Claude Certification Blog

Solution design on CCAR-P: three architectural patterns, not two

The architect professional exam names three architectural patterns where every other Claude track names two, and the extra one is usually the right answer.

17% of the paperThree patternsSix objectives

8 min read

Solution design on CCAR-P is a 17% domain — roughly 11 of 63 items — across six objectives covering translation, end-to-end architecture, pattern selection, multi-agent orchestration, decomposition and business value. The one worth arriving prepared for is pattern selection, because it names three options and the foundations tracks name two.

17%of the CCAR-P paper
11of 63 items
3patterns named, not two
1track that names the third

Six objectives, eleven items

ObjectiveSubjectWhat it turns on
O01Translate business problemsThe stated problem is rarely the real one
O02Design end-to-end architecturesInput, processing, output and feedback
O03Select architectural patternsWorkflow, agentic, or augmented LLM
O04Multi-agent systems and orchestrationWhen more than one is warranted at all
O05Decomposition techniquesDividing ownership, not just steps
O06Align to business value pillarsWhich value this actually moves

One to two items per objective, which is the same arithmetic as the integration domain and points to the same strategy: cover all six, and treat the two least familiar as the best-value hours rather than the ones to skip.

The pattern no other track names

Three patterns, cheapest first, and the one only this exam namesCHEAPEST FIRSTAugmented LLMOne call, given what it needs to answerWorkflowKnown steps in a known order, orchestratedAgenticThe next step depends on the last result
The top row is missing from every other blueprint, which is how a two-way habit meets a three-way question.

An augmented call is one model call supplied with what it needs to answer — retrieval, a tool, the relevant context — and nothing more. No orchestration, no loop, no state between turns. Search the other three blueprints and the pattern is not named; they pose the architecture question as workflow versus agent, which is covered in when not to build an agent.

The consequence for a CCAR-F holder is specific rather than general. You arrive having practised eliminating an agent in favour of a workflow, and the items here often eliminate the workflow too. If a single well-supplied call answers the question, the orchestration was overhead — and an option offering a tidy three-step workflow for something that needs one step is the distractor this domain is built on.

Work up the list, not down it

Start at the cheapest pattern and move up only when the scenario forces you to. A workflow is warranted when there are genuinely several steps with an order; an agent when the next step cannot be known until the last one returns. Starting from the most capable pattern and trimming is how a design ends up with machinery nothing in the scenario required.

Input, processing, output, feedback

The architecture objective names four parts, and the fourth is where answers usually stop short. An input path, a processing design and an output format describe a system that runs once and learns nothing. The feedback loop is what makes it a system rather than a script.

What counts as feedback here is deliberately broad: a correction that reaches the design, a measured outcome that changes a threshold, a route by which a wrong answer becomes a case in an evaluation set. Where those measurements live is the subject of evaluation and testing on CCAR-P.

Business value as an objective

Aligning to business value pillars has its own objective, which reads soft and is not. These items generally describe a design that improves something real, then ask whether that something is what the organisation in the scenario needed moved.

So the recurring wrong answer is a genuine improvement to the wrong quantity: latency reduced for a process nobody waits on, cost saved on a workload that was never expensive, quality raised in an output already good enough. The engineering is sound and the alignment is absent, which is the same shape as the automation trap in workflow integration on CCAO-F, one level up.

Translating a problem

The first objective is translating business problems into Claude-based solutions, and it sits first for a reason: everything after it is downstream of getting the problem right. The examinable version is that the stated problem and the actual one differ, and the strong answers surface the difference before designing.

In practice that means finding what would count as solved before choosing what to build. A scenario describing a request for a specific feature is usually testing whether you ask what the feature is for — which is the discovery objective from the stakeholder domain arriving at the design side of the same conversation.

How to spend the block

A fifth of your study time, matching the weight, split fairly evenly across six objectives. If you are coming from CCAR-F, spend the first hour on pattern selection alone — not because it is hard, but because you have a two-way habit that needs to become three-way, and habits are what fail under time pressure rather than knowledge.

Then spend the second hour on business value, which is the objective a technical candidate is most likely to treat as filler. Between them those two are perhaps four items, and both are decided by a reflex rather than by recall. The rest of the re-weighting between the two architect tracks is in going from CCAR-F to CCAR-P.

Key takeaways

  • Seventeen percent, eleven items, six objectives. One to two items each, so breadth beats depth again.
  • Three patterns, not two. Augmented LLM, workflow, agentic — and only this exam names the first.
  • Work up from the cheapest. A tidy workflow for a one-step problem is the distractor this domain is built on.
  • Feedback completes the architecture. Input, processing and output describe something that runs once and learns nothing.
  • Alignment is examined. A real improvement to a quantity nobody needed moved is still the wrong answer.
  • Get the problem before the design. The stated problem and the actual one differ, and the good answers say so first.

A two-way habit meets a three-way question

This is the domain where prior architect experience actively works against you for about four items, and reading about it is not the same as catching yourself doing it. Timed papers on the published 63-item allocation are where the habit surfaces. Our claude certification study guide covers how to sequence them.

See the CCAR-P blueprint

Questions

Frequently asked

The follow-up questions people search next.

What architectural patterns does CCAR-P name?

Three: workflow, agentic, and augmented LLM. The third — a single model call given the tools, retrieval or context it needs — appears in no other Claude exam blueprint. The foundations tracks pose the choice as workflow versus agent, so a candidate arriving from one has practised a two-way decision that is really three-way.

How big is solution design on CCAR-P?

Solution Design and Architecture is 17% of the exam — about 11 of 63 items — across six objectives. That is roughly one to two items each, which makes it another domain where covering everything beats mastering part of it.

What does an end-to-end architecture have to include?

The objective names four parts: input, processing, output and feedback loops. The last is the one most often missing from otherwise complete answers, and an architecture with no path for what happens after the output is not end to end.

Is business value really examined?

It has its own objective — aligning solutions to business value pillars. In practice these items ask which value a design actually moves, and reject options that improve something real that nobody in the scenario was asking for.

How does this differ from the CCAR-F agent material?

CCAR-F asks whether to build an agent and how to orchestrate one; that is 27% of its paper. CCAR-P asks the same question with a cheaper third option on the table, and then asks which business value the answer serves — architecture judgment rather than construction judgment.

Keep reading

Related posts

Not affiliated with, or endorsed by, Anthropic or Pearson VUE. Details are summarised from publicly published program information and can change — always confirm against the official exam guide before booking.

We use cookies and privacy-friendly analytics to understand usage and improve Cred Farmer. Essential features work either way. See our Cookie Policy.