No two of these projects report the same number. Here is the thing: none of them is wrong.
When you hear that the databases "disagree," you probably picture a mistake. Somebody typed the wrong number. That is almost never what happened. Each project wrote down what it meant by a police killing, then counted honestly against its own rule. The rules are different, so the totals are different. Here are all five, in their own words.
Someone shot and killed by an on-duty officer. That is it. No deaths in custody, no off-duty shootings, and nothing that wasn't a gunshot: no Tasers, no restraint, no cars.[src]
Every person police kill, by any method, including off-duty officers and people who die in custody. Wider than shootings, and still going.[src]
Any death during contact with police, from any cause. The widest net of the five, going back to the year 2000, and the one that got the others started.[src]
Any use of force that kills, badly injures, or points a gun at someone. In theory the widest of all, but reporting is optional, so the total stays sealed until enough departments take part.[src]
Every death at the hands of law enforcement, guns, Tasers, cars, custody. A two-year project that ran alongside the early Post count.[src]
Picture those five rules as circles, one inside the next. A fatal shooting is one kind of killing. A killing is one kind of death in a police encounter. A death in an encounter is one result of a use of force. Each project drew its circle at a different ring, for its own reason. The Post wanted a clean count it could check one shooting at a time. Fatal Encounters wanted everything and lived with the mess. The FBI drew the biggest circle and caught the least.
So there is no single "number of people police kill" sitting out there waiting to be found. There are five careful instruments aimed at five slightly different targets. The space between their answers isn't a bug. It's the whole subject of this series, and the next essay turns it into a number you can hold onto.