This sounds pretty reasonable to me. You could also call this the novel writer's approach, or Nanowrimo approach, since it follows the usual writers' advice of getting an initial manuscript down, then revising.
I can't argue that it's the ideal approach for larger projects, but you'll often find academics (theoretical computer scientists and applied mathematicians) doing something similar. One person thinks of an algorithm, publishes their results, and subsequent research is on improving the method.
My own approach is similar, but I go through more "hand-written" versions--working out the math, processes, etc. on paper. The result is psuedocode. Once I'm happy with it, I code it, then optimize the implementation as much as I can. Writing actual code is tedious, so working on the "baby code", as you call it, on paper first saves time.
If I did a good job planning, by the time the entire game is coded, I won't need to revise much of the actual programming. On the other hand, I may have killed a small forest.