Claude Certification Blog
CCAR-F wrong answers explained: ten questions, worked option by option
CCAR-F wrong answers explained, one option at a time: ten original CCAR-F questions explained in full, written to the five-domain official Exam Guide v1.0 and allocated by the published weights so every domain appears, each with the key and a stated reason why every option holds or fails.
Below are the CCAR-F wrong answers explained for ten questions that are not exam items and were not written from any exam item. Each is an engineering scenario with a measured symptom and a constraint, then four options, or five for a select-two. Nine are single response and one is select-two, the two formats the guide describes. Answer every item cold, record the misses in the map below, and open the reason for every option afterwards. Three items sit in Agentic Architecture and Orchestration because it is the heaviest domain at 27%.
The official CCAR-F exam guide is the authority on format and weights; the Claude Certified Architect - Foundations page carries the blueprint and six more open items. The reading method these items reward is in the Claude certification sample questions hub, and what official material exists is in the guide to Claude certification practice exams.
How are these ten CCAR-F questions weighted?
By the official blueprint: 27% Agentic Architecture and Orchestration, 20% Claude Code Configuration and Workflows, 20% Prompt Engineering and Structured Output, 18% Tool Design and MCP Integration, 15% Context Management and Reliability. The flattest blueprint of the four exams lets ten items reach every domain: three, two, two, two and one. The live exam frames its 60 items with four scenarios drawn from a bank of six; these ten are standalone scenarios, and the free CCAR-F practice exam, a 20-item set, is not built as scenarios either.
Agentic Architecture and Orchestration: who decides, and in code or in prose?
Three items on the heaviest domain. The first asks what a coordinator should do when most queries need one subagent and every query gets four; the second is a refund that must never be issued before the customer's identity is verified; the third is a fixed pipeline applied to an open-ended task. Each asks for a structural change: a coordinator that selects, a gate in code, a plan that adapts. Parallel runs, caching and logging are real techniques, offered here for problems they do not fix.
Your research coordinator always runs four subagents in order: search, analysis, synthesis and report writing. Most staff questions are single-fact lookups, yet each costs as much as a full literature review, and spend is frozen while latency is fine. Which change best fits?
Show the answer and the reason for every option
Answer: B
Pick subagents by query complexity. Task statement O02: Orchestrate multi-agent systems with coordinator-subagent patterns.
A. Parallel Task calls cut latency, which is already fine. Every query still pays for all four subagents.
B (correct). Selecting subagents per query sends simple lookups through search alone and keeps the full pipeline for broad topics, cutting cost where it is spent.
C. Hub-and-spoke routing sends all subagent communication through the coordinator for observability and error handling. Bypassing it trades those away.
D. A cache helps only when the same question repeats. New single-fact questions still run the full four-stage pipeline.
Your returns agent must confirm identity with verify_identity before it calls issue_refund, and the system prompt says so. Logs show about one refund in eleven issued with no verification call, two of them to the wrong account. Which change most reliably enforces the order?
Show the answer and the reason for every option
Answer: D
Enforce required order with a prerequisite gate. Task statement O04: Implement multi-step workflows with enforcement and handoff patterns.
A. Placement and repetition make the rule more visible, but compliance stays probabilistic, and the guide says that is not enough before a financial operation.
B. A daily log review catches the wrong refunds after they are paid. The order is still not enforced at the moment the refund runs.
C. A required parameter only checks that some ID is present. The model can still pass an ID the customer stated, with no verification at all.
D (correct). A prerequisite gate in code blocks the refund until verification has succeeded, which turns the rule into a guarantee instead of a request.
An agent workflow is adding tests to a 900-module legacy codebase by listing modules alphabetically and testing each in turn. After three days it has covered 210 small helpers, while the payment and permission modules behind recent incidents remain untested. How should the decomposition change?
Show the answer and the reason for every option
Answer: B
Map first, then prioritize by impact. Task statement O06: Design task decomposition strategies for complex workflows.
A. Parallel shards speed the same plan up but keep treating every module as equal. The risky modules still wait their turn in alphabetical order.
B (correct). Mapping first, then prioritizing high-impact areas and adapting as dependencies surface, is the decomposition the guide gives for open-ended test work.
