928 of 1,351,870 Polymarket markets got most of their volume from two accounts trading with each other.

In August, Case File 06 found that the two highest-volume contracts in Polymarket’s cartel-arrest event were almost entirely one pair of accounts trading with each other: $117,153.35 of displayed volume for a combined net cost of $171.71, confirmed transaction by transaction on the Polygon blockchain. The natural next question was how common that is. This page answers it, and it answers it with a count over every market rather than a sample.

The mechanism is simple once the object is clear. A Polymarket contract pays $1 if its outcome happens and nothing if it does not. Every market has a Yes contract and a No contract, and exactly one of them pays, so a Yes and a No held together are worth exactly $1 whatever happens. Polymarket creates a new pair whenever one trader pays for the Yes and another pays for the No at prices that add up to $1, and it lets anyone holding a pair hand it back for $1 at any time. Every one of those purchases and sales counts, at the full face value of the contracts, towards the volume figure shown on the market’s page.

Here is what that looked like in the market on whether Ricardo Ruiz Velasco would be arrested. One account bought 30,010 Yes contracts at 96 cents each while a second account bought 30,010 No contracts at 4 cents each, so together they had paid $1 for each new pair. About two minutes later the first account sold its Yes contracts at 26 cents and the second sold its No contracts at 74 cents: in effect the pair handed the contracts back for $1 each, and the two sale prices only decided how that dollar was split between them. The first account lost about $21,000 and the second gained about $21,000; between them they were $139.98 worse off after fees, and neither was left holding anything. The market’s volume had risen by $59,780.04. Those two transactions are 0x1002e4dffb5d02c1a0811b821e3bbe3f527f6d80d17d583bd8bc9835d4481082 and 0x9d25ed69565fb68fc2b853ec79b57dd1e4bb9290bdd2dcbba26b4a4b9f4aa572, 57 Polygon blocks apart, and they are 99.7% of everything that market traded in the window.

We call a market mostly recycled when at least half of its trading volume in the window belongs to episodes of that pattern: two accounts trading with each other, both buying and selling, both ending where they started, at close to no combined cost, within minutes. The rule that identifies an episode was written down and fixed before the count was run; it was tested against the cartel markets, which it flags at 99.7% and 99.5%, and against ordinary markets, which it does not flag. Its parameters are not published on this page.

The count covers every market other than United States elections that traded on Polymarket’s current exchange contracts between April 28 and September 9, 2026: 1,351,870 markets and 463,018,840 individual fills, reconstructed from the exchange contracts’ own on-chain OrderFilled events with every maker leg retained. Because every market was examined, the figures below are counts, not estimates, and carry no sampling margin.

Two things the rule does not do, so that the reader knows what the count leaves out. It only sees loops between two accounts that close quickly; rings of three or more accounts, and loops that take longer to close, are not counted. And it describes trading mechanics, not motive: a trader who continuously quotes both sides of a market and whose quotes are taken by one regular counterparty, or an account chasing Polymarket’s liquidity rewards, can produce the same pattern. Nothing on this page is a finding that any account acted improperly.

By count it is rare. 928 of 1,351,870 markets were mostly recycled, which is 0.07%, or about one in 1,457. By money it is not rare: those 928 markets hold $506.4 million of recycled trading, which is 96.4% of their combined volume, and across the whole platform recycled trading came to $540.6 million of $24.4 billion, or 2.2% of every dollar traded in the window.

The threshold matters less than it might seem, because markets tend to be either mostly recycled or barely touched. Requiring recycling to be at least 90% of volume instead of half still leaves 531 markets. Counting any market with even one recycling episode, however small, gives 8,709 markets, or 0.64%, but most of those episodes are tiny against the market they sit in.

The bigger the market, the more often its volume is mostly recycled. Among the 853,171 markets with less than $1,000 of trading, 138 qualify, about one in 6,182. Among the 2,996 markets with $1 million or more, 50 qualify, about one in 60.

The largest cases are political. The seven markets on who will be Ethiopia’s next prime minister hold $284.0 million of recycled trading between them, more than half of all the recycled dollars in the 928 markets, and in each of the seven recycling made up more than 99% of the volume. Sports markets appear as well, though the sums are smaller.

The full list of 928 markets, with each one’s question, volume in the window, recycled dollars and share, is published beside this page as artificial-volume-census.json. Every figure in the tables above is recomputed from that file when the page is built.

The activity is concentrated in a small number of accounts. 13,369 distinct accounts took part in a recycling episode inside one of the 928 markets. Even so, ten accounts were on one side or the other of 81% of the recycled dollars in those markets, across 163 of them, and three accounts alone were on one side of 50%. The largest single account took part in $147.9 million of recycled trading across 24 markets. The same accounts that dominate the Ethiopia markets also appear in unrelated markets.

The same account appearing in many markets shows that this is repeated behaviour rather than a scattering of one-off episodes. It does not tell us who is behind an account, or whether the two accounts in any pair belong to the same person. This page does not name accounts; the wallet pair in the cartel markets is already identified, on-chain, in Case File 06.

• We are not saying why any account traded this way. Continuous two-sided quoting taken by a single regular counterparty, and trading to earn liquidity rewards, can both produce the pattern, and this page did not attempt to distinguish those from deliberate volume inflation in any market.
• We are not saying that any pair of accounts is controlled by one person. Address recurrence is an observation about addresses.
• We are not alleging fraud, manipulation, coordination, or a breach of Polymarket’s rules by anyone.
• We are not saying the count is complete for every kind of recycled volume. Rings of three or more accounts and slower loops are outside the rule, so 928 is a floor for the broader idea, and exact only for the rule as written.
• We are not saying anything about trading before April 28, 2026, about United States election markets, or about Polymarket’s lifetime displayed volume, which is a different figure from trading inside this window.

What the count does establish is narrower. In 928 markets, the volume figure on the page mostly measures two accounts trading with each other rather than many people taking a view, and the markets where that happens most are the large political ones that draw the most attention.

Source. Every trade is reconstructed from the OrderFilled events emitted by Polymarket’s exchange contracts on Polygon between blocks 86,107,336 and 93,524,955, with every maker leg of every match retained and each fill’s route (a direct exchange of one contract, or the creation or cash-in of a Yes-and-No pair) resolved from the legs themselves. Volume is counted once per fill, at the maker’s cash for a direct exchange and at the pair’s $1 face value for a creation or cash-in.

The rule. An episode is a sequence of trades between two accounts, close together in time, after which both accounts hold what they held before it to within a small tolerance, the pair’s combined cash has changed by only a small fraction of what was traded, and the two accounts were trading almost exclusively with each other during the sequence. The tolerances and the time limit were fixed before the count and are not published. A market’s recycled share is the volume covered by at least one qualifying episode, each fill counted once, divided by its volume in the window.

Coverage. 2 markets in scope had no fills we could attribute, and 64,133 fills, about 0.01% of the total, could not be matched to a market. If every one of those were mostly recycled the headline share would rise to at most 0.14%.

Checks. The count was run over the whole window in independent blocks and reproduced exactly when the blocks were cut differently. It agrees, fill for fill and dollar for dollar, with a separate whole-market measurement on the 2,617 markets of an earlier probability sample, and a second implementation of the rule, written independently from the rule’s text with exact arithmetic, reproduced the census exactly on 30 randomly chosen markets from the list of 928.

Limits. The rule sees two-account loops that close quickly and nothing else, so it undercounts any broader notion of recycled volume. It describes what the trading record shows, not why anyone traded. Shares are measured against trading inside the window, not against the lifetime volume Polymarket displays. Where the record does not answer a question, this page says so rather than inferring.