LINEAGEPATCH

Turn breaking data changes into evidence-backed, validated patches.

Agent path

1 · Read DataHub contextSchema, field lineage, ownership, governance, quality, and documentation.
2 · Calculate blast radiusOne direct and five indirect consumers; each score factor links to evidence.
3 · Build Repair ContractSelect a reversible compatibility alias before generating code.
4 · Generate in isolationPatch buyer_id as customer_id plus governed schema documentation.
5 · Prove the repairThe old dbt project fails; parse, compile, and build pass after repair.
6 · Require human authorityDataHub write-back remains a dry-run until review and explicit approval.

Proof carried with the patch

DataHub MCPOfficial read tools: get_entities, list_schema_fields, get_lineage.
DeterministicStable run IDs, policy scoring, templates, and SHA-256 artifacts.
ValidatedSQL, YAML, secrets, paths, dbt parse, compile, and build.
BoundedRead limits, isolated paths, no shell, dry-run writes by default.
Evidence receipt readyrun_d92c414807fc5f712ac3 · Apache-2.0
Honest boundary: the public fixture proves the end-to-end product. The SDK and official MCP paths are mock-tested; a live server round trip is not claimed on the Docker-less build host.
Build with DataHub — The Agent Hackathon 2026No production metadata was changed by this demo.