Claude Certified Architect Professional (CCAR-P) Practice Test

These questions are part of the practice quiz. Start practicing

• Which concept describes the quantity of input and output units consumed during model interactions and used as a major driver of cost and capacity?
• What is Context Window in model processing?
• Which approach loads most available instructions, documentation, and resources into the model's working context upfront?
• Which term describes high-priority instructions that establish the model's role, behavioral rules, constraints, and general operating context?
• What is the purpose of idempotent operations in integration design?
• In Claude architecture, what design reduces cost for intermittent bursts?
• What is the importance of a 'watchdog' or safe-guard in streaming Claude outputs?
• What is a key benefit of circuit breakers in Claude workflows?
• To minimize hallucinations while preserving task intent in Claude prompts, which approach is recommended?
• Which term best describes an isolated worker with its own instructions, tools, and context used to perform a focused task?
• What is the difference between policy-based and generation-based safety controls in Claude?
• Which description best defines Throughput?
• Which term defines the cache validity duration before content should be considered stale?
• Which heading is described as providing several examples of desired input-output behavior to demonstrate the expected pattern?
• Incremental delivery of generated content as it becomes available rather than waiting for the entire response to finish?
• Which technique helps reduce cold starts for Claude inference in a high-throughput environment?
• In Claude governance, what is the role of version control in policy-as-code?
• What is the term for an attempt to circumvent behavioral or safety restrictions through adversarial instructions or interaction patterns?
• Which metric describes the latency threshold below which 95 percent of measured requests complete?
• Which term refers to malicious or untrusted instructions embedded in input or retrieved content that attempt to override intended application behavior?
• What is Embedding in the context of semantic retrieval?
• Which coordinating component decomposes work, delegates subtasks, checks coverage, handles missing results, and synthesizes outputs?
• Which term is a formal contract specifying service expectations such as uptime and support obligations?
• Which term is a short version-controlled document that records a significant technical choice, its context, considered alternatives, and rationale?
• How can you mitigate data exfiltration risk when Claude outputs are stored?
• Which principle suggests granting the minimum privileges required to perform a task?
• How do you handle failed dependencies in Claude-based workflows?
• Which concept is defined as task-specific input supplied by the person or application requesting a response?
• Which term describes an asynchronous interface for processing large collections of independent requests when immediate responses are unnecessary?
• Which principle involves restricting data collection, processing, and storage to information necessary for the intended purpose?
• What term describes the process of verifying that an individual, application, service, or device is genuinely the identity it claims to be?
• What term describes a packaged collection of specialized instructions and supporting resources loaded for a task requiring domain expertise?
• Which metric is most directly used to measure the propensity of Claude to generate unsupported statements?
• How can Claude adhere to bias mitigation goals in outputs?
• Which interface do developers use to send requests and receive model-generated responses from Anthropic-hosted models?
• What is the role of immutable deployment pipelines in Claude prompt/version control?
• What is the programmatic interface used by applications to send requests to Anthropic-hosted models and receive generated responses?
• Which term refers to an AWS-hosted deployment option for building applications and managed agent workloads with cloud-native integration?
• Which metric measures the volume of tokens consumed during model interactions and influences cost and capacity?
• Which term describes a numerical representation that captures semantic characteristics and can be used to compare related information mathematically?
• Which term corresponds to high-priority instructions establishing the model's role and operating context?
• What is the primary purpose of namespace isolation in a multi-tenant Claude deployment?
• What practice supports auditability when updating Claude prompts and policies in production?
• What authorization framework enables delegated access to protected resources without requiring applications to directly handle a user's primary credentials?
• Which prompting style is described as asking for a task without providing examples of completed responses?
• Which practice involves selecting a model according to required accuracy, reasoning capability, latency, throughput, and cost?
• What describes a data-use contract in Claude governance?
• Which term is primarily used to determine whether a proposed technical approach is feasible before committing to full production, often with a focus on validation?
