Where does turning a business problem into a Claude design start?

State the outcome in the business's own measure, such as hours saved per claim or minutes to first reply, then work backwards to the narrowest task Claude must perform. Starting from the model's capabilities instead produces a demo with no owner.

What are the four stages of an end-to-end Claude architecture?

Input, where data is gathered, cleaned and bounded; processing, where the model and its tools do the work; output, where results are validated and delivered; and a feedback loop that returns corrections and outcomes to improve prompts, retrieval and evaluation sets.

Workflow, agent or augmented LLM: which pattern fits a fixed sequence of steps?

A workflow. When the steps are known before the request arrives, code decides the order and Claude supplies the judgement inside each step. Reserve an agent for work whose path only becomes clear as it unfolds.

Define the augmented LLM building block

A single model call given retrieval, tools and memory so it can look things up, act and carry state within one turn. It is the unit that workflows chain and agents loop; many production systems need nothing more elaborate.

What does an orchestrator send to and receive from each subagent?

It sends a narrow brief: the subtask, the tools permitted and the shape of the answer wanted. It receives a compact structured result, never the subagent's whole conversation. Passing full transcripts between agents multiplies tokens and spreads errors.

Name two decomposition techniques for a task too large for one prompt

Prompt chaining, where each step's output feeds the next with a check in between, and parallel fan-out, where independent parts run at once and a final step merges them. Chain dependent stages; fan out independent ones.

Which business value pillars should a Claude solution's success metric map to?

Efficiency, transformation, productivity, cost and performance service levels. Pick the pillar the sponsor funded and measure that one, in the sponsor's units. A build that saves tokens but misses the promised turnaround has optimised the wrong pillar.

How should an architect choose between Claude model tiers for a production task?

Run the candidate tiers against a representative evaluation set, then take the cheapest and fastest one that clears the quality bar inside the latency and cost budget. Defaulting to the most capable model spends money the task cannot justify.

What belongs in a production system prompt, and what should stay out of it?

In: the role, the task boundary, the output format, rules for uncertainty and refusal, and an instruction to treat retrieved or user-supplied text as data. Out: reference documents that belong in retrieval, and secrets, because a prompt can be coaxed back out.

When does few-shot prompting beat a zero-shot instruction?

When the wanted output follows a house convention that is easier to show than describe: a labelling scheme, a tone, an edge-case ruling. Zero-shot is enough for tasks the model already does well; add examples when instructions alone leave the format inconsistent.

What are the levers for keeping a long agent session inside the context window?

Compact each tool result once it has served its purpose, summarise earlier turns, retrieve the passage that matters rather than pasting the whole document, and keep only the working set in context. A larger window postpones the problem and raises cost.

Prompt caching, modular prompts, Skills: which reuse strategy serves which need?

Caching cuts cost and time to first token for a byte-identical prefix repeated across requests. Modular prompts keep shared blocks in one versioned place so a fix lands everywhere. Skills package instructions and resources that Claude loads only when a task calls for them.

What is capability bloat in an agent's tool configuration?

Giving an agent more tools, or broader tools, than its job requires. Every extra tool is attack surface, extra tokens on every request and another way to pick the wrong action. The fix is removal and narrowing, not more instructions about when to use them.

Where should authorisation be enforced when Claude calls a tool on a user's behalf?

In the tool or the system of record, using the calling user's verified identity carried through the request, never on the model's say-so. A shared service account with broad rights lets any user reach any record the agent can.

How do you justify an accuracy-latency configuration decision to a reviewer?

Name the constraint that binds, such as a response-time budget or an error rate the business accepts, show the measured figures for each candidate configuration, and state which one meets the constraint at the lowest cost. Preference and benchmark rank are not justification.

Which signals does an observability design need to explain a quality drop at scale?

Structured traces per request carrying model, prompt version, retrieval identifiers, token counts, latency and tool calls, plus sampled content and a scored quality signal. Latency and error dashboards alone cannot tell you the answers got worse.

Which chunking strategy suits a RAG pipeline over structured policy documents?

