A daily partitioned job usually runs once a day for the day that just finished. dg.build_schedule_from_partitioned_job(job, hour_of_day=2) builds that schedule from the job's partitions: a cron at 2:00 in the partitions' time zone, and at each tick a run for the latest finished partition. At 2:00 on 9 November it requests 8 November.
The job must be partitioned: an asset job over partitioned assets takes their partitions definition. minute_of_hour sets the minute and the schedule's name is the job name plus _schedule. Like any schedule it starts stopped unless you set default_status.
Each scheduled run carries the partition as a tag, so the partition strip shows which days the schedule filled, and a missed tick shows up as a gap you can backfill.
At 2:00 every day, process the previous day.