The Lava Layer Anti-Pattern (2014)

This one's a tough one - I've worked on a number of existing (legacy/non-greenfield projects) and on most of them I've just gone with the flow (do whatever the current version does) and that was the right thing to do because they were huge codebases with lots of people working on them.

Sometimes codebases do need a v2 though, and doing it iteratively seems like the safest approach, but it's easy to do a v2 just because there are multiple small issues with v1 etc etc