Chunk along the document's own structure, such as sections and clauses, keep enough heading context that a chunk stands alone, and store metadata like version, effective date and owner so retrieval can filter before it ranks.

When should retrieval use lexical matching rather than embedding similarity?

When queries carry exact identifiers: part numbers, error codes, case references, names. Embeddings find meaning, not strings, and will happily return a similar-looking record. Hybrid search runs both and merges the results when the corpus mixes prose and identifiers.

MCP, direct API call or agent-to-agent: how do you choose the integration mechanism?

MCP when many agents and clients share one tool surface under a single point of governance. A direct API or CLI call when one system integrates once with the fewest moving parts. Agent-to-agent when the other side is itself autonomous across a trust boundary.

What is progressive discovery, and when does it beat loading everything up front?

Exposing tool definitions, documents or Skills in stages, a short index first and details fetched on demand. It wins when the full catalogue would crowd the window or distract the model; a monolithic context is fine when the set is small and always needed.

Name the five metric families a Claude evaluation should define before launch

Accuracy against a labelled reference, latency at the tail percentile the user feels, cost per completed task rather than per token, safety such as harmful or off-policy output rates, and security such as injection and data-leak attempts caught. Give each a threshold and an owner.

What does a mixed-methodology evaluation framework combine?

Deterministic checks for anything with a right answer, such as schema validity and rule compliance; a model-graded rubric for scaled judgement of tone or completeness; and a sampled human review that calibrates the grader. Each method covers a weakness of the others.

What makes an A/B test of a prompt change trustworthy?

Random assignment of concurrent traffic to both variants, a metric and sample size fixed before the test starts, and no other change shipping during the window. Comparing this week to last week attributes the season to the prompt.

Prompt failure, hallucination or model mismatch: how do you tell them apart?

Prompt failure shows as wrong format or ignored instructions and improves when the prompt is tightened. Hallucination shows as confident unsupported claims and points at missing grounding. Model mismatch shows as consistently shallow reasoning that a stronger tier fixes on the same prompt.

Which levers cut cost on a Claude workload without touching quality?

Cache the stable prefix, route easy cases to a lighter tier, trim the context to what each call needs, cap output length, and send work that can wait through the Batch API at its discounted rate. Score the evaluation set again after every lever.

Which logging fields let you reconstruct a bad Claude response weeks later?

The request identifier, model and prompt version, the retrieved chunk identifiers, tool calls with their arguments and results, token counts, latency by stage, and the user's follow-up or correction. Without versions you cannot tell whether the prompt or the data changed.

Where do guardrails sit in a Claude system that can take irreversible actions?

Before the model, screening and separating untrusted input; around the tool call, checking authorisation and arguments against policy in code; and after the output, validating schema and content. The model's own judgement is one layer, never the only one.

What separates a preventive control from a detective control in a Claude deployment?

A preventive control stops the action: a tool removed, a schema that rejects malformed output, an authorisation check before a write. A detective control finds it afterwards: logging, sampling, alerts. Detective controls are cheaper to add but never shrink the blast radius.

Name the failure modes specific to an agent that reads documents from outside the organisation

Prompt injection carried inside the document, exfiltration through a tool the injected text steers, confident summaries of forged or stale content, and loops that burn budget. Treat every external document as untrusted input to a system with hands.

How should human-in-the-loop review be allocated across an agent's actions?

By stakes: send a person the actions that are hard to reverse, expensive when wrong or regulated, and let low-stakes reversible ones through with sampling. Reviewing everything trains reviewers to rubber-stamp; reviewing by model confidence trusts the thing being checked.

What does GDPR, HIPAA or FedRAMP compliance change in a Claude architecture?

Where data may be processed and logged, how long it is kept, who can read traces, and which contractual and hosting arrangements are required. The compliance boundary follows the data, so prompts, retrieved content and logs all sit inside it, not only the database.

How does an architect address bias and fairness in a Claude-based decision support system?

Measure outcomes across affected groups on the evaluation set, not by asking the model whether it is fair; document what the system does and does not decide; tell affected people when generated content shaped a decision; and keep a route to contest it.

