CCAR-PAnthropicExam FormatClaudePearson VUE

CCAR-P Exam Format & Structure: What to Expect on Test Day [2026]

Preporato TeamAugust 16, 202614 min readCCAR-P
CCAR-P Exam Format & Structure: What to Expect on Test Day [2026]

The Claude Certified Architect - Professional (CCAR-P) exam is a 63-question, 120-minute, Pearson VUE-delivered test with a 720-out-of-1000 passing bar, and its format rewards candidates who rehearsed the mechanics as carefully as the material. This article walks through the mechanics that shape how you should sit the exam: how scaled scoring works in plain terms, the three question styles you will meet and how each one is built, what "scenario-based applied reasoning" looks like in two worked examples, the Pearson VUE rules for online and test-center delivery (ID, check-in, room requirements, what you cannot use), a pacing plan built on 1.9 minutes per question, how to read the domain-level score report, and the retake and renewal rules as Anthropic publishes them in 2026.

Start Here

If you are new to the certification, read the CCAR-P complete guide first for the seven domains, the study path, and who the exam is for. When you want to rehearse the format described below, Preporato's CCAR-P practice tests give you 6 full-length, 63-question timed exams built on the same 7-domain blueprint, with a multiple-response share that matches the real thing and an explanation for every answer. A free 20-question sampler is at /free/claude-certified-architect-professional/questions.

Exam Quick Facts

Duration
120 minutes
Cost
$175 USD
Questions
63 questions (all scored)
Passing Score
720 out of 1000
Valid For
1 year
Format: Pearson VUE test center or online proctored

The exam at a glance

CCAR-P is the Professional tier of Anthropic's Claude Certification Program, sitting above the Foundations-level CCA-F. There are no formal prerequisites: CCA-F is helpful background and is never required. Registration runs through the Anthropic Partner Academy, which requires free membership in the Claude Partner Network. You purchase the exam there for $175, receive Pearson VUE credentials by email if this is your first Pearson exam, and then schedule a slot either at a Pearson VUE test center or through OnVUE, Pearson's online proctoring platform. Passing candidates receive a digital badge through Credly.

The exam itself is 63 questions, every one of them scored, answered inside 120 minutes. Anthropic's certification FAQ tells candidates to plan for about 135 minutes of total seat time, which covers check-in, the tutorial screens, and the wrap-up. The exam is delivered in English, is closed book, and uses what Anthropic describes as multiple-choice and scenario-based multiple-response questions. Your result appears at the end of the session.

CCAR-P format details

AspectDetail
Questions63, all scored (no unscored pilot items to hide behind)
Time120 minutes to answer; plan about 135 minutes of seat time
Passing score720 on a 100 to 1,000 scaled range
Question stylesSingle-answer multiple choice, multiple-response (Select TWO or THREE), long multi-constraint scenario stems
DeliveryPearson VUE test center or OnVUE online proctoring
RegistrationAnthropic Partner Academy (free Claude Partner Network membership required)
Cost$175 USD per attempt
LanguageEnglish
Validity12 months, with a free non-proctored renewal assessment before expiration
PrerequisitesNone (CCA-F recommended background, never required)

Preparing for CCAR-P? Practice with 390+ exam questions

How scaled scoring works (and what 720 means)

Anthropic reports CCAR-P results as a scaled score between 100 and 1,000, with 720 as the minimum pass for all four Claude certifications. Scaled scoring exists because an exam program maintains more than one version, or form, of the test, and forms are never perfectly equal in difficulty. Instead of reporting the raw number of correct answers, the program converts each candidate's raw performance onto a common scale so that a 720 earned on a slightly harder form represents the same demonstrated ability as a 720 earned on a slightly easier one.

Three practical consequences follow. First, 720 is a position on the scale; Anthropic does not disclose the raw-to-scaled conversion, so no published percentage of the 63 questions corresponds to it. The working rule Preporato uses across its CCAR-P material is to treat roughly 75 to 80 percent accuracy as the bar and to aim above 80 percent on timed practice tests before booking. Second, because all 63 questions count, there are no experimental items you can afford to guess through, and an unanswered question earns nothing, so every item deserves an answer. Third, Anthropic's public certification FAQ states no partial-credit rule for multiple-response items, so the safe preparation assumption is that a "Select TWO" question pays out only when both selections are right.

