Papertrade
Why it happened
Attack class: Oracle manipulationAn instant order-book midpoint with no averaging settled leveraged profits, so moving the book moved the payout.
What would have stopped it
- Price positions with a time-weighted average instead of the instantaneous midpoint
- Add a deviation limit against an independent second price
Not scored before the exploit
BlackHart had not scored Papertrade before this exploit.
On October 10, 2026 an account took about $1.28 million net out of Papertrade's Martingaler liquidity pool on HyperEVM, shortly after the synthetic perpetuals exchange launched. Papertrade priced trades from the instantaneous midpoint of Hyperliquid's ETH order book with no averaging, so trading on Hyperliquid while holding huge leveraged positions on Papertrade moved the price used to settle profits. The team said overnight manipulation attempts hit its BTC and ETH markets, blamed loose launch limits, and said a winning user returned almost all profits, a claim CertiK did not verify. The team has said it is tightening open-interest limits.
The traced trading account opened a leveraged ETH position in one transaction and settled it in another. Positions were priced by the price oracle proxy and paid from the exchange proxy, which paid out five withdrawals, the largest in this transaction and this one.
Attack flow
All 5 on-chain stepsThe attack in 3 steps. Step 1, Root control: Price oracle (source). Step 2, Drain: Trading account (hub), about $1.28M reached this step. Step 3, Outflow: Net outflow (exit).
Step 1: Root control
- SourcePrice oracleinstant bid-ask midpointno averaging0xa9b4c7b7...39cc2e
Step 2: Drain
- HubTrading account500x leverage, about $10M exposure5 withdrawals0x2025137a...70840e
Step 3: Outflow
- ExitNet outflowabout 1.28M USDCteam says most returned
Why each step worked
- No time-weighted averagemissing controlone instantaneous quote
- No independent price checkmissing controlno deviation limit
- Loose launch open-interest capsmissing controlper the team
- Fixed-size haircutdoes not scale with position
- No per-account payout limitmissing controlshared pool exposed to one account
Full forensic detail
Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.
Exploit anatomy
Root cause
Papertrade priced positions off a number that a trader could move cheaply. Per CertiK's analysis and DeFiLlama's description, the protocol's price oracle took Hyperliquid's instantaneous best bid and best ask and returned their midpoint, rejecting only zero prices and crossed quotes. It had no time-weighted average, no limit on how far a price could deviate, and no second price source. A trader who could nudge the order book on Hyperliquid while holding a very large leveraged position on Papertrade could therefore be paid out at the moved price. CertiK also found that the haircut applied to asymmetric price impact used a fixed reference size of 100,000 USDC rather than the actual position size, so payouts grew with exposure while the haircut stayed at about 22 percent. Papertrade synthetic positions were paid from a shared liquidity pool, so a gain for one account was a loss for the pool's providers. The team says loose open-interest caps at launch made the attempts feasible. The deployed source was not verified, so CertiK's oracle analysis comes from decoded bytecode and historical traces.
Prevention analysis
Averaging over a window would have forced an attacker to hold the moved price for many blocks, which costs far more than a brief nudge.
A check against another source such as Hyperliquid's own oracle price would have rejected a midpoint that had moved away from it.
The team attributes the incident to loose launch caps. Tighter caps would have bounded what any one account could take from the pool.
A haircut that grows with exposure would have reduced the payout on very large positions.
Similar incidents
A price that could be moved cheaply on a thin market was trusted by a lending market and borrowed against.
A small trade moved a price input, and the protocol accepted it with no averaging or second source.
A thin market price was moved and then used to extract value from a shared liquidity pool.
Remediation
Timeline
Get your protocol scored across 12 dimensions, or request ongoing coverage.