Papertrade says its overnight winner gave “almost all” his profit back. The records show $1.1M still out

Papertrade is an exchange on HyperEVM, Hyperliquid's smart-contract layer, that offers BTC and ETH positions at up to 1000x leverage. Trades are swaps against a pool of house money, priced at Hyperliquid's best bid and ask midpoint at that moment. Its own docs name the weak spot: "That is a real protocol risk of using BBO directly." Late on 10 October, @btlrtrades posted that the wallet 0x2025…840e had taken $1.28M out of Papertrade by moving Hyperliquid's ETH price in its favor. At 11:42 AM ET on 11 October, Papertrade answered that the user "continued to try again, ultimately donating almost all his profit back to the house via new accounts". We checked both claims against Papertrade's own account records and Hyperliquid's public ledger.

@btlrtrades's numbers hold. Papertrade's cashflow records for the wallet show five deposits totalling $1,200,100 between 3:27 and 10:35 PM ET on 10 October, and five withdrawals totalling $2,479,207 between 5:13 and 10:10 PM ET, the largest $1,543,489 at 6:06 PM. That is $1,279,106 more out than in. Its trade records show 307 closed positions, 296 of them winners, worth $1.76M after Papertrade's win haircut, less $35,456 in win fees, and 30 ETH shorts liquidated for $450,001. The withdrawals reached the trader's Hyperliquid account from 0x6cd5…d0a2, the Papertrade contract that every Papertrade deposit address reports as its exchange.

The 30 shorts show how it worked, and how it stopped working. Between 7:12 and 7:20 PM ET the trader's Hyperliquid account sold 7,057 ETH in 769 fills, and Hyperliquid's ETH price fell from 2,507.3 to 2,504.4. In the same window, between 7:13:14 and 7:13:24 PM, the wallet opened 30 shorts on Papertrade at 2,507.35, each with $15,000 of margin at 660x, about $297M of exposure in all. @btlrtrades describes the earlier wins the same way: a push on Hyperliquid, then a quick close on Papertrade. These 30 never closed. At 10:31:54 PM, with the price at 2,509.95, all 30 were liquidated within the same second, in three transactions (one of them).

The linked account is 0x2024…dff4. The trader's wallet sent it $50,000 at 2:00 PM ET and $250,000 at 7:29 PM ET on 10 October, and had sent it $100,000 on 6 September. Every Papertrade deposit address is a small contract that records whose it is: the one at 0x49cc…c1da reports 0x2024…dff4, and the trader's own, 0xa6a5…2ce4, reports 0x2025…840e. Thirty-three seconds after the $250,000 arrived, 0x2024 moved $182,503 into its Papertrade deposit address. Between 7:14 and 7:30 PM ET, minutes after the shorts got stuck, it opened dozens of ETH longs at 1000x, most with $8,000 or $10,000 of margin. Most were liquidated at 7:42 PM and between 8:25 and 8:31 PM. Across the day it deposited $389,246 and withdrew $212,431, a loss of $176,814.

That part of Papertrade's reply holds: an account tied to the trader kept trading and lost. Two parts don't. The account wasn't new: it first traded on Papertrade at 1:35 PM ET, two hours before the trader's main wallet did. And it didn't give back almost all the profit. Together the two accounts are $1,102,292 ahead. After 10:35 PM ET on 10 October the trader's main wallet sent nothing more out of Hyperliquid, and at 1:19 PM ET on 11 October it held 2,644,299 USDC on Hyperliquid's spot ledger. Its three declared Hyperliquid sub-accounts have no Papertrade activity. Accounts funded some other way, from another exchange or chain, can't be ruled out, so the gap could be smaller than it looks.

Papertrade also wrote that the losses came "without accounting for the cost of manipulating the bbo on hyperliquids side". That cost was real but small next to the winnings. Between 4:53 and 11:00 PM ET the trader's Hyperliquid account traded about $232M of ETH and booked about $44,000 in losses and $52,000 in fees, roughly $96,000 in all. That figure includes closing an ETH position the account already held, so the cost of the price pushes alone is somewhere below it. Net of it, the two accounts are about $1.0M ahead.

@btlrtrades guessed that "the team blocked 0x2025's closes to stop the attack." Public data can't settle that. A close attempt that never executes leaves no record on chain, so the trade history shows only the 7:13 PM opens and the 10:31 PM liquidation. Papertrade's docs say its relayer submits every open and close, so a close the relayer declined to send would look exactly the same. The relayer was processing closes in that window: 0x2024 closed a long for a $1,587 profit at 7:28 PM ET. Papertrade's own explanation points to its launch queue, which it says gave opens and closes "non-deterministic timing". Neither account can be confirmed.

One more detail changes what "donated to the house" means. Papertrade pays losing trades in its own token: PAPER is minted on every loss and liquidation, and staked PAPER earns a share of trading fees. The trader's 30 liquidated shorts minted 17.6M PAPER, about 0.31% of the 5.68B supply, and 0x2024's liquidations minted 7.5M more. We haven't put a dollar value on it, but part of the losses came back to the trader as a claim on Papertrade's future fees.