Even when you can't have a _proof_ of associativity in the type, it's perfectly reasonable to document the associativity requirement in types -- e.g. the Monoid

Even when you can't have a _proof_ of associativity in the type, it's perfectly reasonable to document the associativity requirement in types -- e.g. the Monoid typeclass in Haskell.