When should you build a workflow rather than an agent?

When the sequence of steps is fixed before the run starts and rarely changes. Code calls Claude at set points, so the path is predictable, testable and cheaper. An agent is justified only when the model must choose the next step from what it observes.

What is the supervisor's job in a manager and worker agent hierarchy?

It breaks the goal into bounded subtasks, hands each to a worker with only the context that worker needs, then checks and merges what comes back. Workers never see the whole conversation, so the supervisor owns coherence and the final answer.

What does the Claude Agent SDK provide that a plain Messages API call does not?

The harness behind Claude Code as a library: the agent loop, tools for files and the shell, permission handling, hooks, subagents, MCP connections and automatic context compaction. You supply the task, the allowed tools and the guardrails instead of writing the loop yourself.

Why does a hook enforce a rule more reliably than a system prompt line?

A hook is your own code, run at a fixed lifecycle event such as before a tool call, and it can block or alter the action whatever the model intended. Prose depends on the model noticing and obeying; a hook does not.

Name the steps of a minimal hand-written tool-use loop

Call the API with your tool definitions. If stop_reason is tool_use, run every requested tool, append one tool_result per tool_use id in a single user message, and call again. When stop_reason is end_turn, return the text. A missing tool_result is an API error.

What do frameworks such as LangGraph, Strands or PydanticAI add over a hand-rolled agent loop?

Reusable orchestration: graph or state-machine control flow, typed tool and output schemas, retries, memory stores and tracing. The price is a dependency and an abstraction layer between you and the API, so adopt one for multi-step orchestration, not for a single tool loop.

Which business requirements become infrastructure requirements for a Claude application?

Expected volume and peak concurrency (rate limits, batching), how long users will wait (streaming or batch), where data may travel and how long it is kept (region, provider, logging), and uptime targets. Functional requirements describe the answer; infrastructure requirements describe how it is delivered.

How does a Claude feature move through the systems life cycle?

Requirements with measurable acceptance criteria, a design that treats model, prompt and tools as versioned components, an eval suite before release, a staged rollout with monitoring and a rollback path, then maintenance that tracks model deprecation dates as scheduled changes rather than surprises.

What does stop_reason tell you about a Messages API response?

Why generation ended: end_turn for a natural finish, tool_use when Claude wants you to run a tool, max_tokens when the output limit truncated it, stop_sequence when one of your stop strings appeared. Always check for max_tokens before trusting an answer as complete.

What arrives when you stream a Messages API response?

Server-sent events in order: message_start, then for each content block a start, a series of deltas and a stop, then message_delta carrying the stop reason and output usage, then message_stop. Tool call arguments arrive as partial JSON fragments you must concatenate before parsing.

Which workloads belong on the Message Batches API instead of realtime calls?

Large sets of independent requests where nobody is waiting: overnight classification, backfills, bulk evals. Requests run asynchronously and complete within a window of up to a day at a discount to realtime pricing; you poll the batch and fetch results when it finishes.

What should a code review of Claude-generated changes concentrate on?

Exactly what a review of any contribution checks: does it meet the requirement, is it covered by tests, does it follow the codebase's conventions, and does it touch anything it was not asked to. The model's own summary of the change is not evidence.

Why call the Claude SDK asynchronously inside a web service?

Each call waits on the network for seconds, so a blocking call ties up the worker for its whole duration. An async client lets the process serve other requests, or fan out independent calls together, while waiting. Cap concurrency to stay inside rate limits.

Why should prompts, schemas and tool definitions live in version control?

They change behaviour as much as code does. Version control gives every change a diff, a reviewer, a rollback and a timestamp you can line up against eval scores and incidents. A prompt edited in a dashboard has none of those.

Why does an instruction behave differently in claude.ai, the Desktop app, Claude Code and the API?

The products wrap your words in their own system prompts, tools, memory and formatting rules, so behaviour observed there does not carry over. Through the API you get the bare model: only the system prompt, messages and tools you send shape the response.

How do you mark pasted text as data rather than instructions?

