Everything Britain knows about how often these cars need a human comes from two journalists on their first ride

On Thursday morning fifteen cars began carrying paying passengers through central London with nobody driving them, and a licensed human sitting in the driver's seat of each one in case there was.

The question that decides everything else about this is simple. How often does that person have to act?

Nobody in Britain is required to know.

Two journalists rode on launch day and both wrote it up. The BBC's technology editor, Zoe Kleinman, described a trip that handled traffic near a demonstration outside Parliament, navigated roadworks, and yielded correctly when a cyclist waved the car forward at a mini roundabout. Then the rear doors of a parked van appeared to swing open on a narrow side road. The car braked hard, an internal alarm sounded, and the safety driver took the wheel. Sky's technology correspondent, Rowland Manthorpe, took his own ride the same day and also reported the driver acting.

Neither of them was counting. Neither piece claims to be a measurement, and neither should be read as one. Two journeys is not a sample of anything.

But it is, as far as this desk can establish, the entire published evidence base in the United Kingdom on how often a person has to take over from this system. Not the best evidence. All of it.

Now consider California, which licensed the same category of technology and made a different choice.

Since 2015, every company holding a permit to test autonomous vehicles on Californian public roads has been required to file an annual disengagement report with the Department of Motor Vehicles, due each January. The report states how many disengagements occurred, in what circumstances, at what locations, and across how many miles driven in autonomous mode. A disengagement is defined to cover both a failure of the technology and the safety driver taking immediate control. The DMV publishes the reports. It suspends the permits of companies that fail to file.

The most recent reporting period covers 4,498,066 miles, of which 3,945,171 were driven with a safety driver aboard and 552,895 with nobody in the seat.

The DMV attaches its own caution to all of it, and the caution is worth repeating: the reports are not intended to compare one company against another, or to support broad conclusions about how good the technology is. Companies count differently, test in different conditions, and a low number can mean a cautious route rather than a capable machine.

That caveat matters, and it does not touch the point. California's data is imperfect. It exists. Eleven years of it, published, from every permit holder, with a penalty for silence.

Britain has none, and the reason is not oversight. It is the route chosen. Transport for London licensed these fifteen vehicles as private hire vehicles, under the same regime that governs minicabs. Private hire licensing has no concept of a machine handing control back to a person, and so imposes no duty to record it, because when those rules were written the situation could not arise.

The frame that would ask is the Automated Vehicles Act, and it does not apply until an operator holds a permit. Neither Uber nor Wayve holds one. Uber's head of autonomous mobility said last week that the process is still playing out, and would not put a date on it.

So the vehicles are running under a regime that structurally cannot ask the question, and will not be running under one that can until somebody grants a permission nobody will schedule.

We have asked both companies whether they log the driver taking control, how they classify it, and whether they will publish a count. We have asked Transport for London, by freedom of information, whether it receives one.

Wayve's position on the one documented instance is already on the record: it says its system would have coped and the driver chose to act. That may well be right. There is currently no way for anyone outside the company to know.