I'm not talking about fancy types either. Standard ML's type system is dead simple (no higher-kinded types, no GADTs, no type families, not even type classes),

I'm not talking about fancy types either. Standard ML's type system is dead simple (no higher-kinded types, no GADTs, no type families, not even type classes), and I got a lot more value from it than from Haskell's type system, mostly thanks to: (0) Abstract types, generated by means of opaque signature ascription, which Haskell doesn't have. This ensures that different modules can't “accidentally” each other's internal invariants. (1) The fact variables stand for values (due to ML being strict), rather than potentially diverging computations (due to Haskell being lazy), which makes useful algebraic laws sound even in the presence of nontermination or whatever effects.