In the
data warehouse practice of
extract, transform, load (ETL), an early fact or early-arriving fact, also known as late-arriving dimension or late-arriving data, denotes the detection of a dimensional
natural key
A natural key (also known as business key or domain key) is a type of unique key in a database formed of attributes that exist and are used in the external world outside the database (i.e. in the business domain or domain of discourse). In the rela ...
during
fact table source loading, prior to the assignment of a corresponding
primary key
In the relational model of databases, a primary key is a ''specific choice'' of a ''minimal'' set of attributes (Column (database), columns) that uniquely specify a tuple (Row (database), row) in a Relation (database), relation (Table (database), t ...
or
surrogate key in the
dimension table. Hence, the fact which cites the dimension arrives early, relative to the definition of the dimension value. An example could be backdating or making corrections to data.
A gentle introduction to bitemporal data challenges - Roelant Vos
/ref>
Handling
Procedurally, an early fact can be treated several ways:
* As an error: On the presumption that the dimensional attribute values should have been collected before fact source loading
* As a valid fact, pause loading: The collection pauses whilst the missing dimensional attribute value itself is collected
* As a valid fact, load with dummy keys: A primary key value is generated on the dimension with no attributes (stub / dummy row), the fact completes processing, and the dimension attributes are populated (overwritten) later in the load processing on the new row
* Classify as a Suspense record: Assuming that the associated dimensional attribute was expected by process, move this fact record in a ''Suspense'' table and activate alert/SOPs (reporting mismatch um/count/aggr business/data steward, manual correction etc.) In rare circumstances, the suspense records may also be combined (UNION) with the fact table to ensure the metrics are correctly calculated.
References
{{Data warehouse
Business intelligence terms
Data warehousing