Ah... the infinate wonders of statistical noise...
And because it's a tracking poll... you get two days worth of statistical noise (ok..1.414 days worth if we want to be picky) each and every day as the noise of the sample rolling off gets joined up with the noise of the sample rolling on....
And to add more noise we then reweight the entire sample too..
Ain't life grand
FWIW - I "think" Rasmussen will show Bush up 3ish tomorrow as a sample with a lot of Dems unloads, which shifts his weightings a fair bit, which adds about 1.5% to Bush... unless it doesn't...