Moreover, it varies among the daughter languages; the proximate-obviative distinction seems to have begun in Algic as some kind of topicality or definiteness distinction, which morphed into something more complex, as you describe. In my own Alqonquian-style conlang family, I had it descend from grammaticalized definiteness. But such complexity is the rule with natlangs, I think. Consider, likewise, the Northern Iroquoian gender system, which does many of the same things (disambiguation, e.g.) as the obviative in Algic-Algonquian, but is even more screwed up. Native speakers don't seem to like neat, tidy gender or pronominal systems; as I said in one of my LCC talks, they like having a mess to play around in. If you have a system whose distinctions are vague, and can be bent at will to suit discourse, you can express so much more than with a rigid system. Harder on linguists, to be sure, but oh well... Now in the 3rd person-3rd person, all of a sudden, there doesn't seem to be any agreement about what gets the obviate marker and how to read who did what to who. (I've read it's about topicality, about evidentiality, about pragmatics, about the speakers knowledge or lack of it about the internal mental states of the other people, etc) This makes ergative-absolutive look easy.