CertSafari

    Free Practice Questions for Anthropic Claude Certified Architect – Professional (CCAR-P) - December 7 onwards Certification

    Guide checked for updates:
    20 Nov 2026
    Question bank created:
    18 Nov 2026
    Question bank last updated:
    20 Nov 2026

    Study with 446 exam-style practice questions designed to help you prepare for the Anthropic Claude Certified Architect – Professional (CCAR-P) - December 7 onwards. All questions are aligned with the latest exam guide and include detailed explanations to help you master the material.

    Your progress

    Coverage
    Mastery
    Performance
    Lessons

    LessonsNew

    39 documentation-grounded lessons, one per exam-guide subdomain — every claim cited to the official docs.Based on the official docs as of 18 Nov 2026

    Start Practicing

    Start a quiz

    Practice with randomly mixed questions from all topics

    Question MixAll Topics
    FormatRandom Order

    Exam experiences

    Pass and fail outcomes from candidates who prepared here — advice, scores, and prep time.

    Your saved questions

    Open the list of questions you bookmarked during practice for this exam.

    Study notes

    Private notes per question, grouped by exam domain — opens on its own page, not inline on this overview.

    Quiz History

    Exam Details

    Key information about Anthropic Claude Certified Architect – Professional (CCAR-P) - December 7 onwards

    Official study guide

    View

    Question formats CertSafari offers
    • Multiple choice
    renewal:

    Free, non-proctored renewal assessment on the Anthropic Partner Academy

    exam code:

    CCARP

    exam format:

    Scenario-based multiple choice; each item states how many responses to select

    registration:

    Anthropic Partner Academy, scheduled through Pearson

    guide version:

    1.1 (effective December 7, 2026)

    passing score:

    720 on a scaled range of 100–1,000

    prerequisites:

    None; Claude Certified Architect – Foundations recommended

    retake policy:

    Waits of 14, 30, then 90 days; max 4 attempts per rolling 12 months

    delivery method:

    Proctored via Pearson: OnVUE online or test center

    target audience:

    Solution architects, AI/ML engineers, Technical leads, Senior software engineers

    registration fee:

    $175 USD

    result reporting:

    Pass/fail with scaled score and percent-correct by domain

    time limit minutes:

    120

    number of questions:

    70 (60 scored, 10 unscored)

    certification validity:

    12 months

    recommended experience:

    3+ years in systems architecture or platform engineering and 6+ months with Claude in production

    Exam Topics & Skills Assessed

    Skills measured (from the official study guide)

    1: Solution Design and Architecture

    1.1: Task Delegation Between Claude and Humans

    Determining which components of a described business process should be delegated to Claude, retained by humans, or handled collaboratively, and justifying each assignment against reversibility, stakes, and accountability.

    1.2: End-to-End Architecture Evaluation

    Evaluating whether a proposed end-to-end architecture (input, processing, output, feedback) satisfies stated scalability, maintainability, and access-control requirements, and identifying the component that fails first under a specified load or trust condition.

    1.3: Architectural Pattern Selection

    Selecting among workflow, agentic, augmented-LLM, and Skills-based architectural patterns for a use case with stated reuse, governance, and risk-profile constraints, and justifying the delegation boundary the selected pattern establishes.

    1.4: Orchestrator and Subagent Decomposition

    Evaluating a proposed orchestrator/subagent decomposition for a multi-step workload against stated observability and error-recovery requirements, and identifying where task boundaries, control flow, or result synthesis will fail.

    1.5: Value-Based Architectural Decisions

    Selecting the architectural option that maximizes measurable business outcome under a stated cost or SLA ceiling, weighing competing feature proposals against stated business value pillars such as efficiency, transformation, productivity, solution cost, and performance SLAs.

    1.6: Deployment Surface and Agent Runtime Selection

    Selecting a deployment surface (Anthropic API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry) and an agent runtime (Messages API loop, self-hosted Agent SDK or headless Claude Code, Claude Managed Agents) under data residency, compliance, model-availability, and operational-ownership constraints, and justifying the trade-offs the selection imposes on latency, feature availability, hosting, session state, sandboxing, and data custody.

    2: Claude Models, Prompting, and Context Engineering

    2.1: Model Tier Selection

    Selecting the Claude model tier that satisfies all binding latency, cost, and quality constraints for a stated workload profile, and identifying why each rejected alternative fails.

