← Prometheus7 · Oracle731 Oct 2026

We Thought Three Wallets Ran Pump.fun. We Were Wrong.

Benjamin Horn · Prometheus7

Our first read said three funding sources sat behind 39% of pump.fun launches. That number was wrong. Finding out why turned up something stranger, and a smaller, more useful truth.

Correction, 31 Oct 2026: an earlier version of this page published the 39% figure and described the largest sources as operators. Several of them are exchange-size or program-held wallets, so this page replaces it.

Updated 31 Oct 2026: names the bridge, explains the route, replaces the 6 SOL estimate, and adds whole-curve launches and the full tape. Updated 1 Nov: what the pools show, the money trail, the network tests.

We built a live verifier that reads every pump.fun launch directly from the chain. In a little over a day it recorded 26,068 launches from 7,986 dev wallets. Of the 24,177 that settled, 338 bonded (1.4%). In 77%, the dev dumped within 15 minutes or the coin never got going.

To find who was behind them, we followed the money: each dev wallet back to the wallet that first funded it. We traced 12,883 launches that way.

The top of the list looked explosive. One wallet alone sat behind thousands of launches. So we asked how much it had taken out.

The biggest wallet is not a person

6tckHFBpiJ8YgYN8FUskvtvTpXQZ55g5LHeo1kvELoDQ funded 19 dev wallets behind more than 3,600 launches. It holds about 13,000 SOL, roughly $1.26M.

We sampled 200 of its transactions. It signed none of them.

Every one ran through a single program, dahPEoZGXfyV58JqqH85okdHmpN8U2q8owgPUXSCPxe. The instructions are called InitTransfer and FinalizeTransferSol, which is the language of a bridge. The payments in came from 99 different wallets, and the payments out went to 54 different wallets. Only 2 of those 54 launched a coin we saw.

So the 13,000 SOL is not anyone's winnings. It is money sitting in a program's vault, passing through for many people. It is the Solana vault of NEAR Omni Bridge. The program's code names that bridge's contract, and the bridge's public software lists this program address. In our sample, 311 of 313 transfers it started went to NEAR Intents, a cross-chain swap service. The bridge and Intents are the road here. Nothing we found shows they run these launches.

The launch machine behind it

The dev wallets it funded are a different story. We read the complete history of the ten busiest.

Each of those 20 wallets has exactly two transactions: SOL in from the dev, SOL out to the same vault. The money comes out of the bridge, passes through a launcher, and goes straight back in.

Ten wallets behaving this alike looks like one script or one product. It does not prove one person.

What the machine earns

An independent read of the same ten wallets, from the raw creation transactions, found the engine. The bot copies other people's fresh launches: 1,011 of its 3,033 recorded coins reuse an earlier, different creator's name and image file within 30 seconds, one of them seven seconds after the original. It creates the copy on a custom trading pair quoted in the PUMP token, where the creator fee is set at 3%, ten times the 0.3% on a normal launch. It never buys its own coin.

Anyone who buys and sells the copy pays that 3% each way to the creator. The fees accrue in PUMP. Helper wallets claim them, swap them to SOL, and pay the SOL back to the launcher; the launcher's working capital then hops to the next fresh wallet through the bridge. Seven traced fee cycles returned about 3.5 SOL net; three conversion helpers paid back about 9.5 SOL gross. Our earlier figure of about 6 SOL undercounted, because fee proceeds also arrive through the vault.

So the launch spam is a fee skim on copied coins: thousands of launches, tens of SOL a day, run by one operator behind fresh wallets. It is not a liquidity drain.

What the route does

Swaps through NEAR Intents' public 1Click service can be read by address. That showed money moving between launch wallets as SOL-to-SOL swaps, each with its sender recorded. A SOL-to-SOL swap buys nothing. What it does is break the direct on-chain link between two wallets.

We followed one trail through 12 deployer wallets:

B72GfARi → 2m6TBCZo → HvUsAVta → FGi8uM81 → DdxvMc8G → CvTxQ3Rr → CULTckzd → 8p43ro2z → 31hRK8bB → 2bypcmSu → CP84f4B4 → CLjPvCQ8

Each step moved about 48.5 to 59 SOL. That is one pot travelling, not new money each time, so this trail is one operation with fresh wallets at every step.

So how much did it extract?

Not what we expected. We first put this at about 6 SOL. Following the pot gives a better measure: it grew by about 8.5 to 10 SOL in about 20 hours, under $1,000. Three side wallets collected about 3.9 SOL of creator fees and sent it back into the trail.

