Researchers from Stanford University and Singapore Management University identified signs of manipulation in the settlement of five-minute Bitcoin contracts on Polymarket. This is detailed in the paper Settlement Manipulation in Prediction Markets.
The authors analyzed more than 60 million on-chain records, the platform’s order book snapshots, and high-frequency spot market data. The 821 wallets they highlighted earned $8.22 million during cycles with anomalous activity, while retail and other unclassified participants lost $7.61 million.
How the alleged scheme works
The five-minute contract on Polymarket launched on February 12, 2026. It pays $1 if the Bitcoin price at the end of the window is not below the initial value, and $0 otherwise.
To determine the outcome, the platform uses the BTC/USD price feed from Chainlink. New contracts start every five minutes around the clock.
The researchers described the following alleged scheme: a trader first buys a contract on Bitcoin rising or falling, and in the final seconds before settlement executes large spot trades in the desired direction.
The brief market move affects the price recorded by the oracle. The contract settles in the trader’s favor, after which quotes partially revert to prior levels.
After the five-minute markets launched, directional order flow in the final ten seconds of the window rose by about 50% compared with the period before their introduction. Prices began to reverse within the next ten seconds.
The authors viewed this pattern as a sign of temporary price impact rather than trading on new information, since moves driven by significant news usually persist longer.
In cycles where the outcome remained close to even odds, the activity spike was about 3.9 times the norm. The average volume of directional spot trades in the last ten seconds of purportedly manipulated cycles reached $1.7 million, versus $68,000 in the rest.
About 56% of such episodes occurred during nighttime UTC hours, and another 44% on weekends. The researchers attributed the concentration to lower liquidity: less capital was needed to move the price in these periods.
821 wallets earned $8.22 million
The researchers classified 1,613 cycles as presumably manipulated — the top 10% of observations by intensity of directional trades in the final seconds. They labeled as likely manipulators wallets that participated in at least five such cycles and made aggregate profits of at least $2,000 in them.
A total of 821 wallets met the criteria — 0.34% of roughly 243,000 addresses that traded the contract. They earned $8.22 million in anomalous cycles and only $90,000 in the rest.
The retail/other category lost $7.61 million in the former. This amount corresponds to about 93% of the profits of the group of presumed manipulators. In the overall distribution of losses, retail and other unclassified participants accounted for 70.5%, likely manipulators in cycles that went against them for 20.2%, and market makers for 9.3%.
The authors cautioned that a wallet address does not necessarily correspond to a single person: one participant could use multiple accounts. The study also does not prove intent by address owners.
The classification is based on participation in anomalous cycles and realized profits. The researchers also did not establish a direct link between the accounts executing spot trades and the wallets receiving payouts on Polymarket.
Even near-certain outcomes changed
The most illustrative cycles, according to the authors, were those where Polymarket already assessed one side’s probability of winning at 90–100%. When a directional move occurred against the favorite, the opposite side won in 34.2% of presumably manipulated cycles. Without a recorded spike, this result occurred in only 1% of cases.
In more balanced cycles, where the favorite’s probability was 50–60%, a directional move against it led to the other side winning in 65.4% of cases, versus 41.4% without anomalous activity.
The researchers also considered an alternative explanation — market makers hedging their positions. However, their average maximum exposure was about $1,400 per cycle, while directional spot flow reached roughly $1.7 million.
Moreover, positions on Polymarket built up gradually over five minutes, whereas large spot trades clustered immediately before settlement. The authors deemed this difference in scale and timing inconsistent with ordinary hedging.
Polymarket will change the settlement mechanism
A Polymarket representative told Bloomberg that the platform uses several independent price oracles to aggregate data. The company also plans to move some markets to a mechanism that accounts for prices over a longer period, rather than a single point in time.
In fifteen-minute contracts, the authors found a much weaker activity spike and did not observe a pronounced price reversal after settlement. In their view, lengthening the window reduces the chance that a brief spot trade can change the outcome.
Among other protections, the researchers mentioned settling to an average price over the period, randomizing the sampling time, limiting position sizes, and adding costs for transactions immediately before close.
In April, researchers questioned the “wisdom of the crowd” on Polymarket. They estimated that 3.14% of the platform’s users, together with market makers, capture more than 30% of profits.