• Which term denotes a specifically versioned release used when applications require more predictable behavior?
• Which release is specifically versioned to provide more predictable behavior than automatic upgrades?
• Which practice involves comparing alternative system versions using comparable traffic or datasets to determine which performs better against defined metrics?
• Which data type enables robust evaluation of Claude without exposing individuals’ identities?
• Which statement best describes a data-use contract in Claude governance?
• Which component is the surrounding runtime and control logic that manages model interactions, tool execution, context, iteration, state, and task completion?
• What practice helps handle multi-turn conversations with context continuity while protecting privacy?
• Which term is a unique identifier propagated through related operations so that events belonging to the same request can be connected during analysis?
• Which metric is used to measure user trust in Claude?
• What data handling principle is central to privacy-by-design in Claude deployments?
• Which item is a project-level instruction file used to provide persistent coding conventions, commands, architectural guidance, and repository-specific context?
• Context boundaries in multi-turn prompts refer to what?
• Which dataset type is intended to probe system behavior with inputs that are uncommon or difficult and outside normal conditions?
• Which term is the ability to understand internal system behavior using collected signals such as traces, logs, metrics, request metadata, and errors?
• Which AWS managed service provides access to foundation models, including Claude models, with security and governance controls?
• What deployment strategy releases a new version to a small portion of production traffic before expanding exposure?
• Which storage system is optimized for indexing and searching numerical representations based on semantic similarity?
• How is hallucination rate in Claude outputs quantified?
• Which mechanism helps detect and prevent policy violations in real-time streaming outputs?
• Which practice best protects against data exposure and improper data handling in AI systems?
• Which term describes encryption of data as it moves between devices or services?
• Which approach exposes information or capabilities only when they become relevant during a task?
• Which term describes the AI application or environment that coordinates connections to external capability providers?
• Which practice supports compliance with data localization requirements?
• What term describes the process of validating outputs by humans when the quality depends on nuance and accountability?
• What is the role of an LLM policy in Claude governance?
• What is Context Engineering?
• Which combination of controls reduces cross-tenant leakage in a multi-tenant Claude deployment?
• Which approach helps ensure Claude respects user privacy when training with human feedback?
• Which term is a formal commitment defining measurable service expectations such as availability, response time, or support obligations?
• Which concept describes streaming content incrementally rather than waiting for the entire response?
• Which term denotes information returned to the model after an external operation finishes so reasoning can continue?
• Which term refers to a capability that allows the model to request external functions, APIs, searches, calculations, or other operations needed to complete a task?
• Which term denotes the environment that coordinates connections to external capability providers?
• What role does prompt engineering play in reducing hallucinations?
• How should explainability for Claude outputs be implemented to satisfy regulatory requirements?
• What is prompt injection and how to mitigate it in Claude prompts?
• What is the significance of 'policy-as-code' for Claude governance?
• Which component coordinates, decomposes, delegates subtasks, checks coverage, handles missing results, and synthesizes outputs?
• Which term describes a hosted autonomous execution environment that supplies an agent harness, runtime, tools, session handling, and infrastructure without requiring an application to implement the entire loop itself?
• How would you implement an explainability feature that a non-technical user can understand in Claude?
• Which architectural pattern best supports ultra-low latency Claude inference at high throughput?
• How would you design Claude to comply with age-restriction policies?
• Which practice most directly contributes to reducing hallucinations in generated outputs?
• Which term describes a prompt design approach that uses examples to guide model behavior?
• Which term describes a data-handling arrangement under which eligible API interactions are processed without retaining customer prompts and responses beyond what is necessary for request handling?
• What is the principle of data minimization in Claude workflows?
• What term describes a correlated record showing how one request travels across multiple components, services, model calls, and tools?
• Which threat model best captures Claude's security concerns?
• Which term defines a specific reliability or performance target used to measure service operation?
• Which component maintains a connection between the host environment and an external capability provider?
