Essentially I'm saying nothing more than I can change a function A -> (B, C) to a function (A, X) -> B. Typeclasses provide a way to represent a more general sp

Essentially I'm saying nothing more than I can change a function A -> (B, C) to a function (A, X) -> B. Typeclasses provide a way to represent a more general space of those functions. Existential types are the explicit way of writing the kind of subtyping I'm talking about—so if you need it to be an explicit hierarchy then, yes, you need extensions to Haskell98. The best example of doing this without existential types (though it violates Haskell98 in its own way) the lens library. It absolutely provides a natural subtyping hierarchy without using existential types, though all of the coercions are implicit. I'm very confident in my understanding of the Haskell behavior, but I'm happy to concede that my use of subtyping terminology is off. I cannot see how the behavior you see in lens doesn't provide a stellar example of LSP, though.