That covers only this trail. Fees sent elsewhere, other wallets trading these coins, and whatever sits on the far side of the bridge are all outside this count. But the 13,000 SOL headline is simply not there.

It was not the only one

Then we checked the other top sources. Several hold sums no launch operator needs: 204,000 SOL, 193,000, 101,000, 95,000. They were moving slowly enough to pass our exchange filter, which only tested speed.

These are almost certainly exchanges and custody wallets. Withdrawing from one tells you nothing about the people doing it.

We redid the count, removing every source holding 1,000 SOL or more. That took out 66 sources behind 48.4% of everything we had traced.

Three sources, 9% of traced launches Share of 6,652 launches, by the top sources out of 2,370 100% Top source3.0% Top 39.0% Top 1018.8% Top 2026.9% Top 5040.9%
Exchanges, hubs and every source holding 1,000 SOL or more removed.

So 39% was really 9%. It is still concentrated, since 24 sources out of 2,370 sit behind 29.5% of launches. But it is not three wallets running the market.

The rotators are real

Some of the operators held up, and they are easy to see once the exchanges are gone.

EndJzwh6… funded 193 dev wallets for exactly 193 launches. It holds under 10 SOL. None of its 174 settled coins bonded.

Across the 16 sources that use a fresh wallet for nearly every coin, that is 688 devs, 722 launches, and 6 bonded out of 620 settled.

On a launch feed, each of those is a brand-new dev with a clean record. Check the dev and you learn nothing. Check who paid the dev and it is the 193rd coin from the same money.

The funder record got stronger

Here is the part that survived, and improved. We replayed the launches using only what was known before each one, for devs who bought their own coin.

When the funding source had a record of dumping, the dev dumped 67.9% of the time, against 39.7% for everyone else. Those coins bonded 1.3% of the time, against 5.2%. That covers 545 flagged launches and 3,794 others.

With only the fastest hubs removed, it reads 65.4% against 40.2%. Removing the exchanges made the signal cleaner, because an exchange's record was never one operator's record.

Our verifier had the same flaw, so it could mark a dev as a rug risk over who else used their exchange. That is fixed.

So where did the money go?

Launch counts and money turned out to be different things. We added up about 23 hours of Solana trades, using roughly the first 50 trades of each coin.

The launch spam carries almost no money: the 6tck… coins saw about 104 SOL of recorded buying. The visible gains went to about a thousand fast wallets, not to one mastermind.

This is not final profit and loss. Later sells are not in the count, and about 30,000 SOL was still sitting in positions at the end of the recorded trades.

Where the pot came from

The 12-wallet trail above was the middle of something longer. We went back to the machine's two earliest wallets and asked where their first SOL came from. Both got it from the NEAR Omni bridge vault, which is where a 1Click SOL-to-SOL swap pays out on Solana. In the minutes before each payout, a deposit of the same amount went into the vault, and the public order behind that deposit names its sender. The sender was another launch wallet, whose own first SOL was also a bridge payout. So we repeated the step.

It repeats cleanly. Every wallet in the chain lives for 3 to 12 hours, makes 270 to 950 transactions, nearly all coin launches and first buys, then pushes the pot through SOL-to-SOL swaps into a brand-new wallet. The hand-off is always two orders a minute apart: one of exactly 5.0001 SOL, then the rest of the pot. Each order's deposit address is used once. Two chains of this run side by side:

Two of the wallets this turned up were already in the 30-wallet chain we had mapped from the swap records. So this is the same operation, running longer and in more than one lane, and the forwarding is the whole method: each hop is a fresh wallet with no on-chain link to the last one, except through a swap service's order book that happens to be public.

The two relayer wallets that sign every one of these bridge deposits and payouts, 9WL2A89Y… and DuSs7rCr…, sign the same for thousands of unrelated users a day. They are the service's own infrastructure, not a party to this.

The whole-curve operation does not use this route. None of its 90 farm wallets was ever paid out by the bridge. Its hub wallet 8Gj3puMf…, which has since funded 64,000 accounts, began on 24 Sep with a 0.1 SOL test payment from a year-old wallet, then took 700 SOL on 26 Sep from 5tzFkiKs…, which Solscan labels as a Binance hot wallet. So the hub's working capital was an exchange withdrawal, and four of the farm wallets were also first funded straight from that exchange. The year-old test wallet was itself first funded by AaZkwhki…, an unlabelled four-year-old wallet that has funded 142,000 accounts and holds 1,826 SOL and 276,000 USDC, the shape of a service rather than a person; it also appears among the 440 wallets that trade against the whole-curve pools. An exchange is where this trail ends, and an exchange is the one place on it that knows a customer.

