I believe this group uses a rolling average. They have 3000 people in the poll but only poll about 200 of them each day. So it takes roughly 2 weeks before they go back and poll the same people again. Because each day is only a small sample size, it may be susceptible to larger daily swings (esp if they weigh the more recent responses more heavily, or, if there is something in the news that would push people to one or the other candidate). I suspect it should be slower to move based on news events because it would take several days of 200 responses before a large shift in preference could influence the numbers shown in the entire group of 3000. Overall it should be great for spotting trends. I am unsure how well it will do in predicting outcomes but it sure is an interesting, new way to poll.
I believe it is about 400 per day and they poll over a 7 period not a 14 day period.
It is possible that they weigh the most recent day's poll heavier but I doubt it.
Interestingly, if you analyze the day to day changes you find that some days are much for favorable to Trump than Clinton and some are more favorable to Hillary.
Trump's best days: Thur-Fri-Sat-Sun. Hillary wins Mon-Tue-Wed. But Trumps good days are better than Hillary's.
There is no weighting. The poll rolls in one new sample, and rolls out the 7 day old sample.