Good catch. There must be some typo. Doesn't this mean there's a zero error rate. Seems practically impossible, at least with the current tech.

Or a single error in 638 miles.

Elsewhere in the article "disengagement" is defined as "a human driver has to take over", seemingly excluding natural stops.

> But what's even more significant is how much farther Waymo cars can go before a human driver has to take over. For Waymo, one of these so-called disengagements happened, on average, every 5,127 miles.

Unless it meant they are 2nd in miles driven and stopped "publishing" or driving to not reveal too much.

Or they've only done one test (that they'll share), and it went 638 mi before disengaging?