Some issues here (this sample voted for Biden by 4 points) but what's key is that at 44%, Oz being at 50% is outside of the margin of error. He really needs to win over virtually every undecided, moreso than Republicans normally do, which is not an impossible task (especially given that this was conducted pre-debate) but an uphill battle nonetheless.
Is not just trivial to weight on 2020 recall if they have the data? They weighted on 2016 recall in 2020, but perhaps the midterm electorate is too different to do that?
Not sure I'm understanding the question right - but a huge issue in 2020 was the turnout surge that made the 2016 recall question somewhat less salient (many pollsters also didn't use it as a weight - the ones that did were
somewhat closer to the mark). 2020 polls also significantly overrepresented people who voted by mail which may or may not happen this year given the different circumstances - very hard to tell.