A warehouse pipeline needs both data checks and a clear output contract. The snapshot and change feed in this practice set let you rehearse those checks without writing to Lumen's source datasets. The feed contains repeated customer keys with different revisions, so it must be reduced before an upsert.
Use metadata when the question is about columns or table design. Use source queries when the question is about row values, coverage or duplicate keys. Those are different kinds of evidence. A table can have the right schema and still contain repeated transactions.
The final tasks build a selective order table and verify its physical design. Read the pre-run estimate before executing the source query. All writable outputs live in practice. Keep key counts visible while you work, and make the chosen revision rule explicit rather than depending on the order in which rows happened to arrive.