Investigate a failure

Investigate one hypothesis at a time

The expected symptom is a failed check on the share of customers marked at risk of churn. Start with the error, then use bruin lineage, bruin render, a query against the earlier table, and the logs. The root cause is one filter: stg_orders drops rows whose status is NULL. Those rows include recent orders, so the latest order date moves backward and too many customers appear to have stopped buying.

A diagnosis is complete only when one query confirms the upstream condition and the repaired run passes the same named check. Do not weaken or remove the check.

Your task

Investigate mart.churn_risk in this order: read the failed check, inspect the pipeline lineage, render the SQL, run one query against the earlier orders table, read the logs, then make the smallest fix. Record each hypothesis, command, result, and decision in docs/failure-investigation.md. Do not change a file until the diagnosis is recorded. Rerun validation and the named check after the fix, then record the result.

Check your understanding

• What should you read first in a failed run?
• What is the single root cause of this shipped failure?
• What proves a proposed fix worked?

Do it with your agent

Say next lesson, diagnose before editing, make the smallest fix, then say review my work.

Rubric

• Names churn_risk, the churn-share check, and the exact surfaced failure.
• Names the filter on NULL order_status as the root cause and traces all three hops to the symptom.
• Includes the single confirming upstream query and shows the repaired named check passing.