SchemaShield — Evidence-First Schema Change Preflight

SchemaShield — Evidence-First Schema Change Preflight

A schema change can look harmless in a pull request while breaking models and dashboards one or two hops downstream. RELAUNCH DEPT. built SchemaShield to turn catalog context into a reviewable risk decision before merge.

The public replay evaluates three deterministic schema-change scenarios and generates seven review artifacts: compatibility SQL, a dbt schema patch, an impact report, a pull-request summary, provenance, submission status, and a writeback plan. The rename scenario is classified HIGH and identifies exactly two downstream datasets.

A separate verified local round trip used DataHub OSS v1.6.0: Agent Context Kit read schema fields and performed approval-gated tag and document writebacks; the DataHub Python SDK read two-hop lineage. The hosted replay uses only synthetic data and performs no live writebacks.

Live demo: https://schema-shield.vercel.app/ Source: https://github.com/ceodaradigu/schema-shield

AI disclosure: AI-assisted development and copy. Public claims were checked against reproducible synthetic evidence; no production or private data is used.