Ask HN: What sorts of theory work best in practice for you?

For a lot of the programs I develop, I find I need to break out a lot of theory just to understand what implementation strategies don't provide reliability guarantees or what are bounds on how (in)efficient any algorithm I use will wind up being. This is mostly in the context of graph based heuristics for a grab bag of different problems How often does this happen in your favored domain, and what conceptual tools do you find most helpful?