Summary

Recently we have been making more effort to not run all the tests when unnecessary, particularly for doc only changes (#25145, #25261, #25383). By default CodeCov seems to depend on the tests having been run on both current and previous commit in order to produce the report. If it has different numbers of test reports for the two commits, it fails, showing as "indirect changes" on a lot of files (example).

Proposed fix

This situation is described in the CodeCov docs (point 9 here), and CarryForward flags appear to be the recommended solution. I have not really understood the docs on CarryForward flags though.