Is culling child nodes a valid WURFL search strategy?

I've implemented a WURFL based detection routine based on a similar strategy to the two phase one outlined at http://wurfl.sourceforge.net/newapi/ .

This is working well but I would like to improve the worst case scenario if I can.

In the worst case scenario, at the moment, every device's user agent string is compared against the current user agent string.

What I'm curious about is how valid it would be to search the tree of devices and cull entire branches where device matches don't a minimum match threshold? (Obviously ignoring 'root' devices that don't have user agent strings intended for matching)

Do user agent strings tend to follo开发者_如何学Gow a general pattern of ever closer matches as one decends down the tree... and thus make the aforemention strategy valid? ... Or are user agent strings a completely random beast in terms of parent verse child device matches and I really am forced to search the entire tree every single time?

WURFL-Pro, the company beside wurfl project, has a dual licensing strategy. You can ask them to obtain a GPL-free library.

For know the wurfl java api implementation you can just browse the source available on svn...or ask to the authors.