But a Trump rebound — or another polling miss in the Rust Belt — could blaze a trail for Trump to repeat his victory four years ago, especially as surveys show close contests across the Sun Belt, including all-important Florida.

Here’s a state-by-state field guide on the polls, covering everywhere rated “Lean Republican,” “Toss Up” or “Lean Democratic” in POLITICO’s Election Forecast, going into Thursday’s debate.

Arizona (11 electoral votes):

FiveThirtyEight average: Biden +3.5
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +0.2
2016 result: Trump +3.5

It was 12 days before the 2016 election when Clinton’s campaign announced she would be visiting Arizona — a trip that signaled Democrats were so optimistic about her chances that they were seeking to run up the score in traditionally Republican territory.

Things look even better for Biden with 12 days to go until this year’s election. He’s up by more than 3 percentage points over Trump, compared to a virtual tie at this time four years ago.

Unlike other battleground states, where pollsters have been furiously dropping new surveys this week, the most recent live-interview survey in Arizona came out a week ago, from Monmouth University. That poll found Biden running stronger than in previous surveys, though the race was tight in Monmouth’s “low-turnout” likely-voter model.

Florida (29 electoral votes):

FiveThirtyEight average: Biden +3.8
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +2.3
2016 result: Trump +1.2

The quintessential swing state is leaning slightly toward Biden. In three live-interview polls this week, Biden has led Trump by 3 points, 1 point and 4 points — all within each survey’s margin of error, but showing a persistent slim advantage for the Democrat.

There are reasons to be skeptical of any poll that shows either candidate winning this state in a runaway. Despite the national wave toward Democrats in 2018, the Florida electorate barely budged, allowing GOP candidates for governor and Senate to win very narrow victories amid an otherwise dreadful election for Republicans.

Georgia (16 electoral votes):

FiveThirtyEight average: Biden +1
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Trump +2.9
2016 result: Trump +5.1

The balance of public polling in Georgia points to a photo finish, exemplified by a New York Times/Siena College poll out earlier this week that showed Trump and Biden deadlocked at 45 percent each.

There were two other recent polls showing Biden with larger leads — a Quinnipiac University poll last week and an internal survey from Democratic Senate nominee Jon Ossoff — but most observers believe the race is close. That speaks volumes by itself, both about the state of the 2020 campaign and big changes in the state: Georgia hasn’t gone Democratic since 1992.

Iowa (6 electoral votes):

FiveThirtyEight average: Biden +1.1
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +0.5
2016 result: Trump +9.4

Iowa broke hard for Trump four years ago. Twelve days out from the election, the two candidates were neck-and-neck — but on the eve of the election, a Des Moines Register/Mediacom poll showed Trump opening up a significant lead.

It’s again close with two weeks to go: Polls from The New York Times/Siena College and Monmouth University this week gave Biden very slight leads among likely voters.

Michigan (16 electoral votes):

FiveThirtyEight average: Biden +8.1
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +6.6
2016 result: Trump +0.2

Michigan was the state with the smallest final margin between the two candidates in the 2016 presidential election, but it doesn’t look all that close right now. A Fox News poll out Wednesday gave Biden a 12-point lead, 52 percent to 40 percent.

That is the only live-caller poll there this week. The only surveys in the FiveThirtyEight database since August to show Biden with leads smaller than 5 points were conducted by The Trafalgar Group, a small, Atlanta-based firm that believes traditional polling systematically and persistently favors Democratic candidates, as it did in 2016.

Minnesota (10 electoral votes):

FiveThirtyEight average: Biden +8
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +7.3
2016 result: Clinton +1.5

There hasn’t been a live-caller poll in Minnesota since before the first debate, but two online polls (from SurveyUSAand Morning Consult) showed Biden ahead by single-digit margins this week.

The 2016 race in Minnesota finished closer than pre-election polls suggested, and the state was not central to Trump and Clinton campaign strategies. Trump has made a bigger play for it this year.

Nevada (6 electoral votes):

FiveThirtyEight average: Biden +6.5
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +2.8
2016 result: Clinton +2.4

Unlike some of Great Lakes states, polls in Nevada did not underestimate Trump’s 2016 performance. In fact, the final two live-caller polls in Nevada showed Trump ahead, while surveys using online or automated-phone methodologies were more accurate.

There have only been four polls conducted entirely since the first Trump-Biden debate, showing Biden with a lead ranging from 2 points in a survey conducted by a reputable GOP pollster and commissioned by the Las Vegas Review-Journal — which is owned by Republican megadonor Sheldon Adelson — to a 9-point Biden edge in a survey conducted by a Democratic firm for the liberal website Daily Kos.

New Hampshire (4 electoral votes):

FiveThirtyEight average: Biden +11.5
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +7.6
2016 result: Clinton +0.4

New Hampshire has slipped from the top tier of Electoral College battlegrounds, as polls show Biden up by a wide margin. The most recent polls from The Boston Globe/Suffolk University and the University of New Hampshire showed Biden ahead by 10 and 12 points, respectively.

But there is one sign that the race might be closer: Trump is planning an event in Manchester this weekend.

North Carolina (15 electoral votes):

FiveThirtyEight average: Biden +2.9
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +2.1
2016 result: Trump +3.6

As in 2016, the polls with 12 days to go show Trump narrowly behind in a state where he closed strong last time.

Of the handful of polls there released this week, only one was conducted by live interviewers: an ABC News/Washington Post poll that showed Biden just 2 points ahead of Trump among likely voters, with third-party candidates included.

Ohio (18 electoral votes):

FiveThirtyEight average: Trump +0.8
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +0.1
2016 result: Trump +8.1

Ohio is the rare state where Trump actually appears to be in slightly better position than at this time four years ago, a testament to how Trump-era politics have scrambled coalitions in a number of important states.

A Fox News poll out on Wednesday gave Trump a 3-point lead over Biden, one of a handful of recent surveys showing Trump ahead inside the margin of error.

Pennsylvania (20 electoral votes):

FiveThirtyEight average: Biden +6.3
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +6.4
2016 result: Trump +0.7

Both parties see Pennsylvania as the state most likely to decide the presidency, and so do pollsters, who released four separate live-interview polls on Wednesday.

The surveys — from USA Today/Suffolk University, Quinnipiac University, CNN/SSRS and Fox News — showed Biden leading by margins ranging from 5 to 10 points among likely voters.

The good news for Trump? He’s essentially where he was four years ago, trailing the Democrat by 6 points in the average. But he needs a similar late charge to close the gap.

Texas (38 electoral votes):

FiveThirtyEight average: Trump +0.6
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Trump +6.4
2016 result: Trump +9

Texas is moving quickly toward Democrats — though perhaps not quickly enough for Biden in 2020. There’s only been one live-caller poll since the first debate, a Quinnipiac University poll released Wednesday showing a tied race.

But polling in Texas sometimes underestimates Democrats, especially in anomalous, high-turnout elections.

Wisconsin (10 electoral votes):

FiveThirtyEight average: Biden +6.6
FiveThirtyEight “polls-only” forecast on Oct. 27, 2016: Clinton +6.7
2016 result: Trump +0.8

Like in Pennsylvania, polls in Wisconsin show Biden ahead by margins that closely resemble Clinton’s lead four years ago. That includes a Fox News poll released Wednesday that gave Biden a 5-point lead.

It’s another state where Trump needs history to repeat in the final 12 days to mount another comeback. But like in Pennsylvania, Biden is much closer to majority support than Clinton was in 2016, meaning it could be harder for Trump to close that gap.