In one of his posts he actually provided Haskell-like type signatures (and stated why they weren't sufficient), so that's my basis for concluding that he thought about it in terms of types.
(Perphaps my memory is fuzzy, but I'm 99% sure on this point.)
He did provide them but didn't do a very good job, tbh. Types capture the behavior you want to express in your program and Rich ultimately concluded that types could no achieve this for transducers. On the other hand, many libraries since then have generally shown that types can and do... It's just that Rich's types didn't.
(Perphaps my memory is fuzzy, but I'm 99% sure on this point.)