Checks as automated audit

Make the audit executable

A check is an audit question recorded in the asset. Built-in checks such as not_null, unique, relationships, accepted_values, positive, non_negative, min, max, freshness, and row_count cover common assumptions. A custom check returns a count or value, so you can test less common rules too. A blocking check stops the run when its rule fails.

The shipped reporting table has simple column checks but lacks a one-row-per-key check and a source comparison. Add checks for assumptions that can break the metric. More generic checks only add noise; a useful check names the failure it catches.

Your task

Add a grain custom check and a source-reconciliation custom check to pipeline/assets/mart/weekly_category_revenue.sql. Run bruin validate pipeline and bruin run --only checks pipeline, recording the check names and results in docs/checks.md.

Check your understanding

• Which check tests a claimed one-row-per-key grain?
• What does a blocking check do?
• Why is reconciliation valuable?

Do it with your agent

Say next lesson, add the two checks yourself, then say review my work.

Rubric

• The grain check groups by iso_week, category_name and expects duplicate-group count 0.
• The reconciliation check compares weekly mart revenue to filtered fct_order_lines revenue and expects difference 0.
• bruin validate passes and the two new checks run as blocking checks.