Pipeline · hard · ~14 min
Platform cost review: total warehouse spend is up 3.1x month-over-month, but the account-level dashboard shows only 'compute grew'. Two hundred DAGs share the warehouses, queries carry NO tags, and nobody can say WHICH pipeline is responsible. Six engineer-hours of log archaeology finally find it: daily_customer_snapshot — a MERGE whose runtime went 20 min → 4.5h after its source doubled AND its schedule was accidentally changed from daily to hourly in a refactor.
daily_customer_snapshot
Quantify the damage, then build the attribution system that makes the next one a 10-minute page instead of a 6-hour dig.