Are they upscaling on the server side then encoding it as 4K?

That seems weird. You'd think that would add effort and bandwidth for little (if the have a very fancy upscaler) to no benefit at the cost of bandwidth.

Are they sure its not just semi automatically downscaling the video to meet congestion targets? Like if you watch a 4K stream on YouTube and it's not able to supply it at that Res, then it'll drop down.

I also ever that when HD was introduced, some games text was unreadable. How does the service cope with not knowing what the output resolution is going to be?