One way to think of this is as a file drawer problem.
Essentially as we get closer to the election your poll-based metrics become _less_ reliable, not more, because you don't know what is not being published, and you can't know if most pollsters change their methodologies or weighting slightly.