CCDV-F facts to memorise: 30 recall cards
Thirty facts that are pure recall — the ones you either know cold or lose a mark on. Cover the right-hand column and work down the list.
| Prompt | What you need to know |
|---|---|
| Exam format | 53 scored questions, 120 minutes, 720/1,000 scaled. Multiple choice and multiple response. |
| Biggest domain | Applications and Integration, 33.1% — more than domains 4–8 combined. |
| The core stance | Engineer against the contract (stop_reason, usage, schemas, status codes), never against the content. |
system | A top-level parameter, not a message role. Array form allows cache_control. |
content | Always an array of typed blocks. Never index [0].text blindly. |
stop_reason: max_tokens | Truncated, not finished. Any JSON in it is likely malformed. |
| Other stop reasons | end_turn · tool_use · stop_sequence · refusal · pause_turn · model_context_window_exceeded |
| Retry these | 429 (honour retry-after), 500, 504, 529. |
| Never retry these | 400, 401, 403, 404, 413. |
| Size caps | 32 MB Messages · 256 MB Batch · 500 MB Files. |
| Streaming: final truth | stop_reason and cumulative usage arrive on message_delta. |
| Streaming: tool args | Accumulate input_json_delta; parse only at content_block_stop. |
| Why long chats cost more | The API is stateless — every turn resends the whole history as input tokens. |
metadata.user_id | An opaque hash. Never personal data. |
| Rate limits | RPM, ITPM, OTPM, per model, token-bucket refill — bursts trip limits the average would not. |
| ITPM exemption | cache_read_input_tokens do not count. Caching raises throughput as well as cutting cost. |
| Rate-limit headers | anthropic-ratelimit-{requests,input-tokens,output-tokens}-remaining and matching limit/reset. |
| Cache breakpoints | 4 max, ordered tools → system → messages. Mark the last identical block. |
| Cache pricing | Read 0.1× · write 1.25× (5 min) or 2× (1 hour). |
| Cache invalidation | Tool definitions invalidate everything below; tool_choice and images invalidate messages only. |
| Batches API | 50% off · ≤100,000 requests / 256 MB · ≤24 h · no streaming · match on custom_id. |
| Thinking budget | Minimum 1,024 tokens, must sit below max_tokens, billed as output. |
| Workflow vs agent | Workflow: you write the control flow. Agent: the model directs its own process. |
| The five patterns | Prompt chaining · routing · parallelisation (sectioning, voting) · orchestrator–workers · evaluator–optimiser. |
| Sectioning vs voting | Sectioning splits independent subtasks; voting repeats the same task for confidence. |
| Orchestrator–workers | Use when the subtasks are not known in advance. |
| Subagent inheritance | Nothing but the Agent tool's prompt string. No parent history, tool results or system prompt. |
| Structured output | output_config.format constrains the final response; strict: true constrains tool calls. |
| Unsupported in schemas | pattern, allOf, not, if/then/else, dependentSchemas, recursion. |
| stdio MCP rule | Never write to stdout — it carries the JSON-RPC stream. Log to stderr. |
← Previous: Questions & answers — Domains 6–8 | Next: Five-week study plan →
Explore Other Certification Study Guides
Switch to another exam guide to prepare for your next Claude certification:
Associate
CCAO-F
Claude Certified Associate — Foundations
Entry-level certification for non-coders: prompting, output evaluation, model selection, workflow integration, configuration, governance & responsible use.
Open Guide →
7 Domains
Architect
CCAR-F
Claude Certified Architect — Foundations
Core architectural principles: agentic loops & orchestration, tool design & MCP, Claude Code workflows, prompt engineering, and context reliability.
Open Guide →
5 Domains
Architect
CCAR-P
Claude Certified Architect — Professional
Advanced enterprise architecture: solution design, system integration, evaluation frameworks, governance, safety & developer enablement.
Open Guide →
7 Domains