CCDV-F Free Practice Test: 10 Questions With Answers

Source: the Claude Certified Developer - Foundations Exam Guide, version 1.0, effective July 2026. Anthropic notes that the guide is subject to change without notice.

The Claude Certified Developer - Foundations certification validates that a candidate can build, integrate and ship production-grade applications, agents and workflows with Anthropic's Claude platform at a foundational level. The guide describes the audience as AI and machine learning engineers, technical leads and senior software engineers, typically with one to five years of software engineering experience and at least six months of hands-on experience with Claude or comparable LLM systems. There are no mandatory prerequisites or courses.

10 free CCDV-F practice questions#

These questions come from CertCompanion's CCDV-F bank of 302 questions, the same pool the full practice exam draws from. Choose an answer before you open the explanation.

Question 1#

A developer is debugging a Claude API integration. The response contains a stop_reason of 'max_tokens'. What is the correct interpretation and next step? (Select one!)

• A. The model refused to answer; check the refusal content.
• B. The response is complete; parse the content as the final answer.
• C. The model requested a tool call; execute the tool and resend.
• D. The response was truncated; increase max_tokens or continue the conversation.

Show answer and explanation

Correct answer: D. The response was truncated; increase max_tokens or continue the conversation.

A stop_reason of 'max_tokens' indicates the output hit the token limit and was truncated, so it should not be parsed as complete. The developer should increase max_tokens or send a follow-up request with the truncated content to continue. 'end_turn' indicates completion, 'refusal' indicates a refusal, and 'tool_use' indicates a tool call.

Question 2#

A developer is using prompt caching to reduce cost for a multi-turn chat application. The system prompt and a fixed set of instructions are placed before the cache checkpoint. The developer notices the cache is never being hit. They inspect the code and find that a timestamp is prepended to the user message before the checkpoint. What is the most likely reason the cache is not being reused? (Select one!)

• A. The max_tokens parameter is set too low, preventing the cache from being stored
• B. The timestamp changes on every request, invalidating the cache because the prefix before the checkpoint is not stable
• C. The cache checkpoint is placed too early in the prompt, so the cached prefix is too short
• D. The system prompt is too long, exceeding the maximum cacheable prefix length

Show answer and explanation

Correct answer: B. The timestamp changes on every request, invalidating the cache because the prefix before the checkpoint is not stable

Prompt caching reuses a stable prompt prefix across calls. Placing a timestamp or user ID before the cache checkpoint changes the prefix on every request, so the cache is invalidated each time. The checkpoint position and system prompt length are not the issue; the instability of the prefix is the root cause.

Question 3#

A developer is setting up Claude Code for a team project. They want to ensure that project-specific instructions are version-controlled and shared with all team members, while personal preferences remain local to each developer. Which configuration approach should they use? (Select one!)

• A. Place team instructions in ~/.claude/CLAUDE.md and personal instructions in <project>/CLAUDE.md
• B. Place both team and personal instructions in ~/.claude/CLAUDE.md to keep them in one place
• C. Place team instructions in <project>/CLAUDE.md and personal instructions in ~/.claude/CLAUDE.md
• D. Place both team and personal instructions in <project>/.claude/CLAUDE.md for easy sharing

Show answer and explanation

Correct answer: C. Place team instructions in <project>/CLAUDE.md and personal instructions in ~/.claude/CLAUDE.md

Team configuration lives in the project directory (<project>/CLAUDE.md or <project>/.claude/CLAUDE.md) and is version-controlled, so it is shared with the team. Personal configuration lives in ~/.claude/CLAUDE.md and is not version-controlled, keeping it local to each developer. Mixing them or placing both in one location defeats the purpose of separation.

Question 4#

A developer is building a coding assistant that must handle two types of requests: quick autocomplete suggestions and complex architectural refactoring. The developer wants to use the 'effort levels' model option to control the amount of reasoning the model applies. Which approach best aligns with the intended use of effort levels? (Select one!)

