Unlike some others on FR, I most definitely do not claim to be an expert on polling. However, I've read several books and articles on the subject, including the one I referenced. So, as simply as possible, please let me try to clarify why polls should always be taken, AT BEST, as a broad indicator, not a true predictor of an outcome.
First think of all the many things in life that require true accuracy: things like finances, building, surgery, banking, payroll, sports, and so on. In all of those things, and more, measurements must be exact.
Now contrast true accuracy to how polling results are reported: they give a percentage of a sample, then give a range within which that number can be wrong either high or low by several percentage points.
Then factor in the distance in time between when the poll is conducted and the actual event it is trying to predict. Also factor in variables such as who took the poll, who financed it, what questions were asked, how and in what order, what time of day/week, etc., all of which have a very significant effect on results obtained.
So to answer your question about polls that come close to election results appearing to show some scientific basis: you gave yourself the key by saying "appearing to show." It's an illusion — like sleight-of-hand tricks, we see what the magician (pollster) wants us to see. The closer in time the poll is to the election, the more people have made up their mind as to what they plan to do in the voting booth. More people self-identify as "likely voters," for example. So the polling samples become somewhat more accurate. In addition, the questions tend to become simpler and more straightforward — from "if the election were held today would you vote for A, B, C, or D," to "who do you plan to vote for tomorrow," for example.
However, even with overnight polling just before an election, it is the rare poll that gets the result EXACTLY right: say, 53, 40, 7 to minor party candidates. Rather, the pollsters say, "Here's my prediction, but don't say I was wrong if the actual result is plus or minus a factor of X." It's pretty easy to call yourself a winner when you've given yourself a swing of, say 6 points high or low to play with.
Now focus on polls that are far away in time from the event being predicted and factor in the desire by various parties to try to manipulate the public and get a bandwagon effect going. That's what goes on in all major elections these days. Anyone who thinks, for example, that the various polls about the Dem presidential candidates mean anything right now is dreaming. Same for these early recall polls. Their only purpose is to manipulate, not to inform.