(1) Remove existing statistics for adapter. This is critical to support count.
(2) Run job. The Reducer creates mutations for each statistic, supporting efficient bulk writes.

The argument to the job includes the specific adapter for which statistics are being captured.