Galaxy Finds Retail Polymarket Accounts Lost

Galaxy Research on 1 October 2026 published an onchain study of who makes and loses money on Polymarket's international platform. The headline result is blunt: Galaxy Polymarket retail accounts lose money. Of 2.9 million accounts trading at a human pace, 69.2 percent finished below break-even, and the group was down $338.9 million in aggregate. The study is not about the US app.

What happened

The paper uses settlement data indexed by Stork. Onchain, in plain words, means the researchers worked from records on the public ledger rather than from a private customer file inside a brokerage. Settlement is the moment a contract pays out. An index of those settlements is a way of lining the payouts up so accounts can be scored against what they put in.

Galaxy's choice of source material is part of the result. The figures are only as broad as the ledger and the index the researchers used.

The cut that defines the finding is pace. Galaxy kept accounts that trade at a human pace and set automated accounts aside. The human-pace group is large: 2.9 million accounts. Inside it, 69.2 percent finished below break-even.

Break-even here means the account did not come out ahead after its trades. Below that line, the account lost. The losses are not only a headcount. Added together, the human-pace accounts were down $338.9 million.

The automated accounts were excluded from that tally, and they finished in the other direction. As a group they were up $246.8 million. Galaxy's design therefore splits the platform into two populations and reports a separate result for each. It does not fold the machines into the retail figure, and it does not present one blended number as the experience of a typical person.

The $338.9 million and the $246.8 million answer different questions. One is the aggregate result of accounts trading at a human pace. The other is the aggregate result of the automated accounts that were left out of that group.

The platform under the microscope is Polymarket's international venue. Galaxy said so, and the distinction is not a footnote. The study is not a study of the US app. A reader who carries the percentages across to a different Polymarket product is leaving the sample.

CryptoBriefing reported the same figures on the same day. The primary write-up is Galaxy's paper. The same-day recap is CryptoBriefing's report.

Why it matters

Prediction markets are often described as machines for finding a price on a future event. Contracts pay if the event happens and pay nothing, or pay the other way, if it does not. That description is about the market. Galaxy's paper is about the accounts.

A market can produce a clean price and still distribute gains and losses unevenly across the people who trade it. The October study is a count of that distribution on one platform, over the settlements Stork indexed, for accounts the researchers classed as human-paced.

The scale is what makes the count hard to wave off. These are not a handful of named traders. They are 2.9 million accounts. The paper puts 69.2 percent of the human-pace accounts below break-even.

The aggregate of $338.9 million is the sum of those outcomes, not a typical loss for each account. The study, as reported, does not hand the reader an average loss per account, and this article will not manufacture one by dividing the totals. Aggregate and share answer different questions.

Both are in the paper. A per-account typical figure is not.

Leaving the automated accounts out is a methodological choice with a result attached. Those accounts finished up $246.8 million. The paper therefore does not say that everyone on the international platform lost money. It says that the accounts trading at a human pace, taken together, lost, and that most of them finished below break-even, while the automated accounts that were excluded finished ahead in aggregate.

Specialization is in the title of Galaxy's work for a reason. The split between human pace and automation is the specialization the numbers illustrate. The paper does not, in the facts used here, identify who runs the automated accounts or what strategies they use. Those names and methods are not invented below.

Two limits should travel with the figures. First, the sample is the international platform, not the US app. Second, the window and the market mix are whatever Stork-indexed settlements Galaxy studied.

This article does not add a start date, an end date, or a list of topics the contracts covered. The honest reading is the reading that stays inside those bounds: on the ledger Galaxy examined, human-pace accounts as a group lost $338.9 million, and 69.2 percent of them finished short of break-even.

What's next

Galaxy has published the study. CryptoBriefing has repeated the figures. Neither document, as used here, is a forecast of the next month on the platform, and this article does not become one.

The finding is historical. It describes accounts that have already settled, scored from onchain data, with automation taken out of the retail tally.

The figures to remember are the ones Galaxy printed and CryptoBriefing matched on 1 October 2026. Human-pace accounts: 2.9 million. Share below break-even: 69.2 percent. Aggregate result for that group: down $338.9 million.

Automated accounts, excluded from that group: up $246.8 million. Source of the settlements: Stork-indexed data, read onchain. Venue: the international platform, not the US app.