$33.8 billion of trades on BNB Chain. One wallet, 126 new pools, the same USDT.

01 · A Friday on BNB ChainOne wallet, $2.88 billion in a day

On Friday, August 28, one wallet on BNB Chain recorded $2.88 billion of trades on PancakeSwap. That day, DefiLlama counted $1.13 billion of trading on all the DEXs it tracks on the chain put together.

The wallet never needed $2.88 billion. It moved about $1.2 million of USDT at a time into a small pool and pulled it straight back out, 150 times in each transaction. Every move was recorded as a trade.

It did this every day for a month. In 128 transactions it recorded $33.8 billion of trades, about a third of all the DEX volume we counted on BNB Chain in that time.

Bitquery indexes DEX swaps on BNB Chain. For a count of fake trading on six chains, we read all of them from August 24 to September 22, and this wallet was the biggest single source of volume we found on BNB Chain. So we followed its USDT through the pools, and then the contract its transactions go to. That contract also carries the trades of a farm of 5,500 bot wallets we first ran into on copies of the LAPTOP memecoin.

The trick costs almost nothing, because the USDT ends exactly where it started. At least one big tracker counted it anyway.

02 · Why fake itVolume is the number trackers rank by

PancakeSwap is one of the biggest DEXs on BNB Chain. Its volume, the dollar value of every swap, is the number trackers use to rank pools and tokens. A new token with a big one looks busy and liquid, and that gets it noticed.

So volume is worth faking. Wash trading means trading with yourself so a market looks busier than it is. On Base, we found one wallet that faked most of August's DEX volume with Uniswap V4's flash accounting. The BNB Chain version is plainer. It runs on USDT and on brand-new PancakeSwap V2 pools.

03 · The loopHow to record $300 million of trades in a $22,700 pool

A PancakeSwap V2 pool is two piles of tokens, here USDT and a new token, and the price is the ratio between them. Buy the token with USDT and the price climbs. Sell it back and the price falls.

Take one of that Friday's transactions. Before it ran, its pool held about $22,700 of USDT.

A $1 million buy into a pool that small hands the buyer most of its tokens and sends the price soaring. Selling the same tokens straight back brings the price down to where it started, give or take the fee. Both swaps really happen, and each is recorded at $1 million.

The transaction did that 150 times, and the USDT ended right where it began. So did every other loop, to the cent. In the 12 loops we traced in full, the money only ever moved between the pool and one of 3 wallets that held the USDT and the tokens. We call them the holding wallets.

The pools can be tiny because a swap is recorded at the size of the USDT pushed in, whatever the pool holds. A small pool ties up little money, and the $1 million comes straight back. The median swap was 53 times the USDT its pool held before the loop, and only 7 loops ran through a pool holding $100,000 or more.

04 · The bill$38 of gas for $33.8 billion

PancakeSwap's V2 pools charge 0.25% on every swap. On $33.8 billion that sounds like $84 million. But in a pool about 50 times smaller than the swap, each round trip costs only a small slice of what the pool holds. That is the other reason the pools were tiny.

The loops paid even that slice in their own coin. Each sell handed the pool slightly more tokens than the buy had taken out, so the fee was paid in the new token. At each pool's price before its loop, those extra tokens were worth about $4.2 million. Paid in USDT, the same fees would have come to about $4.0 million.

Most of it went back home. In PancakeSwap's V2 pools, about two thirds of each fee goes to whoever provides the liquidity and the rest to PancakeSwap, according to its docs. We checked three pools, for tokens called MarsCoin, GOLD and pPOLY. In each one the liquidity came from a holding wallet, and the LP tokens went straight back to that wallet. Most of the fees moved from one pocket to another.

The big wallet's own gas for the month came to $38.

05 · The bot farm5,500 wallets, one contract

The big wallet did not work alone. It signs each transaction and sends it to a contract, 0xf9ffcdc5, which in every loop we traced moved the USDT and tokens out of a holding wallet. One of the holding wallets deployed that contract on August 12.

Plenty of other wallets use it too, 5,500 of them, all trading like bots. The larger ones recorded $8.0 billion between them, and the rest traded $1 at a time.

Every one of them was already trading on the first morning of our window and still at it on the last day. The farm wallets do not seem to hold the money they trade. In a farm trade we opened, the tokens came from a holding wallet and the USDT went back to it, and the farm wallet only signed and paid the gas.