What must structured discovery establish before any Claude design work begins?

The decision or task in scope, who owns it today, the data and systems involved and who may access them, the regulatory and authority limits, the expected volume and latency, and how success will be measured. Prototype after these, not before.

How should an architectural trade-off be presented to a non-technical sponsor?

As the constraint behind the decision, the benefit the chosen option buys, the capability sacrificed to get it, and the price of changing course later. A benchmark table or a demo can support that argument but is not the argument.

How do you turn a stakeholder's demand for a perfect answer into an SLA term?

Convert it into a threshold on one named metric, scored against an evaluation set the stakeholder has seen, with a review cadence and an agreed action when the number is missed. A commitment nobody can score becomes a dispute later.

What must an architecture handoff document contain for the team that inherits it?

The decisions and the reasons behind them, the constraints still in force, the evaluation set and how to run it, a runbook for common failures, and who owns each component. A diagram alone leaves the next team guessing which choices are safe to change.

Which lifecycle phase begins at launch, and what does it need?

Monitoring and iteration. It needs a named owner, quality and cost signals with thresholds, a way to feed corrections back into the evaluation set and prompts, and a cadence for reviewing drift as models, data and usage change. Launch is a handover, not an ending.

How should a team standardise its Claude Code configuration?

Keep the shared CLAUDE.md, permission rules and MCP server definitions in the repository so they are versioned and reviewed with the code, and let developers layer personal settings on top. Team rules enforced in a shared file beat rules sent by chat.

What keeps AI-assisted code changes at the team's existing quality bar?

The same gates as any other change: tests, review, linting and a stated intent, with Claude Code asked to run the checks itself before proposing the change. Faster generation raises the volume of review, so automate every check a machine can run.

Name the first three checks when a Claude integration starts failing intermittently

The status codes and error types on the failing calls, which separate throttling from an outage and from a malformed request; whether request size or model version changed; and the tool-call traces for those requests. Start from evidence, not a theory about the model.

No account needed

CCAR-P flashcards: 40 free cards across the seven exam domains

Forty recall cards for the Claude Certified Architect - Professional exam, one deck spread across its seven official domains in proportion to their weights. Flip a card at a time, shuffle the order, or copy the whole deck as tab-separated text for Anki or a spreadsheet. Every front and back is on this page.

40 cardsseven domainsNo accountNothing stored

How the deck is built

The CCAR-P blueprint has seven domains, each with its published objectives. Each card on this page was written from the official Exam Guide version 1.0 (effective July 2026) and the objective it names, then checked against the guide on the date at the foot of the page. The deck is not exam content: nothing is recalled from a sitting, and nothing is drawn from the signed-in study chapters or the gated question bank. The allocation follows the official weights, so the heaviest domain, Integration, gets the most cards and the lightest, Developer Productivity & Operational Enablement, the fewest.

Domain	Official weight	Cards
1. Solution Design & Architecture	17%	7
2. Claude Models, Prompting & Context Engineering	13%	5
3. Integration	19%	8
4. Evaluation, Testing & Optimization	16%	6
5. Governance, Safety & Risk Management	14%	6
6. Stakeholder Communication & Lifecycle Management	14%	5
7. Developer Productivity & Operational Enablement	7%	3

Study one card at a time

Read the front, decide your answer out loud or on paper, then flip. Shuffle once the order stops surprising you. The widget keeps your place only until you reload; nothing is saved.

Card 1 of 40

Tap the card to flip. Arrow keys move between cards.

The whole deck, by domain

All 40 cards, front beside back, grouped by official domain with the objective each one drills. Copy the deck to your clipboard as tab-separated text with a header row.

