In the first instance, both. However, i'd take more code that was better reasoned and easily understood over less verbose code that was smaller for smaller sake.
In terms of minimal change, we refactor when there's a clear business case to permit taking on the risk. Otherwise, we make the most minimal, most stable, least risk change to the existing code even if that code isn't optimal/pretty/well-structured/has-errors/...
Like most other engineering in the world really.