• A. Use a low effort level for all requests to minimize latency.
• B. Use a high effort level for all requests to ensure maximum quality.
• C. Use a low effort level for the autocomplete requests and a high effort level for the refactoring requests.
• D. Use a medium effort level for all requests as a compromise.

Show answer and explanation

Correct answer: C. Use a low effort level for the autocomplete requests and a high effort level for the refactoring requests.

Effort levels allow you to control the amount of reasoning the model applies to a task. For simple, well-bounded tasks like autocomplete, a low effort level reduces latency and cost. For complex tasks like architectural refactoring, a high effort level provides the necessary reasoning. Using a single high or low level for all requests ignores the trade-off and wastes resources or sacrifices quality. A medium level is a compromise that doesn't optimize for either task type.

Question 5#

A developer is building a multi-step workflow that first extracts key data from invoices, then synthesizes a summary that requires deep reasoning across multiple documents. The developer wants to minimize cost while maintaining quality. Which model assignment strategy is most appropriate? (Select one!)

• A. Use Haiku for both extraction and synthesis to minimize cost.
• B. Use Haiku for extraction and Opus for synthesis.
• C. Use Sonnet for extraction and Haiku for synthesis.
• D. Use Opus for both extraction and synthesis to ensure maximum accuracy.

Show answer and explanation

Correct answer: B. Use Haiku for extraction and Opus for synthesis.

Multi-stage workflows should assign cheaper/faster models (Haiku/Sonnet) to routine extraction steps and reserve the strongest model (Opus) for hard synthesis and reasoning-heavy nodes. Using Opus for extraction wastes cost; using Haiku for synthesis risks poor reasoning quality. Sonnet for extraction and Haiku for synthesis reverses the appropriate allocation.

Question 6#

A developer is deciding between self-hosting an agent execution environment and using an Anthropic-hosted managed deployment. The agent handles sensitive customer data and must comply with strict data residency requirements. What is the primary consideration in this decision? (Select one!)

• A. Data residency and compliance requirements may dictate self-hosting to keep data within a specific region.
• B. The choice depends only on the model tier used, not on deployment.
• C. Self-hosting is always cheaper and easier to maintain.
• D. Anthropic-hosted is always more secure and compliant.

Show answer and explanation

Correct answer: A. Data residency and compliance requirements may dictate self-hosting to keep data within a specific region.

Managed deployment models weigh self-hosted versus Anthropic-hosted execution. Data residency and compliance requirements are key factors; self-hosting may be necessary to keep data within a specific region. The other options are absolute and ignore the trade-offs. The decision is not solely about cost, security, or model tier.

Question 7#

A developer is building a system that generates product descriptions. The initial prompt produces descriptions that are too verbose and miss key features. The developer wants to improve the output. Which approach is most aligned with iterative refinement? (Select one!)

• A. Use a higher temperature to increase creativity.
• B. Analyze the failures, adjust the prompt instructions, and test again.
• C. Switch to a larger model to improve quality.
• D. Add more few-shot examples to the prompt.

Show answer and explanation

Correct answer: B. Analyze the failures, adjust the prompt instructions, and test again.

Iterative refinement involves analyzing failures, adjusting the prompt, and retesting. This is the core practice for improving prompt reliability. Switching models or adding examples may help but is not the systematic approach. Higher temperature increases randomness and is unlikely to fix verbosity or missing features.

Question 8#

A developer is implementing a custom tool that allows Claude to look up employee records by ID. The tool schema includes a 'name' field and a 'description' field. The developer writes the description as 'Use this tool to find an employee by their ID.' However, Claude frequently calls this tool with incorrect IDs. What is the most likely cause and the best fix? (Select one!)

• A. The tool name is too generic; rename it to something more specific.
• B. The tool is not registered correctly; re-register it with the API.
• C. The tool schema is missing a required parameter; add a 'department' field.
• D. The description is too vague; rewrite it to include the expected ID format, example values, and when to use it.

