But analysts at two crypto research firms have found evidence of rampant wash trading on Polymarket, even as its odds have been shared widely across social media and mainstream media outlets. Donald Trump currently has a 67% chance of winning according to the platform.

In separate investigations completed by the blockchain firms Chaos Labs and Inca Digital and shared exclusively with Fortune, analysts found that Polymarket activity exhibited signs of wash trading, a form of market manipulation where shares are bought and sold, often simultaneously and repeatedly, to create a false impression of volume and activity. Chaos Labs found that wash trading constituted around one-third of trading volume on Polymarket’s presidential market, while Inca Digital found that a “significant portion of the volume” on the market could be attributed to potential wash trading, according to its report.

While other prediction markets including Kalshi and Robinhood have launched in the U.S. since a pivotal court decision in September legalized electoral betting, Polymarket remains the largest platform by far, thanks in part to its crypto-native design and offshore operations. Polymarket remains inaccessible to U.S. investors. Still, with less than a week until the U.S. elections, the suspicious activity on Polymarket raises questions about the accuracy of the site, which its 26-year-old founder Shayne Copland has touted can “demystify the real world events that matter most to you.”

Polymarket did not respond to a request for comment.

The rise of prediction markets

Founded in 2020 and backed by VCs including Peter Thiel’s Founders Fund, Polymarket tried to launch electoral betting in the U.S. before being forced offshore by the Commodity Futures Trading Commission in early 2022.

Unlike competitors such as PredictIt and Kalshi, which recently prevailed in a lawsuit against the CFTC to operate in the U.S., Polymarket runs its platform on the blockchain Ethereum. Coplan says the crypto element offers greater visibility into its trading activity. “The beauty of Polymarket is it’s all peer-to-peer and transparent,” he recently posted on X.

Polymarket volume exploded during the recent presidential election, with outlets from the Wall Street Journal to Fortune reporting the betting odds on its platform alongside more traditional metrics such as polling data. As a sign of increased credibility around prediction markets, polling star Nate Silver joined Polymarket as an advisor in July.

Polymarket’s crypto design and offshore operations have drawn scrutiny from other quarters. This includes recent reports claiming there is manipulative trading on the site—most notably by a single French trader who allegedly caused Trump’s odds to soar. Polymarket has insisted that the user has “extensive trading experience” and was not acting nefariously.

Wash trading

The evidence of wash trading appears to be a serious sign of misbehavior on the platform. To conduct its analysis, Chaos Labs looked at on-chain data to isolate high-volume traders, filtering out users who were likely engaged in normal activities like market making. It then separated users who exhibited signs of wash trading, examining their ratio of buy and sell orders and taking account of their share holdings compared to their trading volume. Chaos Labs concluded that around one-third of trading volume—and overall users—on the presidential market alone was likely wash trading, along with across all markets.

The practice is common across crypto applications, and especially ones with the potential for future token launches and airdrops, with users often earning tokens based on activity. The Information reported in September that Polymarket has explored launching its own proprietary token.

“The challenges of prediction markets are not unlike those of any other application with a market,” said Omer Goldberg, the founder of Chaos Labs, which is backed by Haun Ventures and develops data integrity software. “Wash trading is not specific to Polymarket.”

Trading volume

Both Chaos Labs and Inca Digital found another anomaly on Polymarket: The purported trading volume on its presidential market, reported in U.S. dollars on Polymarket’s website, does not match the on-chain data. Inca found the actual transaction volume on the presidential betting market to be around $1.75 billion, compared to Polymarket’s reported figure of $2.7 billion.

Chaos Labs attributed this to Polymarket conflating traded shares with U.S. dollars. To put it more clearly, users can buy shares of candidates at different odds. A “yes” share of Hillary Clinton for president costs just $0.01, given the vast unlikelihood that she will be elected, but Chaos Labs found that Polymarket is reporting that share as $1 of volume.

This discrepancy, along with the wash trading, underscores the untested nature of a platform on which many rely for signals about the presidential election.

Polymarket’s decision to operate on blockchain rails, however, also means that researchers such as Chaos Labs and Inca Digital can analyze activity. Coplan has described users’ ability to audit Polymarket as a “feature, not a bug.”

“These companies want to attract real users and build trust in their markets,” Goldberg said. “Identifying and reducing wash trading is critical to ensure that prediction markets are representative of everybody, with market prices and volumes determined by an authentic, enduring user base rather than being muddled by inorganic flow.”