all that a poll that has the democrats ahead by +2 with a 4 point margin of error says is that the outcome is in the margin of error. if the republicans win, that is not a polling error. rather, thinking it is a polling error is an error in reading the polls.

if you have 20 polls that have the democrats ahead but in the margin of error in 20 states and they lose all of them, that is still not a polling error. it's arguably reflective of a slight polling bias, but the polls don't say the democrats are barely ahead. the polls says that the outcome is unclear.

then, when you pile on incumbency effects, get out the vote, voter turnout demographics (age, income) and cultural inertia, these in-the-margin polls in red states all tip back towards the republicans. every single one of these flips is independently extremely unlikely, except perhaps north carolina (where the democratic candidate is popular and has an edge for that reason. that's local, not national. it has nothing to do with trump). that's the right mathematical model - every senate race is an independent experiment and a weighted coin toss in the republicans' favour, and the likelihood that the democrats will win more than two seats becomes extremely remote.