CCDV-F Study Guide: The Blueprint, Weight by Weight

The Claude Certified Developer – Foundations exam (CCDV-F) is 53 questions in 120 minutes, scored on a 100 to 1000 scale with 720 to pass. Its exam guide, version 1.0, effective July 2026, does something the other Claude certification guides do not: it breaks each of the 8 domains into named sub-skills and publishes a weight for every one of them. There are 25 sub-skills, their weights sum to their parent domain, and the domains sum to 100.

That second tier is the most useful thing on this page, because a domain weight on its own can mislead you. Applications and Integration is 33.1% of the paper, which reads like one enormous subject to conquer. The sub-skill table shows it is really six separate subjects, the largest of which, Claude Application Design, is 8.6%. No single sub-skill anywhere in the blueprint is worth more than that, and twelve of the 25 are worth under 3% each while adding up to 30.2% of your score. Revision that chases the biggest heading first spends its days on a label rather than on the topics inside it.

What follows walks the blueprint in descending order of weight, converts each figure into an approximate share of the 53 questions, and closes with a revision sequence that accounts for which sub-skills assume others. Percentages are published weights rather than fixed per-form counts, so treat every question figure here as an expectation rather than a guarantee.

Read the second tier, not the domain heading

A sub-skill weight is a share of the entire paper, not a share of its parent domain. That is what makes the second tier usable: any two entries can be compared directly, however far apart they sit in the blueprint.

With 53 scored questions, one question is worth roughly 1.9% and one percentage point is worth about half a question. That conversion is the whole trick. Claude Application Design at 8.6% comes to about 4.6 questions, which is more than the entire Security and Safety domain at 8.1%. Claude API Mechanics at 6.8% outweighs Claude Code (3.1%) and Eval, Testing, and Debugging (2.6%) added together. Claude Hooks, at 1.0%, is about half a question, which on a given form means one item or none.

Hold these figures loosely. A weight tells you how the blueprint is balanced, not how many items of a given kind your particular sitting will contain. What it does tell you reliably is where an hour of revision has the highest expected return, and that is a far better question to ask than which topic currently feels least comfortable. Discomfort is a poor guide to marks, and the two-tier table is the corrective.

Applications and Integration: 33.1%, and six ways in

A third of the exam sits in one domain, about 17 or 18 questions, and it is also the most subdivided. The six entries are **Claude Application Design** at 8.6% (roughly 4.6 questions), **Software Engineering Foundations** at 7.4% (3.9), **Claude API Mechanics** at 6.8% (3.6), **Configuration Management** at 4.1% (2.2), **Understanding Requirements** at 3.4% (1.8) and **Systems Life Cycle** at 2.8% (1.5).

The three largest come to 22.8% between them. More than a fifth of the paper, from three named topics, is the single strongest argument for reading the sub-skill table before planning anything.

The split also tells you something about scope. Design, API mechanics and engineering foundations are assessed separately, so an item can probe how a request is built and consumed without asking anything about the application around it, and another can do the reverse. The two smallest entries are the ones a hands-on developer is most likely to skim, because requirements and life cycle read as process rather than code. They are still 6.2% together, which beats Claude Code and Eval, Testing, and Debugging combined.

Model Selection and Optimization: 16.8%, and a misleading title

The second-heaviest domain is close to 9 questions, and its internal split is the most surprising figure in the blueprint. Its two largest sub-skills are not about choosing a model at all: **Technical Fundamentals** at 6.1% (about 3.2 questions) and **LLM Fundamentals** at 5.2% (2.8). Those two carry 11.3%, two-thirds of the domain.

The selection and costing work is the smaller half. **Model Selection and Tradeoffs** is 2.7% (1.4 questions) and **Cost and Token Management** is 2.8% (1.5), so the two together are 5.5%, about three questions.

Revise this domain by memorising a comparison table of model names and limits and you have prepared for about a sixth of it. You will also have prepared the part most likely to move between guide versions. The mechanics underneath carry twice the marks and age far better: how tokenisation behaves, what latency and throughput actually depend on, why a given workload costs what it costs, and what a context window does to all three. This is the clearest case in CCDV-F of the second tier rewriting what a sensible plan looks like.

