CCA-F Practice Questions 2026: Current CCAR-F Blueprint, Scenarios, and Answer Strategy

CCA-F Practice Questions 2026: Prepare for the Current CCAR-F Exam

Last updated July 14, 2026. Exam logistics were checked against the current Anthropic certification page and Pearson page; content coverage was checked against the official exam guide.

If you search for CCA-F practice questions, you are preparing for the credential that Pearson now lists as Claude Certified Architect - Foundations (CCAR-F). Older prep pages, discussions, and this article's established URL use CCA-F, but CCAR-F is the code shown on Pearson's current Anthropic certification page. That code change matters when you schedule; it does not make older blueprint-based study work useless.

free Claude Certified Architect practice questionsPractice questions with detailed explanations

Current CCAR-F Facts Before You Practice

The latest active landing pages control current logistics. The older public exam guide remains the detailed source for domains, scenarios, and task statements, but its front matter says Version 0.1, last updated February 10, 2025. Where logistics differ, use the current Anthropic Claude Certified Architect - Foundations page and Pearson's Anthropic certification page.

The score is scaled, so 720 is not proof that 72% correct will pass. Anthropic does not publish a raw-score conversion or public pass rate. Ignore third-party pages that turn 720 into an exact percentage or advertise a guaranteed pass.

Pearson also publishes the current retake sequence: after a failed first attempt, wait 14 days; after a failed second attempt, wait 30 days; after a failed third attempt, wait 90 days. The limit is four attempts at the same exam in a rolling 12-month period. Registration starts in the Anthropic Partner Academy, then moves to Pearson for scheduling.

There is no academic degree or prior certification listed as a formal prerequisite. Access is the real gate: the certification program is open to Claude Partner Network organizations. The exam guide describes the target candidate as a solution architect with roughly six or more months of practical Claude experience; that is recommended experience, not a published eligibility threshold.

Use the Correct Domain Numbers and Weights

The official Claude Certified Architect - Foundations exam guide numbers the domains as follows. Many third-party pages reorder the three middle domains, which makes scenario labels and miss logs confusing.

Do not convert those percentages into promised item counts. The official sources publish 60 total questions and domain weights, but not a fixed number of questions per domain on every form. Use weights to allocate practice blocks, then move extra time toward the domains producing repeat errors.

A sensible first allocation for ten focused practice blocks is three for D1, two each for D2, D3, and D4, and one for D5. That mirrors the blueprint approximately without pretending the plan is an official scoring formula. After a mixed diagnostic, override the allocation with evidence from your miss log.

Map All Six Scenarios to the Right Domains

The exam guide says each exam presents four scenarios selected at random from six. It does not publish an exact question count for each scenario, so claims such as "15 questions per scenario" should be treated as arithmetic assumptions, not exam policy. Prepare all six because you do not select the four you receive.

For each scenario, make a one-page architecture card with five lines: objective, permitted tools, non-negotiable rule, likely failure, and recovery path. Then answer a practice item without looking at the card. If your explanation cannot identify the scenario constraint that defeats each distractor, you have recognized vocabulary but have not yet practiced architectural judgment.

The Five-Step Method for Answering Scenario Questions

The current certification page lists both multiple-choice and multiple-response questions. The older exam guide describes one correct response and three distractors, so practice both formats. For a multi-response item, judge every option independently against the stem instead of stopping when one answer looks good.

Use this five-step method:

1. Underline the required outcome. Is the stem asking for deterministic compliance, better tool selection, lower false positives, recoverability, or faster execution?
2. Find the binding constraint. Words such as must, first, shared with the team, without interaction, and when policy is silent often decide the answer.
3. Name the failure mode. A wrong tool, stale context, ambiguous policy, invalid schema output, and a transient outage require different fixes.
4. Choose the least complex option that solves that failure. Better descriptions may beat a new router; explicit criteria may beat a new classifier.
5. Test every distractor. State what it changes and why that does not satisfy the binding constraint.