Wrap it in a clearly named boundary, such as XML tags, place it in the user turn, and state in your instructions that whatever sits between those tags is material to process, never a command. Structure is what the model can rely on.

What does managing plugins in a Claude application involve?

Treating each plugin as a dependency: record its version, review the commands, agents, hooks, Skills and MCP servers it bundles, note the permissions they need, update deliberately and remove it when unused. One plugin can change a lot of behaviour at once.

What does pinning a dated model ID protect production from?

Silent behaviour change. An alias that names a tier moves to the newest snapshot when one ships; a dated ID stays fixed until it is deprecated. Upgrades then happen when your evals pass and you choose, not when the alias moves under you.

What belongs in settings.json and what belongs in CLAUDE.md?

settings.json is machine-read configuration: permission allow and deny rules, hooks, environment variables, the default model. CLAUDE.md is guidance written for the model: build commands, conventions, architecture notes, things to avoid. Put enforcement in settings and guidance in CLAUDE.md, never the reverse.

Which CLAUDE.md files does Claude Code load, and in what order of scope?

From broadest to narrowest: an enterprise managed policy file, the user's own file in their home directory, the project root file, an untracked local variant for personal notes, and subdirectory files read when Claude works inside them. Narrower files refine broader ones.

How do you tell an integration bug from a model output problem?

Read the trace: request sent, raw response, what your code did next. A well-formed response your code mishandled is an integration bug. A wrong or malformed response to a correct request is a model problem, fixed in the prompt, schema or model choice.

Why can one prompt produce different text on two identical requests?

Each token is sampled from a probability distribution that temperature and top_p reshape, so the path can diverge at any step. Even at temperature zero the platform does not promise identical output, so store the response you acted on rather than regenerating it.

What do extended thinking, adaptive thinking and effort levels each control?

How much the model reasons before replying. Extended thinking reserves a token budget for that reasoning, adaptive thinking lets the model decide per request how much it needs, and an effort setting trades depth for speed and cost. Thinking tokens are billed as output.

What does an official client SDK do on top of the REST API?

Builds authenticated requests, maps JSON to typed objects, raises typed errors per status code, retries transient failures with backoff, and parses streaming events into an iterator. You still own timeouts, concurrency limits, idempotency and what happens once the retries run out.

When does a websocket belong between end users and a Claude integration?

When a browser client needs tokens as they are generated and the API key must stay on the server. Your backend calls Claude with streaming on and forwards each event through the websocket (or server-sent events); the browser never reaches the API directly.

How do you choose between Opus, Sonnet and Haiku for a task?

Start from the task's demands. Haiku for high-volume, latency-sensitive or simple classification work; Sonnet as the default for most coding and agentic tasks; Opus where the hardest reasoning justifies the price. Then run the eval and keep the cheapest tier that clears it.

What counts as a breaking behaviour change between model releases?

A newer model handling the same prompt differently: stricter or looser refusals, changed default verbosity or formatting, a parameter that is deprecated or newly required, different tool-calling habits. Treat a model swap as a dependency upgrade and run the eval suite before promoting it.

How is prompt caching billed, and how do you confirm it is working?

Cache writes cost a premium over input tokens and cache reads cost a fraction, so the saving exists only when a prefix is reused within its lifetime. Confirm hits through cache_read_input_tokens in the usage object; zero means the prefix changed or the cache expired.

What is context drift, and what stops it?

The model's behaviour sliding away from the original instructions as a long session fills with tool output, dead ends and superseded decisions. Prune tool results you no longer need, compact the history into a summary, and restate the goal and constraints after every compaction.

What is context isolation, and when does it pay off?

Running a side task in a subagent or a separate step with its own context window, so the parent receives a short result and never the raw material. It pays off for searches, large file reads and anything that would flood the main window.

What goes in the system prompt, and what goes in the user turn?

System: the role, standing rules, output format and what to do when information is missing. User turn: the task and its data for this request. Per-request data placed in the system prompt breaks caching and blurs which words are instructions and which are content.

