--- title: "Ten CCAR-P Architecture Scenarios Explained | Cred Farmer" url: https://credfarmer.com/blog/ccar-p-practice-questions description: "Ten CCAR-P architecture scenarios explained against Exam Guide v1.0: original items across all seven domains, the key, and why every wrong option fails." last_updated: 2026-10-03 publisher: "Cred Farmer, an independent study platform, not affiliated with or endorsed by Anthropic or Pearson VUE" --- Claude Certification Blog # Ten CCAR-P architecture scenarios explained across the seven Professional domains Ten original CCAR-P architecture scenarios explained option by option, across all seven Professional domains and weighted like the blueprint. Each has its key and a reason every distractor fails, because a scenario earns its place only when every option is explained. 10 worked items · Single and select-two · 7 official domains Published 17 September 2026Updated 3 October 202618 min read By the Cred Farmer editorial team.Checked against the official sources on 3 October 2026. The ten CCAR-P architecture scenarios explained below are original Cred Farmer items written to the CCAR-P Exam Guide v1.0 (July 2026): a short architecture scenario with a measured symptom and a stated constraint, then four options, or five where an item asks you to select two. They cover all seven official domains in rough proportion to weight, every wrong option carries a reason it fails, and none is drawn from a live exam form or from the question bank behind the login. Answer each one cold, then score yourself on the miss map. The [official CCAR-P exam guide](https://anthropic-partners.skilljar.com/page/partner-certifications) publishes three CCAR-P sample questions of its own in Section 8 and says they are not drawn from the live item bank. The [Claude certification sample questions](https://credfarmer.com/blog/claude-certification-sample-questions) hub works one item per code; this post is its Architect Professional spoke. The blueprint and fee are on the [Claude Certified Architect - Professional](https://credfarmer.com/certifications/claude-certified-architect-professional) page, and the [guide to Claude certification practice exams](https://credfarmer.com/guides/claude-certification-practice-exams) covers what official practice material exists. The ten are spread by official weight, so integration, solution design and evaluation get two items each and developer productivity at 7 percent gets one. One of the ten asks for two answers. Read the stated constraint before the options. **Figure:** Seven CCAR-P domains by official weight, from 7 to 19 percent, with one or two items from this set in each - Official weight · items here - D1 Solution Design & Architecture · 17% · 2 items - D2 Models, Prompting & Context Engineering · 13% · 1 item - D3 Integration · 19% · 2 items - D4 Evaluation, Testing & Optimization · 16% · 2 items - D5 Governance, Safety & Risk Management · 14% · 1 item - D6 Stakeholder & Lifecycle Management · 14% · 1 item - D7 Developer Productivity & Enablement · 7% · 1 item Weights from the CCAR-P Exam Guide v1.0, Section 6; the item counts are this post's ten. ## How do you scope a solution to the sponsor's metric, and when do you split one agent? Domain 1, Solution Design & Architecture, is 17 percent of the blueprint, and two of the ten sit here. The first turns on aligning scope to a business value pillar inside a governance constraint: a named metric, a deadline, and a decision type that has not been approved for automation. The second is an orchestration decision about one agent showing three symptoms at once; the stem hands you the fact that decides whether routing can be deterministic. Question 1 · Written for this post · Domain 1 · single response A bank wants contact-center volume cut 30 percent this quarter. Card-status queries, 42 percent of calls, can be answered from an existing system. Autonomous dispute decisions are not approved. What should you scope? - **A.** An assistant that settles disputes on its own, the hardest ones first. - **B.** A general assistant covering every call type across the center from day one. - **C.** A card-status assistant using live lookups, measured on deflection. - **D.** A sentiment score on every call for supervisors. **Reveal the answer and the reason for every option** Answer: C. Scope to the metric inside the governance constraint. **A.** Autonomous dispute decisions are not approved, so this design cannot launch this quarter. **B.** Covering every call type at once is too broad to deliver in a quarter, and it includes disputes. **C.** Correct. It targets the largest call type, uses the existing system live and is measured on the sponsor's metric. **D.