New: Cookbooks and AI ExplanationsStep-by-Step recipes to solve problems connected to Roadmaps and Cheat Sheets. Need more details? Use AI buttons for structured and simple explanations with concrete examples throughout the whole platform.Take a look
A bug in last month's transform means three weeks of numbers are wrong, and today's job can't stop while you fix it.
What you'll have at the end
The three corrected weeks reloaded, with today's totals unchanged before and after
You need
A corrected version of the transform, already tested and proven right on at least one already known bad day, plus a target table split into one physical partition per date, so a single date's rows can be replaced without touching any other date.
Not covered
Working out why the original transform computed the wrong number in the first place: this recipe assumes that diagnosis and fix are already done, and picks up from replacing the wrong history with the corrected one.
Leans on
Your transform looks right until the next edit breaks it
go there first if the fix itself hasn't been proven correct yet; this recipe assumes it already has and starts from replacing history.
Reconcile every row between source and target after a load
reach for its full row by row technique if the totals check here isn't enough to convince a stakeholder every row landed correctly.
Make a pipeline safe to rerun with an upsert
go there instead if the wrong history came from a pipeline that was never made safe to rerun, rather than from a transform bug.
Checked 25 Aug 2026
Part of the ETL (Extract, Transform, Load) cookbook