This is especially true for long-lived reviews. I've been implementing a 1-wire bus. I thought it was ready when I submitted it to phabricator, but a number of comments have surfaced (in my case about the man pages, but in other long-lived reviews it is about the code). As the code changes and evolves, there's a natural motion of the comments people leave form the actual code it affects. This is doubly true if the comments are addressed. Tripply true when the comments may be short, but suggest rather major surgery (eg, you should move from a polled model to an interrupt driven state machine).