Every scheduled run has a logical date and a data interval (data_interval_start to data_interval_end). In Airflow 3 a cron string or preset uses CronTriggerTimetable: the run for a tick happens at the tick, and its logical date, data interval start and data interval end are all that tick. So the @daily run at midnight on 9 November has ds = 2026-11-09.
When a run should cover a period, give the timetable an interval:
from datetime import timedelta
from airflow.timetables.trigger import CronTriggerTimetable
schedule=CronTriggerTimetable("0 1 * * *", timezone="UTC", interval=timedelta(days=1))
The 01:00 run on 9 November then covers 8 November 01:00 to 9 November 01:00, and tasks read data_interval_start and data_interval_end to load exactly that window.
Airflow 2's behavior lives on as CronDataIntervalTimetable: a run per interval, started once the interval has ended, with the logical date at the start, so ds is the day before. Dags upgraded from Airflow 2 often notice this change first.
The 01:00 run on 9 November loads 8 November, the day its interval covers.