--- title: "CCAR-F Practice Questions: Claude Certified Architect Foundations | Cred Farmer" url: https://credfarmer.com/practice/ccar-f description: "20 CCAR-F questions in a free 40-minute mock, no account needed. Sit it timed, quick or untimed, then get a score by domain and a reason for every option." last_updated: 2026-10-29 publisher: "Cred Farmer, an independent study platform, not affiliated with or endorsed by Anthropic or Pearson VUE" --- No account needed # Free CCAR-F practice questions: Claude Certified Architect Foundations 20 questions written for the Claude Certified Architect - Foundations blueprint, sat at the official pace of two minutes an item. No sign-up, nothing saved, nothing sent anywhere. Want more? The post [CCAR-F wrong answers explained](https://credfarmer.com/blog/ccar-f-practice-questions) works 10 more items option by option, and the [CCAR-F study guide](https://credfarmer.com/certifications/claude-certified-architect-foundations#questions) shows 6 more with their keys: 36 distinct items with this paper, none behind an account. Some sites call this exam CCA-F. The official code, on your registration and your Credly badge, is CCAR-F. 20 items · 120 minutes official · 720 to pass · No account **CCAR-F is not in the 7 December 2026 update.** Pearson VUE's notice of an exam update on 7 December 2026 names CCAO-F, CCDV-F and CCAR-P. CCAR-F is not on that list, so this page still follows its Exam Guide v1.0 (July 2026). [Pearson VUE's notice](https://www.pearsonvue.com/us/en/anthropic.html), read on 29 October 2026. **20**Items in this mock **120**Official minutes **720**Scaled score to pass **$125**Official exam fee **5**Official domains Question bank dated 2 October 2026. Checked against Exam Guide v1.0 (July 2026) on 17 September 2026. Three ways to sit it ## Start the CCAR-F mock Answer in this tab and get a raw score by domain with the key and the reason each option fails. Every mode scores the same way and shows the same review afterwards. The clock is the only difference, and it is the difference most candidates underestimate. Or work through the paper below one question at a time, checking each as you go. ### Full, timed All 20 items in 40 minutes, two minutes an item, the official pace. The sitting submits itself when the clock reaches zero. ### Quick, timed 20 items drawn evenly across the 5 domains, in 40 minutes at the same pace. A different draw each time. ### Untimed review Every item with no clock. Flag what you are unsure of, submit when you are done, and read the reasons. ## Filter the paper by domain - [All domains 20](https://credfarmer.com/practice/ccar-f) - [D1 Agentic Architecture & Orchestration 5](https://credfarmer.com/practice/ccar-f?domain=1) - [D2 Tool Design & MCP Integration 4](https://credfarmer.com/practice/ccar-f?domain=2) - [D3 Claude Code Configuration & Workflows 4](https://credfarmer.com/practice/ccar-f?domain=3) - [D4 Prompt Engineering & Structured Output 4](https://credfarmer.com/practice/ccar-f?domain=4) - [D5 Context Management & Reliability 3](https://credfarmer.com/practice/ccar-f?domain=5) Answer as you read ## The CCAR-F paper: every question 20 items, each tagged with its official domain. Work through the paper one question at a time: pick an option to check it, or ask for the key, and read why each option holds or fails. Nothing is saved. For a scored sitting against the clock, use the start modes above. Checked 0 of 20, 0 correct [Sit it timed instead](https://credfarmer.com/practice/ccar-f#pp-start-heading) 1. Question 1 · Domain 1: Agentic Architecture & Orchestration · Single answer Traces from the agentic loop you built on the Claude API show the agent repeating one tool call with identical arguments and ignoring the result. Each iteration sends the original request plus a note that the tool ran, without its output. What should you change in the loop? Select the best answer. - A. Append the tool_use block and a matching tool_result that carries the full output to the history before the next request. - B. Add a system prompt instruction that forbids calling any tool twice with the same arguments. - C. Cache each tool result by its arguments and answer repeated calls from the cache. - D. Replace model-driven tool selection with a fixed sequence that calls each tool once. 2. Question 2 · Domain 1: Agentic Architecture & Orchestration · Single answer Your research coordinator assigns every aspect a request names to its search, analysis and synthesis subagents, but about a third of reports still miss one. The coordinator passes the synthesis output straight to the user without checking it. Which change best closes the gaps? Select the best answer. - A. Raise the synthesis subagent's max_tokens and tell it to write one section per aspect. - B. Run two search subagents on the same query and merge their results before analysis. - C. Have the coordinator check each synthesis output against the requested aspects and re-delegate targeted searches. - D. Give the synthesis subagent a longer, numbered procedure that lists every research step and every aspect it must cover in the report. 