The questions candidates actually ask about Claude Certified Developer — Foundations

Claude Certified Developer – Foundations (CCDV-F) is a real, published, proctored Anthropic certification for developers who build applications on the Claude API. It is one of Anthropic's four Claude certifications, alongside the Associate and the two Architect exams.

AI and machine-learning engineers, technical leads and senior software engineers — typically with one to five years of software engineering and at least six months of hands-on work with Claude or a comparable LLM system.

No. The credential is awarded on exam performance alone.

53 questions in 120 minutes, at $125 per attempt — a shade over two minutes an item.

Through Pearson VUE, online proctored or at a test centre.

720 on a scaled score from 100 to 1,000.

No. It is a scaled score, not a percentage of items answered correctly, and it does not convert to a number of questions you must get right.

No. The cut score is criterion-referenced: it comes from a formal standard-setting study, and every candidate is measured against the same fixed standard, not against the other people who sat that week.

No. You are judged on your total scaled score. The per-domain percentages on your result report are informational and are not used to decide pass or fail.

Pass or fail, your scaled score from 100 to 1,000, and the percentage of items you answered correctly in each domain.

• Applications and Integration — 33.1%
• Model Selection and Optimization — 16.8%
• Agents and Workflows — 14.7%
• Prompt and Context Engineering — 11.0%
• Tools and MCPs — 10.6%
• Security and Safety — 8.1%
• Claude Code — 3.1%
• Eval, Testing, and Debugging — 2.6%

The guide publishes weights, not item counts. Multiplying each weight by 53 gives an expectation of about 18, 9, 8, 6, 6, 4, 2 and 1 items, from Applications and Integration down to Eval, Testing, and Debugging.

Applications and Integration, at 33.1% — a third of the exam, more than the next two domains combined and more than the four lightest together.

No. It carries six weighted sub-objectives: Understanding Requirements, Systems Life Cycle, Claude API Mechanics, Software Engineering Foundations, Claude Application Design and Configuration Management. Requirements gathering and life-cycle placement are graded.

Yes — uniquely in Anthropic's certification family, it weights 25 sub-objectives, each with its own share of the paper. Claude Application Design is the heaviest at 8.6%, more than the whole Security and Safety domain. Software Engineering Foundations (7.4%) outweighs Tool Implementation (4.4%). Claude Hooks is the lightest at 1.0%, about half an item.

• Applications and Integration: turning a requirement into a Claude application that survives production — request and response shape, stop_reason branching, error paths, configuration and versioning.
• Model Selection and Optimization: tokens, the context window as a fixed budget, sampling and non-determinism, cost and token management, and choosing a model versus enabling a reasoning mode.
• Agents and Workflows: workflow versus agent, loops that terminate, memory across sessions, and human-in-the-loop checkpoints where actions are irreversible.
• Prompt and Context Engineering: system prompts, XML structure, few-shot examples, output constraints, compaction, clearing history, subagent handoff and output parsing.
• Tools and MCPs: tool schemas Claude selects correctly, results returned in the expected form, and MCP servers with the right transport and configuration scope.
• Security and Safety: prompt injection from fetched content, jailbreaks, untrusted input, scoped identity, secrets management and data boundaries.
• Claude Code: the explore–plan–code loop, permission modes matched to risk, and durable project context.
• Eval, Testing, and Debugging: defining "done" before deploying, choosing a grading method, and telling retriable errors from terminal ones.

No. The guide publishes no such appendix; its task statements are the whole of the published scope.

Multiple-choice and multiple-response items, each stating how many responses to select. Do not assume four options or a single answer: in the captured courseware items, twelve of seventeen offered four options and five offered three.

Anthropic's courseware uses four-option scenarios, targeted fixes read off a session trace, fill-in-the-blank configuration, matching deployment scenarios to transport-and-scope pairs, free-text corrections compared against a model answer, a multi-layer debug with bugs planted in the schema, streaming, context and memory layers, and "assemble the minimal configuration" items that ask you to leave out what does not belong. These are courseware shapes; nothing states the exam reuses them.

