This is exactly the sort of nerdishness I just can't enough of. FiveThirtyEight are, as far as I've experienced, the most thoughtful pollsters around (actually, they are complete statistics geeks) and last Sunday they released their latest UK model. First of all, they dispense with any assumptions of uniform swing, which means they are already further along than any other model being discussed. They use a full matrix of voting transitions from 2005 to 2010, and build in elements of tactical voting in each seat (that is, the shift from one party to another is more likely in close seats where the transition is to the second place party, for example). They also build in more regional variation, and things like incumbency advantage and shifts between voting and not-voting.

The current results are starkly different to anything else out there. Firstly, there is no chance, in this model, of Labour finishing first in seats with the third largest vote. The predicted seats based on the slowly stabilising vote share poll of polls position (Con 34, LD 28, Lab 27) give the Tories 100 more seats than Labour, although they would be short of a majority. The Liberal Democrats take 120 seats, 80 fewer than Labour. Because of the built-in larger swing between Labour and the other two in seats where they are more likely to win, the model penalises Labour more in terms of seats than, for example, the BBC's simple uniform swing model does. A Lib-Lab coalition doesn't have a majority either, although the margin is tiny.

Obviously, it's our quirky election system, so who really knows, and in a sense its just an academic exercise. But if this is close to what happens on Thursday, what a remarkable month it will have been.