Domain 1: Solution Design & Architecture

17% of the exam, 7 cards

Where does turning a business problem into a Claude design start?: State the outcome in the business's own measure, such as hours saved per claim or minutes to first reply, then work backwards to the narrowest task Claude must perform. Starting from the model's capabilities instead produces a demo with no owner.; Objective O01: Translate business problems into Claude-based AI solutions
What are the four stages of an end-to-end Claude architecture?: Input, where data is gathered, cleaned and bounded; processing, where the model and its tools do the work; output, where results are validated and delivered; and a feedback loop that returns corrections and outcomes to improve prompts, retrieval and evaluation sets.; Objective O02: Design end-to-end architectures (input, processing, output, feedback loops)
Workflow, agent or augmented LLM: which pattern fits a fixed sequence of steps?: A workflow. When the steps are known before the request arrives, code decides the order and Claude supplies the judgement inside each step. Reserve an agent for work whose path only becomes clear as it unfolds.; Objective O03: Select appropriate architectural patterns (workflow, agentic, augmented LLM)
Define the augmented LLM building block: A single model call given retrieval, tools and memory so it can look things up, act and carry state within one turn. It is the unit that workflows chain and agents loop; many production systems need nothing more elaborate.; Objective O03: Select appropriate architectural patterns (workflow, agentic, augmented LLM)
What does an orchestrator send to and receive from each subagent?: It sends a narrow brief: the subtask, the tools permitted and the shape of the answer wanted. It receives a compact structured result, never the subagent's whole conversation. Passing full transcripts between agents multiplies tokens and spreads errors.; Objective O04: Design multi-agent systems and orchestration strategies
Name two decomposition techniques for a task too large for one prompt: Prompt chaining, where each step's output feeds the next with a check in between, and parallel fan-out, where independent parts run at once and a final step merges them. Chain dependent stages; fan out independent ones.; Objective O05: Apply decomposition techniques for complex problem solving
Which business value pillars should a Claude solution's success metric map to?: Efficiency, transformation, productivity, cost and performance service levels. Pick the pillar the sponsor funded and measure that one, in the sponsor's units. A build that saves tokens but misses the promised turnaround has optimised the wrong pillar.; Objective O06: Align solutions to business value pillars

Domain 2: Claude Models, Prompting & Context Engineering

13% of the exam, 5 cards

How should an architect choose between Claude model tiers for a production task?: Run the candidate tiers against a representative evaluation set, then take the cheapest and fastest one that clears the quality bar inside the latency and cost budget. Defaulting to the most capable model spends money the task cannot justify.; Objective O07: Select appropriate Claude models based on trade-offs
What belongs in a production system prompt, and what should stay out of it?: In: the role, the task boundary, the output format, rules for uncertainty and refusal, and an instruction to treat retrieved or user-supplied text as data. Out: reference documents that belong in retrieval, and secrets, because a prompt can be coaxed back out.; Objective O08: Design system prompts, templates, and guardrails
When does few-shot prompting beat a zero-shot instruction?: When the wanted output follows a house convention that is easier to show than describe: a labelling scheme, a tone, an edge-case ruling. Zero-shot is enough for tasks the model already does well; add examples when instructions alone leave the format inconsistent.; Objective O09: Apply prompt engineering techniques (zero-shot, few-shot, chain-of-thought)
What are the levers for keeping a long agent session inside the context window?: Compact each tool result once it has served its purpose, summarise earlier turns, retrieve the passage that matters rather than pasting the whole document, and keep only the working set in context. A larger window postpones the problem and raises cost.; Objective O10: Optimize context windows and manage token usage
Prompt caching, modular prompts, Skills: which reuse strategy serves which need?: Caching cuts cost and time to first token for a byte-identical prefix repeated across requests. Modular prompts keep shared blocks in one versioned place so a fix lands everywhere. Skills package instructions and resources that Claude loads only when a task calls for them.; Objective O11: Implement prompt reuse strategies (caching, modular prompts, Skills)

