Mark Liberman finds a site, Onlinestylebooks.com, that lets you search 43 stylebooks at once:
The whole thing here. (HT John McIntyre's You Don't Say)
Surveying this explicit variety of sources may help to avoid an otherwise-natural confusion. These stylebooks are not about the nature of (the formal written variety of) the English language: each of them documents (aspects of) one organization's policy about how to represent this language in writing, typically covering a limited set of cases that are both reasonably common and somewhat variable in general practice.
Some people are tempted to treat policies of this kind by analogy to the theological differences among religious sects. Believers are convinced that one set of policies is (or should be) God's Truth, with the others to be consumed in the fires of hell, while skeptics think that they're all just different forms of nonsense.
Both of these attitudes seem to me to misread the situation. Once you decide, for whatever reason, that representational consistency is a Good Thing, then you need to deal with the Long Tail of Linguistic Complexity. It's tempting to think that there are a few basic axioms from which the Right Answer could always be logically derived for any question of linguistic analysis — and perhaps some day a future linguistic Peano will give them to us. But as things stand, questions of linguistic representation are more like common law than like set theory. The only known way to achieve reasonably consistent results is to reason from a very long list of precedents, which is always in the process of gradual development, with occasional major revisions. This rational catalogue of worked examples is meant to be consistent with a hierarchy of more general principles, but it's not reducible to them.