Not doing this right now may quickly result in a vast catalogue of heterogeneous routines, transforming Boost.Algorithm into Boost.HereBeDragons.
Tagging is a more flexible way of organizing information than an ontology. Searching becomes the preferred way to find information as the domain grows. To me it is also desirable to able to cherry pick the pieces instead of being forced to adopt the whole boost library.
I have not thought about different algorithms categories and I agree it could be hard (and useless ?) to categorize all the submissions that lands into Boost.Algorithm. Nevertheless, you will most likely need to group similar algorithms as they will certainly use the same basic functions. For sorting this is pretty obvious. The STL toolbox is limited to that aspect considering the large amount of existing strategies. Boost.Algorithm.Sorting could be that toolbox enabling you to arrange strategies to perfectly match your need (and of course providing default choices covering most of the usual cases). -- EA