Having worked with the result of prior generations (circa 2010) of algorithms for that sort of thing in the radar spectrum (I was not privy to the actual algorithms that underpinned it all) I suspect accuracy drops off exponentially once you get away from text based stuff and flagrant body differences (missing mirror, aftermarket spoiler, etc).
I routinely add/remove many of those things so I'm not sure how reliable they are.
Like I alternate between hitch/no hitch/bike rack, add/remove roof rack (it hurts mileage and is easy to swap), swap between my summer and winter rims+tires, and rotating through a set of magnetic bumper stickers would be trivial.
I'm guessing license plate takes precedence over all those things. On the other hand, Flock hasn't shown any particular competence, and is happy to flag a bunch of false positives (look at all those criminals!).
> and is happy to flag a bunch of false positives (look at all those criminals!).
Garrett has explicitly said he'd prefer a false positive to a false negative (very dystopian, no?) given his goal of "zero crime, powered by Brawndo, I mean Flock".
If Shot Spotter is bad at merely identifying the source of a gunshot, I'm not going to trust our security tech bros to cut through all the environmental noise and provide trustworthy identification among 10,000 different Teslas.