• Which term refers to an ongoing lifecycle practice of measuring production performance and iteratively updating prompts, models, architecture, evaluations, or controls based on evidence?
• Which technique replaces directly identifying information with substitutes while keeping separately protected information that can restore the association when required?
• Which retrieval approach improves coverage and precision by combining semantic similarity with keyword-based matching?
• What technique substitutes directly identifying information with substitutes while keeping separately protected information that can restore the association when required?
• Which term describes an AI-powered component that can reason toward a goal, decide on actions, and continue working until completion?
• Which term corresponds to a process where an AI system dynamically decides which actions or tools to use rather than following a fixed sequence?
• How would you implement compliance auditing for Claude usage?
• What best describes A/B testing for model outputs?
• What deployment strategy minimizes risk when releasing a new Claude version?
• Which approach uses a language model to score or compare outputs according to a clearly defined rubric?
• How would you design a red-teaming exercise for a Claude deployment?
• Which term is a design that routes selected decisions, outputs, or actions to a person for review, approval, correction, or escalation?
• Which logging approach best balances observability with privacy for Claude?
• Which term refers to an early architecture activity used to identify the business problem, users, workflows, data, constraints, risks, and measurable outcomes before selecting a technical design?
• Which term is the machine-readable specification that defines the fields, types, and constraints accepted by a callable function?
• What term describes a carefully reviewed set of examples with trusted expected results used as a high-quality benchmark?
• Which term refers to the reuse of stable prompt prefixes across requests to reduce repeated processing cost and latency?
• How do you ensure version control for prompts and configurations in Claude?
• Which term describes a limited implementation used to determine whether a proposed technical approach is feasible before committing to full production development?
• Which term serves as a memory of knowledge, preferences, and conventions that persists across sessions?
• Which concept is described as a rule-based check that yields repeatable results for objectively measurable requirements such as schema validity or exact values?
• How do you design Claude integration with external systems to maintain reliability?
• What term describes the period during which stored reusable request content remains eligible for reuse?
• Which term is described as providing several examples of desired input-output behavior to demonstrate the expected pattern?
• Which term is associated with task-specific input supplied by the requester?
• Which term describes the complete trace of a request across system components, used for analysis?
• Which concept involves inserting fixed guidance and flexible fields to be populated dynamically by applications?
• Which term denotes a capability whose execution is handled by Anthropic-hosted infrastructure rather than by the calling application?
• Which term describes a desired organizational outcome that explains what value a proposed solution must provide?
• Which architecture describes multiple specialized AI workers collaborating to tackle different portions of a larger task?
• Which term describes the set of foundational AI services hosted by AWS that include governance controls?
• Which term refers to persisted information that allows useful knowledge, preferences, project conventions, or prior state to remain available across separate sessions?
• What steps are involved in monitoring Claude for drift in performance?
• What term describes moving an application to a newer model version while retesting prompts, behavior, and compatibility before full deployment?
• Which term describes protecting stored information using cryptographic controls to reduce exposure if storage is accessed improperly?
• Which concept is described as structured delimiters that separate instructions, examples, source material, or other prompt sections to improve clarity?
• How can you test Claude's ability to refuse unsafe requests without harming user experience?
• Which term is an early working representation used to explore system behavior, user interaction, architecture assumptions, and design choices?
• Which concept describes layering multiple independent safeguards so that failure of one does not compromise the system?
• Which term is an open standard for connecting AI applications to external tools, data sources, and services through reusable interfaces?
• Which combination of testing ensures Claude's performance under high load?
• Which retrieval technique finds content based on meaning and conceptual similarity rather than relying only on keywords?
• What should be included to provide explainability and regulatory traceability for Claude outputs?
• Which term describes a process in which an AI system dynamically decides which actions or tools to use rather than following every step through fixed application logic?
• Which term denotes a create-once, reuse-many capability for storing documents and referencing them later through identifiers rather than repeatedly uploading the same content?