    2.2: System Prompt Diagnosis and Revision

    Diagnosing the underspecification in a system prompt that produces degraded or inconsistent output, and selecting the revision that constrains scope, format, and behavior without over-constraining legitimate variation.

    2.3: Prompting Technique Selection

    Applying the prompting technique (zero-shot, few-shot, prompted reasoning versus native thinking, structured output) that addresses an observed failure pattern for a task of stated complexity without unnecessary token cost.

    2.4: Context Composition and Token Economics

    Analyzing context composition and caching behavior against a token budget and a defined call volume to determine where token consumption can be reduced without loss of task-relevant information.

    2.5: Cache Breakpoints and Reusable Packaging

    Applying cache breakpoints and modular or Skill-based packaging to a repeated prompt sequence so that stable content precedes dynamic content and reuse is maximized across the stated call pattern.

    2.6: Reasoning Effort and Context Lifecycle

    Allocating reasoning effort and selecting a context lifecycle strategy (context editing, memory tool, compaction) that sustains task performance across the stated session length for a long-running agent workload with stated latency and accuracy requirements.

    3: Integration

    3.1: Least-Privilege Tool Configuration

    Evaluating an agent or subagent tool configuration against least-privilege and minimal-footprint principles, and determining which tool connections to remove, scope down, or replace with a context-efficient access pattern.

    3.2: Authentication and Authorization Posture

    Analyzing the authentication and authorization posture of an integration design to identify credential exposure and privilege escalation risks, and specifying the mitigation for each.

    3.3: Relieving the Binding Performance Constraint

    Evaluating candidate configuration changes (retrieval depth, model tier, caching, call pattern) for a system whose dominant constraint is accuracy or latency, and selecting the change that relieves the binding constraint at least cost to the other.

    3.4: Instrumentation and Monitoring Gaps

    Analyzing the instrumentation gap in a scaled agentic architecture where a described failure went undetected, and selecting the monitoring strategy that would have surfaced it.

    3.5: Retrieval Architecture Selection

    Selecting among retrieval architectures (pre-built RAG pipeline, agentic search with tools, full-corpus context) based on corpus characteristics, query pattern, and latency and cost limits, and, when indexing is chosen, specifying the contextual-retrieval and fallback configuration that maximizes retrieval quality within those limits.

    3.6: Connection Mechanism Selection

    Evaluating MCP, direct API/CLI, and agent-to-agent handoff against scenario constraints on latency, security, and statefulness, and selecting the connection mechanism the constraints favor.

    3.7: Progressive Discovery vs. Monolithic Context Loading

    Evaluating progressive discovery against loading everything into context up front for a task with stated context demands and latency tolerance, and selecting the strategy whose context economics fit the workload.

    3.8: Production Reliability Controls

    Analyzing the failure pattern in error traces or throughput data from a production integration, and specifying the reliability control (rate-limit handling, retry and backoff, idempotency for side-effecting tools, or fallback routing) that addresses it without introducing duplicate effects.

    4: Evaluation, Testing, and Optimization

    4.1: Evaluation Coverage Gaps

    Analyzing an existing evaluation suite after a failure reached production to determine which metric dimension (accuracy, groundedness, latency, cost, safety, security) or grading target was missing, and specifying the metric that would have detected the failure pre-release.

    4.2: Deriving Evaluations from Business Processes

    Determining which portion of a business process constitutes a valid evaluation for a decision it must support (model change, prompt change, release gate), deriving the success criterion from the business outcome, and selecting the test-methodology mix that covers the deployed distribution.

    4.3: LLM-as-Judge Design and Bias

    Evaluating the design of an LLM-as-judge or rubric-based evaluation whose results are inconsistent with human review, and selecting the correction that addresses the responsible bias (position, verbosity, self-preference, judge drift).

    4.4: Experimental Design for Configuration Comparison

    Applying an experimental design that, within a stated sample and cost budget, yields an interpretable comparison of two candidate prompt or model configurations, including the number of trials per case needed to separate the configuration effect from run-to-run variance.

    4.5: Root-Cause Isolation from Logs and Traces

    Isolating the root cause of a failing Claude system from its logs or traces among prompt failure, retrieval failure, hallucination, tool or subagent error, and model mismatch, and specifying the corrective action.

    4.6: Cost and Latency Optimization

    Analyzing cost and latency telemetry, including cache hit rates, to determine where spend or latency is concentrated, and specifying the optimization that reduces it without violating the stated quality floor.