3. Question 3 · Domain 1: Agentic Architecture & Orchestration · Single answer You define two subagents with AgentDefinition for a Claude Agent SDK coordinator whose allowedTools list holds only your MCP tools and Read. In production the coordinator never spawns either one and does all the work itself. What is the most likely fix? Select the best answer. - A. Describe both subagents, and when to use each one, in the coordinator's system prompt. - B. Give the coordinator every tool the two subagents hold and delete their definitions. - C. Rewrite each subagent's prompt as a numbered procedure. - D. Add the Task tool to the coordinator's allowedTools list. 4. Question 4 · Domain 1: Agentic Architecture & Orchestration · Single answer Your support agent issues goodwill credits through a credit_account tool, and policy says credits over $200 need human approval. The cap is stated in the system prompt, yet a few larger credits still go through each month. Which change enforces the cap while credits under $200 stay automatic? Select the best answer. - A. Add a PostToolUse hook that flags any credit above $200 after it posts and queues it for staff to reverse. - B. Block credit_account calls over $200 with an interception hook and escalate them. - C. Add few-shot examples of refused credits to the prompt and state that the cap is absolute. - D. Remove the credit_account tool and have the agent draft each credit for a person to submit. 5. Question 5 · Domain 1: Agentic Architecture & Orchestration · Single answer You spent two days in a Claude Code session mapping a billing service's modules and call paths, and no code has changed since. You want to explore two refactoring approaches, each starting from that full analysis and neither influenced by the other. What should you do? Select the best answer. - A. Resume the session and explore the two approaches in turn within that conversation. - B. Start two fresh sessions, each seeded with a short summary of the module map. - C. Use fork_session to create two branches from the analysis session and explore one approach in each branch. - D. Ask the resumed session to compare both approaches in a single answer. 6. Question 6 · Domain 2: Tool Design & MCP Integration · Single answer Your agent uses one MCP tool, analyze_document, to extract specifications, summarize care guides and check claims; it takes a file and a free-text instruction and returns whatever text comes back. Code that expects specification fields often fails because the output shape varies. What should you do? Select the best answer. - A. Split it into separate extract, summarize and claim-check tools, each with its own input and output contract. - B. Add worked examples of each instruction style to the analyze_document description and keep one tool for every use. - C. Add few-shot examples to the system prompt that reformat the tool's text before it is passed on. - D. Rename the tool to analyze_supplier_document and keep its interface. 7. Question 7 · Domain 2: Tool Design & MCP Integration · Select 2 Your scheduling agent's MCP booking tool returns "Operation failed" both for timeouts, which clear within seconds, and for calendars its account cannot access. The agent tells patients to call back after a timeout and retries six times on an access failure. Which two changes to the tool's error responses should you make? (Select two.) Select exactly 2.**0/2 selected.** - A. Return every failure with the same isError response and let the agent retry any failure twice. - B. Return timeouts with isError, an errorCategory of transient and isRetryable set to true. - C. Catch every exception in the server and send back an empty success response. - D. Retry every failed call three times inside the server, then send back one generic unavailable status. - E. Return access failures with isError, an errorCategory of permission and isRetryable false. 8. Question 8 · Domain 2: Tool Design & MCP Integration · Single answer Of your agent's four tools for planning applications, extract_metadata must run first, because the other three need the application_id it returns and their inputs cannot change. On about one request in ten the model skips it and the later calls fail. What is the most reliable fix? Select the best answer. - A. Explain the dependency in the system prompt and list extract_metadata first in the tools array. - B. Force extract_metadata on the first turn with tool_choice, then use auto. - C. Set tool_choice to any on every request in the planning workflow. - D. Merge the four tools into one tool whose description lists every step in order. 9. Question 9 · Domain 2: Tool Design & MCP Integration · Single answer Your team added an MCP server to the project's.mcp.json that searches code in all 40 repositories. Claude Code still runs Grep on the local checkout to find a function's callers and misses those in other repositories, and the MCP tool's description reads only "Searches code." What should you change first? Select the best answer. - A. Remove Grep from the allowed tools in the project's shared settings file for every engineer. - B. Move the server from.mcp.json to each engineer's user-level ~/.claude.json file. - C. Add a CLAUDE.md rule that Claude must use the MCP search tool for every search. - D. Expand the tool's description to say what it searches and what it returns. 10. Question 10 · Domain 3: Claude Code Configuration & Workflows · Single answer Your repository holds six packages and one 900-line root CLAUDE.md that mixes every package's standards. Sessions load thousands of tokens of irrelevant guidance, and each package lead wants to own their standards in a separate file. Which approach best fits? Select the best answer. - A. Add a heading for each package to the root CLAUDE.md and tell Claude to apply the matching one. - B. Turn each package's standards into a project skill that engineers run before starting work. - C. Keep the root file and ask each lead to review it every quarter. - D. Give each package its own CLAUDE.md that uses @import to pull in the standards files its lead maintains. 