Review completed with approve. The prior blocker is resolved. The changed-preference feed state now uses accurate, non-causal tooltip wording, with an end-to-end regression test covering both the mismatch label and absence of the former causal claim. No regressions found in the follow-up delta.