Curious what you would use instead of JSON. I am partial to using fixed-width binary formats for whatever you can get away with, but some things are better off with something more readable.

But as you point out they are messy and don't give nice deterministic hashes. Not without a bunch of painful clean-up which opens the door to abuse.

I am doing my own thing as well, but am a bit stuck on what to use for type definition files. All I really care about is the hash of it, but you need something human readable to translate the hash.