But, I'm really not sure why you want to explicitly and only look at Special elections
It's fair. My point in looking at that, is to overall test the assertion that polls are indicative of how people vote. It kind of seems looking at the methodology for the OP article's poll, like if any accurate information came out of the poll about how the election would go, it would be more or less an accident (or a result of the fact that the poll and the election are both general measurements of how people feel politically overall, and not much more resolution than that.)
You could flip what you said around, and say that because the special elections are much less complex, and the polls were done much closer to the actual election than polls today about the election in November, I'd expect the polls to be much more predictive of how the election will go, than the OP article.
So, let's analyze. As you said, it's actually not that hard to find polls and results. I'll follow your lead and look at 538 (for the first three, which is all the effort I feel like investing in it).
Kinda looks like the polls have some methodology problems. I raised some plausible details for some of what those problems might be, and when we check, hey objectively do it seems like there are problems with the output? We find that, hey look, there are problems. Science!
(Incidentally, that poll for Utah claimed a margin for error of 4.26 percentage points, with the use of three significant digits of claimed resolution adding an extra layer of hilarity when it turned out their final answer was off by a factor of 267%.)