Comparison
Data vaultvsMedallion architecture
Data vault
the raw layer is split into keys, relationships and attributes as three kinds of table, and nothing is ever updated or deleted.
A modelling approach that separates business keys, the relationships between them and their descriptive attributes into hubs, links and satellites, all insert-only. It is built for auditability and for absorbing source change without refactoring: a new source adds satellites rather than altering existing tables. Nobody queries it directly — the table count is enormous — so it is always a raw layer with dimensional marts built on top, and adopting it without that second layer is the usual failure.
Full entry →Medallion architecture
the same data exists three times: exactly as it arrived, cleaned and typed, and shaped for a specific report.
Layering a lakehouse into raw arrivals, cleaned and conformed tables, and business-level aggregates. The value is that each layer has one job and one class of failure, so a bad number can be traced by walking down the layers, and reprocessing can start from any of them. The failure mode is ceremony — three copies of a table that only ever gets renamed — which is why the middle layer has to actually do conforming rather than just casting.
Full entry →