C. A fixed two-pass chain suits a predictable review. It still covers the payment and permission modules only when the alphabet reaches them.
D. Deeper tests per module spend more time on the helpers already covered and do nothing to reach the incident-prone modules sooner.
Tool Design and MCP Integration: what does the model select on, and where do secrets live?
Two single-response items. The first is a misrouting case where the tool descriptions are already clear and distinct, so the cause is a system-prompt phrase that ties a whole class of questions to the wrong tool. Rewording that phrase removes the cause; the routing layer, the merged tool and the forced tool call each work around it. The second is scope and secrets: a shared server belongs in the project's committed configuration, and the token reaches it through an environment variable, never through the repository.
Your support agent has two MCP tools, search_orders and search_shipments, with clear and distinct descriptions. The system prompt says "check the order first" before any delivery question, and logs show many delivery questions going to search_orders, which cannot return carrier status. What should you change?
Show the answer and the reason for every option
Answer: C
Prompt keywords can override good tool descriptions. Task statement O08: Design effective tool interfaces with clear descriptions and boundaries.
A. A classifier removes the model's choice and adds infrastructure, while the misleading instruction that causes the misrouting stays in the prompt.
B. A merged tool hides which backend answers and may return an order when the customer needs carrier status.
C (correct). Rewording the instruction removes the keyword that pulls delivery questions to search_orders, and the clear descriptions can then guide selection.
D. Setting any guarantees a tool call, not the right tool. The agent already calls a tool here, just the wrong one.
Your team wants every developer's Claude Code session to reach the ticketing system through one shared MCP server. A developer pasted a token into a personal config, and the same token later turned up committed in a teammate's copy. How should the server be set up so credentials never enter the repository?
Show the answer and the reason for every option
Answer: A
Shared server in .mcp.json, token from the environment. Task statement O11: Integrate MCP servers into Claude Code and agent workflows.
A (correct). A committed .mcp.json shares the server with the team, and environment variable expansion keeps the token on each developer's machine, out of the repository.
B. User scope is for personal or experimental servers. Each developer would configure it alone, and the token would sit in a document.
C. A committed token is a leaked secret from the moment it is pushed. Rotation shortens its life but does not keep it out of the repository.
D. CLAUDE.md gives project instructions and context. It does not configure MCP servers, and writing the token there commits it.
Claude Code Configuration and Workflows: what stays out of the main context?
Two items. The first is a team skill whose findings are worth keeping and whose 60,000 tokens of scan noise are not; the frontmatter option that runs it in a forked sub-agent isolates the noise without making the skill personal or narrowing what it scans. The second is a pull request pipeline with two defects, unparseable output and duplicated findings, and the two fixes the guide names for exactly those: a JSON schema on the output, and the earlier findings fed back into context. Batch processing fails because the check is blocking; session memory fails because the guide prefers explicit context.
Your team's project skill /license-scan reads every dependency manifest and prints its findings. After one run, the main conversation holds about 60,000 tokens of scan output, and Claude loses track of the feature the developer was building. Which frontmatter change helps while the skill stays shared from the repository?
Show the answer and the reason for every option
Answer: B
Isolate verbose skills with context: fork. Task statement O14: Create and configure custom slash commands and skills.
A. allowed-tools limits the tools available while the skill runs. The scan is read-only already, and its output still lands in the main conversation.
B (correct). context: fork isolates the verbose run, and only the findings return to the main thread, which keeps the developer's working context intact.
C. argument-hint prompts for a missing parameter. Limiting each run to one manifest makes the scan less useful without isolating its output.
D. Personal copies stop the skill being shared from the repository, which the stem requires, and each run still floods that developer's conversation.
Your CI pipeline runs Claude Code with -p to review pull requests. Its free-text output breaks the script that posts inline comments in about one build in five, and each new commit brings duplicates of comments already posted. Which two changes fix these problems? (Select two.)
Show the answer and the reason for every option
Answer: A and D
Structured output and prior findings in CI. Task statement O18: Integrate Claude Code into CI/CD pipelines.
A (correct). Structured JSON output that follows a schema gives the posting script a fixed shape, which removes the parsing failures that free text causes.