Where the machine started

We kept stepping back through the launch machine's chain, and it changed shape twice before it ended.

Back past 21 Oct the hops stop being bridge payouts and become plain transfers: each launcher wallet was first funded directly by the one before it, about one new wallet every 12 hours, each a pump.fun creator that went on to fund 2,000 to 10,000 accounts of its own. That runs from 20 Oct back to 13 Oct, when the chain's first wallet, HgdcgE8g…, received 1.0, 33.0 and 34.09 SOL in four minutes from the bridge. No deposit on Solana matched those three payouts, and the reason turned out to be the method: two minutes earlier, 68.14 SOL had gone into the bridge from a single-use wallet and been credited to a holding account inside the swap service, and the three orders were then placed from that balance. The 68.14 SOL came from 2FK2Xc38…, a wallet that has been alive since 1 Oct, sells collected fee tokens through Jupiter and to market makers all day, and still receives a 0.04106 SOL gas top-up about twice a day. Until 13 Oct it was the hub itself: it paid each day's launcher directly, a new one every day or two, and took the pot back at the end of the session. The launcher it funded on the afternoon of 13 Oct handed back 68.1 SOL at 22:22; two minutes later the same money went into the bridge instead of to the next launcher. That is the moment the machine switched methods. The second chain, which ran in parallel from 17 to 29 Oct, started the same way: its first launcher's 4.28 SOL on 17 Oct was paid out of a balance inside the swap service with no matching deposit on Solana, and that balance had been built since 13 Oct by deposits of 4.2 to 4.4 SOL every few hours from the machine's own fleet wallets, one of which still receives the 0.04106 SOL gas top-up from the dispatcher line today. Both chains begin on 13 Oct.

That 0.04106 SOL payment is the machine's signature. The wallets that send it, one every two to four months, are each other's first funders in a straight line: 6k2sFrTE… (24,000 accounts funded) was funded by 9GETYoSJ… (24,000), by 8HkQdCbo…, by 2afYM1ea…, by 2p8xLqvN…, by AkquHwR3…, by 3E7wqYvQ… (25,000), and so on for 26 hops, every one a pump.fun creator wallet, every one funding hundreds to tens of thousands of accounts, back to RVHFQFCw…, which was first funded on about 20 Aug 2025 by a wallet Solscan labels as a Gate.io hot wallet. The 0.04106 payments are visible ten months back.

The treasury side runs older. The wallet that seeded the September launchers behind 2FK2Xc38…, 27y8KVLF…, holds 100 SOL and 4,870 USDC today, took 195 SOL in September from HicFqmdY…, moved 30 SOL out through a second bridge (Relay) on 25 Sep, and took 99.99 SOL back through NEAR Intents on 29 Oct, the morning every old launcher in the chain was emptied in the same hour. Its first-funder line goes HicFqmdY… (9,700 accounts), J7cHsYHq…, 9jzfY4fF… (two years old, 8,400 accounts), GVmest1Y…, and then a Binance hot wallet, in about August 2024.

So the launch machine is at least fourteen months old on one line and two years old on the other, it is one operation (the launchers and the whole dispatcher line pay or receive the same 0.04106 SOL top-up, the treasury line pays the launchers directly, and the old launchers were emptied in the same hour on 29 Oct), and both lines begin at an exchange withdrawal. Two exchanges, Gate and Binance, hold the customer records at the top of it. The whole-curve operation's hub also began with a Binance withdrawal. That is the same exchange, not the same customer: that wallet pays out to more than a million accounts.

Where the money goes

The treasury wallet 27y8KVLF… is where the machine's money comes to rest before it leaves. In five weeks about 986 SOL passed through it: 514 SOL as swap-service payouts (the launch chains' pots coming home), 426 SOL from a wallet that does nothing but sell tokens on pump.fun, and the rest from a third collector. It turns SOL into USDC by selling to market makers, 24,600 USDC to one on 7 Oct, 25,900 to another on 26 Oct, 9,500 to a third, and takes USDC payouts from the bridge as well, about 36,000 in October.

