Because the weather is so much more favorable in those two areas. States with extreme weather, especially snow & ice, are vastly more challenging.

It's like saying you tested your multi-platform app for 10,000 hours, but 9,800 of those hours were on Windows.

Driving condition matter since they vary from place to place. It matters for people, and it most likely matters for autonomous vehicles (either due to training data sets or direct programming of traffic regulations). To choose a mundane example, that is admittedly more likely to affect people, consider how many people try to make a left turn into a (North American) roundabout or who park in a bike lane. (Sometimes it is deliberate, but sometimes it is an out of town person who has never dealt with it before.)

That said, I would expect a San Fransisco decision to be based upon San Francisco data.

True. They ought to have driven every block of every SF street by now.

But of course, things change.