Can Polymarket be manipulated?

The short answer: not one thing, and the evidence is public

The question is usually asked as though manipulation were a single activity, and the honest answer is that it is at least three, each attacking a different component of the market. That distinction matters, because the thing most people picture, a rich trader pushing a price around to make a market say something false, is the one with the least evidence behind it. A prediction-market price is defended by the fact that anyone who disagrees can take the other side at a now-cheaper price, so pushing a price away from consensus costs money and invites the correction. The documented patterns are more interesting than that, and they work by attacking the parts of the system that are not price discovery at all.

Those parts are the settlement source, the resolution vote, and the volume figure. A contract has to settle against something, and if that something is a single price at a single instant, it can be moved. A disputed market has to be resolved by someone, and if that someone is a token vote, the vote can be bought. And volume is a reported number, so it can be manufactured. What follows is what the public record actually shows about each, with the figures dated and sourced, because Polymarket records its activity on a public blockchain and outside researchers have been able to measure all three.

The three documented vectors, next to ordinary large trading

The table sets the three researched patterns against the thing they are most often confused with, which is a large trader simply holding a large position. Reading down the first column is the useful exercise: the vectors differ in what they target, who absorbs the cost, and whether anything has been done about them. Rows that depend on the current state of the platform are marked as of July 2026.

Figures are stated as of July 2026 and come from the working papers and reporting cited in the sections below. The Stanford, SMU, and Columbia papers are working papers that have not been peer reviewed, and their estimates rest on statistical inference about intent from on-chain patterns rather than on any admission or enforcement finding.

Settlement manipulation: moving the thing the contract settles against

The newest and most specific finding concerns Polymarket's very short-dated Bitcoin contracts. A working paper titled Settlement Manipulation in Prediction Markets, posted on July 15, 2026 by David Dai, Ruizhe Jia, and Shihao Yu of Stanford University and Singapore Management University, examined roughly 16,000 five-minute Bitcoin contracts across a two-month sample taken after the product launched. These contracts ask where Bitcoin will be at the close of a five-minute window and settle through a Chainlink price feed drawn from major spot exchanges at a specific timestamp.

The authors reported that in the final seconds before settlement, one-sided order flow on Binance spiked, running near 3.9 times normal levels in contests that were close to even, and that contracts finished on the same side as that Binance flow about 85 percent of the time. The pattern concentrated in low-liquidity periods, nights and weekends, when moving the spot price costs less. They flagged 821 traders whose activity matched the signature and estimated roughly 8.2 million dollars of profit over the two months, drawn largely from retail counterparties. As Yu put it, the contracts settle on a price that traders can move by trading.

The mechanism is worth stating plainly, because it is not a flaw in the prediction market so much as a flaw in the bridge between two markets. Nothing is being done to the Polymarket order book. The cost of moving Bitcoin's spot price for a few seconds is simply lower than the payoff on a contract that settles off that instant, so the two are arbitraged against each other. That framing also explains the paper's most useful control: the pattern largely disappeared in fifteen-minute contracts, where the window is long enough that the manipulation stops paying for itself. Polymarket disputed the paper's characterization while saying it intends to move some markets, over the coming year, to settlement priced over a longer period rather than a single instant, which is the same conclusion the authors reach.

Oracle governance capture: when the resolver is the target

Most Polymarket markets do not settle against a price feed. They settle against the UMA optimistic oracle, where a proposer posts an outcome, a challenge window opens, and a contested result escalates to a vote of UMA token holders. The mechanics are covered in how does Polymarket resolve markets. That design assumes the people voting are more interested in the oracle's long-run credibility than in any single market's payout, and the criticism leveled at it is that the assumption does not always hold.

Two episodes are cited most often. Between March 24 and 25, 2025, a market asking whether Ukraine would agree to a minerals deal before April moved from around 9 percent to 100 percent and resolved Yes despite no agreement having been reached; reporting attributed the result to a single actor controlling roughly a quarter of UMA voting power, on a market worth about 7 million dollars. Then in July 2025, a market on whether Zelenskyy would wear a suit before July, carrying well over 100 million dollars of volume, resolved in a way that contradicted what many observers took the NATO summit photographs to show, and traders publicly accused large UMA holders of coordinating the vote.