• What term refers to the elapsed duration between submitting a request and receiving the first generated output token?
• Which definition best describes a reusable integration endpoint that exposes capabilities or information to compatible AI clients?
• Which security strategy applies multiple independent safeguards so that failure of one control does not expose the entire system?
• Which term refers to a storage system optimized for indexing and searching vector-based representations?
• How do you manage model drift and data drift separately?
• What protects information while it travels between systems by using secure communication protocols?
• What is the defined measure for the average monetary expense associated with completing one meaningful unit of application work?
• Which second-stage retrieval technique scores an initial set of candidates more precisely so the most relevant information is presented first?
• Which concept defines a recurring process that captures production outcomes, reviewer corrections, user signals, and failures to inform future evaluations and improvements?
• Which concept consists of deterministic commands configured to execute automatically at defined points in an AI-assisted development workflow?
• Which concept is described as additional internal reasoning capability intended to improve performance on complex problems requiring deeper analysis?
• What is a 'confabulation' and how to detect in Claude outputs?
• Which term refers to the practice of understanding internal system behavior through traces, logs, metrics, request metadata, and errors?
• What term describes data that can identify or be linked to a specific individual?
• Which API specializes in managing message history, system prompts, and configuration for conversational tasks?
• Which concept is about comparing automated judgments against trusted human judgments and adjusting the evaluation process when systematic differences appear?
• Which term is most directly associated with benchmarking and validating model outputs using a curated baseline of inputs and expected results?
• Which term denotes a documented set of criteria describing how output quality should be scored or classified?
• Which term describes the architecture that retrieves external information to ground the model's responses before generation?
• What is a safe fallback if Claude cannot answer a user query?
• Which term describes a capability or behavior that a system must provide to accomplish a business or user objective?
• What practices support safe production versioning of Claude prompts and policies?
• Which architecture retrieves relevant external information and provides it as grounding context before a model generates its response?
• What is the role of feature flags in Claude deployment?
• What is the language-agnostic software library that simplifies application interaction with remote endpoints?
• Which term best describes the environment that runs autonomously with an agent harness and tools, without requiring the application to implement the entire loop?
• Which practice supports compliance with data localization requirements?
• Which deployment approach intentionally exposes a new version to a small subset of users before full rollout?
• Which design element best enhances explainability for non-technical users?
• In Claude risk assessment, which risk category focuses on data breach risk?
• Which architecture involves multiple specialized AI workers collaborating or operating independently to complete different portions of a larger task?
• Which term describes a measurable quality constraint such as latency, availability, scalability, security, reliability, or cost?
• What credential is used by software to identify and authorize programmatic requests to a service?
• Which term is a machine-readable specification that defines the fields, types, and constraints accepted by a callable function?
• Which term is listed last and typically relates to storing or reusing prompts to improve efficiency?
• Which term denotes a deliberate choice where improving one system characteristic may reduce another, such as increasing accuracy at the expense of latency or cost?
• Which term describes the practice of controlling which information enters the model's working environment to ensure the most relevant instructions are available?
• Which term is the open standard for connecting AI applications to external tools, data sources, and services through reusable interfaces?
• How do you handle model update risk when deploying Claude new version?
• Which dataset type is designed to test a system under conditions that are outside typical usage, challenging its limits?
• In Claude architecture, what is the function of a paraphrase or summarization module?
• Which concept describes the reduction or summarization of accumulated information to allow long-running interactions to continue without storing every token?
• Which metrics are most appropriate for evaluating Claude's quality in a QA assistant?
• Which term describes an agentic development environment that can understand codebases, modify files, execute commands, debug problems, and assist with software engineering workflows?
• What combination constitutes a robust guardrail for a customer support chatbot?
• Which term describes an AI-powered component that can reason about a goal, choose actions, invoke tools, evaluate results, and continue working until a task is completed?
• What is the primary purpose of a log retention policy?
• What is the primary purpose of guardrails at the Claude API boundary?
