Description

For learners studying for Claude Certified Architect – Foundations
What you'll learn
Practice pack content
Domain 1 - Agentic Architecture & Orchestration
Domain 2 – Tool Design & MCP Integration
Domain 3 – Claude Code Configuration & Workflows
Domain 4 – Prompt Engineering & Structured Output
Domain 5 – Context Management & Reliability
Final Exam Simulators
Exam Study Guide
Interactive Flashcards
Try 3 Claude Architect Foundations practice questions
Choose an answer, explore the explanation, and see why the other options are less suitable. Three short scenarios. No sign-in required.
Agentic Architecture & OrchestrationTask 1.6 · Design task decomposition strategies for complex workflows
1. A release-note workflow with fixed dependencies
A release-note service must extract issue IDs from merged pull requests, verify those IDs against the release manifest, and draft notes using only verified entries. The stages are known in advance, and drafting must stop if verification fails. Which architecture best fits these requirements?
- Run extraction, verification, and drafting in parallel, then combine their independently generated outputs.
- Use a fixed prompt chain with application checks, passing verified results forward and stopping on failure.
- Let an autonomous agent choose the task order, then validate the complete release notes after drafting.
- Generate several complete drafts independently, then select the version with the most consistent issue IDs.
Answer and full explanation
Correct answer: B — A fixed prompt chain with checks between dependent stages.
Why B is correct
The workflow has a known sequence, and each later stage needs an earlier result. A prompt chain separates extraction from drafting while the application verifies the extracted IDs against the manifest. The application starts drafting only after that check passes and supplies the verified entries as input. This enforces the dependency instead of asking an autonomous agent to choose the order. Parallel work is useful for independent subtasks, but these stages are dependent.
Why the other options are less suitable
- A — Parallel dependent stages: Parallel execution can shorten tasks that do not need each other’s results. Here, drafting would begin before the verification result is available.
- C — Unconstrained task ordering: An autonomous agent can suit investigations where the next step depends on new discoveries. This workflow already has a required order, and checking only the finished draft does not enforce it.
- D — Voting across complete drafts: Comparing drafts can help identify inconsistent wording or findings. Agreement between drafts does not establish that their issue IDs passed the required manifest check before drafting.
Key takeaway: Use explicit dependencies and application checks when a workflow has a fixed order and required checkpoints.
Read two more sample questions
Tool Design & MCP IntegrationTask 2.2 · Implement structured error responses for MCP tools
2. A tool error that should change the next action
A room-booking agent calls an MCP tool to reserve a conference room for five hours. The tool rejects the request because policy caps each booking at two hours, but its generic error makes the agent repeat the unchanged call. Which result best supports an appropriate next step?
- Return a transient error with the two-hour limit, and recommend retrying the same request after a short wait.
- Return success with an empty booking ID, and include the two-hour policy limit in a separate warning field.
- Return isError: true with the policy reason and metadata marking the unchanged request as non-retryable.
- Return a generic protocol error without policy details, and let the client apply its default retry behaviour.
Answer and full explanation
Correct answer: C — Report the business-rule failure and explain why the same request should not be retried.
Why C is correct
The booking failed because of a business rule, not a temporary service interruption. MCP tool execution failures can be reported with isError set to true and actionable information in the result. Explaining the two-hour limit lets the agent discuss an allowed booking or an authorized alternative. Application-defined retry metadata can tell the client not to repeat the unchanged request. Such metadata is part of this tool’s contract, not an additional mandatory MCP field. Waiting alone will not make the five-hour booking comply with the stated policy.
Why the other options are less suitable
- A — Treating a policy rejection as transient: A delayed retry can be appropriate for a temporary outage. The stated duration still exceeds the limit after a short wait, so repeating it does not address the rejection.
- B — Reporting failure as success: A successful lookup may legitimately return no matching records. A rejected booking is different, and marking it successful misrepresents whether the requested action completed.
- D — Hiding the business-rule explanation: Protocol errors are appropriate for issues such as an unknown tool or a malformed protocol request. Here, the tool handled the call and rejected its business operation, so a generic protocol error loses useful recovery information.
Key takeaway: Distinguish business-rule failures from temporary outages, and make the recovery decision explicit in the tool result.
Prompt Engineering & Structured OutputTask 4.4 · Implement validation, retry, and feedback loops for extraction quality
3. Valid JSON with an invalid date relationship
An application extracts equipment-hire dates from text using a strict JSON schema. One result passes schema validation but places the return date before the collection date, contradicting the source. What should the architect add before accepting records into the scheduling system?
- Accept each schema-valid record after asking Claude for a confidence score about the extracted dates.
- Request another extraction and accept the original record whenever both responses contain valid JSON.
- Require both date strings and check their individual formats, then accept records that pass those checks.
- Check date relationships and source evidence, request corrections with specific errors, and revalidate.
