Flashcards · 38 cards · free
Every Claude Certified Architect — Professional objective, on one card each
One card for each of the 38 objectives the exam grades, in blueprint order. Each says what the objective covers and, more usefully, the distinction its questions turn on. Most items on this exam are a choice between two defensible answers, not a fact you either know or do not. Every card links to practice questions on that objective.
answers are shown, not hidden. This is a reference you read, not a self-test
D1Solution Design & Architecture
17% of the exam · 6 objectives
Objective 1.1 · Translate business problems into Claude-based AI solutions
What does objective 1.1, business problem to solution, cover on the CCAR-P exam?
You are given a stated business problem and asked to turn it into a system built on Claude. That means working out which part of a workflow is genuinely language-shaped, what "good" looks like in the business's own terms, and where a model earns its place over a rule or a search index. Items here punish over-reach. The credited answer is often the narrower scope that solves the stated problem, not the ambitious platform solving a problem nobody asked about.
6 practice questions →Objective 1.2 · Design end-to-end architectures (input → processing → output → feedback loops)
What does objective 1.2, end-to-end architecture, cover on the CCAR-P exam?
End-to-end architecture: input handling, processing, output, and the feedback loop that lets the system improve. The feedback loop is the part candidates skip and the part the exam asks about. Expect stems where a design is complete and functional but has no path from a bad output back into an evaluation set, and where the right answer is the one that closes that loop.
6 practice questions →Objective 1.3 · Select appropriate architectural patterns (workflow, agentic, augmented LLM)
What does objective 1.3, architectural pattern choice, cover on the CCAR-P exam?
Choosing between a fixed workflow, an augmented LLM and an agentic loop. The governing principle is that agency is a cost rather than a feature: an agent that plans its own steps is harder to test, harder to bound and slower. Pick the simplest pattern whose failure modes you can live with. When a stem emphasises predictability, auditability or a fixed sequence, it is pointing at a workflow even if it uses the word "agent".
6 practice questions →Objective 1.4 · Design multi-agent systems and orchestration strategies
What does objective 1.4, multi-agent orchestration, cover on the CCAR-P exam?
More agents does not mean more capability, and this objective exists to test whether you know it. Every hand-off between agents is a place for context to be lost and cost to be duplicated, so the exam rewards the design with the fewest agents that still separates genuinely different concerns. A supervisor pattern applied to a problem a single loop would handle is a recurring wrong answer.
5 practice questions →Objective 1.5 · Apply decomposition techniques for complex problem solving
What does objective 1.5, decomposition, cover on the CCAR-P exam?
Decomposition: breaking a task too large or too vague for one call into steps that can each be prompted, tested and evaluated. Watch for the distinction between decomposing for accuracy, where each step becomes easier to get right, and decomposing for observability, where each step can be inspected when the whole thing goes wrong. Stems usually signal which one they care about.
5 practice questions →Objective 1.6 · Align solutions to business value pillars (efficiency, transformation, productivity, cost, performance SLAs)
What does objective 1.6, business value alignment, cover on the CCAR-P exam?
Every design is supposed to deliver something the business named: efficiency, transformation, productivity, cost, or a performance SLA. This objective is about tying your design to that, and choosing between designs when the pillars conflict. Cost and latency almost always trade against accuracy, and the credited answer matches the pillar the stakeholder in the stem named, and not the one that is best in the abstract.
6 practice questions →
D2Models, Prompting & Context Engineering
13% of the exam · 5 objectives
Objective 2.1 · Select appropriate Claude models based on trade-offs
What does objective 2.1, model selection, cover on the CCAR-P exam?
Model selection across the Claude family, judged on the trade-off the scenario states and not on capability alone. The reasoning has the same shape every time. The largest model is right only when the task genuinely needs it, and a cheaper, faster model is right far more often than instinct suggests, particularly for classification, extraction and routing steps inside a larger pipeline. If a stem gives you a per-request latency budget or a volume figure, that is the constraint deciding the answer.
6 practice questions →Objective 2.2 · Design system prompts, templates, and guardrails
What does objective 2.2, system prompts & templates, cover on the CCAR-P exam?
System prompts, reusable templates and prompt-level guardrails, which is to say establishing role, scope, output format and refusal behaviour in the place that persists across a conversation. The distinction the exam draws is between instructions belonging in the system prompt, which are stable, behavioural and security-relevant, and instructions belonging in the user turn, which are task-specific and variable. Putting per-request data in a system prompt is a recurring wrong answer.
5 practice questions →Objective 2.3 · Apply prompt engineering techniques (zero-shot, few-shot, chain-of-thought)
What does objective 2.3, prompt engineering technique, cover on the CCAR-P exam?
