Altimate dbt PR Review Demo

A small, zero-secret dbt + DuckDB project that demonstrates Altimate dbt PR Review.

Every demo pull request compiles successfully. The point is to show risks that ordinary compilation misses: semantic changes, join fan-out, test removal, PII exposure, warehouse-cost regressions, and unsafe incremental configuration.

Run Locally

python -m pip install "dbt-duckdb>=1.8,<2.0"
dbt seed
dbt build

Review a local branch with Altimate:

altimate review --base main --head HEAD --manifest target/manifest.json

No Altimate account or model key is required. The optional LLM reviewer is disabled in the GitHub workflow; all launch-demo findings are deterministic.

Demo Pull Requests

Branch	Scenario	Expected signal
demo/safe-refactor	Rename and reorganize CTEs without changing behavior	APPROVE or non-blocking comment
demo/join-key-breakage	Join orders to customers on the wrong key	semantic/join risk
demo/test-removal	Remove primary-key tests from a mart	contract/test warning
demo/new-pii-exposure	Add customer email to a public mart	PII/lineage risk
demo/mart-select-star	Replace explicit mart columns with SELECT *	SQL quality/cost warning
demo/incremental-without-guard	Convert a model to incremental without a guard	materialization warning

The scenario definitions live in demo-cases/cases.yml.

Security

The workflow runs on pull_request, not pull_request_target. It uses DuckDB and synthetic CSV seeds, so it needs no warehouse credentials or GitHub secrets. The review action receives only compiled dbt artifacts and the git diff.