One record per item: the global identity, the descriptions, the classification, and the plant-level data that planning and purchasing need.
Today one part carries a different number in each plant's ERP, a manufacturer part number in engineering and a SKU in the web shop, and three descriptions.
With Lima the materials are read where they live, matched on GTIN and manufacturer part number and manufacturer, assembled field by field and published as one golden record per real material, with every source key underneath. Orders, stock, receipts, production orders and forecast rows for the item at every plant, and the BOMs it appears in are one click away, because the transactions sit in the same bed.
Survivorship is per field; the winning source below is the usual one, and a steward's verified edit outranks it.
Attribute-based matching compares the physical attributes (dimensions, material, rating) so two descriptions of the same resistor match even when the words differ.
Pairs above the merge line merge; pairs between the lines go to the steward with the two records side by side and the transactions of each; the rest stay apart. Every merge can be undone. The golden records page explains the steps.
Several structures for the same record, each kept as versions with dates.
| Hierarchy | Levels and use |
|---|---|
| Product hierarchy | family → group → product, for sales and planning |
| Commodity | UNSPSC or category tree for sourcing |
| Engineering | assembly → sub-assembly → component through the BOM |
| Plant | which plants use the item, with their data |
A pivot can roll the record's measures up any of them; a draft version is validated against the facts before it goes live. See hierarchies.
Written in one language, tested on live rows before they are saved.
| Level | Rules |
|---|---|
| Column | GTIN passes the check digit base unit is in the code list net weight is smaller than gross weight |
| Row | an item sold has a sales unit and a tax classification a made item at a plant has a BOM and a routing there a phase-out item has a successor or an end date |
| Business | one active item per manufacturer part number safety stock at a plant is at most 12 weeks of average demand or carries an approval every item in an open order is active at that plant |
| Transactions in the bed | Read from |
|---|---|
| Sales order lines | SAP VBAP; Oracle order lines |
| Purchase order lines | SAP EKPO; Oracle PO lines |
| Stock | SAP MARD · MATDOC; Oracle on-hand |
| Forecast and demand | planning systems |
| Production orders | SAP AFKO · AFPO; Oracle work orders |
Measures above are available in the pivot on any hierarchy level, by month or week, with calculated measures beside them. Drill-down opens orders, stock, receipts, production orders and forecast rows for the item at every plant, and the BOMs it appears in.
Cost and purchase prices masked for sales roles; engineering attributes read-only outside engineering; plant data editable by the plant's planner. The twelve layers apply on every door; see security.
POST /api/v1/golden/material/lookup
{
"manufacturer": "Panasonic",
"mpn": "ERJ-3EKF4700V",
"description": "resistor 470 ohm 1% 0603"
}200 OK
{
"golden_id": "M-1183920",
"confidence": 0.96,
"xref": {
"sap_1000": "7741-X",
"sap_2000": "R-470-0603",
"oracle": "200001941"
},
"plants": 4,
"stock_on_hand": 128400
}The same call creates the record when nothing matches; related facts are expanded through the cross-reference; page sizes and limits are published per endpoint. See the headless API.
The objects with their real names, the keys in that system's fields, the write interface and what to set up first.
Material general data read from MARA · MAKT.
Open the SAP S/4HANA page →Item read from EGP_SYSTEM_ITEMS_B (REST: itemsV2).
Open the Oracle Fusion Cloud page →Product read from Product2. 2 objects, the keys in Salesforce fields, the write path and the prerequisites.
Open the Salesforce page →Material read from CONFORMED.MATERIAL. 3 objects, the keys in Snowflake fields, the write path and the prerequisites.
Open the Snowflake page →Material read from silver.material. 3 objects, the keys in Databricks fields, the write path and the prerequisites.
Open the Databricks page →Names, places and figures are illustrative or anonymised; the pattern is the catalogue's.
A $75B semiconductor equipment maker mastered 1.4 million material records across plants ahead of a planning-system integration. 61,000 duplicates by manufacturer part number were merged, and the planning run stopped double-counting demand on them.
The golden record is a governed view over rows read in place; a governed replica in your lake is used only where a source is too slow or too limited to query live, and the column says so.
One working day for the first two sources; 6 working days to a production-grade version with rules, hierarchies, security and write-back tested.
High-tech and semiconductor, Industrial manufacturing, Consumer goods, Life sciences. The industry pages show the domains each starts with.
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.