And here I was hoping that one of the goals of this was Hindley-Milner type inference for production clojure code (which, for those who didn't look at the code, is explicitly not a goal).
Oh well, cool experiment. I might have to try it sometime.
You really have to design your language around HM if you want to use it, and it would disallow a lot of LISP-style coding that really takes advantage of dynamic typing.
You're aware of core.typed? If not, take a look -- it provides type checking / induction on Clojure code.