** Scoring calls gives supervisors information, but it does not reduce the number of calls. Aligning a solution to business value means choosing the scope that moves the sponsor's metric within its constraints. Card-status queries are 42 percent of calls and already have a system of record, so an assistant that answers them with a live lookup can reach the 30 percent target, measured on deflected calls. Disputes are not approved for autonomous decisions, a general assistant is too broad for a quarter, and sentiment scores move nothing. Question 2 · Written for this post · Domain 1 · single response One agent handles customs, damage and delay exceptions with 38 tools. Tool errors are 14 percent, context passes 90,000 tokens by turn twelve, and resolution lags. Each exception's type is known upfront. Which design fits? - **A.** The same agent, given a longer context limit. - **B.** Three agents that each hold all 38 tools and vote on answers. - **C.** An orchestrator routing each exception type to a subagent with only its tools. - **D.** One agent that summarizes its history every few turns. **Reveal the answer and the reason for every option** Answer: C. Route once, scope the specialist, return structure. **A.** More room does not reduce the 38-tool choice that drives the errors. **B.** Each voter still faces all 38 tools, and voting triples the cost. **C.** Correct. Routing on a known type gives each subagent a small tool set and a focused context, and a structured result keeps the orchestrator lean. **D.** Summaries slow context growth, but all 38 tools are still offered on every turn. Multi-agent orchestration pays off when subtasks are distinct. The exception type is known before work starts, so an orchestrator can route each case to a specialist subagent holding only that type's tools, which cuts selection errors, and receive a structured result, which keeps the context small; both shorten resolution. A larger window or summarization leaves 38 tools in play, and voting agents multiply cost without narrowing the tools. ## What do you cut when the context window has become the cost? Claude Models, Prompting & Context Engineering is 13 percent, and its context objective asks what in the window is doing work on this turn. The select-two below describes a session whose per-turn input has grown until cost has tripled, with a hard requirement that exact text can still be quoted. Two options each remove one named source of growth; the other three raise the ceiling, cut the work short or drop the text the answers must quote. Question 3 · Written for this post · Domain 2 · select two An underwriting assistant resends the full 60-page policy wording and the whole conversation on every turn of a thirty-turn session. Cost has tripled, and answers must still quote the exact clause. Which two changes fit? (Select two.) - **A.** Compact earlier turns into a running summary past a set size. - **B.** Move to a longer-context model and keep resending everything. - **C.** Cap sessions at five turns and have underwriters start fresh after that. - **D.** Retrieve only the relevant clauses from an indexed copy of the wording. - **E.** Drop the wording and answer from the model's general knowledge. **Reveal the answer and the reason for every option** Answer: A and D. Retrieve the clause, summarize the history. **A.** Correct. A running summary keeps what matters from earlier turns while bounding the history. **B.** A longer window holds the same tokens and keeps the tripled cost. **C.** Restarting every five turns loses the session's context and pushes the work onto underwriters. **D.** Correct. Indexed retrieval supplies the exact clause text each answer must quote, without the full wording every turn. **E.** General knowledge cannot quote the exact wording of this particular policy, which answers must cite. Managing the context window means sending what each turn needs. Retrieving only the relevant clauses from an indexed copy keeps exact wording available for quotation without resending 60 pages, and folding older turns into a running summary stops the history from growing without limit. A bigger window keeps the cost, general knowledge cannot quote this policy, and forced restarts lose the session's context. ## Where should authorisation and integration live when agents share systems? Integration is the heaviest domain at 19 percent. Item 4 is a security-gap read: a tool authenticates with one broad credential, and a model instruction is all that stands between a user and data they must not see. An authorisation decision never belongs inside a probabilistic step. Item 5 is the integration mechanism for several agents sharing many systems; the key also limits each agent to its own tools, which guards against capability bloat. Question 4 · Written for this post · Domain 3 · single response A search tool reads every matter through one service account, and the prompt says to use only permitted documents. A paralegal saw an excerpt from a matter she is barred from. What fixes this? - **A.