One common but dangerous shortcut is "programmatic enforcement always wins." The official material is more nuanced. A programmatic prerequisite is correct when a mandatory sequence must never be skipped, such as verified identity before a refund. Clear prompt criteria and examples can be the proportionate first fix when the problem is judgment calibration. Match the strength of the control to the requirement.

Domain 1 Drill: Agentic Architecture & Orchestration (27%)

D1 covers the agent loop, coordinator-subagent systems, explicit context passing, workflow enforcement, hooks, task decomposition, and session state. Build one trace that shows a request, Claude's stop_reason, a tool call, a tool result appended to history, and the next decision. You should be able to explain why tool_use continues the loop and end_turn terminates it.

For multi-agent questions, practice these distinctions:

• A coordinator decomposes, delegates, aggregates, and decides what to retry. Subagents do not automatically inherit the coordinator's full context.
• Parallel work helps independent subtasks. Dependent steps or shared mutable state may require sequencing.
• A fixed chain suits a predictable workflow. Dynamic decomposition suits investigation whose next step depends on findings.
• Resuming a session is useful when prior context remains valid. A fresh session with a structured summary is safer when old tool results are stale.

Worked question: A support agent occasionally calls process_refund before get_customer returns a verified customer ID. What is the best fix?

Answer: Add a programmatic prerequisite that blocks the refund tool until verification succeeds. The constraint is deterministic ordering with financial consequences. A stronger prompt or more examples may reduce errors, but they do not guarantee the required sequence.

Domain 2 Drill: Tool Design & MCP Integration (18%)

D2 tests tool descriptions and boundaries, structured errors, role-scoped tool access, MCP server scope, and built-in Claude Code tools. Create two deliberately similar tools, then improve their names and descriptions until test prompts route reliably. A useful description states purpose, accepted inputs, outputs, boundaries, and when to use the tool instead of its neighbor.

Your error matrix should distinguish transient, validation, permission, and business-rule failures. Include whether the error is retryable, what was attempted, and what partial result remains useful. Also distinguish a failed search from a successful search with zero matches; returning both as an empty list destroys the coordinator's ability to recover.

The guide uses an example in which 18 tools are less reliable than a role-scoped set of four or five. That is an illustration of decision complexity, not a universal maximum. The testable principle is to give each agent the tools its role requires and route cross-specialty work deliberately.

Worked question: get_customer and lookup_order have minimal, overlapping descriptions, and the model chooses the wrong one. What should you try first?

Answer: Expand and differentiate the descriptions with input formats, example queries, edge cases, and boundaries. A separate routing layer adds complexity without first repairing the information Claude uses for tool selection.

Domain 3 Drill: Claude Code Configuration & Workflows (20%)

D3 covers the CLAUDE.md hierarchy, shared commands and skills, path-specific rules, plan mode, iterative refinement, and CI/CD integration. Practice the scope question until it is automatic: user-level configuration is personal; repository configuration is shared through version control; path-specific rules load conventions for matching files.

Build a tiny repository exercise containing a project command, a path-scoped rule, and a skill with restricted tools. Then run a hypothetical change through plan mode and direct execution. A multi-file architectural migration merits exploration and planning. A clear one-function validation fix usually does not.

For CI questions, recognize the need for non-interactive execution with -p or --print, machine-readable output with --output-format json, and a schema when the downstream system requires a fixed shape. The guide also favors an independent review instance over asking the same generation session to critique its own work.

Worked question: A team wants a /review command to appear for every developer after cloning the repository. Where should it live?

Answer: Put the command in the repository's .claude/commands/ directory. A command in ~/.claude/commands/ is user-scoped, while CLAUDE.md supplies instructions rather than command definitions.

Domain 4 Drill: Prompt Engineering & Structured Output (20%)

D4 is about explicit criteria, concrete examples, structured output, validation and retry, batch processing, and independent or multi-pass review. Replace vague prompts such as "be conservative" with category rules that define what to report and what to skip. When prose is being interpreted inconsistently, the guide recommends a small set of concrete input-output examples; it does not publish a magic example count that works for every task.

