Claude Certified Developer - Foundations (CCDV-F) Flashcards¶

120 minutes

53 multiple-choice and multiple-response

720 / 1000

$125 USD

Pearson VUE - online proctored or test center

12 months (free non-proctored renewal via Partner Academy)

the required cap on response length, including thinking and text.

a typed piece of a message: text, image, document, tool_use, tool_result or thinking.

the field that says why generation ended: end_turn, max_tokens, stop_sequence, tool_use, pause_turn, refusal or model_context_window_exceeded.

pins the API contract so old behavior keeps working; new behavior lands behind new versions or opt-in beta headers.

the transport streaming uses; each event is a JSON object on a data: line.

the first streaming event, delivering the initial Message object with empty content.

the server keeps no history, so every request must include the full message history you want Claude to see.

strings that halt generation when Claude emits one; stop_reason becomes stop_sequence.

the content block in an assistant message where Claude asks you to run a tool; it carries an id, a name and an input.

your answer to a tool_use, sent in a user message with a matching tool_use_id.

sets whether Claude may call tools: auto (Claude decides), any (must call one), tool (must call a named tool) or none.

tool_choice any or tool, which forces a tool call; on Claude Opus 5.5, Sonnet 5.5, Fable 5.1 and Mythos 5.1 it returns a 400 error, so use auto with strict tool use or structured outputs there.

stores a prefix of your request server-side and replays it cheaply on later requests with the same prefix.

the marker placed on a content block that makes it, and everything before it, a cache breakpoint.

runs jobs asynchronously at a 50% discount, with results within 24 hours.

your correlation key on each batch request, used to map results back to your records.

lets you upload content once and reference it by ID across many requests.

attribute spans of Claude's response to specific spans of source documents; enabled with citations: {enabled: true} on document blocks.

the content block used to send a PDF, by base64, URL or file_id.

you hit a requests-per-minute or tokens-per-minute ceiling; retryable with backoff.

the official SDK honors it when present on a retried error; a tier spend-cap 429 has no retry-after and keeps failing until access resumes.

after N consecutive failures, stop sending requests for a cooldown and test occasionally with a single request.

the retry pattern for retryable errors, with capped attempts.

code-orchestrated: your program decides the steps and their order, and the model fills in the content of each step.

model-driven: Claude decides which tool to call next, observes the result, and keeps going until it judges the task done.

break a task into sequential API calls, each feeding the next, with optional programmatic gates between steps.

classify the input first, then dispatch to a specialized prompt and often a specialized model.

a central model call decomposes the task dynamically, farms subtasks to worker calls, and synthesizes the results.

one call generates, a second evaluates against criteria, and the generator retries with the feedback until it passes or a retry cap hits.