How should I model processes in RxJava?

Suppose I have some process, such as downloading a file or running a big computation.

The events it fires might look like this:

{ progress: "0%" }
{ progress: "23%" }
{ progress: "78%" }
{ progress: "100%" }
{ content: "Hello. world. " }

Or perhaps:

{ progress: "0%" }
{ progress: "23%" }
{ error: "Access denied. " }

In each case, we have n status-updates (T) followed by either a result (S) or an error (Throwable):

T* ( S | Throwable )

A process is a mix of Observable<T> and Single<S>.

There are a few ways to model a process using the Rx primitives:

• Observable<T> and Single<S>
• Observable<Either<T, S>>
• Observable<Object> and instanceof
• Process<T, S> with toObservable and toSingle methods
• etc...

How has this been successfully modelled in the past?

Which approaches are best and when should they be used?