I assume you are talking about a “real” project that includes documentation, delivery, testing, and so on. For me, I find that these projects don’t develop linearly. As I’m working I think, “I’ve solved this step before” and then go grab code from some “throwaway” analysis which was not documented or tested. Usually I’ve solved a problem three times before it makes it’s way into it’s final shape and delivered to a customer. (See
http://blogs.sas.com/iml/index.php?/archives/36-Tricks-and-Treats.html)

That suggests the question, “When should this process begin?” If I write a small program for a blog or a “back of the envelope” computation, I don’t expend the same energy to document and test it as for a “real” project. And yet, those small programs often contain the germ of an idea or technique that does eventually become production software.

I think my style might be called “iterative programming.” The reproducibility doesn’t become an issue occur until the 3rd or 4th iteration.