You all are right, GND is on USB GND. Having it isolated would be cool, but it also would increase the costs, and maybe the size. I decided to start with the simplest variant, my budget and time is very limited.
Cool, thank you Markino!
The answer was already given, with USB 2.0 we are limited to 500mA @ 5V = 2.5W.
There is a graph showing possible output current down on the Kickstarter page.
I really like the idea having a 15W version based on Type-C, however its the same argument, for me this can only be the second step, and it will again come with additional costs in components.
There are several mechanisms to prevent the USB Host from damages:
On software level, I got good results with an additional interrupt handler on output shortcut situations, that guarantees that input current always fits the USB 2.0 specification.
Additional, when USB input voltage is falling down a specified mark (e.g. 4.75V) the output current is reduced automatically. (A message will be shown that there is a problem with your input). When falling down again, the output will be disabled.
For the situation things go really wrong, I added a PPTC resettable fuse to protect computer from consequences.
Maye I can add a video, showing the noise on USB line with the oscilloscope while doing shortcuts on the output. Will do this next week.
As RNDIS is a Microsoft concept, you have to install the HoRNDIS driver on Apple products:
https://joshuawise.com/horndisI never tested your usb-c/usb adapter combination, but as siru.box acts the same way all other USB devices do, I can't find a reason why it should not work.