What is input sanitisation in prompt engineering?

Bounding text from users or outside systems before it enters the prompt: truncating to a length budget, stripping control characters, escaping or wrapping it in tags and labelling it as data. It prevents accidental breakage; on its own it does not defeat prompt injection.

What does defensive parsing of Claude's output mean?

Treating every response as untrusted input to your code: extract the expected structure, validate it against the schema, handle missing or extra fields, and fail closed when validation fails. Prose requests for JSON guarantee nothing; even schema-enforced output still needs your business rules checked.

Name the two halves of a prompt injection defence

Separation and least privilege. Keep untrusted content (documents, web pages, tool results) structurally apart from trusted instructions and labelled as data, then limit what any injected instruction could achieve by giving the agent only the tools and permissions the task needs.

What does guardrail layering mean in a Claude deployment?

No single control is trusted alone. Input screening, a system prompt with explicit refusals, tool permissions cut to least privilege, output validation and a human approval step for consequential actions are stacked so that when one layer fails another still catches the problem.

Where should an API key live, and where must it never appear?

In a secrets manager or server-side environment variable, one key per environment, rotated on a schedule and monitored for unexpected use. Never in source control, browser code, prompts, logs or error messages. If one leaks, revoke it in the Claude Console before investigating.

What makes a tool description effective?

It states what the tool does, the cases it fits and the cases it does not, what every parameter means with format and an example, and what comes back. Claude selects tools from their descriptions, so vagueness is the usual cause of wrong calls.

Client-side tools versus server-side tools: what is the difference?

A client-side tool runs in your code: Claude returns a tool_use block, you execute it and send back a tool_result. A server-side tool, such as web search, runs on Anthropic's side within the same request, with no round trip through your code.

Which three primitives can an MCP server expose, and over which transports?

Tools (functions the model may call), resources (data the client can read into context, addressed by URI) and prompts (reusable templates a user invokes). A local server talks to its client over stdio; a remote one over streamable HTTP.

Built-in tool, custom tool, Skill or MCP server: what decides between them?

Where the capability lives. Built-in tools cover common needs with nothing to build; a custom tool wraps one application's own logic; a Skill packages procedural know-how and files with no live system behind it; an MCP server exposes an external system many clients share.

No account needed

CCDV-F flashcards: 40 free cards across the eight exam domains

Forty recall cards for the Claude Certified Developer - Foundations exam, one deck spread across its eight 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 cardseight domainsNo accountNothing stored

How the deck is built

The CCDV-F blueprint has eight 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, Applications and Integration, gets the most cards and the lightest, Eval, Testing, and Debugging, the fewest.

Domain	Official weight	Cards
1. Agents and Workflows	14.7%	6
2. Applications and Integration	33.1%	13
3. Claude Code	3.1%	1
4. Eval, Testing, and Debugging	2.6%	1
5. Model Selection and Optimization	16.8%	7
6. Prompt and Context Engineering	11%	5
7. Security and Safety	8.1%	3
8. Tools and MCPs	10.6%	4

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: Agents and Workflows

14.7% of the exam, 6 cards

When should you build a workflow rather than an agent?: When the sequence of steps is fixed before the run starts and rarely changes. Code calls Claude at set points, so the path is predictable, testable and cheaper. An agent is justified only when the model must choose the next step from what it observes.; Objective O01: Agent Architecture
What is the supervisor's job in a manager and worker agent hierarchy?: It breaks the goal into bounded subtasks, hands each to a worker with only the context that worker needs, then checks and merges what comes back. Workers never see the whole conversation, so the supervisor owns coherence and the final answer.; Objective O01: Agent Architecture
What does the Claude Agent SDK provide that a plain Messages API call does not?: The harness behind Claude Code as a library: the agent loop, tools for files and the shell, permission handling, hooks, subagents, MCP connections and automatic context compaction. You supply the task, the allowed tools and the guardrails instead of writing the loop yourself.; Objective O02: Agent Construction with Claude
Why does a hook enforce a rule more reliably than a system prompt line?: A hook is your own code, run at a fixed lifecycle event such as before a tool call, and it can block or alter the action whatever the model intended. Prose depends on the model noticing and obeying; a hook does not.; Objective O02: Agent Construction with Claude
Name the steps of a minimal hand-written tool-use loop: Call the API with your tool definitions. If stop_reason is tool_use, run every requested tool, append one tool_result per tool_use id in a single user message, and call again. When stop_reason is end_turn, return the text. A missing tool_result is an API error.; Objective O03: Agent Patterns and Frameworks
What do frameworks such as LangGraph, Strands or PydanticAI add over a hand-rolled agent loop?: Reusable orchestration: graph or state-machine control flow, typed tool and output schemas, retries, memory stores and tracing. The price is a dependency and an abstraction layer between you and the API, so adopt one for multi-step orchestration, not for a single tool loop.; Objective O03: Agent Patterns and Frameworks

