> The comparison isn't really like-for-like. NHTSA SGO AV reports can include very minor, low-speed contact events that would often never show up as police-reported crashes for human drivers, meaning the Tesla crash count may be drawing from a broader category than the human baseline it's being compared to.

Tesla's own stats don't count any accident without airbag deployment, regardless of severity (and modern airbag systems have a number of factors that play into deployment), and, for some unknown reason, they don't count fatalities in their crash statistics.