A lot of the blogsphere articles related to CQRS (command query repsonsibility) seperation seem to imply that all screens/viewmodels are flat. e.g. Name, Age, Location Of Birth etc.. and thus the suggestion that implementation wise we stick them into fast read source etc.. single table per view mySQL etc.. and pull them out with something like primitive SqlDataReader, kick that nasty nhibernate ORM etc..

However, whilst I agree that domain models dont mapped well to most screens, many of the screens that I work with are more dimensional, and Im sure this is pretty common in LOB apps.

So my question is how are people handling screen where by for example it displays a summary of customer details and then a list of their orders with a [more detail] link etc....

I thought about keeping with the straight forward SQL query to the Query Database breaking off the outer join so can build a suitable ViewModel to View but it seems like overkill?

Alternatively (this is starting to feel yuck) in CustomerSummaryView table have a text/big (whatever the type is in your DB) column called Orders, and the columns for the Order summary screen grid are seperated by , and rows by |. Even with XML datatype it still feeel dirty.

Any thoughts on an optimal practice?