Claude Certification Guide CCDV-F

Practice questions

92 scenario questions across all eight domains, mixed by default so you have to recognise which domain a question belongs to — half the difficulty on the real exam. Every answer explains why the wrong options are wrong, which is where the reasoning actually lives.

0 / 0correct
0current streak
0best streak

Browse all 92 questions

Each question has its own page with the answer and a full explanation.

Applications & Integration 27
  1. A support-ticket classifier's 3,000-token policy document is billed as fresh input on every single call, even…
  2. 40,000 support tickets need classifying overnight for a report due at 9am. No user is waiting. What's the…
  3. A production app's output quality shifted overnight with no code deployed on either side. What's the most…
  4. A team asks for confirmation that a request's system prompt and tool definitions actually hit the prompt…
  5. A clinic's booking assistant needs to show the user typing indicator and stream the assistant's reply as it's…
  6. A chat app resends the full conversation history on every request. By turn 60 the latency and bill have both…
  7. A requirements doc says: 'Clinicians should see the summary appear progressively rather than waiting for a…
  8. An app must process scanned insurance cards that patients photograph with their phones. Which capability is…
  9. A team wants to reduce cost on a workload where the same 8,000-token clinical guidelines document prefixes…
  10. Which statement about the Message Batches API is correct?
  11. A refactor moves the system prompt from a hardcoded string into a versioned file with review required before…
  12. Two users of a multi-tenant assistant occasionally see fragments of each other's conversation context. What…
  13. Where does the CLAUDE.md file sit in a team's configuration, and what is it for?
  14. A nightly job blocks its web request thread while waiting on a long Claude call, and the service times out…
  15. An integration hardcodes the model reference as a rolling alias in production. What risk does this introduce?
  16. A compliance requirement states claim data must never leave the hospital's network. Which design choice…
  17. A response comes back with stop_reason of 'refusal'. What should production code do?
  18. A team is choosing between calling Claude directly and going through a third-party vendor that resells…
  19. A product spec says: 'Clinicians must be able to upload a scanned referral letter and have the assistant pull…
  20. A requirement states claim data may not leave the hospital network, and separately that the team would like…
  21. A team wants to know whether their prompt cache is actually working. What is evidence, rather than assumption?
  22. Which of these belongs in version control alongside application code? Select the TWO that do.
  23. An endpoint calls Claude synchronously and holds the request thread for the full generation. Under morning…
  24. A team's assistant occasionally surfaces a fragment of another user's conversation. What has gone wrong?
  25. Which statement about streaming is accurate?
  26. A response returns with stop_reason 'max_tokens'. What has actually happened?
  27. Where should retrieved document content be placed relative to your system instructions?
Model Selection & Optimization 15
  1. A workload classifies 2 million short product tags overnight; results aren't needed until morning. Select the…
  2. A teammate insists setting temperature to 0 will make two identical requests return byte-identical output. Is…
  3. A nightly pipeline defaults every task — simple and complex alike — to Opus with maximum thinking effort…
  4. A team wants production behavior to stay stable across Claude version updates rather than silently pick up a…
  5. A budget-constrained pipeline is over cost target. Put these in the order the exam expects you to try them.
  6. A request's system prompt, conversation history, retrieved documents, tool results, and the model's response…
  7. A task requires deliberate multi-step reasoning through an ambiguous clinical case, and accuracy matters more…
  8. Before scaling a pipeline from 1,000 to 500,000 daily requests, what does the blueprint expect you to do…
  9. A prompt carefully tuned against one model version produces noticeably different output after the team…
  10. A long prompt has several stable sections and one volatile section in the middle. What technique places reuse…
  11. A team is 40% over budget on a pipeline with a large, unchanging system prompt and heavy tool output in…
  12. Which task profile best fits routing to the smallest tier?
  13. What does the context window include?
  14. A prompt tuned on one model version behaves differently after an upgrade. What is the correct…
  15. Before scaling from 1,000 to 500,000 daily requests, what does the blueprint expect first?
Agents and Workflows 13
  1. A document-ingestion pipeline runs the same four fixed steps for every document, regardless of content…
  2. An orchestrator delegates three independent research subtasks to three subagents instead of running them in…