B. A session that wrote the review keeps its own reasoning context and is less likely to question it. It also does nothing for the parsing failures.
C. Batches have no latency guarantee and suit overnight work. A pull request review blocks merging, and batching fixes neither problem.
D (correct). Prior findings in context let Claude skip what it already reported and raise only new or unaddressed issues, which removes the duplicates.
E. Without -p, Claude Code waits for interactive input and the CI job hangs. The output also stays free text.
Prompt Engineering and Structured Output: what fixes both symptoms with one change?
Two single-response items. The first is a self-review that declared its own defect intentional; the guide's answer is an independent instance with no access to the generation conversation, not more thinking, more passes or a per-table split inside the same session. The second is a review whose two pages of instructions still produce six layouts and a 45 percent false-positive rate; a strict schema fixes only the layout, a conservative instruction fixes neither, and a few targeted examples fix both.
Claude writes database migration scripts and then reviews each one in the same session with "check your work carefully." Several scripts reached staging with defects the self-review had called safe, one of them a null-handling error it called intentional. What change most improves defect detection before staging?
Show the answer and the reason for every option
Answer: D
An independent instance reviews better than self-review. Task statement O24: Design multi-instance and multi-pass review architectures.
A. The guide says independent review beats extended thinking for this. The model still reasons inside the context that produced the defects.
B. Three passes in one session share the same reasoning context, and they tend to agree on the same blind spots instead of catching them.
C. Per-table passes help attention on a large review, but each pass still runs inside the context that wrote the script and shares its blind spots.
D (correct). A separate instance has none of the generator's reasoning to defend, which the guide says makes it better at catching subtle defects.
Your automated code review prompt has two pages of instructions on output format and on which patterns to flag. Findings still arrive in six layouts, and 45 percent of them flag accepted team conventions, such as a shared retry wrapper. Which single change fixes both problems?
Show the answer and the reason for every option
Answer: A
Few-shot examples fix format and judgment. Task statement O20: Apply few-shot prompting to improve output consistency and quality.
A (correct). Examples show the exact layout and teach the line between accepted conventions and real issues, which addresses both problems with one change.
B. A schema gives one layout, but the review would still flag the shared retry wrapper. It solves only half the problem.
C. The guide says general instructions like "be conservative" fail to improve precision, and shorter instructions do not show the expected layout.
D. Temperature affects randomness, not what the model believes the format and the team's conventions are. Both problems remain.
Context Management and Reliability: is an empty list a result or a failure?
One item on the lightest domain, and the sharpest distinction in the set. A search subagent that returns the same empty list for no matches and for a timeout leaves its coordinator unable to tell a gap in the literature from a broken API. The closest miss adds a fixed status that separates the two but strips the context a retry decision needs; terminating on any empty list is an anti-pattern the guide names outright.
Your search subagent returns an empty list both when a query matches nothing and when the search API times out. The coordinator treats every empty list as "no published work" and moves on, and some reports wrongly say a topic has none. How should the subagent report results?
Show the answer and the reason for every option
Answer: C
Tell access failures apart from empty results. Task statement O27: Implement error propagation strategies across multi-agent systems.
A. Retrying every empty result wastes time on queries that truly found nothing, and a timeout still reaches the coordinator looking like an empty list.
B. The guide names terminating a whole workflow on one failure as an anti-pattern. Most empty lists are valid results that need no alarm.
C (correct). A structured error marks the timeout as a failure and carries what the coordinator needs to decide on a retry, so an empty list again means no match.
D. This tells failure from no match, but the guide says generic statuses hide what the coordinator needs: the failure type, the query and any partial results.
