Tensor Auto vs WeRide
These two cannot be compared on crash rate. 5 of 5 conditions fail, and the reasons are more informative than the comparison would have been. Both are measured on the 2026 Q2 filings, against California driverless mileage.
Comparability
Five conditions must hold before two operators' crash rates mean the same thing.
| Condition | What the record shows | |
|---|---|---|
| Both have reports in the period | Tensor Auto filed no reports in this period, so there is nothing to rate. | |
| Both have published exposure | Tensor Auto publishes no mileage anywhere, so no per-mile rate exists for it. This is a disclosure gap, not a safety result. | |
| Divisors measure the same thing | Not testable: Tensor Auto has no exposure figure, so there is no second quantity to match against. | |
| Each rate covers most of its record | Only 0% and 0% of reports are rateable. A rate covering half the record cannot be set against one covering all of it. | |
| Both file crashes the same way | One operator has too few filings to characterise its reporting practice, so any difference in counts could be practice rather than events. |
The two rates
Shown because withholding them would be its own distortion, but read the verdict first.
no rate computable
No CPUC driverless deployment VMT filed for this operator and period. Source: https://www.cpuc.ca.gov/regulatory-services/licensing/transportation-licensing-and-analysis-branch/autonomous-vehicle-programs/quarterly-reporting
no rate computable
No CPUC driverless deployment VMT filed for this operator and period. Source: https://www.cpuc.ca.gov/regulatory-services/licensing/transportation-licensing-and-analysis-branch/autonomous-vehicle-programs/quarterly-reporting
Do not put these two numbers in a sentence together. Tensor Auto filed no reports in this period, so there is nothing to rate.