Here is Marist's track record going back to every vaguely competitive race since 2016 and recording their final poll (or any poll) that was done in the September to November period.
2016
President
Arizona: Trump +5
Florida: Clinton +1
Georgia: Trump +1
New Hampshire: Clinton +2
Nevada: Trump +1
North Carolina: Clinton +6
Ohio: Tie
Pennsylvania: Clinton +12
Senators
Arizona: McCain +16
Florida: Rubio +2
Nevada: Heck +7!!
North Carolina: Tie
Pennsylvania: McGinty +4
Notice how Republicans fared better in polling for Congress vs Trump, yet they didn't end up performing as much better as polling showed.
2018
Senators
Arizona: Sinema +6
Florida: Nelson +4
Indiana: Donnelly +3
Missouri: McCaskill +3
Nevada: Heller +2
Ohio: Brown +13
Tennessee: Blackburn +5
Wisconsin: Baldwin +14
Governors
Florida: Gillum +4
Georgia: Kemp +2 (Got it right for once!)
Nevada: Laxalt +1
Ohio: Tie
Wisconsin: Evers +8
So, what matters more to people. A pollster's track record, or their 538 rating which is clearly not rated on accuracy? I have a feeling this poll is going to end up very siimiliarly to most of these...
These numbers are meaningless unless you provide a date as to when they were taken/released as they totally ignore the fact that there were major events that took place during this time period (debates, Access Hollywood Tape, Comey letter) that radically moved numbers. Same thing happened to a lesser extent in 2018 (Kavanaugh hearings). The last Marist PA poll was released on 10/6/2020, yet you are saying they are awful because they were wrong about an election that occurred a month later.