Depends on the task, but if your task is amenable to it, I'd consider writing tests which basically say, at a very high level: "Is the output for a given set of

Depends on the task, but if your task is amenable to it, I'd consider writing tests which basically say, at a very high level: "Is the output for a given set of inputs the same?" Even if you're wrapping the entire application. We're not looking for "does this code do the right thing", just "does this code do the same thing" (as the initial code). At which point you'll at least have some more confidence that you aren't damaging the codebase due to having not understood a part of it.