Consider a grocery store scenario (I'm making this up) where you have FACT records that represent a sale transaction, where the columns of the Fact table include
SaleItemFact Table------------------CustomerID ProductID Price DistributorID DateOfSale Etc Etc Etc Even if there are duplicates in the table when you consider ALL the keys, I would contend that a surrogate running numeric key (i.e. identity column) should be made up, e.g., TransactionNumber of type Integer.
I can see someone arguing that a Fact table might not have a unique key (though I'd invent one and waste the 4 bytes, but how about a dimension table?
