[AccessD] OT: The Great Primary Debate

For practical reasons, I prefer Robert Stewart's Surrogate/Business key
approach.

Regarding Celko's emphasis on natural keys, I would suggest the
counter-example of catalogs, where many disparate items, each with their
own, non-overlapping attributes, are given unique codes. Putting an ID on a
blob of data is a common real-world thing to do. If that ID is random, or is
a counter, you are very close to the Autonumber concept.

On the other hand, the Autonumber is supposed to be a 'meaningless' unique
ID. But, consider an auto-generated date dimension table, with consecutive
date records. Think an Autonumber key is meaningless in this situation?
Think again - it really represents the number of days since Day Zero (the
earliest date record in your table). Sometimes, the Autonumber IS a natural
key!

-Ken