Zero-shot, few-shot and chain-of-thought, and knowing which the situation calls for. Few-shot earns its tokens when the task has an idiosyncratic format or an edge case that a description cannot convey. Chain-of-thought earns its latency when the task has genuine intermediate reasoning. Applying either reflexively is the error being tested, and you should expect stems where the right answer is a clearer instruction instead of more examples.
5 practice questions →Objective 2.4 · Optimize context windows and manage token usage
What does objective 2.4, context window & tokens, cover on the CCAR-P exam?
Context-window management: what to put in, what to leave out, and what to do when the material exceeds the budget. Summarisation, truncation strategy and ordering effects all appear here, as does the fact that a full context window is not a free one. Cost and latency both scale with it, and recall over very long contexts is not uniform. A stem that says "we are passing the entire document set" is usually pointing at retrieval instead.
5 practice questions →Objective 2.5 · Implement prompt reuse strategies (caching, modular prompts, Skills)
What does objective 2.5, prompt reuse & caching, cover on the CCAR-P exam?
Prompt reuse means caching stable prefixes so they are not re-billed and re-processed on every call, composing prompts modularly, and packaging repeated procedures so they can be invoked rather than restated. One mechanical detail decides most of these questions: ordering. A cache only helps if the stable part comes first and the variable part last, so a design that interleaves them gets no benefit however reusable it looks.
5 practice questions →
D3Integration
19% of the exam · 8 objectives
Objective 3.1 · Evaluate tool/agent configuration for capability bloat
What does objective 3.1, capability bloat, cover on the CCAR-P exam?
Capability bloat is what happens when an agent is given more tools than the task needs. Every additional tool consumes context, adds a way to select wrongly, and widens the blast radius of a mistake. Expect scenarios where an agent is misbehaving and the credited fix is removing or narrowing its tools instead of improving the prompt that chooses between them.
5 practice questions →Objective 3.2 · Analyze authentication and authorization requirements to identify security gaps
What does objective 3.2, authn & authz gaps, cover on the CCAR-P exam?
Authentication and authorisation across an agentic system: whose identity a tool call runs as, how credentials reach the tool, and where privilege can escalate. The recurring gap is an agent holding a single service credential broad enough to act for any user, which turns a prompt injection into a data breach. Look for answers that carry the end user's authority through to the tool and scope it down, and be suspicious of any that trust the model to restrict itself.
5 practice questions →Objective 3.3 · Evaluate accuracy-latency trade-offs and justify configuration decisions
What does objective 3.3, accuracy-latency trade-offs, cover on the CCAR-P exam?
Accuracy against latency, and being able to justify where you landed. Model choice, reasoning depth, retrieval breadth, and whether a verification pass is worth its round trip all sit here. What the exam wants is a decision tied to a stated budget: a stem that hands you a p95 target is telling you the most accurate option is the wrong one.
5 practice questions →Objective 3.4 · Analyze observability challenges and select monitoring strategies at scale
What does objective 3.4, observability at scale, cover on the CCAR-P exam?
Observability at scale asks what to log, what to trace, and how to make a non-deterministic system debuggable after the fact. Traces have to span the whole chain, from retrieval through tool calls to model turns, because the failure is usually in the step nobody instrumented. Sampling strategy comes up, so does the cost of logging full prompts and completions, and so does redaction.
5 practice questions →Objective 3.5 · Design a RAG pipeline with appropriate chunking and indexing strategies
What does objective 3.5, rag chunking & indexing, cover on the CCAR-P exam?
Designing a retrieval-augmented generation pipeline: chunking, embedding, indexing, and the choices inside each. Chunk size trades recall against precision. Overlap protects meaning that straddles a boundary. Structure-aware splitting beats fixed-length splitting on documents that have structure. The most common wrong answer is a larger chunk size offered as a fix for poor retrieval quality, when the real problem is chunking that ignores the document's structure.
5 practice questions →Objective 3.6 · Apply retrieval strategies matched to data shape and query pattern
What does objective 3.6, retrieval strategy, cover on the CCAR-P exam?
Matching the retrieval strategy to the shape of the data and the shape of the query. Dense vector search handles paraphrase and concept. Keyword and lexical search handle identifiers, part numbers and exact terms. Hybrid retrieval with reranking handles the common case where both matter. If a stem mentions codes, SKUs, error strings or names, it is signalling that pure semantic search will miss.
5 practice questions →Objective 3.7 · Evaluate connection protocols and select the appropriate integration mechanism (MCP, API/CLI, agent-to-agent)
What does objective 3.7, integration protocol choice, cover on the CCAR-P exam?
