Another one of my favorites was by Brenda Baker called struct, which took Fortran programs and converted them into a structured programming language called Ratfor, which was Fortran with C syntax. This seemed like a possible undertaking, like something you do by the seat of the pants and you get something out. In fact, folks at Lockheed had done things like that before. But Brenda managed to find theorems that said there's really only one form, there's a canonical form into which you can structure a Fortran program, and she did this. It took your Fortran program, completely mashed it, put it out perhaps in almost certainly a different order than it was in Fortran connected by GOTOs, without any GOTOs, and the really remarkable thing was that authors of the program who clearly knew the way they wrote it in the first place, preferred it after it had been rearranged by Brendan. I was astonished at the outcome of that project.

"Computing History at Bell Labs" http://research.swtch.com/bell-labs