TL;DR: The Safety Score is crap.
So, after having read half a dozen pages of this thread (not quite willing to read all 35 of them), I'm going to let off a bit of steam. Like 80% of Americans, I believe that my driving ability and safety is well above average. I know that I drive faster than the speed limit, but my comfortable and normal following distance is about 2 seconds, and I tend to not change lanes. I don't text when I drive, half a million miles on a motorcycle trained me to always be paying attention to all the cars around me, and I tend to be considerate of other drivers.
I'm really appreciative of the transparency of Tesla's calculation -
Safety Score Beta tells you precisely what G-levels correspond to "extreme force" in braking, or turning. One of the problems that I personally have with most insurance efforts to put a transponder in vehicles is a lack of transparency - what are they measuring, and how are they judging? It's great that I can find that out.
But, let's talk. They don't apparently score hard accelerations. Doing 1+G accelerations in your M3P on the street doesn't impact your safety score, but 0.3G braking does - and in my vehicle, full regen is sufficient to ding me for hard braking. A 0.4G turn is a ding, and I do those everytime I make a 90 degree right turn out of a side street onto a main street and try to stay within the curb lane's boundaries. AP disengagements seems like it's trying to measure attentiveness - and a forced AP disengagement really does imply unattentiveness in my book - but a great proxy for non-AP attentiveness would seem to be Lane Departures (which wouldn't need Lane Departure Warnings enabled) which isn't measured. AP disengagements is obviously only a factor when on AP, LDW would measure on a finer time scale and when not on AP. I'm sorry, but if you regularly are ping-ponging off the lane lines when driving manually you're either not paying attention or your driving skills are so poor that you shouldn't be driving. Perhaps "lane changes per mile" would be a great criteria - I don't have the data that Tesla has, but it seems to me that those people slaloming traffic doing ten lane changes per mile are more likely to be in an accident than me driving two seconds behind the car in front of me in the middle lane for 10 miles.
It appears to me that Tesla made some decisions as to what their evaluation criteria should be, some of which (like acceleration) are questionable, then ran it through their big data statistical analysis to create their PCF function. That's great and all, but you could put in any criteria and crunch the data to determine it's impact on PCF - for example, you could use the color of the car, or the color of the driver's skin, and calculate the appropriate coefficients to add it into the PCF; that doesn't make it a great criteria.
I'd really hate to see what a weekend at the Autocross would do to my score, even though I believe that it makes me a better, safer driver.