Domain 3: Integration

19% of the exam, 8 cards

What is capability bloat in an agent's tool configuration?: Giving an agent more tools, or broader tools, than its job requires. Every extra tool is attack surface, extra tokens on every request and another way to pick the wrong action. The fix is removal and narrowing, not more instructions about when to use them.; Objective O12: Evaluate tool/agent configuration for capability bloat
Where should authorisation be enforced when Claude calls a tool on a user's behalf?: In the tool or the system of record, using the calling user's verified identity carried through the request, never on the model's say-so. A shared service account with broad rights lets any user reach any record the agent can.; Objective O13: Analyze authentication and authorization requirements to identify security gaps
How do you justify an accuracy-latency configuration decision to a reviewer?: Name the constraint that binds, such as a response-time budget or an error rate the business accepts, show the measured figures for each candidate configuration, and state which one meets the constraint at the lowest cost. Preference and benchmark rank are not justification.; Objective O14: Evaluate accuracy-latency trade-offs and justify configuration decisions
Which signals does an observability design need to explain a quality drop at scale?: Structured traces per request carrying model, prompt version, retrieval identifiers, token counts, latency and tool calls, plus sampled content and a scored quality signal. Latency and error dashboards alone cannot tell you the answers got worse.; Objective O15: Analyze observability challenges and select monitoring strategies at scale
Which chunking strategy suits a RAG pipeline over structured policy documents?: Chunk along the document's own structure, such as sections and clauses, keep enough heading context that a chunk stands alone, and store metadata like version, effective date and owner so retrieval can filter before it ranks.; Objective O16: Design a RAG pipeline with appropriate chunking and indexing strategies
When should retrieval use lexical matching rather than embedding similarity?: When queries carry exact identifiers: part numbers, error codes, case references, names. Embeddings find meaning, not strings, and will happily return a similar-looking record. Hybrid search runs both and merges the results when the corpus mixes prose and identifiers.; Objective O17: Apply retrieval strategies matched to data shape and query pattern
MCP, direct API call or agent-to-agent: how do you choose the integration mechanism?: MCP when many agents and clients share one tool surface under a single point of governance. A direct API or CLI call when one system integrates once with the fewest moving parts. Agent-to-agent when the other side is itself autonomous across a trust boundary.; Objective O18: Evaluate connection protocols and select the appropriate integration mechanism
What is progressive discovery, and when does it beat loading everything up front?: Exposing tool definitions, documents or Skills in stages, a short index first and details fetched on demand. It wins when the full catalogue would crowd the window or distract the model; a monolithic context is fine when the set is small and always needed.; Objective O19: Evaluate progressive discovery vs. monolithic context strategy

Domain 4: Evaluation, Testing & Optimization

16% of the exam, 6 cards

Name the five metric families a Claude evaluation should define before launch: Accuracy against a labelled reference, latency at the tail percentile the user feels, cost per completed task rather than per token, safety such as harmful or off-policy output rates, and security such as injection and data-leak attempts caught. Give each a threshold and an owner.; Objective O20: Define evaluation metrics (accuracy, latency, cost, safety, security)
What does a mixed-methodology evaluation framework combine?: Deterministic checks for anything with a right answer, such as schema validity and rule compliance; a model-graded rubric for scaled judgement of tone or completeness; and a sampled human review that calibrates the grader. Each method covers a weakness of the others.; Objective O21: Design evaluation datasets and test frameworks using mixed methodologies
What makes an A/B test of a prompt change trustworthy?: Random assignment of concurrent traffic to both variants, a metric and sample size fixed before the test starts, and no other change shipping during the window. Comparing this week to last week attributes the season to the prompt.; Objective O22: Conduct A/B testing and iterative improvements
Prompt failure, hallucination or model mismatch: how do you tell them apart?: Prompt failure shows as wrong format or ignored instructions and improves when the prompt is tightened. Hallucination shows as confident unsupported claims and points at missing grounding. Model mismatch shows as consistently shallow reasoning that a stronger tier fixes on the same prompt.; Objective O23: Diagnose system issues (prompt failure, hallucinations, model mismatch)
Which levers cut cost on a Claude workload without touching quality?: Cache the stable prefix, route easy cases to a lighter tier, trim the context to what each call needs, cap output length, and send work that can wait through the Batch API at its discounted rate. Score the evaluation set again after every lever.; Objective O24: Optimize token usage, latency, and cost-performance trade-offs
Which logging fields let you reconstruct a bad Claude response weeks later?: The request identifier, model and prompt version, the retrieved chunk identifiers, tool calls with their arguments and results, token counts, latency by stage, and the user's follow-up or correction. Without versions you cannot tell whether the prompt or the data changed.; Objective O25: Monitor system performance using logging and observability tools

