Ah very interesting. I guess my question was mainly about and, though I hadn't thought about the reasons for rec. It makes sense that that removes a class of errors.

So as a follow-up question: why is type inference impossible without and or what is it about and that makes it possible? For instance, what would happen if I (out of laziness or something) would just insert and everywhere, even for functions that aren't mutually recursive. So, e.g.

  let rec f x = x + 1
  and g y = y * 2

Would type inference somehow fail on that? If not, what's stopping me from writing a version of OCaml where all top level lets are automatically converted to let ... and ... and? (Other than a complete lack of necessary compiler writing skills obviously)