> I would prefer to look at it by police encounter or by crime or something like that
How would that fix anything? Look at the numbers of white vs black people who were picked for stop-and-frisks (read: number of white vs black people who were suspected of crime) as well as the innocence ratings: https://www.nyclu.org/en/stop-and-frisk-data
If 5% of a population ("Young black and Latino males between the ages of 14 and 24") accounts for 38% of the stop-and-frisk and is innocent 80% of the time, doesn't that sound like targetting? If you look harder for crime from certain demographics, you'll find more crime from them. Replace the targetted demographic with another demographic and you'll find way more crime in that other demographic. But get this - once you target people and get stats on them, you now have "statistics" and "facts" that show you were right and should continue to target them - even though your statistics are garbage collected on unrepresentative samples! This is why per-capita is the best metric to use - any other metric is tainted.
> total number of killings for white on white or black on black is roughly the same despite the 15% number you quoted.
What does this have to do with anything? Just because there are a lot of "black on black" murders doesn't justify the amount of black people killed by cops. Honestly, I don't even care if the officers are white, black, or any other race or ethnicity. It's the victims who matter. Just because an innocent black person is killed by a black police officer, does not mean it had nothing to do with the victim's race, nor does it mean we can ignore how disproportionately policing is applied.
If you adjust by per-police encounter you've already shot yourself in the foot - we know that police stop black people at a far greater percentage than white:
