I also find that phrase super misleading. I've been using a different heuristic that seems to work better for me - "comments should add relevant information that is missing." This works against redundant comments but also isn't ambigous about what "why" means.

There might be a better one that also takes into account whether the code does something weird or unexpected for the reader (like the duplicate clear call from the article).

I like this framing, but might add to it: "comments should add relevant information that is missing and which can't easily be added by refactoring the code".