Nate Silver, the statistician who built his name forecasting American elections and now runs the newsletter Silver Bulletin, told readers on Tuesday that the final numbers coming out of Polymarket matched his own model almost exactly as voters headed to the polls. Writing on X and in Silver Bulletin on the morning of November 3, 2026, United States midterm Election Day, he described "pretty insane agreement between prediction markets and final FLIPR forecasts."
The point he was making is concrete and, within a day or two, checkable against real results. FLIPR, the forecasting model Silver runs that outputs probabilities for individual states and for control of the House and Senate, gave Democrats an 80% chance of winning the Senate and treated a Democratic House majority as close to a certainty. Polymarket, the crypto-funded prediction market where traders buy and sell contracts on real-world outcomes, had settled on nearly the same picture. "The average state is only 3 points different," Silver wrote, meaning the typical distance between his model and the market price was small enough to shrug off.
Why a forecaster grading the crowd is the real story
Silver spent more than a decade as the most visible name in American polling analysis, and the question of whether a market of self-interested traders can out-forecast a careful statistical model has followed him the whole way. For most of that time the two camps talked past each other. A model like FLIPR digests polls, fundamentals and historical error and spits out a probability. A prediction market like Polymarket does something stranger: it lets thousands of people put money behind their guesses and reads the resulting price as the crowd's probability.
What makes this Election Day different is that Silver, the model builder, publicly held his own work up against the market and said the two had converged. That is not a small admission from someone whose business is selling the output of a model. When the modeler says the crowd landed in the same place he did, the interesting question stops being "who is right" and becomes "what does it mean that they agree."
Where the model and the market shook hands
The headline convergence is on control of Congress. FLIPR's 80% for a Democratic Senate and its near-certain call on the House were, by Silver's account, within a few points of where Polymarket's contracts were trading. In practice that means a trader reading the market and a reader following Silver's model would have walked into Election Day expecting the same outcome: Democrats favored to take both chambers, with the Senate the closer of the two.
Across the individual states the pattern held. A roughly 3-point average gap between a model and a market, spread over dozens of races, is the kind of alignment that rarely happens by accident. It suggests the market was absorbing the same polling and fundamentals Silver's model runs on, and pricing them with similar discipline.
The quiet disagreements
The agreement was not total, and Silver was careful to flag where the crowd and the model parted ways. Polymarket traders were more bullish on Democrats than FLIPR in Alaska and Missouri, pricing a better Democratic chance than the model would allow. In Iowa, Kansas and Maine the market leaned the other way, more skeptical of Democrats than the model.
Those last three are telling. Iowa, Kansas and Maine have a history of polls understating Republicans, and a market that discounts Democratic poll leads in exactly those places is doing something a naive model does not: remembering where the polls have burned forecasters before. Whether that instinct was right is one of the things today's results will resolve.
Nevada, where they openly split
The sharpest disagreement was the Nevada governor's race. Polymarket put roughly an 81% chance on incumbent Republican Governor Joe Lombardo losing to his Democratic challenger, Aaron Ford. FLIPR had the same outcome at just over 50%, a coin flip. That is a gulf, not a rounding difference, and it is the one place where following the market and following the model would have led to very different expectations.
Silver put the gap down to traders pricing in early-voting data that his model was not weighting as heavily. Early turnout in Nevada showed Democrats running roughly 6 points ahead, and the market appears to have treated that as a strong signal that Lombardo was in real trouble. The model, built to be cautious about reading too much into early votes, held closer to even. Nevada is the cleanest test on the board: one side of this argument is about to be proven wrong.
A record that is strong, but not spotless
Silver's broader point was that markets have earned the benefit of the doubt without earning blind faith. He pointed back to 2024, when Polymarket and Kalshi, the United States-regulated prediction-market exchange, both had Donald Trump favored to win weeks before Election Day while traditional polling still showed a toss-up. In that cycle the markets read the race better than the polls did, and that memory is part of why their 2026 prices carry weight.
But he did not pretend markets are oracles. The cautionary example sits in the academic data: the Iowa Electronic Markets, a small university-run prediction market, gave the Democratic candidate an 85.7% chance of winning the national popular vote on September 29, 2024. Trump went on to win the popular vote by about 1.5 points. A market can be confident and wrong, and a price near certainty is still only a probability.
Silver underlined the uncertainty with his own money, or at least the idea of it. Reaching for a card-table metaphor, he said that if he were placing bets he "might put some chips on the 'Dems slightly underperform polls but win the Senate anyway' spot on the felt and some on 'Republicans narrowly hold the Senate,' but also a few on an epic Democratic blowout." The spread of his imaginary chips is the honest version of an 80% forecast: likely, not locked.
What Nevada will tell us by morning
For once the markets-versus-models debate has a clean referee. On nearly every race Polymarket and FLIPR agreed, so nearly every result will confirm both or neither, and prove little about which method is sharper. Nevada is where they disagreed out loud, an 81% market against a 50% model, and the reason for the split is documented: the market trusted Nevada's early-vote numbers and the model did not. Whoever is standing in the governor's mansion when the counting ends will have settled not just a race but a small, specific argument about whether a crowd watching ballots come in can see something a model is built to ignore.