Answer and full explanation
Correct answer: D — Validate meaning and source consistency, then retry with targeted feedback.
Why D is correct
Schema compliance does not establish that extracted values match the source. The existing checks accept the individual fields but miss the invalid relationship between them. The application should check that the return date is not earlier than the collection date and verify the field assignments against the source. A correction request should include the original text, the failed extraction, and the specific validation error. Revalidate the correction before accepting it; unresolved cases should be held for review rather than retried indefinitely.
Why the other options are less suitable
- A — Using confidence as acceptance evidence: Confidence can be one signal in a calibrated review process. It cannot override a detected contradiction between the extracted dates and the source.
- B — Checking only whether another response parses: A second extraction can provide material for comparison. Two parseable responses still do not establish that either assigns the correct dates to the correct fields.
- C — Validating fields without their relationship: Required fields and date-format checks help reject missing or malformed values. Two individually valid dates can still be reversed or inconsistent with the source document.
Key takeaway: Treat schema validation, cross-field checks, and source verification as separate acceptance requirements.
Preview complete
You have tried all three samples
Review the explanations or try the samples again. This short preview covers only a small part of the exam objectives.
Your sample score resets when this page reloads.
Your practice pack includes
More ways to make each study session count
Focus on a topic, practise under a time limit, or return to the supporting resources before your next attempt.
Domain-based practice
Timed practice sets
Study guide
Flashcards
Requirements
- Anthropic describes a typical candidate as having six months or more of practical experience with Claude APIs, Agent SDK, Claude Code and MCP. This is recommended background, not a stated mandatory experience requirement.
- For unfamiliar topics, combine question practice with official documentation and a small project of your own. Before booking the official exam, confirm eligibility through the Claude Partner Network registration process.
- Use an internet-connected computer, tablet or phone, a current web browser and an ExamsDigest account to access this online practice product.
Know the exam you’re studying for
These details describe Anthropic’s official Claude Certified Architect – Foundations exam.
What our students say
“I kept choosing the more complicated solution because it sounded better. The ExamsDigest questions made me slow down and think about what the scenario actually needed, especially when choosing between a fixed workflow and an autonomous agent.”
Adrian Foster
✓ Verified Review
“The explanations were the most useful part for me. Understanding why the other options didn’t fit helped me spot gaps that I hadn’t noticed while reading my notes.”
Hannah Lewis
✓ Verified Review
“I used this pack alongside my Claude Architect Foundations study notes. The questions about tool errors and output validation gave me specific things to revisit and test in my own project. I liked having to think through what should happen when something goes wrong.”
Marcus Bennett
✓ Verified Review
Your questions, answered.
Which Claude Architect exam does this pack target?
This practice targets Claude Certified Architect – Foundations (CCAR-F), using the July 2026 exam guide, version 1.0. Match that code to your intended certification before purchasing. Check the product’s listed resources and access terms separately, and confirm your official registration eligibility. A Foundations practice product should not be treated as preparation for a different Architect exam.
Who should use CCAR-F practice questions?
This practice is for technical learners who need to reason through Claude application design, not just write everyday chat prompts. Familiarity with agents, tool calls, project instructions and JSON-based data exchange will make the scenarios easier to follow. Use unfamiliar terminology as a signal to return to the relevant documentation before attempting similar questions again.
How can I combine questions with practical work?
Pair the questions with a small project you control, such as a release-note workflow or a text-to-JSON extractor. Use synthetic data, define the expected behaviour, and test both successful and failing inputs. After answering a question, explain how your chosen approach would affect the project. These suggested exercises are separate from the practice questions; they do not imply a bundled lab environment.
What should I do with incorrect answers?
Treat each incorrect answer as a decision to revisit, not just an item to repeat. Identify whether you missed a dependency, confused an error type, trusted an unsupported value or chose the wrong configuration scope. Read the explanation, check the relevant documentation, and try a new example. Use the pattern of mistakes to plan your next study session rather than treating a score as a pass prediction.
Does the official exam use this three-question format?
No. The official exam uses multiple-choice and multiple-response items, and its questions are grouped around selected scenarios. Each item specifies how many answers to choose. This preview offers three standalone, single-answer examples, so it does not reproduce the full exam structure. Check the paid product’s description for the formats and resources it actually includes.
Does completing this practice earn the certification?
No. Practice completion does not award Anthropic certification or schedule an official exam. Register through Anthropic Partner Academy and follow the Pearson VUE booking process, subject to your eligibility. Pearson VUE currently describes certification as open to Claude Partner Network organizations. Check the product’s purchase terms for any voucher arrangements; do not assume practice access includes an official booking.
30-day conditional refund policy
Eligible requests must be made within 30 calendar days, with no more than 20% of any included simulator completed. The purchase must not already be refunded or under a payment dispute or chargeback. Other provisions apply; mandatory consumer rights remain unaffected.
Useful links for your preparation



