If everything is annotated then obviously you don't have predictability problems. I was talking about relying on automatic optimizations. In practice, do Haskell users rely mostly on the explicit annotations rather than the 'magic' optimizations?