A 95% favourite lost by half a point, and the case for prediction markets got harder

Polymarket and Kalshi both gave Francesca Hong roughly a 95% chance of winning the Wisconsin governor's Democratic primary, and David Crowley won it by less than half a percentage point ^[2][3]. That matters because both platforms are in the process of selling their prices to newsrooms as a superior class of information: CNN has named Kalshi its official prediction-markets partner, and Dow Jones plans to carry Polymarket data across The Wall Street Journal, Barron's and MarketWatch ^[10][11].

The defence offered by the operators is arithmetically correct and beside the point. Kalshi chief executive Tarek Mansour noted that a 95% price explicitly allowed the underdog roughly a one-in-20 chance, and co-founder Luana Lopes Lara wrote that "5% is not 0%" ^[6][7]. True: the price carried an implied 5% upset probability, and upsets at that frequency are the cost of doing business ^[15]. Polymarket's own conduct suggests less confidence in the framing, since it deleted a post advertising Hong at 96% and noting she had no endorsement from Bernie Sanders or Alexandria Ocasio-Cortez ^[5].

The interesting number is not 95. It is the fact that, according to Fortune, the contracts on both platforms tracked the sentiment of traditional polling firms, which also expected an easy Hong win ^[4]. The commercial case for event contracts rests on the claim that money aggregates information polls cannot reach. That case was made in 2024, when the platforms called every state in the presidential race and gave Donald Trump an edge while many conventional polls showed a tossup ^[8][9]. In Wisconsin the markets did not diverge from the polls, so they supplied no independent signal, and the miss was correlated with the miss everyone else made. A market that agrees with the polls is polling with a settlement date and a fee.

This is not the platforms' first bad race, but it is the first at a price high enough to test them. In June, Kalshi and Polymarket gave Spencer Pratt about a 75% chance of advancing out of Los Angeles's nonpartisan mayoral primary, and he came third; the month before, Kalshi put Thomas Massie at a similar level in Kentucky's 4th District Republican primary, and he lost to Trump-backed Ed Gallrein ^[12][13]. Each of those implied a one-in-four upset, which is not a scandal in a single case ^[18]. Three publicly identified misses in three months, one of them at 95%, is the beginning of a calibration record rather than an anecdote ^[16]. Nate Silver's line is the operator's version of the point: markets are worth watching but are not magic and are not a substitute for polls ^[14].

Also worth holding onto: Crowley's margin was under half a point, so a swing of fewer than a quarter point of vote share would have handed the platforms a win they had not earned by any better process ^[17]. Precision at the top of the probability range is mostly luck at these margins.

What to watch: whether either platform publishes a calibration curve with volume attached, so buyers can see how 90-plus-percent contracts have actually resolved rather than reading a highlight reel; whether CNN and Dow Jones attach any disclosure about market-poll correlation when they run the numbers ^[10][11]; and whether thin-volume primaries continue to be priced and promoted with the same confidence as national races.