• Which term refers to a reusable instruction structure containing fixed guidance and variable fields used by applications?
• Which term describes a documented framework outlining how outputs should be scored and classified across evaluations?
• Which term describes a reusable external interface that exposes capabilities or information to AI clients?
• Which term refers to an isolated worker assigned a focused task with its own instructions, tools, and context?
• Which concept involves re-running established tests after a change to detect deterioration in previously acceptable behavior?
• Which retrieval strategy combines semantic similarity with lexical or keyword-based matching to improve coverage and precision?
• What is synthetic traffic in performance testing?
• Which term is commonly used to capture a significant technical choice, its context, alternatives considered, and rationale in a version-controlled document?
• Which method involves testing through user-level comparisons to identify the superior version under similar exposure?
• How can you handle multilingual prompts in Claude?
• What process involves removing or masking sensitive values before information is displayed, transmitted, logged, or processed further?
• Which term refers to assessment performed by qualified reviewers when quality depends on expertise, judgment, nuance, or accountability?
• What is the recommended approach to handling drift in Claude's inputs vs model behavior?
• Which principle limits collection, processing, transmission, and storage to information necessary for the intended purpose?
• Which term is an authorization framework that enables delegated access to protected resources without requiring applications to directly handle a user's primary credentials?
• Which term describes an ongoing process of using evidence to fine-tune prompts, models, and system architecture?
• What is the primary conversational interface for supplying message history, content, tools, and other request configuration?
• Which term represents a capability where the requested operation is executed by the hosting infrastructure rather than the client application?
• Which term is a specific reliability or performance target used to measure whether a service is operating at an acceptable level?
• Which term is metadata describing a callable capability, including its purpose, parameters, and expected input structure?
• Which term refers to contextual information or data exposed for an AI application to read and incorporate into its work?
• What are the explicit and measurable conditions that define acceptable system performance before testing begins?
• Which concept is defined as using a fixed rubric to score outputs in a deterministic, repeatable manner?
• Which approach to rate-limiting Claude endpoints provides robust protection with user and IP quotas?
• What is the role of synthetic data in Claude training and evaluation?
• Which element is essential for making explainability accessible to non-technical users?
• What is data lineage and why is it important for Claude governance?
• Which file contains repository-specific context and persistent coding conventions guiding a project?
• Which concept corresponds to a packaged collection of domain-specific instructions that can be loaded when needed?
• Which phrase defines a reusable interaction template for a particular workflow or task within external integrations?
• Which process systematically determines whether an AI solution meets defined quality, safety, business, and operational requirements?
• Which term refers to information returned to the model after an external operation finishes so reasoning can continue?
• In performance metrics, which latency measure indicates the value below which 95 percent of response times fall?
• What is an executable operation made available through a standardized external integration?
• What is multi-tenant isolation and how to implement in Claude?
• Which principle involves granting identities, applications, and tools only the permissions necessary to complete their authorized tasks?
• Which term describes a predefined sequence of operations where application logic controls exactly which steps execute and in what order?
• Which approach defines data retention for Claude's training data?
• In a multi-tenant Claude deployment, which isolation strategy reduces cross-tenant data leakage?
• Which term describes the overall technical design that connects models, applications, data sources, controls, infrastructure, and operational components to satisfy business requirements?
• What protects stored information using cryptographic controls to reduce exposure if storage is accessed improperly?
• Which term refers to a language-specific software library that simplifies application interaction with remote endpoints and handles common request and response operations?
• What is the AWS-hosted deployment option for building applications and managed agent workloads while integrating with cloud-native controls?
• What determines which resources and operations an authenticated identity is permitted to access?
• Soak testing is best described as?
• Which concept is described as a model's responses constrained to a defined format or schema so downstream software can process them reliably?
• Which term describes a function whose requested operation is executed by the surrounding application or infrastructure?
• A reusable interaction template exposed by an external integration for a particular workflow or task is called what?