11. Question 11 · Domain 3: Claude Code Configuration & Workflows · Single answer Your team shares a project skill, /release-notes, and one engineer wants a terse, bullet-only version for an internal channel. She edited the shared SKILL.md, and the next day everyone's customer notes came out as bullets. How should she keep her format without affecting anyone else? Select the best answer. - A. Revert the shared skill and add a section for internal notes to the project CLAUDE.md file. - B. Revert the shared skill and create a personal skill in ~/.claude/skills/ under a new name. - C. Revert the shared skill and add context: fork to its frontmatter. - D. Copy the skill to ~/.claude/skills/ under the same name. 12. Question 12 · Domain 3: Claude Code Configuration & Workflows · Single answer You must replace a deprecated date library used in 64 files. Two candidate replacements handle time zones differently, and on a past migration a hidden dependency forced the team to discard days of edits. How should you run the task in Claude Code? Select the best answer. - A. Use plan mode to map dependencies and choose the library, then execute the plan directly. - B. Start direct execution with the first prompt, naming all 64 files and the chosen replacement library. - C. Stay in plan mode for the whole migration and ask for a fresh plan before each file you edit. - D. Edit the files in direct-execution batches of about ten. 13. Question 13 · Domain 3: Claude Code Configuration & Workflows · Single answer Reviewers reject many tests that a CI job writes by running Claude Code with -p: they cover trivial getters or build fixtures by hand that the shared test harness already provides. The prompt holds only the diff and "write thorough tests." Which change best fits? Select the best answer. - A. Add the existing test files for the changed modules to the prompt context. - B. Add a cleanup step that deletes any generated test whose name matches an existing one. - C. Document testing standards and the harness fixtures in CLAUDE.md. - D. Feed rejected tests from other modules into the prompt as examples of what to avoid. 14. Question 14 · Domain 4: Prompt Engineering & Structured Output · Single answer You extract claim details from adjuster reports with tool_use and a JSON schema that marks every field required. Some reports never state a policy end date, yet the extraction always returns a plausible one, and billing cannot accept an invented date. What should you change while keeping schema enforcement? Select the best answer. - A. Tell the model to return a date only when it is certain of it. - B. Make policy_end_date nullable and tell the model to return null when the report does not state one. - C. Drop the schema and ask for free-text JSON that may omit unknown fields. - D. Retry any extraction whose date fails a plausibility check, with the error attached. 15. Question 15 · Domain 4: Prompt Engineering & Structured Output · Single answer Your extraction pipeline retries with validation feedback up to three times, which fixes date-format failures on the second try. Most first-pass failures, though, are a missing tax ID that appears only in a separate customs form never sent with the request, and they still fail on the third try. What should you change? Select the best answer. - A. Raise the retry limit to six for every failing document and keep the same feedback message. - B. Remove the retry loop and accept whatever the first pass returns. - C. Tell the model to search each document harder for the tax ID. - D. Stop retrying a missing tax ID, and attach the customs form or record it as null. 16. Question 16 · Domain 4: Prompt Engineering & Structured Output · Single answer Your nightly Message Batches job audits 30,000 documents. About 2 percent of requests fail because their documents exceed the context limit, and you resubmit the whole batch, which doubles that night's cost. What should you do instead, keeping the batch discount? Select the best answer. - A. Move the audit to the synchronous Messages API with high concurrency and retry each failure at once. - B. Enable multi-turn tool calling in each batch request to fetch the rest of a long document. - C. Resubmit only the failed requests by custom_id, with the oversized documents chunked. - D. Raise max_tokens on every request in the next batch. 17. Question 17 · Domain 4: Prompt Engineering & Structured Output · Single answer Developers dismiss most comment-accuracy findings from your automated pull request review but few of its security findings, and many now skip the review altogether. Telling the review to report only high-confidence findings changed nothing. What should you do first? Select the best answer. - A. Turn off the comment-accuracy category while you rewrite its criteria. - B. Require a confidence score of 90 percent or higher on every finding the review reports. - C. Switch the whole review off until every category's prompt has been rewritten and tested. - D. Keep all categories and ask developers to record a reason for every dismissal. 