The score report you see at the end shows your overall scaled score, a pass or fail result, and the percentage of items you answered correctly in each section. The sections correspond to the seven exam domains, which makes the report a useful diagnostic even when you pass, and essential when you do not.

Scaled scoring in practice

QuestionWhat Anthropic publishesWhat to do about it
What range is reported?100 to 1,000, with 720 as the pass mark for all four Claude certificationsTreat roughly 75 to 80 percent accuracy as the working bar and aim above 80 percent on timed practice tests
How many raw answers make 720?Not disclosed; the raw-to-scaled conversion is privateDo not chase a percentage; build a buffer across every domain
Do unanswered items cost anything?All 63 questions are scored and a blank earns nothingAnswer every item, even when guessing
Is there partial credit on Select TWO or THREE?No partial-credit rule is statedAssume the item pays out only when every selection is right
What does the score report show?Overall scaled score, pass or fail, and percent correct per sectionRead the sections as the seven domains and use them as your diagnostic

The three question styles you will meet

The official description names two formats, multiple choice and scenario-based multiple response. In practice candidates experience three styles, because the length and density of the stem changes how you work the question, and all three appear across every domain.

Single-answer scenario questions

These are the backbone of the exam. A stem describes an enterprise situation (a company, a workload, a constraint or a failure) and asks which design, diagnosis, or next step is MOST appropriate. Four options follow, and the wrong ones are legitimate techniques applied to a scenario they do not fit. The capitalized qualifier matters: several options would work, and you are ranking them against the constraints in the stem.

Multiple-response questions (Select TWO or THREE)

Roughly a quarter of the 63 items ask you to select two, or occasionally three, options from the list. These take longer because you are effectively making two or three independent true-or-false judgments, and because the correct pair almost always covers two different requirements written into the stem (one security control and one monitoring control, for example). A pair that leaves a stated requirement uncovered is the classic trap. If the interface offers a flag-for-review function (standard on Pearson VUE delivery), use it and confirm on your final pass that every multiple-response item has exactly the requested number of selections.

Multiple-Response Rule

No partial-credit rule is published for Select TWO or THREE items, so assume the point arrives only when every selection is right. The correct pair almost always covers two different requirements written into the stem, one to one, and a pair that leaves a stated requirement uncovered is the classic trap.

Long multi-constraint stems

This style is the professional-tier signature. Some stems run to a paragraph and carry three to five constraints at once: a compliance regime, a latency SLA (service-level agreement, the response-time commitment the business has made), a cost target with a deadline, a team skill profile, a stakeholder demand. The options are engineered so that each satisfies some constraints and violates one. Your job is to identify which constraint discriminates between the two finalists before you commit. These items can be single-answer or multiple-response; careful reading of the stem is most of the work.

CCAR-P question styles compared

StyleWhat it looks likeApproximate shareTime budgetTactic
Single-answer scenarioEnterprise situation, one MOST appropriate option out of fourAbout three quarters of itemsAbout 90 secondsFind the dominant constraint before reading options
Multiple-responseSelect TWO or Select THREE from the listAbout a quarter of itemsUp to 150 secondsJudge each option independently; map selections to stem requirements one to one
Long multi-constraint stemParagraph-length scenario with 3 to 5 stated constraintsA subset of both styles aboveUp to 3 minutesList the constraints, then eliminate any option that violates one

What scenario-based applied reasoning means: two worked examples

Anthropic labels the exam scenario-based, and the phrase has a specific meaning. A recall question asks whether you know what prompt caching is. An applied-reasoning question gives you a system with a budget problem and a latency constraint and asks what you would change first. The points come from matching the technique to the constraints. The two examples below are written in the exam's register (Preporato originals, shorter than the longest real stems), one in each major style.

Question 1

Domain: Claude Models, Prompting & Context Engineering (13%)

A retail bank runs a customer-support assistant on Claude that answers roughly 40,000 chats per day. Every request resends the same 6,000-token policy and tone preamble ahead of a short customer message. Finance wants input cost cut by a third this quarter, and operations will not accept any increase in first-token latency. Which change is MOST appropriate?

  • A. Serve the preamble as a stable prompt-cached prefix so repeated tokens hit the cache.
  • B. Route every chat through the Message Batches API to collect the asynchronous discount.
  • C. Drop to a smaller, faster model tier while keeping the full 6,000-token preamble unchanged.
  • D. Compress the policy preamble to about 500 tokens and remove the tone guidance entirely.

