On the subject of judges, the reason that they're appointed is that they shouldn't have to be accountable to anybody. The judges are supposed to be fair, and not have any obligations or needs coming from any person or group of people. Of course, it is certainly true that appointment isn't the perfect system for selecting fair and impartial judges, but I do believe it is better than elections.
After all, when you have open elections, it suddenly becomes apparent that these people will believe in image and funding above all else. The judges would just become another congress, and there would be little reason to have the Congress separate from the judicial branch of government.
Granted, long terms for judges can help them to be less accountable (and they do have very long terms currently) to the people that elect them, but cannot eliminate the problem.
Additionally, if it is true that we, as a people, have elected remotely-good people in government, then those people will attempt to appoint those that do try to be fair. Don't forget that all presidential appointments are overseen by the Senate, so the people as a whole do have some say, though indirect, over which judges get appointed.
Anyway, back to Cg.
I think it's excellent how it is now obvious that though the backend is closed, other developers can write their own backends if they so choose. Granted, it might take a while for anybody else to come forward with a decent compiler for their own hardware (nVidia's been working on Cg for a year), but I believe this is certainly enough for Cg to become a standard. The only things that are left are:
1. Will nVidia (or Microsoft...as it seems they've been rather, um, IP-grabby) attempt to charge other companies for using Cg? This would be very bad.
2. Will Cg gain wide industry acceptance? This I'm neutral on. If Cg is good compared to other HLSL's, then I hope it is accepted. I further would like to see OpenGL 2.0 use Cg, unless OpenGL uses a better language.
I'm still wondering about what JC said about HLSL's needing to do multipass. If no big HLSL's turn out to do automatic multipass rendering for today's hardware, I really do hope that all future (DX9+) hardware has completely unlimited program sizes. Otherwise the dream of having newer hardware take advantage of new features immediately upon recompilation of a Cg program would be shot (given that the Cg program was compiled at runtime, which should become common if Cg takes off).