Highlights
- Declare the grain before anything else — Adamson makes grain the non-negotiable first act of every fact table design; it is a modeling rule in the purest sense, and getting it wrong invalidates everything generated downstream.
- The fact-table catalog is finite — transaction, periodic snapshot, and accumulating snapshot cover essentially all measurement shapes; a closed catalog of patterns is exactly what makes dimensional marts a prime target for model-driven generation.
- Slowly changing dimensions are enumerable responses to change — the SCD types form a decision table (what happens on change of attribute X?) that can be captured as metadata and generated, rather than hand-coded per dimension.
- Conformed dimensions are the enterprise consistency contract — shared dimensions are what let separate stars add up to one truth; without that agreement layer, "self-service BI" fragments into silo arithmetic.
- Bridges, hierarchies, and multi-valued cases have standard resolutions — even the awkward edge cases resolve to documented patterns, reinforcing that dimensional design is craft applied within a pattern language, not invention.
- The star is a presentation layer, not the archive — Adamson's marts answer questions fast; keeping full auditable history is a different job (the Data Vault side of the house), and a healthy architecture generates the star from the vault rather than choosing between them.
Highlights on this page are generated with the help of AI.