Domain 5: Governance, Safety & Risk Management

14% of the exam, 6 cards

Where do guardrails sit in a Claude system that can take irreversible actions?: Before the model, screening and separating untrusted input; around the tool call, checking authorisation and arguments against policy in code; and after the output, validating schema and content. The model's own judgement is one layer, never the only one.; Objective O26: Implement guardrails and safety controls
What separates a preventive control from a detective control in a Claude deployment?: A preventive control stops the action: a tool removed, a schema that rejects malformed output, an authorisation check before a write. A detective control finds it afterwards: logging, sampling, alerts. Detective controls are cheaper to add but never shrink the blast radius.; Objective O26: Implement guardrails and safety controls
Name the failure modes specific to an agent that reads documents from outside the organisation: Prompt injection carried inside the document, exfiltration through a tool the injected text steers, confident summaries of forged or stale content, and loops that burn budget. Treat every external document as untrusted input to a system with hands.; Objective O27: Identify risks, limitations, and failure modes of LLM systems
How should human-in-the-loop review be allocated across an agent's actions?: By stakes: send a person the actions that are hard to reverse, expensive when wrong or regulated, and let low-stakes reversible ones through with sampling. Reviewing everything trains reviewers to rubber-stamp; reviewing by model confidence trusts the thing being checked.; Objective O28: Apply human-in-the-loop validation strategies
What does GDPR, HIPAA or FedRAMP compliance change in a Claude architecture?: Where data may be processed and logged, how long it is kept, who can read traces, and which contractual and hosting arrangements are required. The compliance boundary follows the data, so prompts, retrieved content and logs all sit inside it, not only the database.; Objective O29: Ensure compliance with regulations (e.g., GDPR, HIPAA, FedRAMP)
How does an architect address bias and fairness in a Claude-based decision support system?: Measure outcomes across affected groups on the evaluation set, not by asking the model whether it is fair; document what the system does and does not decide; tell affected people when generated content shaped a decision; and keep a route to contest it.; Objective O30: Address ethical AI considerations (bias, fairness, transparency)

Domain 6: Stakeholder Communication & Lifecycle Management

14% of the exam, 5 cards

What must structured discovery establish before any Claude design work begins?: The decision or task in scope, who owns it today, the data and systems involved and who may access them, the regulatory and authority limits, the expected volume and latency, and how success will be measured. Prototype after these, not before.; Objective O31: Conduct structured discovery and requirement gathering
How should an architectural trade-off be presented to a non-technical sponsor?: As the constraint behind the decision, the benefit the chosen option buys, the capability sacrificed to get it, and the price of changing course later. A benchmark table or a demo can support that argument but is not the argument.; Objective O32: Communicate architectural decisions and trade-offs
How do you turn a stakeholder's demand for a perfect answer into an SLA term?: Convert it into a threshold on one named metric, scored against an evaluation set the stakeholder has seen, with a review cadence and an agreed action when the number is missed. A commitment nobody can score becomes a dispute later.; Objective O33: Manage stakeholder feedback loops and expectation alignment (including SLAs)
What must an architecture handoff document contain for the team that inherits it?: The decisions and the reasons behind them, the constraints still in force, the evaluation set and how to run it, a runbook for common failures, and who owns each component. A diagram alone leaves the next team guessing which choices are safe to change.; Objective O34: Document architectures and provide implementation guidance
Which lifecycle phase begins at launch, and what does it need?: Monitoring and iteration. It needs a named owner, quality and cost signals with thresholds, a way to feed corrections back into the evaluation set and prompts, and a cadence for reviewing drift as models, data and usage change. Launch is a handover, not an ending.; Objective O35: Support lifecycle phases (discovery, design, handoff, monitoring, iteration)

