The polls this year followed the same trajectory that polls in Collins' past races followed, with the difference this year being that the race was significantly more competitive than ever before.
Copying and pasting this post from a past thread on a similar topic:
2008
RCP average: Collins +14.6
Actual result: Collins +23
https://www.realclearpolitics.com/epolls/2008/senate/me/maine_senate-564.html2014
RCP average: Collins +29.2
Actual result: Collins +37
https://www.realclearpolitics.com/epolls/2014/senate/me/maine_senate_collins_vs_bellows-4168.htmlIn both races [2008 and 2014], the Democrat performed almost exactly at their [polling] average and Collins wound up consolidating the undecided/uncommitted voters (probably Dem-leaners who didn't want to commit to voting Republican but were always more likely than not with Collins).
It looks like Gideon's polling suffered from both that Collins consolidation phenomenon, as well as pollsters generally modeling for a 2018-style environment, which likely made her average vote share a bit too high. If we assume that Gideon's average was a few points too high throughout the race, then the same "Dem hitting their average/Collins consolidating" pattern probably holds in this race as well. The fact that Collins' overperformance was roughly the same 8 points both times, and her final average (probably lowballed for the same reason Gideon's was overshot) this year was around 42% before she won with 51%, fits into this scenario almost perfectly.
A good (though obviously imperfect) historical comparison is NV-Sen 2010: incumbent is a multi-term institution with long history of consolidating in the end, polls modeling a different turnout than actually happened, giving the incumbent a lower share and the challenger a higher share than was the truth.