The attackers in the Jelly-My-Jelly incident didn’t rely on complex contract exploits or cryptographic tricks. Instead, they identified and weaponized a mathematical flaw in the mark price mechanism—small data sources, median aggregation, fragmented liquidity—and combined it with the platform’s liquidation logic. This wasn’t a hack; it was a "compliant attack" on systemic rules, turning Hyperliquid’s core safety feature into a tool for mass liquidation.

In March 2025, JELLY, a low-cap token with less than $2M daily volume, triggered a $10M+ liquidation storm on Hyperliquid. Shockingly, the attackers didn’t breach smart contracts or exploit code vulnerabilities. They simply gamed the system’s transparency, exposing how mark prices—intended as neutral safeguards—can become weapons in illiquid markets.

This article dissects the systemic risks of perpetual contract mark pricing, revisits the JELLY attack, and reveals how oracle design flaws and Hyperliquid’s innovative (but vulnerable) HLP Vault created a perfect storm.

Mark prices are calculated using a median-of-three formula:

1. Index Price × (1 + Funding Rate) (anchored to spot markets).

2. Index Price + Moving Average Basis (smoothed for volatility).

3. Last Traded Price (on the derivative platform).

The median is meant to filter outliers—but this assumes sufficient, independent data sources. For illiquid altcoins, however, controlling just 2 of 3 exchanges lets attackers "pollute" the index price. Transparency backfires: the clearer the rules, the easier they are to weaponize.

Liquidations trigger based on mark price, not the actual market price. Worse, platforms often:

• Preemptively liquidate positions to avoid insolvency.

• Keep surplus funds from early liquidations (instead of returning them to users).

This creates a misalignment: platforms profit from conservative triggers, while traders face "phantom liquidations."

For large-cap assets, median aggregation works. But for altcoins:

• 3 data sources? Manipulate 2, and the median is compromised.

• 2 data sources? The median becomes a simple average—zero outlier resistance.

Result: A "decentralized" index becomes a single point of failure.

Oracles report prices faithfully—but they can’t judge fairness. Attacks like Mango Markets (2022) and JELLY exploit this:

• Not oracle exploits, but market manipulation.

• Low liquidity = low cost to manipulate.

1. Target selection: Pick a thinly traded altcoin listed on derivatives.

2. Capital raise: Use flash loans for cheap funding.

3. Spot market blitz: Pump the price on monitored exchanges.

4. Oracle contamination: Let the oracle "honestly" report distorted prices.

5. Mark price infection: Trigger mass liquidations via derivatives.

Hyperliquid’s transparency backfired: Attackers knew exactly which exchanges to target and how much capital was needed.

Hyperliquid’s HLP Vault is both a market-maker and liquidation backstop. Users deposit USDC to earn yields, but:

• It automatically absorbs toxic liquidations (no human discretion).

• No isolation between market-making and liquidation pools.

Attackers could predict exactly who would buy their bad positions: the HLP Vault.

The Auto-Deleveraging (ADL) system failed because:

• The HLP Vault’s "health" appeared fine (thanks to shared collateral).

• Losses concentrated in the vault instead of spreading across the market.

• Attackers built a $4M short position on Hyperliquid.

• Used wash trading to inflate open interest (OI) without moving prices.

• JELLY’s 1% market depth was just $72K—easy to manipulate.

• Coordinated buys spiked the price 500% in 1 hour.

• Polluted oracles fed into Hyperliquid’s mark price.

• The $4M short was force-liquidated onto the HLP Vault.

• ADL never triggered; losses socialized to LPs.

• Binance/OKX listed JELLY perpetuals mid-attack, worsening losses.

• Hyperliquid delisted JELLY and compensated users.

• HLP Vault’s unrealized losses hit $12M (official loss: $700K).

The JELLY attack revealed:

1. Oracle fragility: "Multi-source" indexes rely on the same few illiquid markets.

2. Median weakness: Useless against small-sample manipulation.

3. Blind trust in mark prices: Platforms treat them as gospel, even when poisoned.

Solution: Design mechanisms that account for game theory, not just math. Until then, mark prices remain weapons waiting to be wielded.

Key Terms:

• Mark Price: The "fair" price used for liquidations (vs. last traded price).

• HLP Vault: Hyperliquid’s pooled liquidity system.

• ADL (Auto-Deleveraging): A mechanism to socialize losses during extreme volatility.

