The issue of whether to hold partisan percentages constant, or allow them to flip, is an interesting. I tend to be an anti flipper type, but which is best, I really have no idea. Pity that the final polls, when the rubber meets the road, will tend to converge (folks coming home to their previous party affiliation?), so we will never really know the answer to this interesting question.
My theory is if you hold partisan percentages constant, you automatically undo unintentional things that might skew results by party -- like perhaps polling on days or hours favorable to one side or the other . . . in terms of whether or not they are home to answer the phone. This also, btw, addresses selection of zipcode. Holding a mix constant is a valid methodology for a constant registration mix in the population.
But what it won't do is spot some surge in registration. Holding partisan mix constant would intentionally ignore a surge in registration.
Rasmussen holds mix constant and he's showing very little change all year long. He'll miss a registration surge, if there is one.
Those allowing mix to float . . . well, they will catch such a surge. I do not think this recent closure is evidence of such a surge, because Dems were in a majority of samples in July, then became a minority, and now are majority again. The registration drives were ongoing all along. The Dem majority should have grown consistently if they were consistently out-registering Republicans.
We'll know more in a week or so.