The thing I would note in regards to point 3 is that this factor has an n>3. The trend whereby polls overstate liberal/left-wing performance is traditionally called the "Shy Tory Effect", getting its name from when it was first seen in Britain in the 1990s. It also happened recently in Brazil, with polls greatly underestimating the strength of Bolsonaro and his allies.
Granted, nobody knows what exactly causes the Shy Tory Effect, and so a reversal is in principle possible based on obscure factors. But I think the body of evidence supporting a polling error that benefits the GOP is stronger than you realize. There's definitely also been neutral elections, but I don't know that it is historically supportable to say that a polling error benefiting Democrats is as likely as one benefiting Republicans.
The one thing here is that (unlike in 2020) everyone already seems to be pricing in that sort of polling error. To use one example, if Oz loses (which I don't expect), the surprise will not be that "there was a polling error that benefitted the Democrats", the surprise will be "there was no polling error". The "general expectation" already is a polling error leading to an Oz win by a couple points; Oz outperforming expectations would be a case where the GOP-favored error not only existed but was fairly large.
It may still be the less likely option but Democrats matching/beating expectations becomes a much more viable possibility when that can be done even without beating (and perhaps even while underperforming) public polling.