The structural point critics make is an arithmetic one rather than an accusation about any individual. When the money resting on a single market approaches or exceeds the cost of acquiring decisive voting power in the token that adjudicates it, the incentive that is supposed to keep the oracle honest weakens. Whether that gap has ever actually been exploited is contested; that the gap exists is not. This is the part of the system with the least resolution as of July 2026, and it is a reasonable input when weighing how much confidence a price on a subjectively worded market deserves. Related reading on the platform's broader track record sits in is Polymarket legit.

Wash trading: manufacturing the volume figure, not the price

The third pattern does not try to move a price at all. Wash trading is a trader transacting with themselves, or with a coordinating partner, to manufacture the appearance of activity. In November 2025 a Columbia Business School paper posted to SSRN estimated that artificial trading averaged about 25 percent of Polymarket volume over three years, flagging 14 percent of the platform's 1.26 million wallets. The concentration is the interesting part: roughly 45 percent of all-time sports volume was classified as likely wash trading, against 17 percent in election markets, 12 percent in politics, and only 3 percent in crypto. Weekly peaks ran far higher, near 95 percent in election markets in late March 2025.

This was not the first such estimate. In October 2024, Chaos Labs told Fortune that wash trading made up around one third of volume on the presidential market, and Inca Digital separately found the on-chain transaction volume on that market to be near 1.75 billion dollars against a reported 2.7 billion. The usual explanation offered for the behavior is incentive farming, activity manufactured in anticipation of rewards or a future token distribution, which fits the observation that the share fluctuates with the calendar rather than with news.

What this changes for a reader is narrow but real. Wash trading does not make a price wrong, because a trader transacting with themselves nets to no view. It makes the volume figure next to that price a weaker signal of genuine interest than it appears, and volume is one of the two figures most people use to decide how much weight a price deserves. That is a good reason to lean on liquidity, meaning what is actually resting in the book to absorb an order, alongside volume, which is a distinction laid out in volume vs liquidity on Polymarket. It is also worth noting that these are estimates inferred from wallet behavior, not confessions, and that a market maker quoting both sides can resemble a wash trader in the data.

A large position is not automatically manipulation

The word manipulation gets attached to any large wallet that moves a market, and the distinction is worth defending. The 2024 US election is the standard example: one trader accumulated an enormous position, visibly moved prices, was accused of manipulating the market, and turned out to be right. That is a forecast expressed with size, not a distortion. The same is true of most of the activity on the traders leaderboard, where large notional reflects conviction and capital rather than any attack on the mechanism.

The line the three documented vectors cross is that they profit from distorting something other than the forecast. Settlement manipulation profits from moving the reference price, oracle capture from deciding the resolution, and wash trading from the appearance of activity. None of them require being right about the event. A large directional position, by contrast, only pays if the outcome actually arrives, which is precisely the incentive a prediction market is built on. A thin market is a separate matter again: a modest order can move a price a long way there simply because there is nothing behind it, an effect described in why do Polymarket prices move, and that is thinness rather than intent.

How this sits against the accuracy record

It would be easy to read the sections above and conclude the prices are not worth reading, and that would overshoot the evidence. The same public record that documents these patterns also shows Polymarket's prices calibrating well in aggregate, with a self-published Brier score near 0.06 and strong performance in liquid markets close to resolution, which is covered in how accurate is Polymarket. Both things are true at once, and they are true of different markets. The documented manipulation clusters where the structure invites it: five-minute contracts settling off one timestamp, subjectively worded markets that reach a contested vote, and reward-farmed categories like sports.

That pattern is itself the practical takeaway, and it is a reading heuristic rather than a warning. The conditions that make a price informative, deep liquidity, an unambiguous resolution criterion, a settlement source that cannot be moved in a few seconds, are close to the inverse of the conditions under which manipulation has been measured. A deep, plainly worded, long-dated market is where the accuracy research applies and where the manipulation research mostly does not. Every market on this site carries its volume and liquidity next to its price for that reason, and how we collect those figures is set out in the methodology.

Frequently asked questions