No. None of the 32 courseware items visible to the course carries a recorded correct answer, so anyone claiming to know the official answers is not reading them from the courseware.

The absolutist ("always", "never", "only"); true but not the question; right action at the wrong layer; the over-engineered option; and an answer to a different question. The developer versions add a plausible-but-wrong API shape — real fields that fix a different class of defect than the error names — and an anti-pattern wearing best-practice clothing, such as rotating a key but leaving it in the credentials file.

Evidence first: classify the defect as schema validation, message ordering, transport or authentication, and name the layer it lives in. Then read the options and discard everything outside that class.

Only when you can name what you missed on the first read — never because your confidence wobbled.

Not published. Nothing says whether you can flag and revisit items, whether delivery is adaptive, whether partial credit or negative marking applies, or whether breaks and scratch paper are allowed. Confirm anything your pacing depends on when you register.

Valid, unexpired, government-issued photo ID whose name matches your registration exactly. You must accept the NDA before the exam begins; declining ends the session with no refund.

Cancelling less than 24 hours before your slot forfeits the fee, as does not showing up.

The retake waits are 14 days, then 30, then 90, with a maximum of four attempts in a rolling 12 months, and the fee is payable each time. Keep your per-domain breakdown — it is your plan for the retake.

Appeals must be filed within 14 days. Neither the standard-setting outcome nor the content of an individual item is appealable.

12 months. Renewing on time is a free, non-proctored assessment on the Partner Academy; if it lapses, you sit the full exam again at full fee.

Yes. CCDV-F is one of the exams that count towards Claude Partner Network eligibility.

The heaviest ones, because only the total score counts: an hour that lifts Applications and Integration by ten points is worth about 1.8 items, against about 0.1 for Eval, Testing, and Debugging. The exception is a domain where you are near zero — security items are decidable once you know the failure modes and undecidable otherwise.

An hour each. Together they are 5.7% of the exam, about three items — the two most enjoyable topics and the two worst places to spend a third of your revision.

No. The newest models reject non-default temperature, top_p and top_k with a 400, and where temperature is accepted, 0 makes output more repeatable without guaranteeing identical text. Tests should assert properties, not exact prose.

No. Streaming changes when the pieces of one response arrive; it does not reduce total generation time or raise throughput. Fast mode raises output tokens per second at premium pricing, and does not improve time to first token.

No. Claude streams over server-sent events on an ordinary HTTP response. A websocket can legitimately sit between a browser and your server; the hop to Claude is HTTPS.

No. Alloy is independent preparation and is not affiliated with Anthropic. Passing CCDV-F is what earns the credential; finishing the course confers none.

Thirteen lessons: an exam-strategy lesson, then twelve domain lessons in a teaching order — fundamentals before API mechanics, mechanics before what is built on them. Applications and Integration gets four lessons, Model Selection and Optimization two, and each other domain one. Every domain lesson is built against the exam's published objectives.

With the free diagnostic, before you study anything: 24 questions, about 20 minutes, drawn from the same question bank as the course and weighted like the exam. It reports each domain as questions right out of questions asked — about three per domain, so a pointer, not a grade.

A free account needs no card and has no trial clock. It includes the full diagnostic, the first two lessons with their podcasts — the exam-strategy lesson and the lesson on what a request costs, why it varies and how it reaches Claude — and a place to store and verify certifications you already hold.

Yes. The exam-strategy lesson is worth a 10–20% score bump: it targets getting wrong a question you actually knew.

Because certain-and-wrong is what fails people: an answer you were sure of never gets re-checked. Alloy tracks the gap between confidence and correctness, counts it towards readiness, and can hand back exactly the questions you were confident about and got wrong.

Yes, hosted by Morgan and Casey. "How to Take This Exam" (about 24 minutes) goes with the exam-strategy lesson; the second lesson has a short episode and a question walkthrough.

Yes. Alloy's Full membership covers all four Claude certification courses, and every course that opens while you are a member. Prices are on the pricing page, and credits you buy never expire.