> when trying to track down buggy behaviour, it's very useful to know that some component couldn't have been the problem.

> However, writing only pure code is quite limiting. Also, we still need to audit all IO code.

Quite the turn-around.