For extraction, separate syntax from meaning. A JSON schema can enforce shape and types, but business validation still has to check relationships such as totals, dates, and permitted values. When validation fails, feed the specific error into a controlled retry. Track which failures can be repaired by retry and which require source clarification or human review.

Worked question: An invoice may legitimately omit due_date. The pipeline currently requires the field and invents dates. What should change?

Answer: Make due_date nullable or optional, preserve the absence explicitly, and validate the resulting object before downstream use. Requiring unavailable data creates pressure to fabricate; silently deleting the field hides an important condition.

Domain 5 Drill: Context Management & Reliability (15%)

D5 covers critical-fact preservation, escalation, error propagation, large-codebase context, human review, and provenance. Practice with a long support history that contains amounts, dates, identifiers, and customer expectations. Extract those into a persistent facts block while trimming irrelevant fields from verbose tool results. The goal is not merely a shorter prompt; it is reliable retention of decision-critical information.

Escalation also requires precise triggers. An explicit request for a human, a policy exception or gap, and inability to make meaningful progress are stronger signals than negative sentiment or an uncalibrated self-reported confidence score. In research and extraction, preserve claim-to-source mappings and dates so later summarization does not erase provenance or turn time-separated measurements into false contradictions.

Worked question: After a long interaction, a support agent remembers the issue but loses the exact refund amount and order date. What is the most direct repair?

Answer: Persist transactional details in a structured case-facts block included in subsequent prompts, and trim tool results to relevant fields. Repeatedly summarizing the entire conversation risks compressing the exact values again.

Turn Every Practice Miss Into a Repair Task

the OpenExamPrep CCAR-F question bankPractice questions with detailed explanations

Run the loop in four stages:

1. Diagnose: Take a mixed set without notes. Include all five domains and more than one scenario.
2. Repair: Read only the relevant official task statement, build or sketch the smallest artifact that demonstrates it, and answer fresh questions.
3. Explain: For every option, say why it fits or fails the constraint. Correct guessing does not count as mastery.
4. Retest later: Use a new mixed set after a delay. A repair is working when the same misconception stops recurring in a different scenario.

For timing, the official average is two minutes per question. In a 60-question simulation, check whether you have reached question 30 near the 60-minute mark, but do not force equal time on every item. Flag a slow scenario, move on, and reserve a final review window that you have rehearsed. Because the exam guide says unanswered questions are incorrect and guessing has no penalty, finish every item.

A Readiness Check Without Inventing a Pass Percentage

Do not use an unofficial "75% means ready" rule as a substitute for coverage. Anthropic publishes a scaled passing score, not the raw percentage required for a particular form. A more defensible readiness check asks whether you can:

• map every scenario to its official primary domains without mixing up D2, D3, and D4;
• explain why each distractor fails, not merely identify the keyed answer;
• complete new mixed sets without repeating the same failure pattern;
• produce one practical artifact for each domain;
• finish a 60-question, 120-minute simulation without blanks; and
• verify current registration and retake rules immediately before scheduling.

Use CCAR-F flashcards for fast recall, then move back to scenarios; flashcards cannot replace trade-off practice. Keep the CCAR-F cheat sheet beside your miss log, and use the Claude Certified Architect foundations overview for broader context while treating the official live pages above as the authority for current logistics.

Official Sources and Next Steps

Use these sources in order:

1. Check the current Anthropic certification page for price, validity, delivery, format, score, and domains.
2. Check Pearson's Anthropic program page for the CCAR-F code, partner access, scheduling, and retake rules.
3. Use the official CCAR-F exam guide PDF for task statements, scenario definitions, and sample reasoning.
4. Read Anthropic's Claude Partner Network announcement for the credential's March 12, 2026 launch context and partner positioning.

free CCAR-F practice questionsPractice questions with detailed explanations