At this moment, polls show a strong upswing for Donald Trump’s campaign. He’s made real ground on Hillary Clinton, closing the gap between them by consolidating the Republican Party behind his candidacy. Clinton, meanwhile, is fighting a contentious primary, which keeps her from doing the same. There’s still a strong case to make that Donald Trump is a long shot for the White House—that he’s too unpopular with too many Americans outside the Republican Party and that if the Obama coalition holds, he loses.

At least, that’s what I think. But humility is a virtue, and life is volatile, even if elections often aren’t. That Trump is a poor bet for November doesn’t mean he couldn’t win. He could. Which raises an obvious question: How would it happen? And finding an answer here isn’t too difficult. You just need to reverse-engineer the conditions of a Trump loss. What are those conditions?

At the top of the list is demographics. Trump is deeply unpopular with black and Latino voters, as well as with women and young people. But Clinton is also unpopular, albeit by a smaller magnitude. Given two unpopular candidates, it’s possible that “popularity” just won’t matter as much, that voters will take both candidates’ worst qualities as a given and move on.