How do we trust robotaxi safety data?

With robotaxi services expanding to more cities, I'm trying to figure out how we should evaluate their safety record. Some companies publish disengagement reports, but I'm not sure those tell the whole story. What about near-misses or how often a remote operator has to intervene in ambiguous situations? Also, regulators are approving deployments while the software changes almost weekly via OTA updates. How can oversight keep up with that pace?