They appear to have correctly filtered out the "Non-Shooting Incident" sort of results from it (which your picks appear to mostly be) to arrive at the ~900 numbers.
I've also been looking at the export, but I don't see any way to filter down to "Shooting incidents". Just sampling the first 10, 10% of 2000 is looking like way less.
Were you able to find a way to validate the 900 number?
They appear to have correctly filtered out the "Non-Shooting Incident" sort of results from it (which your picks appear to mostly be) to arrive at the ~900 numbers.