Chaining source freshness and build

Source freshness is the first thing /run_transformation does. If raw is older than the threshold, the build does not run. Better to skip the build than to publish dashboards built on incomplete data.

The pattern is intentional: freshness is a soft gate (warn) for some sources, a hard gate (error) for others. Hard-gate sources block the build. Soft-gate sources log a warning that surfaces in dbt docs and Slack.

No. dbt produces a manifest of failed models. dbt build --select result:error reruns only the failed nodes. Combined with on-failure callbacks in Airflow, this is how you turn intermittent failures into self-healing pipelines.