** Strengthen the prompt so the model checks the user's matter list. - **B.** Run searches under the user's delegated token. - **C.** Run a classifier over answers to strip text from barred matters. - **D.** Have each user list the matters they may see at the start of every session. **Reveal the answer and the reason for every option** Answer: B. Let the system of record enforce access. **A.** A prompt cannot enforce access; the service account still returns barred documents. **B.** Correct. Delegated identity lets the document system enforce its existing walls before any text reaches the model. **C.** Filtering after retrieval relies on catching every leak, and the barred text was already fetched. **D.** A list typed by the user is a self-report, not an enforced permission. Authorization belongs in the system of record, not in the prompt. The tool should call the document system with a token delegated from the authenticated user, so the system's own permissions decide what a search can return before the model sees anything. A stronger prompt, an output classifier or a user-supplied matter list all leave the service account able to fetch documents the user may not see. Question 5 · Written for this post · Domain 3 · single response Four internal agents each have their own wrappers for the same nine plant systems: 36 integrations with three ways of handling credentials. The platform team wants one integration surface. Which approach fits best? - **A.** One merged agent with every tool for all nine systems. - **B.** Agent-to-agent calls so each agent can borrow the others' integrations. - **C.** One shared credential for every agent. - **D.** Shared MCP servers that register only each agent's own tools. **Reveal the answer and the reason for every option** Answer: D. A shared protocol surface with narrow registration. **A.** One agent with every tool brings capability bloat and loses the separation between teams. **B.** Borrowing through other agents adds hops and still leaves 36 integrations to maintain. **C.** One credential for every system widens each agent's access and removes per-agent scoping. **D.** Correct. Shared servers remove the duplicate integrations and handle credentials once, and narrow registration keeps each agent scoped. When several clients need the same systems, an MCP server per system gives one governed surface: each integration is built once, authentication is handled in the server, and every agent connects to it. Registering only the tools each agent needs keeps least privilege. Merging the agents creates a bloated tool set, agent-to-agent calls add hops, and a shared credential widens access. ## How do you prove a change worked, and cut the bill without touching quality? Evaluation, Testing & Optimization is 16 percent. Item 6 is a causal question with a confound: something else changed the day after the prompt did, and a decision is due by Friday. Only one design isolates the prompt from everything else that week. Item 7 is cost optimisation with quality held fixed: the summaries must not change. The lever for a high-volume, latency-tolerant job is not the lever for a long stable prefix. Question 6 · Written for this post · Domain 4 · single response A team moved all traffic to a new recommendation prompt on Monday; conversion is 2.1 points above last week. A promotion began Tuesday, and a decision is due Friday. How should the team evaluate it? - **A.** Run both prompts at once on randomly assigned sessions. - **B.** Credit the new prompt with the full 2.1-point gain and keep it. - **C.** Compare with the same week last year to remove seasonality. - **D.** Subtract the promotion's usual lift and credit the rest to the prompt. **Reveal the answer and the reason for every option** Answer: A. Randomize concurrently to isolate the cause. **A.** Correct. Concurrent random assignment exposes both prompts to the same promotion, isolating the prompt's effect. **B.** The promotion began during the same week, so the gain cannot be credited to the prompt. **C.** Last year's week had different traffic and promotions, so the comparison is still confounded. **D.** A typical lift is an estimate; the result depends on an assumption, not on a test. An A/B test isolates a change by running variants at the same time on randomly assigned traffic. The promotion started a day after the switch, so a before-and-after comparison mixes the two effects. Running both prompts concurrently, with conversion as the declared metric and a sample size set in advance, separates the prompt's effect. Year-over-year or adjusted comparisons still rest on assumptions. Question 7 · Written for this post · Domain 4 · single response A university summarizes about 3,500 lecture transcripts every evening for its accessibility service. Nobody reads them until the next afternoon, the bill is over budget, and summary quality must not change. What fits best? - **A.