimed at defending a choice rather than recalling a setting**

- Design a retrieval pipeline and write down why you chunked and indexed it that way
- Compare MCP against direct API or CLI calls against agent-to-agent, and state when each
  wins
- Put real numbers on an accuracy against latency trade-off
- Audit an agent for capability bloat: tools its role never needed
- Sketch monitoring and observability for a system running at scale
- Treat GDPR, HIPAA and FedRAMP as design constraints you can name, not as legal reading
- Run a discovery conversation, then write the service level commitment that comes out
  of it

## When it breaks

**"I am an engineer and I lost the exam on the soft domains."** The most common failure
shape here, and it is visible in the blueprint before you sit. Nine questions on
stakeholder and lifecycle management is not a rounding error.

**"I took CCAR-F, so I expected an overlap."** Different exam, different verbs. Recall
of file paths and parameter values is not what this one is measuring.

**"I no-showed and lost the fee."** The costliest no-show in the program. Cancel or
reschedule at least 24 hours ahead. Canceling in Pearson VUE frees the slot but does
not refund you; refunds go through Anthropic certification support.

Registration, ID, managed-device and badge problems work the same as the rest of the
program and are written out in full on
[the CCAO-F page](/guides/claude-certification/cert-ccao-f/).

<div class="source-note">
<span class="source-note__label">Where these facts come from</span>
<p>The exam guide (version 1.0, effective July 2026) and the Partner Academy FAQ are the sources of record. This page restates them in plain language and adds nothing confidential.</p>
<p>Weights and community reporting are cross-checked against <a href="https://claude.termidy.com/ccar-p/">claude.termidy.com</a>, an independent prep site, and policy and scheduling detail against <a href="https://www.pearsonvue.com/us/en/anthropic.html">Pearson VUE</a>. Confirm on the Partner Academy before you pay.</p>
</div>