Checks tell you the data is right; freshness tells you it is recent enough. Since Dagster 1.12, dg.FreshnessPolicy is the way to say it, on the asset: @dg.asset(freshness_policy=...).
dg.FreshnessPolicy.time_window(fail_window=timedelta(hours=24), warn_window=timedelta(hours=12)): the asset must have been materialized within the last 24 hours, and turns amber after 12. The warn window must be shorter than the fail window.
dg.FreshnessPolicy.cron(deadline_cron="0 9 * * *", lower_bound_delta=timedelta(hours=6)): by each 9:00 deadline, there must be a materialization from the six hours before it. Use it when a report is due at a set time.
The asset's health shows PASS, WARN or FAIL (or unknown before the first materialization), and alerts can follow. A freshness policy does not launch runs; it describes the promise, and schedules or automation keep it.
Amber after 12 hours, red after 24.