I was talking about both. Sometimes even in a problem space time constraints demand that you utilize something off the shelf (whether you use part of it or build on top of a custom version of it).

Tools aside, I think everyone who has 10+ years can think of a time they had a prototype go well in a new problem space only to realize during the real implementation that there were still multiple unknown unknowns.

Yeah, typically when you start thinking something through and actually implementing stuff you can notice that some important part of the behaviour is missing and it might also be something that means that the project is no longer feasible