A group of researchers from Stanford University and the Singapore Management University has presented evidence of systematic manipulation on the five-minute Bitcoin contract market of the Polymarket platform. An analysis of over 60 million on-chain records, order book data, and high-frequency spot flows identified 821 wallets that earned $8.22 million from anomalous cycles, while retail and unclassified participants lost $7.61 million.
Mechanics of the Alleged Scheme
The five-minute contract, launched on February 12, 2026, pays $1 if the Bitcoin price at the end of the window is not lower than the initial value, and $0 otherwise. The calculation uses the BTC/USD price feed from Chainlink. Researchers identified a pattern: traders first open a position on the contract, and then, seconds before the calculation, execute large spot trades, temporarily moving the market in the desired direction. The oracle records the altered price, the contract settles in their favor, after which quotes partially return to their original level.
After the launch of these contracts, the directional order flow in the last ten seconds of the window increased by 50% compared to the period before their introduction. In cycles where the outcome remained close to equally probable, the activity surge was 3.9 times stronger than normal. The average volume of directional spot trades in the last ten seconds of allegedly manipulated cycles reached $1.7 million, compared to $68,000 in others. About 56% of such episodes occurred during nighttime hours UTC, and another 44% on weekends, explained by lower liquidity requiring less capital to shift the price.
Scale and Consequences
Researchers classified 1,613 cycles as allegedly manipulated—the top 10% by intensity of directional trades. Wallets that participated in at least five such cycles with a total profit of no less than $2,000 were categorized as likely manipulators. 821 wallets met the criteria (0.34% of the 243,000 addresses that traded the contract). They earned $8.22 million in anomalous cycles and only $90,000 in others, while the retail/other category lost $7.61 million in the former—this represents 93% of the profit of the alleged manipulator group.
Particularly telling were cycles where Polymarket estimated the probability of one side winning at 90–100%. With directional movement against the favorite, the opposite side won in 34.2% of allegedly manipulated cycles, whereas without anomalous activity, it won only 1% of the time. The authors ruled out the alternative explanation of hedging: the average maximum exposure of market makers was about $1,400 per cycle, while the spot flow was $1.7 million, incompatible with normal hedging.
Platform Response and Conclusions
A Polymarket representative stated that the platform uses several independent price oracles and plans to transition some markets to a mechanism that considers prices over a longer period. In fifteen-minute contracts, the activity surge was significantly weaker, and no pronounced price reversal was observed. Other protection options include calculating based on the average price over a period, randomly selecting the fixation moment, limiting position sizes, and imposing additional costs for operations before closure.
My comment: This situation is a classic example of how institutional players exploit technical vulnerabilities to profit at the expense of retail traders. Polymarket urgently needs to implement more robust calculation mechanisms, otherwise trust in the platform will be undermined. Prediction markets should be fair, not an arena for manipulation.