When a tick needs logic, write a function with @dg.schedule(cron_schedule=..., job=...). It receives a dg.ScheduleEvaluationContext; context.scheduled_execution_time is the tick's time in the schedule's time zone, which is the right "now" to compute dates from.
Return a dg.RunRequest to launch the job, with run_config for per-run settings and tags for filtering. Return dg.SkipReason("why") to skip: the tick is recorded as skipped with your reason, which is far easier to debug than silence. You can also return a list of run requests.
A run_key makes a request idempotent: if the same schedule already launched a run with that key, the request is skipped. Using the processed date as the key means that evaluating the same tick twice never loads the same day twice.
At each 6:00 tick, load the previous day.