CCAR-P practice › Claude Models, Prompting & Context Engineering · Question 30 of 110
Claude Models, Prompting & Context Engineering
A team enabled prompt caching with a cache_control breakpoint after the system prompt. Usage data shows cache write tokens on nearly every request and almost no cache read tokens. The prompt order is: tool definitions, then a system prompt whose first line is 'Session: {session_id}, opened {timestamp}', then a long policy document, then the user message. What is the most likely cause?
Show answer and explanation
Answer: A — Per-request session and time values before the breakpoint make every prefix unique, so none is reused
(a) is right: caching only hits on an identical prefix, and a session ID and timestamp near the top change it on every request, so each call writes a new cache entry that is never read. Move volatile values after the breakpoint. (b) would show no cache writes at all, not writes on every request. (c) is false; usage data reports cache reads and writes. (d) is false; tool definitions are part of the cacheable prefix.
Keep going
More Claude Models, Prompting & Context Engineering questions
- A support agent handles long troubleshooting conversations. To control tokens, older…
- An enterprise assistant's system prompt carries 25 departmental procedures (expense…
- A nightly account-research agent's cost has doubled. A token breakdown shows most input…
- A personal-finance app categorizes every card transaction with Claude, millions of…
- A contract-analysis assistant on the mid tier keeps giving wrong termination dates…