
The four developer decisions, computed as quarters from the guide's own skill weights, and the two elimination rules validated against all three official sample questions. The coverage map: all 25 skill statements, owned.
One support ticket becomes one API call. Tokens, context windows, sampling, non-determinism, and the Messages API shape: roles, content blocks, stateless multi-turn.
Streaming and when it matters, vision input for screenshot tickets, extended thinking on hard tickets, and what the SDKs actually wrap.
Works-on-my-laptop becomes a real service. Requirements from business asks, the systems life cycle, version control, code review, refactoring. The heaviest combined weight of any lecture on the exam.
The biggest single skill on the exam. How Claude behaves across surfaces, content boundaries, schema design, session hygiene, and plugin management.
CLAUDE.md files and their hierarchy, settings.json, model version pinning, prompt versioning, and Claude Code operation.
Instruction clarity, placement of system versus user content, few-shot examples, and iterating on a prompt without guessing.
Context window management, drift and bloat, pruning, compaction, and isolation.
Structured output, response validation, defensive parsing. Then debugging: when output is garbage, whose fault is it, and what the traces tell you.
Model selection as tier logic, thinking modes, prompt caching, and cost modeling you can defend on an exam item.
The Batch API against realtime, and where third-party vendor platforms fit.
Tool use and function calling: schemas, description writing, error shapes, client versus server execution, and approval.
MCP server development, resources, tools and prompts, and the decision between built-in tools, custom tools, Skills, and MCP.
Workflow versus agent as a decision with criteria, manager and supervisor patterns, and where frameworks fit.
The Claude Agent SDK, custom loops, harnesses, hooks for deterministic actions, and hosted versus self-hosted deployment.
Prompt injection, jailbreak defense, untrusted input, guardrails, safety hooks, identity, and secrets.
The finished desk walked end to end, which doubles as a sweep of the whole blueprint. Then worked items at exam pace and a readiness read before you book.
The Claude Certified Developer Foundations (CCDV-F) exam does not reward recall. It hands you a scenario, a support desk that must answer ten thousand tickets by morning, a dashboard that chokes on prose, a ticket with hidden instructions, and asks you to make a developer's decision.
Every item on it asks one of four decisions. What runs. How does it call Claude. What does Claude see and say. Will it survive production. The quarters are real: computed from the skill weights in the current official exam guide, each decision covers roughly a quarter of the 53 items. The course stamps the decision on every worked question, so by the end you read stems the way the item writers built them.
Two elimination rules do the rest, and both are extracted from the official sample rationales rather than invented. The constraint rule: every stem names its binding constraint, cost, overnight, reuse, untrusted input, and the credited answer is the mechanism built for exactly that constraint. Any option that turns a generic knob, temperature, max_tokens, a bigger model, one more line in the prompt, is a distractor. The guarantee rule: when a stem says must, never, always or cannot, guidance loses to mechanism. A polite system-prompt line is advice. A hook that blocks the action, or a schema the output is validated against: that is enforcement. All three published sample questions fall to one of these two rules.
Roughly a quarter of this exam is not about Claude's API at all. Requirements, the systems life cycle, software engineering discipline, configuration management. It is the least marketed fact about the exam and it is where candidates who only studied the API lose points. A full lecture owns the core of it, and the mocks weight it the way the blueprint does.
The course is a running build. One customer support desk gets built on screen across seventeen lectures, from a single API call to an agent that survives untrusted input. Each lecture is the moment the system hits something it cannot do, and the exam skill is the decision that resolves it. All 25 skill statements in the current guide are covered, across all eight scored domains, at the guide's published weights.
Then the course tells you where you stand. Two timed 53-question mocks, matched to the blueprint at skill level, every item tagged twice: by domain and by which of the four decisions it tests. Your score comes back on both axes, so a weak area sends you to specific lectures instead of a vague "study more". A downloadable cheat sheet carries every number that can change, fee, time limit, cut score, validity, so the facts you memorize stay current even when the program moves.
What the exam itself looks like, per the current official guide: 53 items in 120 minutes, multiple-choice and multiple-response, with each item stating how many responses to select. Scoring is scaled 100 to 1,000 and 720 passes. The fee is 125 USD per attempt and the credential is valid for 12 months.
One access rule to know before you spend anything. Registration and scheduling run through the Anthropic Partner Academy and Pearson VUE, and access currently requires an organization in Anthropic's partner network; joining the network at its entry tier is free of charge for an organization. Check the current registration page before you schedule.
This is the developer track, for people who build, integrate, and ship applications on Claude's API, agents, and tooling. If you use Claude at work without building on it, the associate track tests that instead.
AI-use disclosure: AI tools were used in the production of this course, including the on-screen animation system and parts of the research and production workflow. The instructor designed the curriculum, verified every exam fact against the current official exam guide, and reviewed every lecture before publishing.
If you are sitting a different exam in this program, the other courses in this series are Claude Certified Associate (CCAO-F), Claude Certified Architect Foundations (CCAR-F), and Claude Certified Architect Professional (CCAR-P).