Show answer and explanation

Correct answer: D. The description is too vague; rewrite it to include the expected ID format, example values, and when to use it.

Tool descriptions act as prompts that guide the model's routing and argument generation. A vague description like 'find an employee by their ID' does not specify the ID format (e.g., numeric vs. alphanumeric) or provide examples, leading to incorrect calls. The best fix is to improve the description with concrete details. Renaming the tool may help but is secondary; adding a parameter or re-registering does not address the root cause of poor description quality.

Question 9#

A developer is building a system that uses an AI model to generate code. They want to prevent the model from executing destructive commands. Which approach is the most reliable for enforcing this constraint? (Select one!)

• A. Add a system prompt line stating that the model must never execute destructive commands.
• B. Use a hook that intercepts the model's tool calls and blocks any command that matches a list of destructive patterns.
• C. Ask the model to confirm before executing any command.
• D. Use a stronger model that is less likely to make mistakes.

Show answer and explanation

Correct answer: B. Use a hook that intercepts the model's tool calls and blocks any command that matches a list of destructive patterns.

Programmatic enforcement, such as hooks that intercept tool calls, is deterministic and reliable for hard security constraints. Prompt instructions have a non-zero failure rate and are insufficient for preventing destructive actions. A stronger model may still make mistakes. Asking the model to confirm is a prompt-level instruction and not a robust control. Hooks run at lifecycle boundaries and can block actions before they occur.

Question 10#

A developer wants to run a Claude Code session that provides real-time output as the model generates responses, but still allows interactive input. Which Claude Code feature should the developer enable? (Select one!)

• A. Auto-mode
• B. Streaming mode
• C. Headless mode
• D. Session management

Show answer and explanation

Correct answer: B. Streaming mode

Streaming mode provides real-time output during interactive sessions, allowing the developer to see responses as they are generated. Headless mode is non-interactive, auto-mode involves agent-driven pauses, and session management handles conversation state. Streaming mode is the correct feature for real-time output with interactivity.

CCDV-F domains and weights#

The exam guide splits the blueprint into eight domains. Per-domain percentages on your score report are informational; pass or fail depends on the total scaled score.

The two largest domains, Applications and Integration and Model Selection and Optimization, together account for 49.9% of scored items. The three smallest, Claude Code, Eval, Testing, and Debugging, and Security and Safety, together account for 13.8%.

Frequently Asked Questions#

How many questions are on the CCDV-F? The exam has 53 items to complete in 120 minutes. Items are multiple-choice and multiple-response, and each item states how many responses to select.

What is the CCDV-F passing score? The passing score is a scaled score of 720 on a scale of 100 to 1,000. Your result shows pass or fail, the scaled score and the percentage correct in each domain.

How much does the CCDV-F cost? The exam fee is $125 USD per attempt.

What is the retake policy? After a failed attempt you wait 14 days before the second attempt, 30 days before the third and 90 days before the fourth. You can take an exam up to four times in a rolling twelve-month period, and the fee applies to each attempt.

How long is the credential valid? The credential is valid for 12 months from the date it is awarded. On-time renewal is a free, non-proctored assessment on the Anthropic Partner Academy. If the credential lapses, you retake the full exam at the full fee. If exam content changes significantly, Anthropic may require holders to retake the full exam instead.

Are there prerequisites? No. The guide states that no courses or credentials are required. It recommends one to five years of software engineering experience, at least six months of hands-on experience with Claude or comparable LLM systems, proficiency in Python and/or TypeScript, and fluency with REST APIs and CLI tools.

Who is the exam not for? The guide states that the certification is not intended for non-technical or casual users of Claude-based applications, for people without hands-on software development experience, or for roles limited to prompt writing or other isolated tasks.

Next steps#

Work through the full CCDV-F practice exam for every domain with explanations. For the architecture track, see the CCAR-F free practice questions, and browse every Anthropic certification practice exam to compare the tracks.

Practice all eight CCDV-F domains with full explanations at CertCompanion's CCDV-F exam page.