I can also see how compounded errors can creep into polling with both bias and unintended consequences.
1. Polling samples might be, or are often crap. The increasing lack of landlines. Statistical bias to urban zip-codes because that's where the density is if you try to randomize/buckshot it. Republicans/Productive Class people are at work. Then, at best, when pollsters try to "normalize" the data for oversampling urban vs. rural, Democrats vs. Republicans etc. it's at least one more source of error, at worst, they pick the model that conforms to preconceived notions and selection bias. Plus, if a candidate inspires a bunch of unlikely voters, or first-time voters to come to the polls, are they captured at all?
2. Polled people lying. It's a constant barrage from the costal Leftist MSM elites and Hollywood that voting for Trump, much less being a supporter is "uncool" at best, and you're a redneck racist misogynist hick at worst. Did "Pussygate" really depress Trump's polling numbers that badly, or did it just deepen the percentage of people unwilling to voice Trump support/votes? We know this is real, from Obama's elections always having a few percentage points lower in actual voting returns vs. polling, probably for fear of "looking racist". And we know it was a real thing in the Brexit referendum too.
3. Do the poll-takers ever fudge/lie? Maybe a certain caller pisses them off, and they mark them down as being a Clinton supporter so "their voice isn't heard". I can't imagine this kind of call center work pays well, or is very interesting, and probably comes with a consummate lack of professionalism and rigor too.
While polling often aligns well with electoral returns, I can see error-stacking like this, and a chain of people all thinking they're the only ones "fudging just a bit", when perhaps many or all of them are creating wide swings that just have no bearing on reality.
My hunch is that if you could conduct a real honest and well controlled poll right now, Trump would indeed still be down a few points. However, if you look at who has the most to lose to the "Murphy's Law Factor", it's screaming for a Trump win. If only for the Lulz, and the apoplectic WTF'ery and screaming/crying from all sorts of establishment types, and personally, as a rebuke to Clinton's decades long monomania to become POTUS.