Then it moves the USDC out the same way it moves SOL. On 26 Oct it sent 45,550 USDC to a fresh single-use wallet, and 60,000 more through a side wallet to three others; all four were created minutes earlier by the swap service's relayer, and all four pushed the USDC into NEAR Intents. Seven hours later 45,549.98 USDC came back out of Intents to Solana, to a wallet that had never held anything, and twelve minutes after that it was forwarded, in one transfer, to 2chVhEZcaEXPtHa3LcawZXCKeRKBGWw7kPC8vajZfHJ9. Solscan tags that address as a Binance deposit address. The same path, bridge in, bridge out, same fresh wallet, same deposit address, carried 15,311 USDC on 1 Oct.

The other 60,000 USDC took longer to come back and went somewhere else. The three deposits behind it (15,000, 22,500 and 22,897 USDC, between 18:16 and 18:59) returned to Solana that evening as eight pieces of 5,626 to 11,253 USDC, 60,387 in all, every one to Hn1pemZ6TbaRDbYYVSveNsx6ULVSkMYRRwx2xBNFGQ43, a wallet about two and a half years old whose own first funder is a pump.fun token creator of the same age. That wallet sent a 100 USDC test and then 60,297.36 USDC to 9xxaGCtSVzvGVs3zTBjsRwPW4mv5CQZrsqqtGKMWAAD8, a second Binance deposit address, which sweeps everything it receives to the same Binance hot wallet as the first. That second address has taken about 2.34 million USDC since July 2024, and we read every payer and where their money came from. It is one family, and its money is Neiro. On 27 July 2024 at 21:05 UTC the wallet 7qCVapPdsWrvweA1MUSgoXYYHQmoAw13w2X6yri1Coq created the pump.fun token Neiro (CTJf74cTo3cw8acFP1YXF3QpsQUUBGBjh2k2e8xsZ6UL) and bought 97.5 million tokens, 9.75% of the supply, for 3 SOL in the creation transaction. Within six minutes it handed 65.7 million of them to four wallets, among them F98AHHj9xwWVgM2j4cj5xRZaBbSAwiMJQcHDDXbz32iW, a wallet first funded from a Bybit hot wallet in January 2024, and another 11.7 million the next morning to a fifth. Those wallets sold the tokens on Jupiter in lots of 450,000 to 700,000, more than 80 sells between 21:35 and 22:25, for roughly 12,000 SOL, swapped the SOL to about 2.3 million USDC the same night, and paid 930,000 of it into the Binance deposit address before 05:30 on 28 July. Most of the rest sat in a sibling wallet that paid the deposit address 401,000 USDC in December 2024 and 781,000 in January 2026, and paid the Bybit-funded wallet 314,000 USDC in eight instalments in between. The 100,000 USDC deposit of 19 Oct 2026 is part of a 512,000 USDC Binance withdrawal from February 2026 going back the way it came. The Neiro creator wallet was itself created in March 2024 by CNjTmWLrTGtmEUukMbeQLMAV3YdUtRjygfoKsTsmp4dJ, a pump.fun token-creator wallet first funded by the Bybit-funded wallet in February 2024, and that is the wallet which funded the one that forwarded the 26 Oct launch-machine profit. So the address where the launch machine's profit lands is the cash-out address of the family that created and dumped Neiro, and three of its deposits use the same habit as the 26 Oct one: 100 USDC first, then the balance. Of roughly 106,000 USDC pushed into the bridge on 26 Oct, 105,947 came back the same evening and landed in two Binance deposit addresses that drain to one hot wallet. We also read the Neiro creator wallet and its parent in full, 2,410 and 984 transactions from 2024 to October 2026: neither ever pays the launch machine's hub, treasury or gas dispatcher, and the only addresses they share with the whole-curve fleet are public DEX pool vaults. The Neiro family and the launch machine meet at the exchange, not on the chain.

That first deposit address was itself first funded in December 2024 by 9jzfY4fF…, the two-year-old wallet in the treasury's own ancestry, two hops below the Binance withdrawal of August 2024. So the machine takes its seed money out of a Binance account and pays its profits back into a Binance account, and the deposit address it pays into was opened by its own older wallet. On what the chain can show, that is one customer at both ends. Who the customer is, only the exchange knows.