The attackers in the Jelly-My-Jelly incident didn’t rely on complex contract exploits or cryptographic tricks. Instead, they identified and weaponized a mathematical flaw in the mark price mechanism—small data sources, median aggregation, fragmented liquidity—and combined it with the platform’s liquidation logic. This wasn’t a hack; it was a "compliant attack" on systemic rules, turning Hyperliquid’s core safety feature into a tool for mass liquidation.

In March 2025, JELLY, a low-cap token with less than $2M daily volume, triggered a $10M+ liquidation storm on Hyperliquid. Shockingly, the attackers didn’t breach smart contracts or exploit code vulnerabilities. They simply gamed the system’s transparency, exposing how mark prices—intended as neutral safeguards—can become weapons in illiquid markets.

This article dissects the systemic risks of perpetual contract mark pricing, revisits the JELLY attack, and reveals how oracle design flaws and Hyperliquid’s innovative (but vulnerable) HLP Vault created a perfect storm.

Mark prices are calculated using a median-of-three formula:

1. Index Price × (1 + Funding Rate) (anchored to spot markets).

2. Index Price + Moving Average Basis (smoothed for volatility).

3. Last Traded Price (on the derivative platform).

The median is meant to filter outliers—but this assumes sufficient, independent data sources. For illiquid altcoins, however, controlling just 2 of 3 exchanges lets attackers "pollute" the index price. Transparency backfires: the clearer the rules, the easier they are to weaponize.

Liquidations trigger based on mark price, not the actual market price. Worse, platforms often:

• Preemptively liquidate positions to avoid insolvency.

• Keep surplus funds from early liquidations (instead of returning them to users).

This creates a misalignment: platforms profit from conservative triggers, while traders face "phantom liquidations."

For large-cap assets, median aggregation works. But for altcoins:

• 3 data sources? Manipulate 2, and the median is compromised.

• 2 data sources? The median becomes a simple average—zero outlier resistance.

Result: A "decentralized" index becomes a single point of failure.

Oracles report prices faithfully—but they can’t judge fairness. Attacks like Mango Markets (2022) and JELLY exploit this:

• Not oracle exploits, but market manipulation.

• Low liquidity = low cost to manipulate.

1. Target selection: Pick a thinly traded altcoin listed on derivatives.

2. Capital raise: Use flash loans for cheap funding.

3. Spot market blitz: Pump the price on monitored exchanges.

4. Oracle contamination: Let the oracle "honestly" report distorted prices.

5. Mark price infection: Trigger mass liquidations via derivatives.

Hyperliquid’s transparency backfired: Attackers knew exactly which exchanges to target and how much capital was needed.

Hyperliquid’s HLP Vault is both a market-maker and liquidation backstop. Users deposit USDC to earn yields, but:

• It automatically absorbs toxic liquidations (no human discretion).

• No isolation between market-making and liquidation pools.

Attackers could predict exactly who would buy their bad positions: the HLP Vault.

The Auto-Deleveraging (ADL) system failed because:

• The HLP Vault’s "health" appeared fine (thanks to shared collateral).

• Losses concentrated in the vault instead of spreading across the market.

• Attackers built a $4M short position on Hyperliquid.

• Used wash trading to inflate open interest (OI) without moving prices.

• JELLY’s 1% market depth was just $72K—easy to manipulate.

• Coordinated buys spiked the price 500% in 1 hour.

• Polluted oracles fed into Hyperliquid’s mark price.

• The $4M short was force-liquidated onto the HLP Vault.

• ADL never triggered; losses socialized to LPs.

• Binance/OKX listed JELLY perpetuals mid-attack, worsening losses.

• Hyperliquid delisted JELLY and compensated users.

• HLP Vault’s unrealized losses hit $12M (official loss: $700K).

The JELLY attack revealed:

1. Oracle fragility: "Multi-source" indexes rely on the same few illiquid markets.

2. Median weakness: Useless against small-sample manipulation.

3. Blind trust in mark prices: Platforms treat them as gospel, even when poisoned.

Solution: Design mechanisms that account for game theory, not just math. Until then, mark prices remain weapons waiting to be wielded.

Key Terms:

• Mark Price: The "fair" price used for liquidations (vs. last traded price).

• HLP Vault: Hyperliquid’s pooled liquidity system.

• ADL (Auto-Deleveraging): A mechanism to socialize losses during extreme volatility.