Malcolm Gladwell, author of Blink, helps one understand variability and how grouping things into clusters can be used to better understand which is the best pickle and what the best spaghetti sauce is. The answer is that there is no best pickle or spaghetti sauce. But there are best clusters of pickles and spaghetti sauces. Watch this TED video to understand how to meet peoples needs.
So, I am sure you are asking what the heck this has to do with XBRL. Profiles of XBRL is basic a clustering of user needs.
There is no one right approach to using XBRL. It is also true that each system making use of XBRL and inventing their own specific approach to using the general functionality which XBRL provides. Each system creating its own architecture is simply too inefficient.
The middle ground is to cluster needs of a system. Is extensibility needed? If yes, then you take the pieces from XBRL which cause the least issues relating to extensibility. You wrap your set of characteristics into a bundle, that is your profile. Or, rather than building your own XBRL profile, even better is using someone else's XBRL profile which is known to work well.
Watch the video and you will better understand how clustering user needs into profiles can help you see XBRL just a little differently.