Answer: A

The waste here is a large, stable, repeated prefix, which is exactly what prompt caching (storing a stable prompt prefix so repeated tokens are processed at reduced cost and latency on later calls) is designed for; it cuts input cost without changing the model or the assistant's behavior. B fails the latency constraint: the Message Batches API (asynchronous bulk processing that returns results later at a discount) cannot serve a live chat. C changes model capability with no evaluation evidence and leaves the 6,000 repeated tokens in place, so the saving is partial and the quality risk unmeasured. D buys savings by altering system behavior, trading an unstated quality regression for the stated budget goal.

Prefix caching
req 0
prefix · prefilled
req 1
prefix · prefilled
req 2
prefix · prefilled
req 3
prefix · prefilled
req 4
prefix · prefilled
prefill computed prefix reused (skipped) unique query
system prompt1,800 tok
requests5
off: 9,150 prefill tokenscached: 1,95079% of prefill skipped
Compute the shared part once. When the first 1,800 tokens are identical across requests, their keys and values are identical too, so there is no reason to recompute them. Cache the prefix once and every later request skips straight to its own short tail. The longer the shared prompt and the more requests share it, the bigger the win, which is why chat products with long system prompts lean on this hard.

Drag the system prompt to its maximum and add requests: the striped prefix rows are the repeated preamble from Question 1, computed once and skipped on every later call, which is why option A cuts cost without adding latency.

Question 2

Domain: Governance, Safety & Risk Management (14%)

A hospital network is piloting a Claude assistant that drafts discharge summaries from patient records. The clinical sponsor wants drafts written directly into the electronic health record to save clinician time, and the compliance office has flagged that logs currently store full prompts. The system must comply with HIPAA. Which TWO controls should the architect require before launch? (Select TWO)

  • A. Require clinician review and sign-off before any draft is committed to the patient record.
  • B. Minimize protected health information in stored logs and tightly restrict who can read them.
  • C. Move the workload to the most capable model tier so drafts need less human correction.
  • D. Add a disclaimer to every draft stating the summary was generated by an AI system.

Answer: A and B

The stem states two gaps: an autonomous, hard-to-reverse write into a medical record, and logs that retain PHI (protected health information, the patient data governed by HIPAA, the US health-privacy law). A closes the first with a human-in-the-loop gate placed where the stakes justify it, and B closes the second with data minimization and access control on the logs. C is a model-selection change that addresses neither gap and adds cost without evidence. D is a legitimate transparency practice, but a disclaimer neither prevents an unreviewed write nor removes PHI from logs. Notice that the correct pair covers the stem's two requirements one to one, the pattern to look for on every multiple-response item.

Question 2, built

Build the human-in-the-loop gate from Question 2

A permission callback that holds the irreversible write until a human approves it, plus an audit trail of what the agent did, is the control option A describes; this project has you build the callback, the trail, and the escalation path.

Both examples show the same discipline: read the stem for the constraints, decide which one discriminates, and only then evaluate the options. The CCAR-P practice questions with explanations article gives you 20 more in this format across all seven domains, and the common CCAR-P exam mistakes article catalogs the ways strong architects still lose these points.

Pearson VUE logistics: online proctored vs test center

Both delivery modes present the identical exam, so the choice is about your environment. Online proctoring through OnVUE removes travel and widens your choice of appointment times, and it puts the burden of a compliant testing space on you. A test center provides the environment and the hardware and asks you to arrive early and follow its check-in procedure. Either way, you can reschedule or cancel free of charge up to 24 hours before your appointment; inside that window, or if you do not show, the fee is forfeited.

Online proctored (OnVUE) requirements

Pearson VUE publishes an Anthropic-specific online testing page, and the requirements below come from it. Failing any of them on exam day can mean immediate cancellation and a forfeited fee, so treat this list as a pre-flight checklist.