** Move the job to the fastest, cheapest model tier. - **B.** Shorten each summary to save output tokens. - **C.** Cache the short shared system prompt. - **D.** Run the unchanged job through the Message Batches API. **Reveal the answer and the reason for every option** Answer: D. Latency-tolerant volume is what batch is for. **A.** A different model can change summary quality, and the stem says quality must not change. **B.** Shorter summaries change the output that the accessibility service relies on, which is ruled out. **C.** Only the short system prompt repeats; each transcript differs, so caching saves very little. **D.** Correct. Batch pricing lowers cost for work nobody waits on, and keeping the model and prompt holds quality steady. Cost-performance trade-offs start with what binds. Nobody waits for these summaries, so latency does not matter, and asynchronous batch processing, priced below synchronous calls, cuts cost with the same model and prompt, keeping quality unchanged. A cheaper tier or shorter summaries change quality, and caching a short shared prompt saves little when each transcript is different. ## Which controls stop an injected instruction from moving money? Governance, Safety & Risk Management is 14 percent. The item below reconstructs an incident: an instruction arrived inside a customer message, and an irreversible tool executed before any check ran. The key puts the check where the refund happens. The distractors are the controls teams reach for first, and each is downstream of the harm or a probability rather than a control anyone can audit. Question 8 · Written for this post · Domain 5 · single response A fare-card agent can issue refunds. Text hidden in a customer message made it refund a card named in that text, and the output filter ran only after the refund. What control fits best? - **A.** A stronger output filter that blocks replies confirming a refund to another card. - **B.** A prompt rule telling the agent to ignore embedded text. - **C.** A tool-side check that the card belongs to the signed-in account. - **D.** A daily refund report for finance to review. **Reveal the answer and the reason for every option** Answer: C. Authorize before the call, not after the reply. **A.** An output filter only sees the reply, and by then the refund has already gone through. **B.** The attack was injected instructions; another instruction is not a reliable defense. **C.** Correct. The refund tool refuses cards outside the authenticated account, denying by default, so injected text cannot move money. **D.** A report finds the bad refund a day later instead of stopping it. A guardrail must act before an irreversible action, not after. A deterministic check inside the refund tool that the destination card belongs to the authenticated account, denying by default and logging denials, stops the refund whatever the message says. An output filter sees the reply after the money has moved, a prompt rule can be overridden by the same injected text, and a daily report only finds the loss. ## What do you leave behind at handoff, and how do you keep a whole team inside the rules? Stakeholder Communication & Lifecycle Management is 14 percent and Developer Productivity & Operational Enablement is 7, the smallest domain, and they share a theme: enabling the people who inherit or operate the system. Item 9 asks what a handoff is missing when nobody on the team can say why a design choice was made. Item 10 asks how a team's Claude Code rules should be enforced once individual configuration has drifted. In both, the key outlasts any one person: a written record of why, and rules the tooling applies to everyone. Question 9 · Written for this post · Domain 6 · single response An architect leaves a prior-authorization project in two days. The team gets the evaluation harness, a diagram and a dashboard, but nobody knows why the pipeline uses a legacy pricing tool. What is missing? - **A.** A second, more detailed diagram of every service the pipeline calls. - **B.** A decision log of each choice and its assumptions. - **C.** A recorded walkthrough of the dashboard and its alerts. - **D.** Another week of the departing architect's time for questions. **Reveal the answer and the reason for every option** Answer: B. A handoff must record why, not only what. **A.** A diagram shows that the tool is called, not why it was chosen. **B.** Correct. A log of each choice, the alternatives rejected, its assumptions and its owner preserves the reasoning behind the pricing tool. **C.** A dashboard walkthrough covers day-to-day operations, not the reasons behind design choices. **D.