On balance, after a few days of driving the Safety Score, I feel like it's pretty good overall as a screening tool, though it definitely induces unnatural driving. They say they may tweak it. If they do, the first thing I think they may tweak is the hard braking - I really do think it is far too easy to trigger it. 0.3g is actually a decent slowing force, but I do think transient decelerations often exceed that as people transition to braking. And it's so relatively restrictive that it leads to quite unnatural driving behavior - moreso than any of the other metrics. Maybe make it 0.4g? Unsafe following, which isn't even weighted very much, is not hard to do decently at, and it's great to enforce. Aggressive turning is a little sensitive, but it's also not all that punitive. I do feel like they could potentially increase the g limit there slightly as well, maybe to 0.35g. I've had a couple of minor dings there and I really can't figure out where they came from - I think primarily they came freeway interchanges where I was below the speed limit (but probably above the yellow recommendation), but I'm not sure.
It is a little weird that they only look for intervals (times) exceeding a certain g value, and do not look at the peak g values, etc. Maybe it all comes out well correlated, but it seems like there could be a better way to grade people's smoothness and measure the real panic of a stop (which is one thing they're trying to check I assume). Maybe it just turns out that lower g's are well correlated with people that don't speed, so maybe it all makes sense.
Still, I'm not convinced it's optimally balanced, and it sounds like Tesla isn't ready to call it optimal either.
It does sound like from others (at this point) that AP use masks all of the traffic lights (I guess there is still uncertainty but certainly a few people say it takes care of it), so that would take care of most hard braking, but it is kind of a bit silly as using AP in those situations is kind of annoying.