One of the TR maintainers (samth?) told me on IRC a while ago that it is very infeasible to do global type inference (top level) for TR. So I'm not sure how much you could get done in so short a time.
Yep that's correct. I want a tool that infers a rough approximation of top-levels to accelerate the process of porting untyped code to be typed. The programmer would inspect the annotations manually and fix any inaccurate ones, and then run the type checker.
I believe it has to do with features like subtyping and method overloading which cause havok for global type inference. Found a paper here: http://ropas.snu.ac.kr/lib/dock/HoMi1995.ps