The Democratic primary in Wisconsin's governor race delivered a shock that rippled through both political circles and the burgeoning world of prediction markets. Progressive candidate Francesca Hong, widely seen as the frontrunner, lost to rival David Crowley by a margin of less than half a percentage point. The upset was particularly jarring for Polymarket and Kalshi, two platforms that had built a reputation for near-perfect accuracy after correctly forecasting the outcome of every state in the 2024 presidential election. In the days leading up to the primary, both markets had assigned Hong roughly a 95% probability of winning, mirroring the sentiment of traditional polling firms. When the votes were tallied, however, Crowley emerged victorious, a result that left many questioning the reliability of these forecasting tools.

Polymarket quickly deleted a social media post that had boasted of Hong's 96% chance of winning, a post that also noted her lack of endorsements from prominent Democratic socialists like Bernie Sanders and Alexandria Ocasio-Cortez. The deletion itself became a point of criticism, as observers noted the platform's willingness to amplify its predictions when they seemed strong but retreat when they proved wrong. Kalshi's leadership took a different approach, defending the probabilistic nature of their forecasts. In a social media post, CEO Tarek Mansour explained that a 95% probability still implies a one-in-20 chance of the underdog winning—a rare but explicitly acknowledged possibility. Co-founder Luana Lopes Lara echoed this sentiment, writing simply, “5% is not 0%.” Their defense highlights a fundamental tension in the prediction market space: the public often interprets high probabilities as certainties, even when the underlying math says otherwise.

This week's result is not the first time prediction markets have mispriced an election. In June, both Kalshi and Polymarket gave reality-TV personality Spencer Pratt about a 75% chance of advancing from Los Angeles's nonpartisan mayoral primary to the general election. He finished third. The month before, Kalshi assigned Rep. Thomas Massie a similar chance of winning the Republican primary in Kentucky's 4th Congressional District. Massie lost to Trump-backed challenger Ed Gallrein. In each of those cases, the probabilities left more room for an upset, but the Wisconsin outcome—with its 95% odds—was a far more dramatic miss.

The timing is particularly awkward for prediction markets, which have seen their credibility surge since the 2024 election. Both Polymarket and Kalshi were widely credited with accurately signaling Donald Trump's edge in a race that many conventional polls portrayed as a tossup. Their influence has since expanded into mainstream media: CNN named Kalshi its official prediction-markets partner, and Dow Jones plans to incorporate Polymarket data across The Wall Street Journal, Barron's, and MarketWatch. These partnerships have elevated prediction markets as a complement to traditional polling, but this week's failure may give critics ammunition.

Statistician and political forecaster Nate Silver weighed in on the broader debate in a recent social media post. “I think prediction markets are cool. But people should stop treating them as magic, and I don’t think they’re a good substitute for polls, or belong in models,” he said. Silver's comment captures the cautious optimism that many analysts hold: prediction markets offer valuable real-time sentiment, but they are not crystal balls. For the progressive wing of the Democratic Party, the loss of Francesca Hong is a setback. For the prediction market industry, it is a humbling reminder that even the most sophisticated algorithms and aggregated bets can be wrong. As the platforms continue to integrate into media and political forecasting, the question remains whether users will learn to interpret probabilities correctly—or whether they will continue to expect perfection.