The problem is that with a L2 system like the current autopilot where you have to take over when you exit a highway and encounter for example a traffic light, you trigger effectively a "disengagement" while operating with the expected flow of the system. It means that the "disengagement metric" gets very noisy (I'd say useless) in this case.

I don’t mean that Tesla will an anamolous number of disengagements, I mean that that they will file a report at all, which they haven’t done in 2017 because presumably they stopped autonomous driving on real roads. These reports don’t cover systems like the current Autopilot, as it is not considered autonomous driving.

Heh, that also points to the severely limited use of the autopilot itself. Certainly it’s salient data to publish.

Most self-driving vehicle projects have moved to Arizona which does not require disclosing these numbers.

Has Tesla done this?