There is no reason why query should be limited to merely set operations on tabular data.
And no reason why decision support should be limited to preconfigured cached/non cached queries a la OLAP.
It is my view that from a base set of "Don't repeat yourself" (DRY) data (i.e. things that actually happened, and can't change.) then basically anything that you can logically derive from this data, is also unchangeable, and true. It doesn't matter if it didn't happen, if it's a factual derivation then it's as valid as a SQL query.
Not only does deductive inductive queries like this (I call my view of it UQ - unified query, because it allows for relational, hierarchical, object, network and inductive queries like functional (function query is induction, because the function models the nature of the man you speak Clarice) - all current databases being just a subset;) but the mode of derived query gives rise to all sorts of other paradigms such as the enterprise replicated dataset whereby derivation sets can be published, but the base data hidden. (The simplest possible example being ANPR, one such set could be the average speed of each vehicle - it's perfectly possible to have an immutable view - i.e. a derived dataset which is the average speed of each car yesterday, without disclosing any details of ANPR sitelocations. The average speed dataset is just as valid as the point to point observations and can be replicated. Other examples "people who couldn't be in London at the moment,")
I think this work is the most interesting place to be at the moment, and just wish I could find someone to pay me to do it. What I could make of SQL Server with a freehand, I just can't describe. It shouldn't be a relational database. It should be a database that among other things, also implements the relational paradigm.