** Answers given in person leave nothing behind once the architect is gone. Implementation guidance has to record why, not only what. A decision log that lists each significant choice, the alternatives rejected, the assumptions it depends on and who owns revisiting it would answer the pricing-tool question and protect the design after handoff. More diagrams and walkthroughs show what exists, and extra time from the departing architect ends when they leave. Question 10 · Written for this post · Domain 7 · single response Sixty engineers configured Claude Code individually, and one session ran a migration that dropped a staging table. A survey found eleven permission setups. Leadership wants the speed gain kept. Which configuration fits best? - **A.** Committed project rules plus managed policy for destructive commands. - **B.** A wiki page of safe commands that each engineer copies into personal settings. - **C.** Turning off saved permissions, with manual approval for every command. - **D.** A ban on Claude Code in repositories with database access. **Reveal the answer and the reason for every option** Answer: A. Enforce team rules in the configuration hierarchy. **A.** Correct. A committed project settings file gives shared rules, and managed policy enforces the critical denials where nobody can override them. **B.** Copying a list by hand recreates the eleven different setups the survey found. **C.** Approving every command slows the team and trains people to click through prompts. **D.** A ban removes the speed gain that leadership has explicitly asked the team to keep. Team configuration for Claude Code uses the settings hierarchy. A project settings file committed to each repository gives everyone the same allow and deny rules and hooks, and managed policy settings, which personal settings cannot override, block destructive database commands. A wiki depends on copying, manual approval of every command slows the team and breeds careless clicks, and banning the tool loses the gain. ## Which of the CCAR-P architecture scenarios explained above caught you? A miss map by domain Score the ten, then count misses by domain rather than in total. Ten CCAR-P questions cannot produce a percentage that means anything against a scaled cut of 720, and the score report gives percent-correct by domain, so that is the unit to practise reading. Each row names the items here, the objective to reread if you missed one, and the free mock's untimed filter for that domain. | Domain | Items here | Your misses | Objective to reread | Untimed drill | | --- | --- | --- | --- | --- | | D1 Solution Design & Architecture (17%) | 1 and 2 | __ of 2 | O06: scope aligned to a business value pillar; O04: multi-agent orchestration design | [Domain 1 untimed](https://credfarmer.com/practice/ccar-p?domain=1) | | D2 Claude Models, Prompting & Context Engineering (13%) | 3 | __ of 1 | O10: context window and token optimisation | [Domain 2 untimed](https://credfarmer.com/practice/ccar-p?domain=2) | | D3 Integration (19%) | 4 and 5 | __ of 2 | O13: authentication and authorisation gaps; O18: choosing the integration mechanism | [Domain 3 untimed](https://credfarmer.com/practice/ccar-p?domain=3) | | D4 Evaluation, Testing & Optimization (16%) | 6 and 7 | __ of 2 | O22: A/B testing a change; O24: token, latency and cost trade-offs | [Domain 4 untimed](https://credfarmer.com/practice/ccar-p?domain=4) | | D5 Governance, Safety & Risk Management (14%) | 8 | __ of 1 | O26: guardrails and safety controls | [Domain 5 untimed](https://credfarmer.com/practice/ccar-p?domain=5) | | D6 Stakeholder Communication & Lifecycle Management (14%) | 9 | __ of 1 | O34: documentation and implementation guidance | [Domain 6 untimed](https://credfarmer.com/practice/ccar-p?domain=6) | | D7 Developer Productivity & Operational Enablement (7%) | 10 | __ of 1 | O36: configuring Claude tools for a team | [Domain 7 untimed](https://credfarmer.com/practice/ccar-p?domain=7) | **Figure:** Count misses by domain: none means move on, one means reread the objective, two means run the domain filter untimed - Score the ten items - then count misses by domain - 0 misses - Move on - The full mock rechecks it - 1 miss - Reread the objective - Then redo the item cold - 2 misses - Run the domain filter - Untimed, in the free mock A Cred Farmer reading of a ten-item score, not an official rule. Domains with one item here cap at one miss, so treat a miss there as a reread, not a verdict. Two misses in the same domain is the pattern worth acting on: run that domain's filter untimed, then sit the full timed form. The [two-week CCAR-P study plan](https://credfarmer.com/blog/two-week-study-plan-ccar-p) schedules the rereads, the [CCAR-P cheat sheet](https://credfarmer.com/blog/ccar-p-cheat-sheet) compresses each domain to its