18. Question 18 · Domain 5: Context Management & Reliability · Single answer Your support agent summarizes long conversations after turn 20, and in some sessions it later refunds a different amount than it quoted, such as 84.50 quoted and 85.00 refunded. Each order lookup adds over 40 fields, and the context budget cannot grow. Which change best protects the quoted amounts? Select the best answer. - A. Put amounts, dates and order numbers in a case facts block every prompt carries, and trim lookups to return-related fields. - B. Turn summarization off and send the full history, including every lookup result, on each turn. - C. Ask the summarizer for longer summaries that quote every number exactly. - D. Have the agent repeat the quoted amount to the customer on every turn. 19. Question 19 · Domain 5: Context Management & Reliability · Single answer Your extraction system reaches 97 percent field accuracy overall on a validation set of referral letters, discharge summaries and lab reports. A manager wants to skip human review for any extraction with confidence above 0.9. What should you require before any extraction skips review? Select the best answer. - A. Raise the confidence threshold for skipping review from 0.9 to 0.95 for every document. - B. Ask the model to justify each confidence score in one sentence. - C. Check accuracy by document type and by field, and confirm each segment holds up. - D. Automate the highest-volume document type first and add the others later. 20. Question 20 · Domain 5: Context Management & Reliability · Single answer Your research system's brief says a sector grew 4.1 percent, but the analysis subagent had read 4.1 percent in a 2024 annual report and 6.3 percent in a 2025 update, and synthesis kept one figure with no source or date. Disagreements between sources must stay visible. Which change best fits? Select the best answer. - A. Tell the synthesis agent to keep the most recent figure whenever two sources disagree on a value. - B. Add a review pass that deletes any figure not found in at least two of the source documents. - C. Ask the subagents to return their full reasoning chains with each finding. - D. Require a source and date on each figure, and present conflicting values side by side. Before you ask ## Before you sit the mock **Is this the real CCAR-F exam?** No. Every item here was written for this page from the official CCAR-F Exam Guide and its published domains and objectives. Nothing is recalled from a sitting, nothing is drawn from another provider, and Cred Farmer is not affiliated with Anthropic or Pearson VUE. The items are short scenarios, about the length of the guide's sample questions. Of the 20, 19 are single-answer with four options and 1 asks you to select two of five, saying so in the stem. The questions themselves are ours, and their difficulty is not calibrated against the exam. **How is the mock scored?** You get the raw count of correct answers out of the items you sat, and the same count for each domain beside that domain's official weight. A select-two item counts only when both picks match the key. The exam reports a scaled score with a 720 cut, and no practice percentage converts to it, so this page never shows a scaled score, a pass verdict or a percentile. **Do I need an account to take it?** No. Choose a start mode and the sitting runs in your browser tab. There is no sign-up, no email field and no cookie that follows you. An account only matters if you want the larger signed-in library, and the link for that sits at the bottom of the page rather than in your way. **Is it actually free, or is there a catch?** No catch. Everything on Cred Farmer is free, this page included: no card, no trial, no limit on how many times you sit it. An account costs nothing either; it adds the five timed forms, saved progress and a readiness score capped by your first sitting of each form. The one limit here is inventory: the no-account paper is one paper, so sit it a second time and you will remember some items, and that is when the signed-in forms become the next step. **How many questions are in the mock?** 20 in the full sitting, allocated across the 5 official domains in proportion to their weights. The clock runs at the official pace of 60 items in 120 minutes, two minutes an item, so the full sitting gets 40 minutes. The quick sitting draws 20 of them evenly across the domains in 40 minutes, and the untimed review takes the whole paper without a clock. **What happens to my result?** It stays in this tab. Nothing is sent to a server, nothing is stored on your device, and nothing you answer here feeds a readiness score or any aggregate. Close or refresh the page and the result is gone, which is also why it is worth reading the per-option reasons before you leave. ## This page is 20 items. The account has 300. Everything on Cred Farmer is free. An account adds the five timed forms, saved progress and a readiness score: CCAR-F has 5 forms aligned to the current blueprint, 60 items each, 300 in total, plus drills by objective, and your first sitting of each form sets the band the score can reach. No card, no trial clock. [Create a free account for the five timed forms](https://credfarmer.com/start/ccar-f?to=practice) Every item on this page was written for it from the official CCAR-F Exam Guide, not recalled from a sitting and not copied from anyone else's set. Material sold as recalled questions can cost you the credential: read [why Claude certification exam dumps are a bad bet](https://credfarmer.com/blog/claude-certification-exam-dumps). Items were last checked against the guide on 17 September 2026. The guide is published on Anthropic's [CCAR-F certification page](https://anthropic-partners.skilljar.com/claude-certified-architect-foundations-certification). Not affiliated with, or endorsed by, Anthropic or Pearson VUE. Exam facts are summarised from publicly published program information and can change. Always confirm against the official exam guide before booking.