Choosing the connection mechanism, whether that is the Model Context Protocol, a direct API or CLI integration, or agent-to-agent communication. MCP is the answer when you want a standard, discoverable interface reusable across clients and tools. A direct API call is the answer when the integration is one-off, latency-critical or already built. Reach for agent-to-agent only when the other side genuinely needs autonomy, and not when it is a function you could call.
4 practice questions →Objective 3.8 · Evaluate progressive discovery vs. monolithic context strategy
What does objective 3.8, progressive discovery, cover on the CCAR-P exam?
Progressive discovery against loading everything up front. Handing an agent a small entry point and letting it fetch what it needs keeps context small and scales to corpora that would never fit; loading everything is simpler and lower-latency when the material is genuinely small and bounded. The exam tests the judgement rather than a preference, so a stem describing a fixed, small reference set is not asking for progressive discovery.
4 practice questions →
D4Evaluation, Testing & Optimization
16% of the exam · 6 objectives
Objective 4.1 · Define evaluation metrics (accuracy, latency, cost, safety, security)
What does objective 4.1, evaluation metrics, cover on the CCAR-P exam?
Before you can measure whether something works you have to define what working means, which is what this objective covers: accuracy, latency, cost, safety and security, and the fact that a metric nobody can compute is not a metric. Expect items where several proposed metrics are reasonable and the credited one is the only one both observable in production and tied to the business outcome in the stem.
6 practice questions →Objective 4.2 · Design evaluation datasets and test frameworks using mixed methodologies
What does objective 4.2, eval datasets & frameworks, cover on the CCAR-P exam?
Building evaluation datasets and test frameworks with mixed methods: exact-match and programmatic checks where the answer is checkable, model-graded evaluation where it is not, and human review where the stakes justify it. Dataset design matters as much as grading here. An eval set drawn only from happy-path traffic will report a system as healthy right up until it is not.
5 practice questions →Objective 4.3 · Conduct A/B testing and iterative improvements
What does objective 4.3, a/b testing & iteration, cover on the CCAR-P exam?
A/B testing on a non-deterministic system, which means holding everything but one variable steady, running long enough to separate signal from sampling noise, and resisting the urge to change the prompt and the model in the same experiment. Items often describe an experiment whose result cannot be attributed to anything and ask what was wrong with the design.
5 practice questions →Objective 4.4 · Diagnose system issues (prompt failure, hallucinations, model mismatch)
What does objective 4.4, diagnosing failures, cover on the CCAR-P exam?
Diagnosis: telling a prompt failure from a hallucination from a model mismatch from a retrieval miss. The discriminating question is what the model was actually given. A confident wrong answer whose supporting document was never retrieved is a retrieval bug, and no amount of prompt work will fix it. Expect stems that describe a symptom and reward the answer isolating the layer before changing anything.
5 practice questions →Objective 4.5 · Optimize token usage, latency, and cost-performance trade-offs
What does objective 4.5, cost & latency optimisation, cover on the CCAR-P exam?
Optimising cost and latency once the system works. Caching, model right-sizing per step, trimming context, batching, and streaming to improve perceived latency without changing total time. The framing that scores well is measuring first: the exam dislikes answers that optimise a component nobody has shown to be the bottleneck.
5 practice questions →Objective 4.6 · Monitor system performance using logging and observability tools
What does objective 4.6, logging & monitoring, cover on the CCAR-P exam?
Running the system in production, where the interesting problems appear over weeks instead of in a test. Logging, metrics, alerting and drift. What a useful alert looks like for a probabilistic system, where its thresholds come from, and the difference between monitoring the model and monitoring the product outcome it is supposed to move.
6 practice questions →
D5Governance, Safety & Risk Management
14% of the exam · 5 objectives
Objective 5.1 · Implement guardrails and safety controls
What does objective 5.1, guardrails & safety controls, cover on the CCAR-P exam?
Guardrails and safety controls in layers: input filtering, system-prompt constraints, output validation and tool-level permissions. The point the exam presses hardest is that a prompt instruction is not a security control. Anything that must hold under adversarial input has to be enforced outside the model, and answers that defend against prompt injection by asking the model not to comply are reliably wrong.
6 practice questions →Objective 5.2 · Identify risks, limitations, and failure modes of LLM systems
What does objective 5.2, risks & failure modes, cover on the CCAR-P exam?
Knowing the failure modes and naming the one that fits the architecture in front of you: hallucination, prompt injection, jailbreaking, data leakage through context, non-determinism, and the ways a system degrades quietly instead of loudly. The risk the exam is looking for is usually the one created by a component the design added for some other reason.
6 practice questions →Objective 5.3 · Apply human-in-the-loop validation strategies
What does objective 5.3, human-in-the-loop, cover on the CCAR-P exam?