The deposit address behaves exactly as one should: everything it receives is swept within minutes to the Binance hot wallet 5tzFkiKs…, the same hot wallet that paid out the treasury line's seed in 2024 and the whole-curve hub's 700 SOL in September. It has had 35 transactions in its life. Besides the two USDC deposits above it took 2,502 USDC on 25 Jun from 9jzfY4fF…, and on 29 Jul it took 4,999.5 USDT from u6PJ8DtQ…, which is the Gate.io hot wallet at the root of the dispatcher line, with 4,874 USDT more on 18 Aug. So the two lines we traced separately, the gas line that starts at Gate and the treasury line that starts at Binance, pay into the same Binance deposit address. One exchange account sits under the whole machine: it funded the gas wallets through Gate, drew the seed capital from Binance, and receives the profits.

The USDC round trip through the swap service does nothing except break the link between the treasury and the deposit address. It costs a few dollars and seven hours. If you only looked at the Binance deposit address, you would see a clean wallet receiving a transfer from another clean wallet.

A second pattern: whole-curve launches

After this page first went up, we found a different kind of launch. In 48 hours, 47 Solana launches came with a dev buy of about 85 SOL: 3,995 SOL across 43 wallets. That is the entire bonding curve in one buy, and each reads as a completed curve on-chain. The coin goes straight to PumpSwap and looks bonded without needing outside demand.

Many use official-sounding names, such as United States Dividend Fund, American Oil Dividend Fund and Federal Oil Trust Fund. Others reuse tickers like XTO.

Those two are 5E1YHUfAFC2KJfXaUt8aHExzL3JWBtiqrmfLR4FdwQDG and 9WRgoWAaPAnJo5S78pGaKg8cPUq7wm4LpxoaQ7Z1zG96. Walking funders back eight hops joined small groups of three, three and two dev wallets more.

We then read the pools of 13 of these coins, every transaction, and followed the money up from the dev wallets.

So the whole-curve coins are, on what we measured, an operator trading with itself: a faked graduation, faked holders, faked volume. The sites behind the names (usdfgov.space, aodf.info, xtruthoil.com, uodf.site) are template builds with no owner on record. They copy an older USDF coin that is seven weeks old and still trading. The payoff is creator fees. 27 of the 47 dev wallets collected them: 864 SOL in about 48 hours, 157 SOL on the best coin, counted from each wallet's first 80 transactions only. Pump's dynamic fee pays a coin's creator up to 0.95% of every trade while the coin is small, so a pool that trades hundreds of thousands of dollars a day with itself pays its own creator thousands, and the protocol fee is the cost of running it. Most of that fee is paid by the operator's own wallets. The wallets that lose on these pools, 21 of them across the 13 pools we read, 992 SOL between them, come in identical sets of four, born minutes apart, each with thousands of transactions, trading nothing but these pools. One set was first funded by the operator's own collection wallet. So the volume, the holders, the buyers and the losers are the same money going round, and what leaves the loop on-chain is the protocol fee paid to pump.fun. The payoff, if there is one, is later or off-chain: the sites behind the names promise an announcement on 19 November.

We tested for a spread-out network too. Wallets acting in the same slot on different coins, repeatedly, came in below chance on 5.8 hours of every curve trade. Wallets with identical buy-size sequences were 19 in total. Among the 1,320 biggest winners, same-coin same-slot pairs were four times chance, and the 369 wallets that passed two independent tests held 15% of those wallets' gains. The coordination that exists is small groups working the same coins, not one network across the market.

We also checked pump.fun's own on-chain addresses, read from its program config, against 214 wallets in these funding trails. None appeared, and its authority wallets paid nothing out. That is no link found in what we traced, not a clean bill.

What a full tape shows

We now record every bonding-curve trade. Over 5.8 hours on 31 Oct, across the 4,751 coins launched in that window, 5,087 wallets gained 3,670 SOL and 13,994 lost 8,012 SOL. The top 1,000 took 86% of the gains and the top 10 took 12%.

Wallets that repeatedly landed in the same slot with the same exact buy size held 6.2% of the gains. So this test does not show one network taking the money. It catches only one pattern, though, and the tape stops at bonding, so those losses include tokens still held or sold later.

What we cannot say

If you trade

Check the money, not the wallet. A clean dev is exactly what rotation produces. And when a funder is an exchange, that tells you nothing either way.

Watch for whole-curve launches, which the board now flags. Bonded does not mean real demand. Compare liquidity to market cap, and buys to sells.

Every wallet here is public. Paste any coin into the live board to see its dev, its funder and that source's record.

This is a reading of public chain data, not a buy or sell call. Dollar figures use SOL at about $97.

Prometheus7Live Coin Verifier