It occurs to me if you have a A1 | A2 | A3 ... where Ai <= A and you call a method m defined on A but each Ai may override with a more specific return type Ri then you should get as a result R1 | R2 | R3 ... This wouldn’t require any reflection since it is just propagating static information about the refined override return types.
Then, we can consider Int as having 2^(32) subtypes, and each have different values for def +[N <: Int](n: N): scala.compiletime.ops.int.+[this.type, N], but that could be done virtually at typechecking time.
Something like that…