One record per vendor, with the legal entity, the sites that ship, the bank details that are verified and the terms that apply.
Delta tables are read in place through Unity Catalog with filters and aggregates pushed down; nothing is copied out.
| Object | Read from | Fields the domain uses | Freshness |
|---|---|---|---|
| Supplier | silver.vendor | vendor_key · source_system · source_id · name · country · vat_id · iban_hash | as fresh as the landing pipeline; live for governed tables |
| Purchase order | silver.purchase_order · silver.purchase_order_line | po_id · vendor_key · order_date · net_value · status | as fresh as the landing pipeline; live for governed tables |
| Goods receipt | silver.goods_receipt | receipt_id · po_id · receipt_date · quantity | as fresh as the landing pipeline; live for governed tables |
The first object above carries the strong keys; the standardised attributes come from the same rows. A verified IBAN is a strong key; a name match with a different verified IBAN opens a review; the merge waits for a person.
Golden records and their history are written with MERGE INTO into a governed catalog, registered in Unity Catalog with lineage.
| Interface | Used for | Result |
|---|---|---|
MERGE INTO gold.vendor on the golden key; history in gold.vendor_history; registered in Unity Catalog | golden values approved for write-back: identity; sites; bank and payment | the id or error Databricks returns, stored beside the change |
Write-back is off per object until a test write succeeds, and approval rules can require a person per object. Reconciliation compares the source row with the golden value on a schedule.
Unity Catalog grants and row filters are honoured; Lima adds its own layers on top.
Supplier on
Other domains on Databricks
The golden record reads the objects above in place. History the pivot needs at scale is landed in your lake by the Backbone, under the same keys, security and audit.
Grant the read access and the write interfaces listed above and agree the refresh method. Nothing is installed in Databricks.
Yes. Write-back stays off per object until a test write succeeds; the golden view, the API and the pivots work from day one on reads.
Bring two systems that hold the same customers, suppliers or materials. In one working day we connect them, run the match on live rows, review the queue with your steward and publish the first golden view with its API.