Almost every token the larger wallets traded was one the big wallet also looped, and they split their dollars about evenly between buying and selling. In the half minute before the MarsCoin loop, a stream of $1 trades hit that pool, buys and sells, paid from the holding wallet that had seeded it.

We first wrote about this farm in September, when it traded copies of the LAPTOP memecoin around the real token's launch on Base. That piece counted 43 tokens in seven days, and it described one address as sending the farm's trades. Each wallet in fact signs and pays for its own. What they share is the contract they send them to. Over our month, the contract carried $41.8 billion of recorded trades, the loops and the farm together. Who funds the holding wallets we have not traced. The trail can be followed with the Bitquery MCP.

06 · The tokens126 tokens, most gone within a day

The loops and the farm went after the same targets, a stream of brand-new tokens. The big wallet used 126 pools, each for one new token, and rarely came back to one. The tokens were traded for a median of 7.7 hours from first trade to last, most of them for less than a day.

Every one of the tokens had its first recorded trade less than 3 minutes before its first loop. We checked five launches by hand. Each token was deployed in the same transaction as its PancakeSwap pool, about 2 to 4 minutes before the loop. In the two where we traced the supply, all 1 billion tokens went straight to a holding wallet.

Sixteen tokens were called LAPTOP, after a memecoin that launched on Base, and none of them was the real one. Others went by names such as MarsCoin, GOLD, ARC and pPOLY, and 44 had Chinese-language names. Whoever deploys a token picks its name, and a name says nothing about who is behind it.

Why do it? The chain does not say. A token with hundreds of millions of dollars of trades on its first day can climb tracker lists, and a crowd of thousands makes it look popular. Whatever the aim, few others joined in. The big wallet and the farm made at least 97% of all the trading we recorded in the 126 tokens, and we did not measure whether the rest lost money. Whether faked volume pulls in real buyers is a question we tested on Solana, in our study of wash trading and bots.

07 · Who counted itOne tracker counted the loops, one did not

GeckoTerminal, the DEX tracker run by CoinGecko, counted the loops. We checked three of the biggest pools, and for each one its volume on the loop day matches the total we recorded for the token that day to within 0.3%. The loops made most of those totals. Its API still listed the MarsCoin pool on September 23, when the three pools held between 3 cents and $14.

DefiLlama's totals are too small to hold them. On August 28 the wallet alone recorded $2.88 billion, more than twice DefiLlama's figure for the whole chain. Over the month DefiLlama counted $36.5 billion on BNB Chain. Had the loops been in it, everyone else on the chain would have traded $2.7 billion in a month, less than this one wallet recorded on its busiest day.

Bitquery's own DEX data records these swaps like any other trade, and so do most feeds that read swap events straight from the chain. The figures here come from pulling them apart.

08 · How to spot itThree signs of a loop

Each of these can be checked from public data.

• The swap is bigger than the pool. A $1 million buy in a pool holding $20,000 multiplies the price by thousands. Nobody buying a token for itself trades like that.
• The token is minutes old. Each of these tokens recorded millions of dollars of trades within 3 minutes of its first trade.
• The crowd shares one private contract. A public router proves nothing, because everyone uses one. Here 5,500 wallets sent every trade to one contract that nobody else used, and a wallet that funds the pools deployed it. A pool's top traders, and where their transactions go, show it.

On September 23, the day we finished counting, a wallet from the farm was still sending trades to the contract.

09 · How we measuredWhere the numbers come from

We read every DEX swap on BNB Chain from August 24 to September 22. For this wallet we also checked each pool's USDT before and after its loop from the pool events, traced 12 loops in full, pulled the gas for every transaction from a public BNB Chain node, and checked five token launches and three liquidity deposits on-chain. Dollar values come from the USDT side of each swap. GeckoTerminal and DefiLlama figures come from their public APIs, read on September 23.

In our six-chain count of fake trading, loops like these sit in their own bucket. A no-cost loop worth $1 million or more, with 10 or more swaps in one transaction, counts as neither fake nor real there, because one wallet this size would swamp every other number. The larger farm wallets' trading counts as fake, and the $1 wallets fall under our floor. Our BNB Chain totals also leave out trades on launchpad bonding curves, whose dollar values we could not check.

10 · The recordThe addresses behind the story