[IshKebab]

Very briefly, I resisted implemented features that wouldn't work perfectly. The memory use example was a real one (I was working on a profiler of some complex AI hardware). My boss would say things like "can we report how much memory this operation uses", and I would say "no because some of it is shared with other operations or only live for part of the program run etc.".

He didn't really say anything to change my mind, he just kept asking for things that would be useful to customers and eventually I realised that even if we can't give an answer that makes perfect sense or doesn't work all the time, we can still do better than nothing. Very often something that is roughly right and can be shown some of the time is better than something that doesn't exist at all.

It kind of sounds obvious when I put it like this, but you'd be surprised how often you see "we can't do this very useful thing because <minor flaw that means it won't always work perfectly>".