LineageShield

Deterministic BLOCK decision with visible evidence
Submit a proposed schema change
Affected assets enriched with DataHub metadata
Auditable Agent Context execution and lineage fallback
Generated migration, compatibility, testing, and rollback safeguards
Disabled-by-default write-back preview
Verified DataHub write-back receipt

Inspiration

A database column rename can look like a tiny change in a pull request, yet quietly break dashboards, scheduled pipelines, transformation models, and production features.

The problem is not that lineage information is unavailable. Platforms such as DataHub already contain much of the required context. The problem is that reviewers must manually search through lineage graphs, metadata, ownership, quality signals, and downstream dependencies while deciding whether a change is safe to merge.

We built LineageShield to turn that scattered context into one clear, explainable pre-merge decision:

Know what will break before you merge.

What it does

LineageShield is a DataHub-powered schema-change review agent.

A developer proposes a column operation such as:

Rename a column
Drop a column
Change a data type
Add a column

LineageShield then:

Resolves the affected DataHub asset and schema field.
Checks downstream column-level lineage first.
Transparently falls back to dataset-level lineage when fine-grained lineage is unavailable.
Enriches affected assets with available ownership, platform, schema, tags, glossary terms, structured properties, and quality evidence.
Calculates a deterministic risk score.
Returns an explainable ALLOW, REVIEW, or BLOCK decision.
Identifies required owner approvals.
Generates review-only migration safeguards.
Optionally previews a tightly controlled DataHub write-back.

Every point in the risk score is visible. Missing metadata remains unavailable rather than being invented, and an AI-generated explanation cannot override the deterministic decision.