Robotaxi safety record

Why this is a record, not a per-mile safety rate (the denominator rule)

A per-mile crash rate is only honest where the miles are a filed, audited denominator. Waymo publishes vehicle-miles through the California CPUC and its own periodic safety reports, so a rate can be computed for it; those figures are Waymo-published, validated in part by the reinsurer Swiss Re, and are not regulator-audited. Tesla's robotaxi runs under a chauffeur-service permit rather than an autonomous-vehicle testing permit, so it does not appear in California DMV autonomous-mileage data and has no clean filed denominator. DEPLOY therefore renders robotaxi incident counts here and does not publish a cross-operator “Nx worse” rate. Where an operator has no audited mileage, the honest answer is that the rate is not yet knowable.

Reading these counts against NHTSA's aggregate (the reported-tier fence)

Under NHTSA's Standing General Order, AV operators report every crash involving their automated-driving systems, a set that runs to hundreds of reports and captures minor contact a human driver would never file. Those manufacturer-reported figures are cited at a reported tier, beside the DEPLOY-verified counts above, never merged into them. The great majority of an operator's SGO-reported incidents involve a human-driven vehicle hitting the robotaxi; “in a crash” is not “caused a crash.” Our counts are smaller because every row is individually verified.