Business problem
Data Foundation & Governance
Landing ERP data in a lakehouse without losing deletes, history, or control of who reads it.
Every metric elsewhere in this library assumes the data underneath it is right, and that assumption is where large analytics investments usually fail. A hard delete in the source system leaves no row to notice, so a change-data-capture feed that only reads inserts and updates carries a deleted order forever. An inventory snapshot taken as a daily copy of the current on-hand cannot be restated once a backdated movement lands, so the history quietly disagrees with the ledger it came from. And a governed lakehouse that nobody can query is as unusable as an ungoverned one that everyone can.
This domain is guide-led by design — there is no KPI to define and no dashboard to pattern, because the work is a set of decisions made once at the pipeline layer that every downstream number inherits.
The guides cover them in the order they bite: what breaks when a source row disappears and how each ERP lets you detect it, which ingestion path preserves the history each source overwrites in place (and what the lakehouse must snapshot itself), which layer of a medallion pipeline owns each fix so a quirk is corrected once rather than per-report, the checks that catch a bad load before a dashboard does, and how Unity Catalog is laid out so governance is a property of the catalog rather than of a review meeting.
Cross-ERP guides
Modeling guides that hold across every source system rather than inside one.
- ERP Deletes & Change TrackingHow an ERP signals a deleted row — in-band soft deletes, tombstone ledgers, CDC delete operations, and delete-blind feeds — and the Databricks landing pattern each one needs.
- Source History & Ingestion PathWhich ingestion posture fits a source's change semantics — change-feed replication, watermark pulls, or snapshot-per-run exports — what a source keeps versus overwrites in place, and what the lakehouse must snapshot itself.
- Medallion Layer Contracts for ERP DataWhich layer of a Databricks medallion pipeline owns each fix ERP data needs — bronze as a faithful, reconcilable current-state copy, silver as the one place dates, decimals, codes, and keys are converted, gold as the star schema where business decisions live — with the layer contract and the misplaced-fix traps for each of the eight references.
- ERP Data Quality ChecklistThe checks an analytics team should actually run on ERP data landed in Databricks — declared-key, referential-integrity, partition, encoding, freshness, and reconciliation checks — where each belongs in a medallion pipeline, and where the thresholds come from.
- Unity Catalog Governance PatternsHow to govern ERP data landed in Databricks with Unity Catalog — catalog and schema layout for a multi-ERP medallion, ownership and grant patterns, row filters on the org columns, column masks on the fields each of the eight references flags, a working tag vocabulary, and the system tables that answer where a number came from and who read it.
Reference builds
Reference builds for this business problem, implementing the definitions above wherever one exists, running over frozen, simulated data. Not a resource and not a case study — no client data, and no client named.
The other business problems
- Inventory & Working Capital — What you're holding, why it's there, and what it costs to hold it.
- Demand Planning & Forecasting — How wrong the forecast is, in which direction, and where the error is worth fixing.
- Customer Service & Fulfillment — Whether the order arrived complete, on the date you committed to.
- Procurement & Supply — What's on order, when it lands, and how the supplier actually performs.
- Manufacturing & Production — Whether the plant built what the plan said it would build.