  • Technology: Windows 10 or macOS 14 or newer; a working webcam, microphone, and speaker (headphones and headsets are prohibited); one display only, with any second or touchscreen display disconnected and covered; at least 6 Mbps down and 2 Mbps up; and every application except OnVUE closed. Virtual machines, beta operating systems, VPNs, and corporate or public networks are prohibited. Run and pass the system test on the same device and network you will use on the day, and restart the machine beforehand.
  • Testing space: a private, quiet room where you remain alone for the whole session and nobody can see your screen. Your desk must be completely empty except for the testing computer and a beverage in an unmarked container; books, notes, paper, pens, electronics, food, bags, and wallets all leave the desk and arm's reach, and whiteboards in the room must be cleared. Bathrooms, offices, libraries, and coffee shops are prohibited spaces.
  • Check-in: begin 30 minutes before your appointment. You complete the technology checks, photograph yourself and your ID, and perform a 360-degree room scan. If a requirement is not met at this point, you cannot test and the fee is forfeited.
  • During the exam: the session is recorded and monitored by a proctor. You may not leave the webcam view unless the exam has confirmed you are on an approved break (not every exam offers breaks), may not speak or read aloud, and may not touch your phone unless a proctor explicitly permits it. Proctors are reachable through in-exam chat, but they cannot pause or extend the exam or fix your device.
  • Allowance: for Anthropic exams, candidates have access to a digital whiteboard inside the exam. That allowance does not extend to physical whiteboards or writing materials of any kind.

Test center delivery

At a Pearson VUE center you present the same government-issued ID, store personal items where the center directs, and sit at a workstation the center controls. Pearson's general guidance is to arrive early enough to settle in and to leave study materials at home; the how to pass CCAR-P first attempt guide suggests arriving 30 minutes ahead. Ask the administrator at check-in what note-taking surface, if any, the program allows, and follow the center's rules about breaks and leaving the room.

OnVUE online vs Pearson VUE test center

AspectOnVUE (online proctored)Test center
Testing environmentYou supply a private, quiet room with an empty desk and a compliant machineThe center supplies the room, the workstation, and storage for personal items
Check-inStarts 30 minutes early: system checks, photos of you and your ID, a 360-degree room scanArrive early enough to settle in and follow the center procedure
IDValid, unexpired government photo ID matching your Pearson VUE profile exactlySame government-issued ID requirement
Notes and scratch spaceDigital whiteboard inside the exam only; no paper or physical whiteboardAsk the administrator what the program allows
ReschedulingFree up to 24 hours before the appointment; inside that window the fee is forfeitedSame 24-hour rule
Trade-offWider appointment choice and no travel; the compliant space is your responsibilityThe environment is handled for you; you travel and arrive early

ID rules that catch people out

You need a valid, unexpired, government-issued photo ID whose name exactly matches the name on your Pearson VUE profile and booking. Accepted forms include an international passport, a plastic driver's license, and national, state, provincial, or EU ID cards. Expired, digital, damaged, copied, or privately issued IDs are refused, and so are birth certificates and naturalization papers. The most common failure is a mismatch between the name you typed at registration and the name printed on the ID (a missing middle name, a maiden name, an accent). Fix the profile before the day; you cannot fix it at check-in.

What you cannot use

The exam is closed book. Notes, documentation, browser translation tools, and AI assistants are all off limits. Phones, tablets, headphones, earbuds, styluses, watches, and smart or connected devices with recording or AI features (smart speakers, smart glasses) are prohibited from the testing space. There is no second monitor. Everything you bring into the exam has to be in your head, which is why the CCAR-P cheat sheet is built for final-week memorization.

Master These Concepts with Practice

Our CCAR-P practice bundle includes:

  • 6 full practice exams (390+ questions)
  • Detailed explanations for every answer
  • Domain-by-domain performance tracking

30-day money-back guarantee

Timing math and a pacing plan

The arithmetic is simple: 120 minutes across 63 questions is about 1.9 minutes, or roughly 114 seconds, per question. Questions are unequal, though. A crisp single-answer item can be settled in a minute, and a paragraph-long Select TWO with five constraints can legitimately need three. A pacing plan lets the fast items subsidize the slow ones, so you never discover at question 50 that you have 12 minutes left.

Here is the plan Preporato recommends, built for a first pass that finishes with a buffer.

