Isn't one part of pattern matching to ask for a type and the other to ask which values the (instance?) of that type holds?

When a language has algebraic data types -- i.e., data types that have multiple, disjoint constructors -- pattern matching generally means determining which constructor was used to create the value being matched, and branching on that. I'm guessing this is the kind of pattern-match that TypeScript users would be eager to have.

What I don't see is how this requires the addition of a "runtime" on top of the JS model to make it work. I don't use TypeScript, and don't know how it implements its data types, so there may be something obvious here that I'm missing.