Towards a more perfect union type

We present a principled theoretical framework for dealing with union types, and show its work in practice on JSON data structures.

The framework poses a union type inference as a learning problem from multiple examples. The categorical framework is generic, and easily extensible.