Wait just a second here. Haven't 538 (and most election prognosticators, for that matter) endlessly opined about how likely voter polls are more reliable than registered voter polls, especially when the election is ~2 months away? And yet now we're supposed to disregard that, just for this specific poll mind you, for what reason? Because of Harry Enten's gut feelings? Because it shows a Democrat doing better, and that "can't be right"?
Read the article. With this specific pollster, Marquette University, there is no conclusive evidence that its likely voter screen consistently makes it more accurate.
Plus, republicans clearly have an enthusiasm advantage this cycle, and midterm election turnout is ALWAYS lower than presidential election turnout regardless of whether a gubernatorial election exists or not in a given state, and the lower the turnout is, the likelier a republican victory becomes.