A team asks for confirmation that a request's system prompt and tool definitions actually hit the prompt cache, not just that the flag was set. Where do they check?

Answer: A — response.usage.cache_read_input_tokens on the response

Setting cache_control is a request, not proof it worked. usage.cache_read_input_tokens on the response is the actual evidence caching engaged — never assume it did just because the field was present in the request.