Bill Inmon, author of Building the Data Warehouse (1992), defined a warehouse as "a subject-oriented, non-volatile, integrated, time-variant collection of data in support of management's decisions." His Corporate Information Factory works top down: an enterprise warehouse in third normal form integrates every source first, and dimensional data marts are derived from it for each department.
Data Vault, conceived by Dan Linstedt in the 1990s and published in 2000 (Data Vault 2.0 followed in 2013), models the integration layer for auditability and change. Hubs hold business keys (customer 1, book 2), links hold relationships between hubs (an order line links an order, a book and a customer), and satellites hold descriptive attributes with their full history, insert-only, like a Type 2 dimension split by source. New sources add satellites without remodeling, and every value stays traceable to its load.
| Aspect | Kimball | Inmon (CIF) | Data Vault |
|---|---|---|---|
| Build order | Bottom up, per process | Top down, enterprise first | Integration layer first |
| Core layer | Conformed stars | 3NF warehouse | Hubs, links, satellites |
| Strength | Fast to deliver, easy to query | One integrated model | Auditable, absorbs change |
| Weakness | Conformance discipline | Slow first delivery | Many joins; needs marts on top |
In practice the schools combine. A typical dbt 37,942 project (Transforming Data with dbt) has staging models, an optional integration layer, and Kimball stars as the marts analysts query. The medallion layers of Data Lakes and Swamps follow the same pattern.