One bias I have, and admit to, is for database knowledge as it applies to interactive CRUD applications. That is what I know. I have done enough reading to know there is a whole different world out there called OLAP, which is where you are coming from here. You have brought the point home. In a static database used for reporting, slicing and dicing, the rules of referential integrity do not apply. The data is fixed and the issue shifts to accessibility and performance.

That is very true, there's an OLTP world and an OLAP world, and things would certainly be cleaner and easier if they were two distinct realms.

But for every one environment where you have a clear distinction and a series of ETL processes in between, you have "Y" number of hybrid environments that evolved over time because the organization didn't have the time or money or in-house expertise to build a textbook environment.

That's one of the (many) reasons for the success of products/tools like PowerPivot, QlikView, Vizubi, and others. These products don't have a distinct identity - they are quasi-fringe products, quasi-platforms for quasi-environments.

The concept of "static database" is becoming somewhat obsolete.