Imagine you’re a person who reviews cars for a living. You’re part of a website that takes new vehicles, tries them out for various periods of time, and reports on the Internet about what you found.

Now imagine you’re driving a $155,000 luxury SUV with manufacturer license plates on it. These plates are different from regular license plates, but look very similar. (A cop should know the differences, but as you’ll see, not all of them do.) You’re in Minnesota, and because someone in California entered a very similar plate into NCIC as stolen, you get pulled out of the SUV, not quite at gunpoint, but very nearly so.

As you’ll see in this video, the police figure out that it was the wrong plate, but it took a lot more work to prove that than it took to alert them that this stolen plate/vehicle was in the Kohl’s parking lot.

In the video, a Flock executive says, paraphrasing, that since their error rate is less than 1%, it’s all good. Any mistakes aren’t their fault. It’s the cops. (20 billion plate reads a month translates to 200 million errors in that 1%. Two hundred million chances to get shot by a cop during a felony stop. For a mistake.)

This incident ended without an arrest or bloodshed. This time.

In California, a license plate can be entered into CLETS (the California Law Enforcement Telecommunications System) as “lost or stolen”, and then CLETS in turn enters that into the NCIC (National Crime Information Center, ran by the FBI). That entry, which is different from a stolen vehicle entry, does not indicate directly if the plate is lost, or has been stolen. A police officer in Minnesota may not know that a return of “lost or stolen” on the plate doesn’t mean the vehicle was stolen.

Two hundred million “errors” means it’s only a matter of time before someone is killed because of an “error” in a system that should never have been allowed to become so pervasive. Indeed, it may have already happened, and it’s been covered up.

Two hundred million chances a month.

That’s an inevitability.