Would you then recommend prototyping in a dynamically typed language, then re-implementing that in Haskell once you know more about the solution space? Exploration first, hardening second?