Wisdom about interpreting polls

This DailyKos post by "DemFromCT" has a lot of wisdom about reading polls. Some of the points there (Look at all the polls, Polls are a snapshot in time) are fundamental to the work I'm trying to do here. I think my only small quibble is that I believe that a snapshot in time today can be boundedly predictive of future results, since public opinion has a measurable amount of volatility.

The statement that state polling is more difficult than national polling is surprising to me, and I'm curious why DemFromCT thinks that, or what the evidence is. Anecdotally, from things I've been involved in, I can remember both the Star Tribune's poll on the 1998 MN gubernatorial race and Terry Madonna's poll on the 2006 PA-08 congressional race being way off the mark (speaking of Terry Madonna, who died and made him Grand Guru of Pennsylvania Politics?), but the dual of "anecdote" is not "data." I actually would have guessed that a national poll or a poll in a large heterogeneous state would be more difficult than a poll in a small homogeneous state. In a place like CA, a pollster has to account for different populations of people having different access to phones, speaking different languages, having different propensities to answer surveys, etc.; while in a place like ND, there's basically just one undifferentiated population.

One thing that DemFromCT emphasizes, but that I've never paid much attention to, is the difference between LV and RV (and, god forbid, A) polls. Now that I've worked through parameter-setting, something I might try to figure out is how to measure the uncertainty implied by the difference between LV and RV results. For now, my methodology is simplistic: If a pollster only gives one result, I use it; if a pollster gives both LV and RV results, I use RV for polls before the conventions end, and LV for polls after the conventions end.