Agents and Workflows (14.7%) with Tools and MCPs (10.6%)

Agents and Workflows is roughly 8 questions and the flattest domain on the blueprint: **Agent Construction with Claude** at 5.3%, **Agent Patterns and Frameworks** at 4.9%, **Agent Architecture** at 4.5%. Under a point separates the largest from the smallest, so there is no prioritising to be done inside it. Cover all three, or accept losing marks in whichever you drop.

Tools and MCPs adds about 5 or 6 questions on a distinctly uneven split: **Tool Implementation** at 4.4%, **Agentic Customization** at 4.1%, and **MCP Server Development** at only 2.1%. That last number deserves a pause. MCP is prominent enough in the ecosystem to invite weeks of study, and server development is worth about one question, while making tools that a model can call reliably is worth twice as much.

Taken together the two domains are 25.3% of the paper, roughly 13 questions, and they overlap in practice, since an agent is largely a loop over tools. Studying them as one block, agents first and then the tool surface they act through, wastes less time than treating them as unrelated subjects a fortnight apart.

Prompt and Context Engineering (11.0%) and Security and Safety (8.1%)

Prompting is close to 6 questions, divided into **Prompt Engineering** at 4.6%, **Context Engineering** at 3.8% and **Output Handling** at 2.6%. The blueprint separates these deliberately: one is what you write, the second is what you assemble around it, and the third is what happens to the response once it arrives. Schema work, parsing and validation live in that 2.6%, small enough that a developer who handles strict output contracts daily can often bank it cheaply.

Security and Safety is about 4 questions and carries the finest subdivision on the paper: **AI Application Security** at 3.2%, **Guardrails and Safe Deployment** at 2.3%, **Identity, Secrets, and Key Management** at 1.6%, **Claude Hooks** at 1.0%. Nowhere else are so few marks split so many ways, which changes the economics of preparing it: a first hour spent on credential handling is likely to return more than a fifth hour spent on injection defence.

The pair comes to 19.1%, around 10 questions. That is more than Model Selection and Optimization carries on its own. Neither is skippable and neither deserves a fortnight.

The tail: single sub-skill domains and the twelve small entries

Two domains contain exactly one sub-skill each, which makes their scope unusually legible. **Claude Code** is 3.1%, all of it Claude Code Operation. **Eval, Testing, and Debugging** is 2.6%, all of it Debugging and Error Handling. Between them, 5.7% and about 3 questions. The name is the syllabus, and there is no second topic hiding behind it.

Misallocation runs both ways here. Spend your working day inside Claude Code and it is easy to assume the exam must weight it heavily, which over-prepares a 3.1% slice. Glance at 2.6% and it is just as easy to drop debugging altogether, forfeiting a question or two that an afternoon would have recovered.

The general pattern is worth internalising. Twelve of the 25 sub-skills are worth under 3% apiece, and collectively they are 30.2% of the exam: nearly a third of your score arriving in one-question and two-question pieces. The big headings cannot carry you, either. A flawless performance across the five largest sub-skills is 34.2%. These small entries are cheap to collect and cheap to lose, and their size is precisely why they keep getting postponed.

A revision order built from weight and dependency

Weight sets the priority, dependency sets the sequence. Four blocks, each close to a quarter of the paper.

**Block one (26.9%).** Claude API Mechanics (6.8%) and Software Engineering Foundations (7.4%) before Claude Application Design (8.6%), because design questions assume both, then Configuration Management (4.1%). One connected body of material, and the largest quarter available.

**Block two (23.0%).** Model Selection and Optimization end to end (16.8%), opening with the two fundamentals sub-skills that carry 11.3% of it, then Understanding Requirements (3.4%) and Systems Life Cycle (2.8%) to finish Applications and Integration off.

**Block three (25.3%).** Agents and Workflows (14.7%) and Tools and MCPs (10.6%) in a single pass.

**Block four (24.8%).** Prompt and Context Engineering (11.0%), the four corners of Security and Safety (8.1%), then the two single sub-skill domains (5.7%).

The ordering is front-loaded on purpose: run out of time and you run out of it inside the 5.7%, not the 26.9%. Re-rank the blocks against your own results after a first full practice run. The weights exist to price your gaps, not to be obeyed mechanically.

Common questions