Human-in-the-loop: where to put a person, how much they can realistically review, and how to route only what needs them. Review-everything does not scale and review-nothing is not a control, so credited answers tend to involve confidence thresholds, sampling, or gating on irreversibility, with a human on the actions you cannot undo instead of the ones you can.
5 practice questions →Objective 5.4 · Ensure compliance with regulations (e.g., GDPR, HIPAA, FedRAMP)
What does objective 5.4, regulatory compliance, cover on the CCAR-P exam?
Regulatory compliance, from GDPR to HIPAA to FedRAMP, treated as a constraint on architecture: data residency, retention, audit trails, and what a processor agreement does and does not cover. One question comes up more than any other, so it is worth pre-loading: deleting a source document does not delete the vectors derived from it unless something was built to make that happen.
6 practice questions →Objective 5.5 · Address ethical AI considerations (bias, fairness, transparency)
What does objective 5.5, ethical ai, cover on the CCAR-P exam?
Bias, fairness and transparency as engineering concerns rather than sentiments. Where bias enters, whether through training data, prompt framing, evaluation sets or the humans reviewing output. How you would detect it. What you disclose to the people affected. Expect stems that ask for the measurement, not the intention.
5 practice questions →
D6Stakeholder Communication & Lifecycle
14% of the exam · 5 objectives
Objective 6.1 · Conduct structured discovery and requirement gathering
What does objective 6.1, discovery & requirements, cover on the CCAR-P exam?
Structured discovery: the questions that determine whether a project is feasible, what data exists, who owns it, what the current process looks like day to day, and how success would be measured. Items reward the question that would change the design if it were answered differently, and penalise gathering requirements that no design choice depends on.
6 practice questions →Objective 6.2 · Communicate architectural decisions and trade-offs
What does objective 6.2, communicating trade-offs, cover on the CCAR-P exam?
The audience named in the stem is the whole question here. Communicating an architectural decision means naming the trade-off, the alternatives considered and the reason this one won, in terms of the listener's stake rather than the system's internals. The same decision is explained differently to a CFO, a security reviewer and the engineer who will maintain it.
6 practice questions →Objective 6.3 · Manage stakeholder feedback loops and expectation alignment (including SLAs)
What does objective 6.3, feedback loops & slas, cover on the CCAR-P exam?
Managing expectations over time: feedback loops with stakeholders, and SLAs on a system whose output quality is probabilistic. What you can honestly commit to is availability, latency and a measured accuracy band on a defined eval set. What you cannot commit to is per-response correctness. This objective also covers renegotiating when measurement contradicts the original promise.
6 practice questions →Objective 6.4 · Document architectures and provide implementation guidance
What does objective 6.4, documentation & handoff, cover on the CCAR-P exam?
Documentation and handoff, which is mostly about what a design document usually omits: the prompts themselves, the eval set, the failure modes seen in testing, and the reasoning behind each constraint so it is not silently optimised away later. Decision records and their rationale come up more often than diagrams do.
5 practice questions →Objective 6.5 · Support lifecycle phases (discovery, design, handoff, monitoring, iteration)
What does objective 6.5, lifecycle phases, cover on the CCAR-P exam?
The whole lifecycle, from discovery through design, build, handoff, monitoring and iteration, and knowing which phase a described situation belongs to. A common item shape puts a team in one phase doing work that belongs to another: tuning prompts before an eval set exists, or committing to an SLA before anything has been measured.
5 practice questions →
D7Developer Productivity & Enablement
7% of the exam · 3 objectives
Objective 7.1 · Configure Claude tools and environments for teams (e.g., Claude Code)
What does objective 7.1, team tooling & environments, cover on the CCAR-P exam?
Setting up Claude tooling for a team instead of an individual: shared configuration, project context that lives in the repository, permissioning, and the environments developers actually work in. The theme is making the good path the default one, which usually means a checked-in file rather than a document asking people to remember something.
5 practice questions →Objective 7.2 · Improve developer workflows using AI-assisted tooling
What does objective 7.2, developer workflows, cover on the CCAR-P exam?
Improving developer workflows with AI-assisted tooling. Where it reliably helps, which is review, tests, migrations and unfamiliar code. Where it needs a human gate. And how to measure whether it helped instead of assuming that it did. Items favour changes that shorten a feedback loop over changes that generate more code.
5 practice questions →Objective 7.3 · Support debugging and operational issue resolution
What does objective 7.3, debugging & operations, cover on the CCAR-P exam?
Debugging and operational support with these systems in the loop: reproducing a non-deterministic failure, using traces and logs to find the step that broke, and knowing when the answer is a fix to the system rather than to the prompt. The smallest reproduction is the goal, and with a probabilistic component "smallest" includes pinning down what was in context at the time.
4 practice questions →
Cards tell you what is on the exam. A timed mock tells you whether you can answer it.
Sit a timed mock →