Domain 2: Applications and Integration

33.1% of the exam, 13 cards

Which business requirements become infrastructure requirements for a Claude application?: Expected volume and peak concurrency (rate limits, batching), how long users will wait (streaming or batch), where data may travel and how long it is kept (region, provider, logging), and uptime targets. Functional requirements describe the answer; infrastructure requirements describe how it is delivered.; Objective O04: Understanding Requirements
How does a Claude feature move through the systems life cycle?: Requirements with measurable acceptance criteria, a design that treats model, prompt and tools as versioned components, an eval suite before release, a staged rollout with monitoring and a rollback path, then maintenance that tracks model deprecation dates as scheduled changes rather than surprises.; Objective O05: Systems Life Cycle
What does stop_reason tell you about a Messages API response?: Why generation ended: end_turn for a natural finish, tool_use when Claude wants you to run a tool, max_tokens when the output limit truncated it, stop_sequence when one of your stop strings appeared. Always check for max_tokens before trusting an answer as complete.; Objective O06: Claude API Mechanics
What arrives when you stream a Messages API response?: Server-sent events in order: message_start, then for each content block a start, a series of deltas and a stop, then message_delta carrying the stop reason and output usage, then message_stop. Tool call arguments arrive as partial JSON fragments you must concatenate before parsing.; Objective O06: Claude API Mechanics
Which workloads belong on the Message Batches API instead of realtime calls?: Large sets of independent requests where nobody is waiting: overnight classification, backfills, bulk evals. Requests run asynchronously and complete within a window of up to a day at a discount to realtime pricing; you poll the batch and fetch results when it finishes.; Objective O06: Claude API Mechanics
What should a code review of Claude-generated changes concentrate on?: Exactly what a review of any contribution checks: does it meet the requirement, is it covered by tests, does it follow the codebase's conventions, and does it touch anything it was not asked to. The model's own summary of the change is not evidence.; Objective O07: Software Engineering Foundations
Why call the Claude SDK asynchronously inside a web service?: Each call waits on the network for seconds, so a blocking call ties up the worker for its whole duration. An async client lets the process serve other requests, or fan out independent calls together, while waiting. Cap concurrency to stay inside rate limits.; Objective O07: Software Engineering Foundations
Why should prompts, schemas and tool definitions live in version control?: They change behaviour as much as code does. Version control gives every change a diff, a reviewer, a rollback and a timestamp you can line up against eval scores and incidents. A prompt edited in a dashboard has none of those.; Objective O07: Software Engineering Foundations
Why does an instruction behave differently in claude.ai, the Desktop app, Claude Code and the API?: The products wrap your words in their own system prompts, tools, memory and formatting rules, so behaviour observed there does not carry over. Through the API you get the bare model: only the system prompt, messages and tools you send shape the response.; Objective O08: Claude Application Design
How do you mark pasted text as data rather than instructions?: Wrap it in a clearly named boundary, such as XML tags, place it in the user turn, and state in your instructions that whatever sits between those tags is material to process, never a command. Structure is what the model can rely on.; Objective O08: Claude Application Design
What does managing plugins in a Claude application involve?: Treating each plugin as a dependency: record its version, review the commands, agents, hooks, Skills and MCP servers it bundles, note the permissions they need, update deliberately and remove it when unused. One plugin can change a lot of behaviour at once.; Objective O08: Claude Application Design
What does pinning a dated model ID protect production from?: Silent behaviour change. An alias that names a tier moves to the newest snapshot when one ships; a dated ID stays fixed until it is deprecated. Upgrades then happen when your evals pass and you choose, not when the alias moves under you.; Objective O09: Configuration Management
What belongs in settings.json and what belongs in CLAUDE.md?: settings.json is machine-read configuration: permission allow and deny rules, hooks, environment variables, the default model. CLAUDE.md is guidance written for the model: build commands, conventions, architecture notes, things to avoid. Put enforcement in settings and guidance in CLAUDE.md, never the reverse.; Objective O09: Configuration Management

