Change-impact decision engine · for dbt

Know what breaks —
and what to rebuild.

A change-impact decision engine for dbt. It categorizes breaking vs cosmetic changes, gates PRs on your policy over any dbt meta, and follows impact past dbt’s edge into your BI tools — offline, from your artifacts, no warehouse, no dbt run. Built on column-level lineage.

1breaking change
1exec dashboard hit
BLOCKpolicy verdict
impact · account_holder
raw_accounts stg_accounts accounts raw_transactions stg_transactions transactions finance_dashexposure
The problem

Change one column in a large dbt project and you’re guessing. Which models recompute? Which dashboards break? Should CI even let it merge? Without column-level impact, every refactor is a risk you can’t measure — and every CI gate is all-or-nothing.

The decision engine

From blast radius to a decision.

Breaking, or just cosmetic?

Diff the SQL expression, not the text. A provably-equivalent refactor doesn’t block; a change that shifts meaning — or can’t be proven safe — fails safe.

Your rules, your gate.

Write rules over any dbt meta. The tool ships the engine; you ship the policy. Block, warn, or schedule a selective rebuild — nothing about your taxonomy is hardcoded.

Past dbt’s edge.

Follow a column change into your BI layer — which dashboard, which field — so “what breaks” includes your BI, not just your models. Metabase is the first supported connector.

All surfaced in an interactive explorer and as machine-readable JSON for your agents — on top of the column-level lineage it’s built on. New here? Start with how it works.

Quick start

One pip install. No dbt run required.

Every command reads your manifest.json and catalog.json — offline, zero-credential, it never touches your warehouse.

bash
# install
$ pip install dbt-col-lineage

# explore lineage + impact in the browser
$ dbt-col-lineage --explore

# turn a PR into a decision: gate on your policy
$ dbt-col-lineage impact --base-manifest base/manifest.json \
    --policy policy.yml --fail-on policy
dbt-col-lineage MIT · a decision engine for dbt teams who ship with confidence