A dbt Model Silently Redefines 'Active Customer' — How Do You Catch It Faster?

Three weeks after a dbt pull request merged, someone in Finance notices that "active customers" on the executive dashboard has been meaningfully lower than expected since the merge. Investigation shows an analytics engineer's PR narrowed the definition of is_active in dim_customers (from "ordered in the last 90 days" to "ordered in the last 60 days") as an incidental side effect of an unrelated refactor, and it passed code review and every existing dbt test.

1. Explain why the existing dbt tests described in the case study (not_null, unique, accepted_values, relationships, row-count reconciliation) did not and could not have caught this.
2. Propose the specific tests or process changes that would have caught it, and say at which layer/stage each one operates.
3. Given the ownership model in the case study, whose responsibility was this gap, and what does that imply about the PR review process going forward?