  3. A team wants a rule that blocks any agent from running a destructive shell command, no matter what the model…
  4. A team is choosing how to build a multi-step research agent. They want structured state management and are…
  5. A startup wants to ship an agent quickly with minimal operational burden and is comfortable trading some…
  6. An agent loop keeps calling tools and re-evaluating without ever producing a final answer. What design…
  7. A team needs their agent running inside their own VPC with full control over the runtime, and has the ops…
  8. In the agent loop, the model requests a tool call but your application code decides whether to actually…
  9. A manager agent delegates to four subagents. One subagent's task involves reading a long, noisy log file. Why…
  10. An agent runs a fixed sequence of four steps on every input, never varying. What should it be?
  11. What is the request-execute gap?
  12. An agent loop calls tools and re-evaluates indefinitely without producing a final answer. What is missing?
  13. A startup needs an agent shipped quickly with a small team and no dedicated ops capacity. Which deployment…
Prompt & Context Engineering 11
  1. In a long-running chat, the assistant starts ignoring formatting instructions given at the very start of the…
  2. A pipeline calls a JSON parser directly on model output and crashes whenever the model wraps its JSON in a…
  3. An instruction only sometimes applies — like a special discount policy only for orders over $500. Where does…
  4. A model reliably follows a format on 3 worked examples in training conversations but drifts on live traffic…
  5. An agent's search tool returns 40 results; the agent needed 3. All 40 stay in the conversation history for…
  6. A constraint applies only when one specific tool runs — not on every turn. Where should it live?
  7. A summarization prompt says 'don't be too long, don't leave out anything important, avoid being vague.'…
  8. An instruction reads: 'Don't be too verbose, don't omit key findings, avoid vagueness.' Why does this…
  9. A rule applies only when one particular tool runs. Where should it live?
  10. Model output occasionally arrives as valid JSON wrapped in a sentence of prose. What handles this correctly?
  11. A long session shows rising cost, slower responses, and details being missed that are demonstrably present in…
Tools and MCPs 11
  1. An agent has two tools with nearly identical descriptions and starts calling the wrong one intermittently…
  2. A capability needs to be discoverable and reusable across six different internal Claude apps built by…
  3. A team's MCP server runs as a local subprocess launched by the client; a second one runs remotely and serves…
  4. A one-off internal script needs a single bespoke capability used by exactly one application, with no plan to…
  5. An MCP server exposes a read-only patient directory that clients can fetch from. Which MCP primitive is this?
  6. An agent has grown to 22 tools and selection accuracy has degraded across the board — not between any…
  7. A team needs to package a reusable set of instructions and procedures — no executable integration, just…
  8. An MCP server exposes a read-only document store clients can fetch from. Which primitive?
  9. A maintained community MCP server already covers a team's PostgreSQL query needs. They'd prefer to build…
  10. An agent with 20 tools picks slightly wrong ones across the whole set, with no consistent confusable pair…
  11. A capability is packaged instructions and procedures the model should follow — no code execution. What is it?
Security and Safety 7
  1. An agent processes a scraped webpage that contains hidden text instructing it to email its conversation…
  2. An autonomous agent runs with the full API credentials of the engineer who launched it, including production…
  3. A support agent's tool output routinely includes customer phone numbers and partial card numbers, and the…
  4. Where should a production application's Claude API key actually live?
  5. A retrieved document contains embedded text targeting the model, while a user in chat directly asks the model…
  6. An agent authenticates successfully on every request while being manipulated into enumerating an entire…
  7. A team argues broad agent credentials are acceptable because the agent is internal-only. Is that reasoning…
Claude Code 4
  1. A team keeps manually repeating the same five-step code-review checklist in Claude Code every PR. What's the…
  2. A team wants Claude Code to run as part of an automated nightly CI job with no one at a keyboard to respond…
  3. A team wants Claude Code to refuse to suggest any code that writes directly to the production database…
  4. Which Claude Code component fits: 'never suggest code that writes directly to the production database'…
Eval, Testing & Debugging 4
  1. An assistant confidently gives a wrong answer after a tool call silently returned a week-old cached value…
  2. A production integration occasionally gets a transient 529 (overloaded) response from the API. What's the…
  3. A production integration starts receiving 429 responses during a traffic spike. What's the appropriate…
  4. An integration receives a 400 response. What is the correct action?