    4.7: Model Version Migration

    Evaluating migration readiness from a model deprecation timeline and regression results for a candidate version, and selecting the response (pin, re-tune prompts, re-baseline evaluations, or defer) that manages behavioral drift within the timeline.

    5: Governance, Safety, and Risk Management

    5.1: Guardrail Placement

    Analyzing guardrail placement in a system design against a described misuse path, and selecting the control that closes the path without degrading a stated legitimate use case.

    5.2: Prompt Injection Surface and Controls

    Analyzing the injection surface of an architecture whose model context includes retrieved documents, tool results, MCP server responses, or third-party Skills, and specifying the controls that prevent untrusted content from being treated as instruction.

    5.3: Risk Assessment and Ranking

    Analyzing a proposed deployment for risk categories (attack surface, failure modes, token and quota exhaustion, dependency and supply-chain risk) and ranking them by exposure.

    5.4: Human-in-the-Loop Checkpoints

    Analyzing where human-in-the-loop checkpoints must sit in an agentic workflow containing irreversible or high-stakes actions, and specifying gates that pause execution without stalling acceptable-risk paths.

    5.5: Architectural Controls for Compliance Obligations

    Evaluating candidate architectural controls (data residency and deployment surface, retention and zero-data-retention configuration, PII detection and redaction, audit logging, tenant isolation) against a stated compliance obligation, and selecting the control set that evidences the obligation.

    5.6: Bias Source Identification and Mitigation

    Evaluating the evidence for bias when outputs or a dataset show disparate results across groups, and selecting the mitigation appropriate to its source, distinguishing data bias from prompt-framing and evaluation-design bias.

    6: Stakeholder Communication and Lifecycle Management

    6.1: Unsupportable Commitments in Requirements and SLAs

    Identifying a commitment in discovery notes or a draft SLA that the proposed Claude architecture cannot meet, and specifying the requirement clarification or architectural change that resolves it.

    6.2: Architecture Decision Records

    Evaluating a draft architecture decision record against a stakeholder constraint it fails to address, and selecting the revision that makes the trade-off and its residual risk explicit without overstating certainty about model behavior.

    6.3: Operational Handoff Completeness

    Identifying the missing assumption, dependency, or runbook element in a handoff package that would block the team receiving a Claude system entering operation, and specifying the documentation that closes the gap.

    7: Developer Productivity and Operational Enablement

    7.1: Packaging and Distributing Claude Code Workflows

    Evaluating how to package and distribute a Claude Code workflow proven on one team across an organization, including plugin contents (skills, hooks, subagents, commands, MCP configuration), marketplace, and managed settings, and deciding which controls are enforced centrally versus left to teams so the workflow stays safe when it runs unattended.

    7.2: Agent SDK and Claude Code Configuration

    Configuring the Claude Agent SDK or Claude Code (permission modes such as default, acceptEdits, plan, auto, dontAsk, and bypassPermissions; hooks; subagent definitions; headless or CI operation) to support a team workflow within stated safety and automation guardrails.

    7.3: Accountability for AI-Generated Code

    Analyzing the review and accountability process of a team adopting AI-assisted development, and identifying the control that preserves human ownership of AI-generated code.

    Techniques & products

    Claude Developer Platform
    Anthropic API
    Messages API
    Claude model tiers
    Claude Code
    Claude Agent SDK
    Claude Managed Agents
    Agent Skills
    Model Context Protocol (MCP)
    Amazon Bedrock
    Google Vertex AI
    Microsoft Foundry
    Prompt caching
    Cache breakpoints
    Extended thinking
    Reasoning effort
    Context editing
    Memory tool
    Compaction
    Structured output
    Few-shot prompting
    Retrieval-augmented generation (RAG)
    Contextual retrieval
    Agentic search
    Orchestrator/subagent architectures
    Agent-to-agent handoff
    Progressive discovery
    Rate-limit handling
    Retry with exponential backoff
    Idempotency keys
    Fallback routing
    LLM-as-judge evaluation
    Rubric-based grading
    Evaluation suites
    Observability and tracing
    Model version migration
    Guardrails
    Prompt injection defenses
    Human-in-the-loop approval gates
    Zero data retention
    PII redaction
    Audit logging
    Tenant isolation
    Data residency
    Least-privilege tool access
    Secrets management
    Architecture decision records
    Operational runbooks
    Claude Code plugins and marketplace
    Hooks
    Managed settings
    Permission modes
    CI/CD pipelines

    CertSafari is not affiliated with, endorsed by, or officially connected to Anthropic PBC. Full disclaimer