key points, and [why exam dumps are worth so little](https://credfarmer.com/blog/claude-certification-exam-dumps) explains what a memorised answer letter costs. ### Ten items are a sample, not a score Two misses in the same domain is a pattern; one miss anywhere is noise until the full 63-item form confirms it. Nothing on this page is a scaled score, a pass verdict or a readiness figure, and it does not predict an official result. Question bank dated 2 October 2026. Checked against the CCAR-P Exam Guide v1.0 (July 2026) on 17 September 2026. Cred Farmer is not affiliated with Anthropic; the [CCAR-P](https://credfarmer.com/certifications/claude-certified-architect-professional) page records what the official sources say and when they were last read. ## Key takeaways - **Read the constraint before the options.** Every item here turns on a stated fact: an unapproved decision type, a quality that must not change, a credential's scope. - **Count misses by domain, not in total.** Ten items give no percentage worth quoting against a scaled cut of 720. - **The bigger knob is usually the trap.** A larger window, a stronger prompt or a stricter filter each treats a symptom the scenario has already explained. - **Deterministic decisions stay out of the model.** Authorisation, pre-execution checks and team-wide rules hold only when a system enforces them. - **Practice evidence is not a pass probability.** The band of a Cred Farmer readiness score comes only from first sittings of the aligned forms, and it does not predict an official result. ## Sit the whole 63-item form under the clock The free CCAR-P mock is the full 63 items in 120 minutes, no account, with a reason for every option and a raw count by domain beside the official weight. Signed in, five blueprint-aligned timed forms add 315 more items. The first sitting of each one decides your Cred Farmer readiness band, and that band does not predict an official result. [Start the free CCAR-P practice exam](https://credfarmer.com/practice/ccar-p) Already scored the mock? [Sign in for the five timed forms](https://credfarmer.com/login?next=/c/ccar-p/practice). Questions ## Frequently asked Short answers to the questions this post leaves open. **Are these real CCAR-P exam questions?** No. All ten were written fresh for this post from the CCAR-P Exam Guide v1.0 and its published objectives; none comes from a live exam form or from the gated Cred Farmer bank. The guide's own three sample items are illustrative too, and it says they are not drawn from the live item bank. **How many questions are on the CCAR-P exam?** The CCAR-P has 63 items in 120 minutes, multiple-choice and multiple-response, and each item states how many responses to select. The result is a pass or fail with a scaled score from 100 to 1,000; the cut is 720. The fee is $175 and the credential is valid for 12 months. **Is there a free CCAR-P mock exam?** Yes. The free Cred Farmer mock at /practice/ccar-p runs the full 63-item form under a 120-minute clock with no account, then shows a raw count by domain and a reason for every option. It never reports a scaled score or a pass verdict, because no practice percentage converts to the 720 cut. **How should I score myself on these ten CCAR-P questions?** Count misses by domain, not in total, and treat a select-two item as a miss unless both picks match the key. One miss in a domain means reread the objective in the miss map and redo the item cold a day later; two means run that domain's untimed filter in the free mock first. **What score do I need on practice questions to pass the CCAR-P?** No practice percentage converts to the exam's scaled cut of 720. Cred Farmer measures with its own readiness score instead. Signed in, Exam-ready (90) needs your first sittings of the 63-item forms to average 800 or more on our practice scale, which is 49 correct (78%). The free items on this page never feed it, and the scale has never been checked against real exam results. It is a signal, not a promise. Keep reading ## Related posts Article ### CCAO-F worked scenarios: ten practice items across the seven Associate domains Ten CCAO-F worked scenarios written to Exam Guide v1.0: original business items across all seven domains, the key, and why every wrong option fails. Read the post → Article ### CCAR-F wrong answers explained: ten questions, worked option by option CCAR-F wrong answers explained on ten original questions written to Exam Guide v1.0: scenario stems in all five domains, the key, and a miss map by domain. Read the post → Guide ### Claude certification sample questions, explained Six original Claude certification sample questions with worked explanations, the official format, clearly labelled mock data, and a practical reading method. Read the post → 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.