CCAR-P pacing checkpoints (120 minutes, 63 questions)

ClockWhere you should beWhat you are doing
0 to 30 minThrough question 18First pass: answer everything, cap single-answer items at 90 seconds, flag anything unresolved after a best guess
30 to 60 minThrough question 36Same discipline; give multiple-response items up to 150 seconds
60 to 90 minThrough question 54Same discipline; do not reopen flagged items yet
90 to 105 minThrough question 63Finish the first pass with every item answered
105 to 115 minFlagged itemsSecond pass: re-read the stem, find the discriminating constraint, commit
115 to 120 minFinal checkConfirm nothing is blank and every Select TWO or THREE has the right number of selections

Three rules make the plan work. First, always leave an answer behind when you flag; a flagged blank is a zero if time runs out. Second, spend your extra seconds on the stem, since most mis-answered scenario items trace to a misread constraint. Third, treat a Select THREE as three independent calls that must jointly cover the stem, make them, flag it if you are unsure, and move on; its point value is identical to the easiest question on the exam. If you have not rehearsed under a real clock, do it before you book: the 6-week CCAR-P study plan schedules timed full-length practice tests in the final weeks for exactly this reason.

Reading your score report by domain

The section percentages on your report line up with the seven exam domains, whose weights set roughly how many of the 63 items each contributes: Solution Design & Architecture 17 percent (about 11 items), Claude Models, Prompting & Context Engineering 13 percent (about 8), Integration 19 percent (about 12), Evaluation, Testing & Optimization 16 percent (about 10), Governance, Safety & Risk Management 14 percent (about 9), Stakeholder Communication & Lifecycle Management 14 percent (about 9), and Developer Productivity & Operational Enablement 7 percent (about 4).

Largest report section

Build the Integration domain that carries the most items

About 12 of your 63 items sit in Integration; designing a scoped tool set and then building a full MCP server that Claude connects to covers the mechanism-selection, tool-scoping, and capability-bloat questions in that section.

Read the report with those counts in mind. A 50 percent in the 4-item Developer Productivity section means two questions and is weak evidence of anything, whereas 50 percent in the 12-item Integration section is a real gap to build a retake plan around. If you passed, the report still shows which domains to reinforce before the renewal assessment a year later. If you did not, prioritize the heaviest domains where you scored lowest, and use the domain-tagged results from Preporato's practice tests to confirm the gap has closed before you rebook.

Retakes: what Anthropic publishes in 2026

If you do not pass, you can retake the exam, paying the full $175 fee each time. As of this writing, Anthropic's certification policies and Pearson VUE's Anthropic program page state the same waiting periods: 14 days after your first failed attempt, 30 days after your second, and 90 days after your third, with a maximum of four attempts per exam in any rolling 12-month period. The policy page adds that waiting periods reset when a new exam version is released. Because these rules can change, confirm the current version on the Partner Academy policies page and the Pearson VUE Anthropic page before you book a retake, and remember that rescheduling is free only up to 24 hours before the appointment.

Use the waiting period on purpose. Your score report has already told you which sections underperformed; the common CCAR-P exam mistakes article covers the format-driven errors (missed Select TWO instructions, ignored cost and latency constraints, over-engineered pattern choices) behind many narrow misses, and a second timed practice run under the pacing plan above will show whether the fix has taken.

Retake waiting period

Rebuild the weak domain before you rebook

Over-engineered pattern choices and unmeasured prompt changes sit behind many narrow misses; building the canonical workflow patterns and proving a prompt against an eval set fixes both inside a 14-day waiting period.

Renewal after you pass

The credential is valid for 12 months from the date you earn it. Before it expires, you can renew by passing a free, non-proctored, open-book online assessment on the Anthropic Partner Academy, which extends the certification for another 12 months and costs nothing. If you let the credential lapse, that path closes and you retake the full proctored exam at full price. Put the renewal window in your calendar on the day you pass; the renewal assessment is a far lighter lift than a fresh proctored sitting.

Frequently asked questions

Key Takeaways

0/11 completed

Next steps

Rehearse the format before you meet it for real. Take one of Preporato's 63-question CCAR-P practice tests under a strict 120-minute clock using the pacing checkpoints above, then read the explanation for every item, including the ones you got right. To sample the style first, the free CCAR-P question sampler is open to everyone; the full six-test set and the 500-card flashcard deck are included in Preporato Pro (see /pricing). For the study side, the CCAR-P complete guide covers all seven domains in depth.

Sources:

Ready to Pass the CCAR-P Exam?

Join thousands who passed with Preporato practice tests

Instant access30-day guaranteeUpdated monthly
CCAR-P
6 Practice Exams
Detailed Explanations
Performance Analytics
Get Full Access - $19.99Try Free Questions →