I've spent some time thinking about this, and for me (and a lot of other people I've talked with) the line in the sand is "enough for us to be able to build new

I've spent some time thinking about this, and for me (and a lot of other people I've talked with) the line in the sand is "enough for us to be able to build newer or versions of the userland and kernel." This does not require that we're able to modify every bit of code. For instance, the Raspberry Pi has a binary blob of firmware that contains a lot of the code for using with the graphics hardware. While we can't modify or recompile this part, the (minimalistic) interface to this firmware is an open source kernel module. This allows you to build newer or modified kernels, and build other distributions for the RPi and still have access to the graphics acceleration. I consider this "open enough." While Android is fairly open, there is a large part of it still that is closed source. This includes the front-ends of Google's own services (I consider back ends out of scope), binary drivers (graphics), binary firmware (radio) and binary libraries (OpenGL implementations, closely linked to graphics drivers). The goal of Canonical's project is also to make the drivers and OpenGL part open source, but I've not seen anything about the radio bits.