Thanks for the report! I think I might know what causes the 120% here: if the stream has a length, the default progress function currently divides the number of annotated examples in the current session by the length of the stream. However, the internal count of "examples in the current session" is incremented whenever new anwers are received and doesn't distinguish between generated session IDs. That's how you end up with 6 "session" annotations.
The custom progress function you can provide receives the number of session annotations and the number of annotations in the whole dataset – but it currently doesn't receive the session ID that's assigned at runtime (when you access the app with a ?session=). You could make it return None to make the progress less confusing.
We'd have to find a better general progress logic internally and I can't think of an easy workaround for now, since it's kinda complex. But I'll update if I can think of something you can implement yourself in the meanti,e!