CCAR-P questions and answers — Domain 1: Solution Design and Architecture

11 practice questions on Solution Design and Architecture, the domain weighted at 17% of the CCAR-P exam. Every answer explains why the distractors fail — that is the part that transfers to questions you have not seen before.

Question 1

A business asks for an agentic system to process insurance claims end to end. Claims are handled in a known five-step sequence, and the regulator requires a documented decision trail. What should you recommend?

1. An autonomous agent, because it adapts to unusual claims
2. A workflow implementing the five steps with gates between them, and human review where the regulator requires it
3. A multi-agent system with one agent per step
4. An agent with a strict system prompt describing the five steps

Answer: B. The steps are known, so a workflow gives predictable cost, testable stages and a decision trail that can be audited. Autonomy here buys flexibility nobody asked for and costs exactly the properties the regulator requires. A system prompt describing the steps guides but does not enforce them.

Question 2

Which question should be answered before choosing an architecture?

1. Which model tier has the best benchmark scores
2. What the cost of a wrong output is, and what hard constraints apply
3. Whether the team prefers Python or TypeScript
4. Which framework the organisation already uses

Answer: B. Stakes set the review model and the guardrails; constraints such as residency, latency and retention determine which designs are permissible at all. Benchmarks, language and framework are downstream implementation choices that cannot be made sensibly first.

Question 3

Two designs both meet the functional requirement. One is a multi-agent system your team built; the other is a single workflow another team could operate. The stem notes the system will be handed over next quarter. Which do you recommend?

1. The multi-agent system, since it is already built
2. The workflow, because operability by the receiving team is part of the requirement
3. Either, since both meet the functional requirement
4. The multi-agent system, with extra documentation

Answer: B. The handover is stated in the stem, so it is a constraint rather than a preference. Between designs that both work, the one another team can operate is the better architecture. Documentation mitigates but does not remove the operational burden of the more complex design.

Question 4

A proposed design saves 30% on processing cost by removing a human review step that a sector regulation requires. How should this be assessed?

1. Accept it, since the saving is material and the model is accurate
2. Reject it — the saving is not available, because the control is a constraint rather than an optimisation target
3. Accept it with additional automated guardrails in place of the review
4. Accept it as a pilot and reinstate review if problems appear

Answer: B. A regulatory control is an input to the design, not a cost line to be traded off. Automated guardrails do not substitute for a required human decision, and a pilot that removes a mandated control is a compliance breach with a friendlier name.

Question 5

Which element is most often missing from an otherwise complete architecture, and is a frequent focus of exam questions?

1. The choice of programming language
2. Defined failure behaviour — what happens on timeout, refusal, rate limit or degraded quality
3. The number of subagents
4. The embedding model

Answer: B. Graceful degradation is a design decision, and systems that never specify it fail unpredictably under exactly the conditions where behaviour matters most. The other items are either implementation detail or components rather than system properties.

Question 6

A stakeholder asks for a design that is "as capable as possible". The stem specifies a 2-second latency budget and a fixed cost per transaction. What is the correct response?

1. Use the most capable model tier available
2. Design to the stated latency and cost constraints, and show what capability is achievable within them
3. Ignore the constraints, since capability was the stated priority
4. Propose relaxing the latency budget

Answer: B. Stated numeric constraints outrank a general aspiration. The professional move is to design within them and show the achievable capability, which converts a vague ask into an informed decision. Proposing to relax the budget may be legitimate later, but only with the trade-off quantified.

Question 7

When is a multi-agent design genuinely justified?

1. Whenever a task has several steps
2. When subtasks are independent and need different tools, expertise or context isolation
3. When one agent is producing poor output
4. When the organisation wants to showcase agentic capability

Answer: B. Independence plus differing tool or context needs is what multi-agent structure actually buys. Sequential steps are a workflow; poor output from one agent is usually a prompt, grounding or tier problem that multiplying agents will not fix.

Question 8

How should business value be expressed when defending a design?

1. As the sophistication of the technical approach
2. As unit economics — cost per transaction against the alternative, including human time added or removed
3. As the number of tasks the system can perform
4. As the model tier used

Answer: B. Unit economics is the framing that survives a budget conversation, and it must include the review time the design adds as well as the work it removes. A design that costs more per unit than the process it replaces has failed regardless of its sophistication.

Question 9

A design routes all requests to a frontier model for consistency. Volume is high and most requests are simple. What is the architectural objection?

1. Frontier models are less accurate on simple tasks
2. It pays frontier prices and latency for work a lighter tier handles, with no stated requirement justifying it
3. Consistency is not achievable across requests
4. Frontier models cannot be used at high volume

Answer: B. Proportionality. Without a requirement that only the frontier tier satisfies, the design spends cost and latency for nothing. The usual professional answer is a tiered arrangement where a cheap model carries the volume and escalates the hard cases.

Question 10

A system must remain useful when the model API is unavailable. Which is the appropriate design response?

1. Retry indefinitely until it succeeds
2. Define degraded behaviour explicitly — queue for later, fall back to a simpler path, or fail with a clear message and no silent data loss
3. Cache all previous responses and serve those
4. Treat availability as the provider's responsibility

Answer: B. Degradation must be designed. Indefinite retry converts an outage into a backlog with unbounded latency; serving stale cached answers to different questions is wrong output presented as right; and your availability commitment cannot exceed your dependencies, which is precisely why the fallback must be specified.

Question 11

Which best describes how the workflow-versus-agent decision changes in a regulated context?

1. It does not change; the technical criteria are the same
2. Agentic systems are harder to bound, audit and cost-forecast, which often makes a workflow the defensible choice on governance grounds alone
3. Agents are preferred because they can be instructed to follow regulations
4. Regulation prohibits agentic systems

Answer: B. Regulation rarely bans a pattern outright, but auditability, boundedness and predictable cost are governance properties that workflows have and agents have to be engineered into. That is frequently decisive before any performance comparison is made.

← Previous: Cheat sheet  |  Next: Questions & answers — Domain 2 →