Domain 3: Claude Code

3.1% of the exam, 1 cards

Which CLAUDE.md files does Claude Code load, and in what order of scope?: From broadest to narrowest: an enterprise managed policy file, the user's own file in their home directory, the project root file, an untracked local variant for personal notes, and subdirectory files read when Claude works inside them. Narrower files refine broader ones.; Objective O10: Claude Code Operation

Domain 4: Eval, Testing, and Debugging

2.6% of the exam, 1 cards

How do you tell an integration bug from a model output problem?: Read the trace: request sent, raw response, what your code did next. A well-formed response your code mishandled is an integration bug. A wrong or malformed response to a correct request is a model problem, fixed in the prompt, schema or model choice.; Objective O11: Debugging and Error Handling

Domain 5: Model Selection and Optimization

16.8% of the exam, 7 cards

Why can one prompt produce different text on two identical requests?: Each token is sampled from a probability distribution that temperature and top_p reshape, so the path can diverge at any step. Even at temperature zero the platform does not promise identical output, so store the response you acted on rather than regenerating it.; Objective O12: LLM Fundamentals
What do extended thinking, adaptive thinking and effort levels each control?: How much the model reasons before replying. Extended thinking reserves a token budget for that reasoning, adaptive thinking lets the model decide per request how much it needs, and an effort setting trades depth for speed and cost. Thinking tokens are billed as output.; Objective O12: LLM Fundamentals
What does an official client SDK do on top of the REST API?: Builds authenticated requests, maps JSON to typed objects, raises typed errors per status code, retries transient failures with backoff, and parses streaming events into an iterator. You still own timeouts, concurrency limits, idempotency and what happens once the retries run out.; Objective O13: Technical Fundamentals
When does a websocket belong between end users and a Claude integration?: When a browser client needs tokens as they are generated and the API key must stay on the server. Your backend calls Claude with streaming on and forwards each event through the websocket (or server-sent events); the browser never reaches the API directly.; Objective O13: Technical Fundamentals
How do you choose between Opus, Sonnet and Haiku for a task?: Start from the task's demands. Haiku for high-volume, latency-sensitive or simple classification work; Sonnet as the default for most coding and agentic tasks; Opus where the hardest reasoning justifies the price. Then run the eval and keep the cheapest tier that clears it.; Objective O14: Model Selection and Tradeoffs
What counts as a breaking behaviour change between model releases?: A newer model handling the same prompt differently: stricter or looser refusals, changed default verbosity or formatting, a parameter that is deprecated or newly required, different tool-calling habits. Treat a model swap as a dependency upgrade and run the eval suite before promoting it.; Objective O14: Model Selection and Tradeoffs
How is prompt caching billed, and how do you confirm it is working?: Cache writes cost a premium over input tokens and cache reads cost a fraction, so the saving exists only when a prefix is reused within its lifetime. Confirm hits through cache_read_input_tokens in the usage object; zero means the prefix changed or the cache expired.; Objective O15: Cost and Token Management