Domain 7: Developer Productivity & Operational Enablement

7% of the exam, 3 cards

How should a team standardise its Claude Code configuration?: Keep the shared CLAUDE.md, permission rules and MCP server definitions in the repository so they are versioned and reviewed with the code, and let developers layer personal settings on top. Team rules enforced in a shared file beat rules sent by chat.; Objective O36: Configure Claude tools and environments for teams (e.g., Claude Code)
What keeps AI-assisted code changes at the team's existing quality bar?: The same gates as any other change: tests, review, linting and a stated intent, with Claude Code asked to run the checks itself before proposing the change. Faster generation raises the volume of review, so automate every check a machine can run.; Objective O37: Improve developer workflows using AI-assisted tooling
Name the first three checks when a Claude integration starts failing intermittently: The status codes and error types on the failing calls, which separate throttling from an outage and from a malformed request; whether request size or model version changed; and the tool-call traces for those requests. Start from evidence, not a theory about the model.; Objective O38: Support debugging and operational issue resolution

How to use the cards in a study week

Run the whole deck once a day for a week, unshuffled on day one so the domains stay together, shuffled from day two so the domain tag stops giving the answer away. Say the back before you flip; a card you can only recognise, not produce, is not learned yet. Pull the cards you miss into their own list and read the matching domain section of the CCAR-P cheat sheet, which holds the decision rules the cards compress, then work through the ten CCAR-P practice questions with the reasoning written out. Then sit the timed CCAR-P mock: recall is the floor, and the exam tests the judgment built on top of it. The week-by-week study plan says where the cards fit in a longer run.

What a flashcard cannot tell you

The exam reports a scaled score with a 720 cut and a percent-correct by domain. Knowing every card here says nothing about either, because the items are scenarios that ask you to choose between two defensible options on a stated constraint. Flashcard reviews are not measured answers, so they never move a readiness score on Cred Farmer; only measured answers do.

Where to go next

The deck is an independent study aid written from the published exam guide. The CCAR-P exam is 63 items in 120 minutes for a $175 fee, and these cards are not its questions. Read the originality note on why no page here carries recalled exam content.

Cards learned? Test the judgment

The CCAR-P mock sits 63 scenario items at the official pace and gives you a raw score by domain with a reason for every option. No account, nothing stored.

Sit the CCAR-P mock, no account

Frequently asked

Quick answers to the follow-up questions.

Do I need an account to use the CCAR-P flashcards?

No. All 40 cards are on this page, every front and every back, with nothing held back and no limit on how often you come back. Everything on Cred Farmer is free. An account adds the five timed forms, saved progress and a readiness score. This deck is complete as it stands and nothing here unlocks later.

Is anything I do on this page stored?

No. Flip, shuffle and copy work in your browser without an email address or a cookie that follows you. Nothing you do on this page is sent anywhere or stored, which is also why your position in the deck resets when you reload.

Do the cards come from the live CCAR-P exam?

No. Every card was written for this page from the official CCAR-P Exam Guide version 1.0 and its published objectives. Nothing is recalled from a sitting and nothing is copied from another provider. Disclosing live exam content breaks the candidate agreement, and Cred Farmer does not use, solicit or accept it.

How many flashcards do I need for CCAR-P?

Fewer than you would think. The blueprint lists its objectives under seven weighted domains, and these 40 cards give each domain a share in proportion to its weight. A deck you can finish in twenty minutes and repeat daily beats a deck of several hundred that you review once.

Can I download or print the flashcards?

Use the copy button to put the whole deck on your clipboard as tab-separated text with a header row, then paste it into Anki, Quizlet, Google Sheets or Excel. The page also prints cleanly: the study widget is hidden and the full list, front beside back, is what comes out.

Are flashcards enough to pass CCAR-P?

On their own, no. The exam asks scenario questions that test judgment between two defensible options, and recall cards only prove you know the terms. Use the deck to fix the vocabulary and decision rules, then sit the timed CCAR-P mock, no account needed, to see how that knowledge holds up under the clock.

Checked against the official exam guides on 17 September 2026.

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.