// the take
A dbt package that hooks into dbt's on-run-end lifecycle to capture artifacts, run results, and metadata into warehouse tables, then runs anomaly detection tests on your data as native dbt tests. It's the open-source core of the Elementary commercial product, designed for analytics engineers who want observability without leaving the dbt ecosystem.
The hook-based architecture is genuinely clever — zero separate pipeline, your warehouse is the observability store. Cross-DB support is thorough: Snowflake, BigQuery, Redshift, Databricks, Spark, Trino, ClickHouse, and more, each with dedicated macro implementations. The anomaly detection tests (volume, freshness, column distributions, schema changes) run as first-class dbt tests, so they show up in your existing test infrastructure rather than a parallel system. Integration test suite covers real warehouses via Docker Compose, not just mocked SQL.
The package is a feeder for a commercial product, and that tension shows — the free tier gives you tables and tests, but actionable alerting and lineage push you toward Elementary Cloud. Anomaly detection requires sufficient history to baseline, so it's effectively useless for the first few weeks on any new model. The macro layer is massive and sprawling; debugging a failure in the SQL generation chain means navigating hundreds of Jinja files, and the cross-DB abstraction adds an extra indirection layer that makes SQL errors hard to trace. No way to tune sensitivity per-test beyond the config vars — if your data has seasonal patterns, you'll get noise until you configure `seasonality` everywhere manually.
We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →