What
The prompt cache lets Claude Code reuse parts of a request it has already sent, instead of sending them again in full. When building a request, Claude Code now checks the most recent messages that the cache does not cover. It reports the reasons and an estimate of the tokens involved; a token is a small piece of text. This is internal reporting.
Why
It measures how often, and why, requests miss the cache.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear where this reporting is sent or whether a switch controls it.