In my style of writing WL, I typically use pure functions inline, without assigning them to a name, as when writing the function argument to Map. For functions that I want to be part of my package, I use function definitions (the SetDelayed option).

Function definitions are generally my go-to default, because they have many advantages: * pattern-matching, which you mentioned; this doubles as a sort of documentation for the function, similar to the way types can hint at how to use a function * attributes (Protected, HoldAll) can be assigned to a symbol; there is a way to assign attributes to pure functions but I find the syntax awkward and can't recall ever having done it * because of the function definitions use pattern-matching, it lends itself to writing overloads of a function, which I use a lot

Pure functions, when used anonymously (without assigning them to a simple), are good when a relatively simple transformation needs to be done, as with Map, Fold, or SortBy (or any of the other *By functions). Although even with Map and Fold I sometimes find the function gets bigger than I first expected, and I move it to be a function definition assigned to a named symbol.