Trump consistently getting 80%+ support among conservatives. The problem is that most pollsters do not weight by ideology, which is the single most important determinant in voting preference. For example, the final
YouGov poll had a top-line of 49-40 with registered voters and 53-43 with likely voters.. However, the sampling on this poll was hilariously skewed with liberals outnumbering conservatives. h
This is a drastic contrast to Gallup's polling data which has long-held that conservatives make up 36-38% of adults, liberals make up 24-26%, and the remaining moderates are 36-40% of the population. Say we split the difference and have a split of 37-25-38. That gives a margin of 47.5-42 and 51-45, respectively. Both obviously are much more in line with final of 51-47.
I'm inclined to buy this line of reasoning because it fits the narrative around both candidates. Biden had the highest appeal of any candidate in the last 8 presidential election cycles among moderates, even more than Obama's in 2008. The reason Obama's win was bigger was because he had higher liberal turnout+actually won a high percentage of conservatives for a Democrat. The popular vote remained closer than expected because Trump won conservatives by a historically large margin, fitting the explanation that polls were missing conservatives due to COVID, not passing likely voter screens, etc.
This also falls in line with the partisan splits. Poll screens tended to have more Democrats than independents, whereas most partisan surveys of the nation at-large tend to have independents slightly outnumber Dems (Gallup+Pew for example).
Much like 2012 when most pollsters failed because they underestimated Obama's GOTV operation, I think the most likely explanation is that Trump's GOTV operation got more conservatives to the polls than anticipated.
Tl;dr:poor turnout models didn't factor for high GOP enthusiasm, or as I like to call it, "variable-T"
