This sounds really interesting (and also eerily similar to haskell's Coercible and roles), is there any material on the matter somewhere? I'd love to see what solution they're thinking of working towards, seeing as this is a pain point that I encounter somewhat frequently.