A deep analysis of on-chain data and high-frequency trading has revealed systematic manipulation on the Polymarket prediction platform. A team of researchers from Stanford University and the Singapore Management University discovered that a group of 821 wallets managed to extract $8.22 million in profits by artificially influencing the settlement price of five-minute Bitcoin contracts.

The attack mechanism is elegant in its simplicity. A trader opens a position on an "up/down" contract seconds before its expiration. Then, using a large spot order on an exchange, they push the Bitcoin price in the desired direction. The Chainlink oracle records the altered price, the contract settles in the manipulator's favor, after which the market partially recovers. Researchers recorded that directed order flow in the last 10 seconds of the window increased by 50% after these contracts were launched, with a price reversal beginning within the next 10 seconds.

The Price of Liquidity and Night Attacks

Most indicative is the behavior during cycles with high uncertainty. When the outcome was assessed as 50/50, the spike in manipulative activity was 3.9 times higher than normal. The average volume of spot trades in the last seconds of such cycles reached $1.7 million, compared to just $68,000 in normal ones. Meanwhile, 56% of attacks occurred during nighttime hours and 44% on weekends, when market liquidity is at its lowest. This confirms the thesis: shifting the price during periods of low liquidity requires significantly less capital.

The statistics on "nearly resolved" outcomes look particularly cynical. In cycles where Polymarket assessed the probability of one side winning at 90-100%, a manipulative move against the favorite led to its defeat in 34.2% of cases. Without anomalous activity, such a reversal occurred in only 1% of cases. This is direct evidence that a short-term spot impulse can overturn market consensus.

Who Pays for Manipulation?

The economics of the scheme are simple: retail traders and unclassified participants lost $7.61 million on these anomalous cycles, accounting for 93% of the profit of the suspected manipulator group. In the overall loss structure, retail accounted for 70.5%, cycles unfavorable to manipulators for 20.2%, and market makers for 9.3%. It is important to note that researchers do not claim direct intent, and the classification is based solely on statistical patterns of participation in anomalous cycles and profits earned.

Polymarket has already announced plans to change the calculation mechanism, moving to price averaging over a longer period. Fifteen-minute contracts, as the study showed, exhibit a significantly weaker spike in activity and no pronounced price reversal after settlement. This is logical: the longer the window, the harder it is for a single spot trade to change the outcome.

My expert opinion: This study is not just academic work but a warning signal for the entire DeFi industry. The problem is not with Polymarket per se, but with a fundamental design flaw: using a single price snapshot to settle contracts with high volatility. Solutions such as TWAP (Time-Weighted Average Price) or random settlement timing should become the standard for all prediction platforms; otherwise, the "wisdom of the crowd" will be constantly undermined by the "power of the whales."