Domain 6: Prompt and Context Engineering

11% of the exam, 5 cards

What is context drift, and what stops it?: The model's behaviour sliding away from the original instructions as a long session fills with tool output, dead ends and superseded decisions. Prune tool results you no longer need, compact the history into a summary, and restate the goal and constraints after every compaction.; Objective O16: Context Engineering
What is context isolation, and when does it pay off?: Running a side task in a subagent or a separate step with its own context window, so the parent receives a short result and never the raw material. It pays off for searches, large file reads and anything that would flood the main window.; Objective O16: Context Engineering
What goes in the system prompt, and what goes in the user turn?: System: the role, standing rules, output format and what to do when information is missing. User turn: the task and its data for this request. Per-request data placed in the system prompt breaks caching and blurs which words are instructions and which are content.; Objective O17: Prompt Engineering
What is input sanitisation in prompt engineering?: Bounding text from users or outside systems before it enters the prompt: truncating to a length budget, stripping control characters, escaping or wrapping it in tags and labelling it as data. It prevents accidental breakage; on its own it does not defeat prompt injection.; Objective O17: Prompt Engineering
What does defensive parsing of Claude's output mean?: Treating every response as untrusted input to your code: extract the expected structure, validate it against the schema, handle missing or extra fields, and fail closed when validation fails. Prose requests for JSON guarantee nothing; even schema-enforced output still needs your business rules checked.; Objective O18: Output Handling

Domain 7: Security and Safety

8.1% of the exam, 3 cards

Name the two halves of a prompt injection defence: Separation and least privilege. Keep untrusted content (documents, web pages, tool results) structurally apart from trusted instructions and labelled as data, then limit what any injected instruction could achieve by giving the agent only the tools and permissions the task needs.; Objective O19: AI Application Security
What does guardrail layering mean in a Claude deployment?: No single control is trusted alone. Input screening, a system prompt with explicit refusals, tool permissions cut to least privilege, output validation and a human approval step for consequential actions are stacked so that when one layer fails another still catches the problem.; Objective O20: Guardrails and Safe Deployment
Where should an API key live, and where must it never appear?: In a secrets manager or server-side environment variable, one key per environment, rotated on a schedule and monitored for unexpected use. Never in source control, browser code, prompts, logs or error messages. If one leaks, revoke it in the Claude Console before investigating.; Objective O22: Identity, Secrets, and Key Management

Domain 8: Tools and MCPs

10.6% of the exam, 4 cards

What makes a tool description effective?: It states what the tool does, the cases it fits and the cases it does not, what every parameter means with format and an example, and what comes back. Claude selects tools from their descriptions, so vagueness is the usual cause of wrong calls.; Objective O23: Tool Implementation
Client-side tools versus server-side tools: what is the difference?: A client-side tool runs in your code: Claude returns a tool_use block, you execute it and send back a tool_result. A server-side tool, such as web search, runs on Anthropic's side within the same request, with no round trip through your code.; Objective O23: Tool Implementation
Which three primitives can an MCP server expose, and over which transports?: Tools (functions the model may call), resources (data the client can read into context, addressed by URI) and prompts (reusable templates a user invokes). A local server talks to its client over stdio; a remote one over streamable HTTP.; Objective O24: MCP Server Development
Built-in tool, custom tool, Skill or MCP server: what decides between them?: Where the capability lives. Built-in tools cover common needs with nothing to build; a custom tool wraps one application's own logic; a Skill packages procedural know-how and files with no live system behind it; an MCP server exposes an external system many clients share.; Objective O25: Agentic Customization

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 CCDV-F cheat sheet, which holds the decision rules the cards compress, then work through the ten CCDV-F practice questions with the reasoning written out. Then sit the timed CCDV-F 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 CCDV-F exam is 53 items in 120 minutes for a $125 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 CCDV-F mock sits 53 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 CCDV-F mock, no account

Frequently asked

Quick answers to the follow-up questions.

Do I need an account to use the CCDV-F 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 CCDV-F exam?

No. Every card was written for this page from the official CCDV-F 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 CCDV-F?

Fewer than you would think. The blueprint lists its objectives under eight 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 CCDV-F?

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 CCDV-F 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.