Baffled

I am epistemically baffled by the failure of contemporaneously given polls to provide a consistent result. Polls taken over the same window show Thompson up by one and Giuliani up by about ten. I remember a period after the GOP convention in 2004 when a bevy of polls were failing to coalesce around any reasonable margin in the presidential election.
I can understand that certain periods may pose more difficult problems for creating an appropriate model, but the explanation I remember from that period was that wildy unstable polling was common during periods of transition, when the numbers were moving.
What? But there was, on any given day, a certain number of real Americans leaning to Bush or Kerry, right? And if it were moving from a reality of Kerry +5 to Bush +5, why wouldn't the values in the middle occur on every day in between?
Kina gives me chills, people.