Most often I look back at the original, work out how I did it then and then re-code from scratch. The results are (mostly) a big improvement

An experience I share; I am continuously reminded of the old adage that you only really understand a problem after you've solved it. By definition this also means that you continuously improve your understanding of a problem with every iteration.

I have found that even recoding once more yet usually yields another substantial improvement - third time's the charm, so to say.

Makeshifts last the longest.