Managing transformation as code.
WHAT THEY PROVIDE
Transformations defined as select statements, materialised as tables or views Dependencies resolved automatically Testing built in Documentation generated Version control of the whole model
WHY DEPENDENCY RESOLUTION MATTERS
Models reference each other, and the framework determines the correct build order.
WHAT THAT REPLACES
Manually ordered scripts, which break whenever a dependency is added.
WHAT MATERIALISATION OPTIONS EXIST
Views, computed at query time Tables, rebuilt fully Incremental tables, updated with new data only Ephemeral, inlined into consumers
WHAT TO USE WHEN
Views for small, cheap logic Tables where rebuild is affordable Incremental where rebuild is not
WHAT INCREMENTAL REQUIRES
A reliable way to identify new or changed rows Handling of late-arriving data Periodic full rebuild to correct drift
WHY THE PERIODIC REBUILD
Incremental logic accumulates small errors over time.
WHAT TESTING PROVIDES
Assertions about uniqueness, nulls, relationships and accepted values.
WHAT TO TEST ON EVERY MODEL
That the key is unique and not null.
WHY THAT ONE ALWAYS
Duplicate keys silently inflate every downstream aggregate.
