The presidential race polls, at this point, are kind of boring: Vice President Kamala Harris is leading by 2 percentage points in Pennsylvania. Yawn. Shes tied with former President Donald Trump in North Carolina? Same as it ever was. Trumps up by a point in Arizona? Yeah, thats what we expected to hear.

But four weeks before Election Day, a very steady and excruciatingly close presidential race is masking a still wide-open election, where even a slight polling error could lead to a variety of truly unexpected outcomes.

The election analysis firm Split Ticket gives Democratic nominee Harris a 57% chance of winning the presidency , based on its proprietary analysis of polling and other data. The firms forecast hasnt changed much since she entered the race, despite dozens of additional polls, with Harris maintaining a tiny edge in Pennsylvania and Midwestern states while slightly trailing in the Sun Belt and North Carolina.

But Split Ticket co-founder Lakshya Jain said the seeming stability of the race as a toss-up is an illusion. Its entirely possible Trump outperforms his 2016 margin of victory or that Harris wins all seven swing states an outcome that shows up in 25% of the firms election simulations.

Nothing would really surprise me, Jain told HuffPost. I wouldnt be surprised if Kamala Harris lost Pennsylvania but won North Carolina, right? These types of things happen.

In Split Tickets model, for instance, there is a 54% chance at least one of the Democratic-leaning states of New Hampshire and Minnesota, or the GOP-tilted Texas and Florida ends up flipping to the other party. None of the four states has seen extensive advertising in the presidential race.

Polling error isnt consistent either. Wisconsin, another battleground state, has seen perhaps the largest polling errors in the past eight years. Polling in the Upper Midwestern state predicted that both Hillary Clinton in 2016 and Joe Biden in 2020 would do, on average, about 6 percentage points better than they ultimately did.

The accuracy of this years battleground state polling in Wisconsin and elsewhere could hinge on whether pollsters are overcorrecting for their mistakes in 2016 and 2020.

To account for the reluctance of many Trump-aligned or conservative voters to participate in polling, most major polls have asked voters to recall how they voted in the 2020 election and then used their answers to weight the polls sample in such a way that it reflects the makeup of the 2020 electorate. Since the makeup of electorates changes considerably from one presidential election to the next, and people sometimes do not accurately remember how they voted, this kind of weighting is potentially inaccurate, argues New York Times top polling expert, Nate Cohn .

And this year, the polls that use this practice weighting for recall vote have Trump doing better against Harris than those surveys, including those conducted by the Times, that do not.

Polling, unlike punditry, was quite accurate in the 2022 midterm elections, though that may be because the midterm electorate skews more educated and thus more likely to respond to pollsters.