How to use the CCAR-F wrong answers explained here: the miss map
Score exactly: a select-two item counts only when both letters match. Then write each miss into its domain row rather than adding up a total, because ten items cannot estimate a scaled score. What they can do is point. One miss in a domain means reread the task statement named in that row. Two or more means the domain link, which opens the free set's untimed review filtered to that domain, and then that domain's study section.
| Domain | Items | Your misses | Task statement to reread | Practise the domain |
|---|---|---|---|---|
| D1 Agentic Architecture and Orchestration | 3 | ___ of 3 | O02 Orchestrate multi-agent systems with coordinator-subagent patterns; O04 Implement multi-step workflows with enforcement and handoff patterns; O06 Design task decomposition strategies for complex workflows | Domain 1 in the free set |
| D2 Tool Design and MCP Integration | 2 | ___ of 2 | O08 Design effective tool interfaces with clear descriptions and boundaries; O11 Integrate MCP servers into Claude Code and agent workflows | Domain 2 in the free set |
| D3 Claude Code Configuration and Workflows | 2 | ___ of 2 | O14 Create and configure custom slash commands and skills; O18 Integrate Claude Code into CI/CD pipelines | Domain 3 in the free set |
| D4 Prompt Engineering and Structured Output | 2 | ___ of 2 | O24 Design multi-instance and multi-pass review architectures; O20 Apply few-shot prompting to improve output consistency and quality | Domain 4 in the free set |
| D5 Context Management and Reliability | 1 | ___ of 1 | O27 Implement error propagation strategies across multi-agent systems | Domain 5 in the free set |
If it must happen every time, it is not a prompt
The refund item offers a restated rule where the stem describes an order that must hold on every run. Ask what happens on the run where the model ignores the sentence. If nothing stops it, the option is not a control.
The set pairs with the two-week CCAR-F study plan and the CCAR-F cheat sheet, whose rules these items exercise; how the four-of-six scenario framing maps to the domains is in the CCAR-F exam scenarios post. The CCAR-F blueprint sets the weights used here. Checked against the official CCAR-F Exam Guide v1.0 (July 2026) on 17 September 2026.
Key takeaways
- Ten original items, all five domains, by weight. The flattest blueprint of the four means no domain rounds to zero, and none can be skipped.
- Answer cold, then reveal. Reading the reasons before committing turns a diagnostic into a reading exercise.
- Score by domain, not by total. Ten items cannot estimate the scaled score; a miss pattern in one domain can name the task statement to reread.
- Enforce in code what must always happen. Restating the rule is the refund item's trap, and the guide gives prompt-only enforcement a non-zero failure rate.
- The free set is a paper, not a replica. Twenty items on the five-domain guide, no scenario framing, a raw count only.
Twenty more items under a clock, then the study sections
The free CCAR-F set is 20 timed items on the five-domain guide with a reason for every option, no account needed. From there the next step is the CCAR-F study sections, the drills by objective and five timed forms built to the five-domain guide, 300 items in all. Nothing reports a scaled score or a pass probability.
Start the free CCAR-F setQuestions
Frequently asked
Short answers to the questions this post leaves open.
Are these real CCAR-F exam questions?
No. All ten were written fresh for this page from the published CCAR-F Exam Guide v1.0 and its 30 task statements, and none is drawn from, or paraphrased from, any exam form or any gated Cred Farmer form. Anthropic keeps its live items confidential; a page offering CCAR-F exam questions that claims otherwise is selling a breach.
How many questions are on the CCAR-F exam?
The official guide sets 60 items in 120 minutes across five domains, framed by four scenarios drawn from a published bank of six. Items are single-answer and multiple-response, and every multiple-response stem states how many options to select. The pass mark is a scaled 720 on a 100 to 1,000 scale; the fee is $125.
Is there a free CCAR-F mock exam?
Yes, with a caveat. The Cred Farmer set at /practice/ccar-f is 20 timed items written to the five-domain guide, no account needed, with a reason for every option after you submit. It is not built as scenarios and is not a replica of the exam, and it reports a raw count by domain, never a scaled score.
Where can I find CCAR-F sample questions?
The official exam guide publishes its own sample items with rationales, and they are the only official ones. Cred Farmer adds six open items on the certification page and the ten worked items here, all written to the same five domains and none drawn from the live bank.
Do the Cred Farmer CCAR-F timed forms match this exam?
In length, yes: 60 items in 120 minutes, built to the five official domains at the published weights. In structure, they follow the official layout: each presents four of the six published scenarios and groups its items under them. They are Cred Farmer items with difficulty not calibrated against the exam, and no result on them predicts a pass.
Keep reading
Related posts
Not affiliated with, or endorsed by, Anthropic or Pearson VUE. Details are summarised from publicly published program information and can change. Always confirm against the official exam guide before booking.