Imagine placing a wager on a coin flip after you already know how it landed. That is, in essence, what 821 accounts were pulling off on Polymarket — and researchers have now attached a figure to it: $8.2 million extracted from settlement windows they flagged as likely manipulated.
Polymarket has since overhauled the way its short-dated crypto contracts settle. The single-price snapshot has been scrapped in favour of a time-weighted average price, or TWAP. Getting there required months of trader complaints plus an academic paper.
The replacement formula is precise. According to the platform, five-minute markets will rely on a 30-second average, while 15-minute and four-hour markets will use a 60-second average. It added that Chainlink Data Streams will supply the data.
There was nothing clever about the exploit
Any trader active in these markets could have walked you through it. Variance Lover, a pseudonymous onchain analyst, spelled the whole thing out in a detailed post dated May 21 — long before the study appeared.
“By now, most people are aware that market manipulation has become a major problem on Polymarket’s 5-minute crypto markets. The mechanism is simple: accumulate a large position on Polymarket, then move the price on Binance during the settlement window to force the market to resolve in your favor.”
That is the entire play. Load up on your preferred outcome in the prediction market, then push the underlying spot price in that direction for a handful of seconds. Because the contract reads a single number at a single instant, whoever owns that instant owns the payout.
What the researchers actually found
Working out of Stanford University and Singapore Management University, the authors reviewed roughly two months’ worth of five-minute bitcoin contracts. They went hunting for outsized Binance orders in the closing seconds before settlement — and found them, trailed by rapid price reversals in bitcoin.
That reversal is the giveaway. When a price lurches hard and then snaps straight back the instant there is no money riding on it, the market is not discovering anything.
“The vulnerability is structural,” the researchers wrote. “An asset-price contract settles on a financial price, and that price can be moved by trading the underlying market itself.”
And here is the sentence that ought to unsettle anyone who traded these contracts: “A bet the market treated as near-certain was overturned one time in three,” the authors wrote.
The paper is upfront about its limits
Intent was never proven, and the study stopped short of directly showing that the spot-market orders came from the same parties holding Polymarket positions. That is a candid caveat and deserves to be said outright.
Still, where the damage landed is difficult to dismiss. Set market makers aside, and 93% of the losses inside windows classified as manipulated were absorbed by retail traders.
Polymarket did not respond to an email requesting more information.
Back in May, a developer said not to worry
Traders were flagging this months ahead of the study. On May 11, an Axis Robotics contributor posting as Christine on X pointed out that manipulation in Polymarket’s five-minute bitcoin market was deteriorating, pointing to “precise reversals in the last few seconds.”
Polymarket developer Josh Stevens responded: “we are looking into this a bit deeper – don’t worry.”
The fix landed only after the paper did.
Polymarket is spending to keep the order books stocked
“To protect market integrity in our crypto up/down markets, we’re updating how these markets resolve,” Polymarket said in the X post laying out the changes. “To support liquidity through this transition, we’re adding $1M in liquidity rewards across all impacted markets through the month of August.”
It is that second sentence worth dwelling on. Rewriting settlement mechanics mid-stream generally spooks the people quoting prices, and $1 million is Polymarket’s estimate of the bill for keeping them seated through August.
Kalshi claims immunity. An X user disagrees
Kalshi developer IcoBeast.eth posted on X that “this problem doesn’t exist on Kalshi fwiw.” Tomdnc shot back: “it literally does happen on Kalshi. I have seen with my own eyes.”
A Kalshi spokesperson said the venue settles its markets against a regulated CF Benchmarks price index, and that investigating suspicious behaviour is easier there because every trader is identity-verified.
Offshore markets can move prices, the spokesperson conceded. But the company argued that its 60-second moving average, drawn from regulated exchanges, makes short-lived price pushes “significantly harder and more expensive” than on venues relying on instant snapshots. Artificial moves, Kalshi added, get corrected quickly by arbitrageurs.
Look closely at that figure. Kalshi’s safeguard is a 60-second average — precisely what Polymarket has now adopted for its 15-minute and four-hour markets. Its five-minute markets receive half as much.
The enforcement figures are where the gap really shows
Kalshi said it has run 150 to 250 material investigations each quarter and has made roughly 40 to 50 referrals to the Commodity Futures Trading Commission so far this year. Those totals cover the company as a whole and are not confined to its short-dated crypto markets.
Identity checks plus a regulator you can hand cases to amount to a different species of deterrent than a wider averaging window. A TWAP drives up what an attack costs. It does not tell you who launched one.
What changes for you if you trade these markets
With a 30-second average on a five-minute contract, a manipulator has to sustain an artificial price for a full 30 seconds rather than clipping one tick. That demands a far bigger position, held far longer, in the face of arbitrageurs happy to take the other side. The economics deteriorate quickly.
None of this makes the attack impossible. It makes it costly — and costly is the only real defence a permissionless market has.
If you trade five-minute crypto contracts anywhere, the questions to ask are how long the averaging window runs and where the price feed originates. A snapshot lifted from a single exchange is a settlement price with a lever bolted on, and the researchers have now priced what pulling it is worth: $8.2 million, spread across 821 accounts, taken mostly from retail.