A new study of the $641 million forced-sale event that hit Hyperliquid on Oct. 10 breaks down exactly how the exchange's backstop soaked up the selling. The preprint, which hasn't been peer-reviewed, says about $576 million went to the backstop while just $64 million reached the order book — and that the cascade moved fast, with 87.8% of forced selling after the onset done in half an hour.
How the numbers line up
At 21:19 UTC on Oct. 10, a wave of liquidations swept through Hyperliquid. The backstop absorbed 62.6% of forced-sale value off-book after the onset, according to the paper. The study tracked $733 million of book-directed forced-sale value across a 15.7-hour post-onset window, including $644 million during the initial nucleation phase. The branching ratio — a measure of how much a cascade feeds on itself — stayed below 0.2 in every measured regime, hitting 0.195 during nucleation and 0.140 at peak.
The backstop's role
Hyperliquid's liquidation rules first try to fill orders on the market. If that would cause too much slippage, the liquidator vault can step in. That vault is a component strategy within the HLP protocol vault. It's what swallowed the $576 million, keeping the selling pressure from slamming the order book. Without it, the paper argues, the cascade could have been worse.
What the study doesn't cover
The paper looks only at Hyperliquid, not the broader crypto market. It's also the second part of a larger research effort — Part I found no event-invariant early-warning variable for such cascades. Part II digs into the mechanism. The authors note that Hyperliquid's fill-log archive only begins May 25, 2025, making the October event the only in-flight case study they could examine so far.
The study places the Oct. 10 event in the context of seven major Bitcoin perpetual futures cascades from 2022 through 2025. The authors hypothesize that venues without a backstop would see higher realized branching — meaning a cascade would amplify more — and they plan to test that on other exchanges. For now, the preprint remains unreviewed, and the authors haven't released a timeline for the next round of testing.
