14 June 2022, Baden-Wuerttemberg, Rottweil: The candlestick chart of the cryptocurrency Terra Luna with the sell-off is seen on the screen of a computer in an office. (Shot with zoom effect). Photo: Silas Stein/ (Photo by Silas Stein/picture alliance via Getty Images)
dpa/picture alliance via Getty Images
“There are so many products in the market,” said Margie Feng, marketing lead at Solayer, the company behind the Solana perpetual futures venue Margin Trade. “And basically doing the same thing.”
That is an unusual thing for a marketer to say out loud, and it is accurate. Onchain perpetual futures exchanges have multiplied to the point that the product itself no longer separates them. Hyperliquid, Aster, Lighter, EdgeX, Paradex, Drift and a long tail of newer venues all offer leveraged bets that settle onchain. Feng’s own list of companies she watches includes Backpack, whose marketing she admires, and the Solana order book Phoenix.
So the competition has moved somewhere most traders never look. Two places, specifically: what an exchange does to you when you win, and which markets it will list that nobody else will.
The switch that closes winning trades
The first of those is a piece of plumbing called auto-deleveraging, or ADL. It works like this. When a leveraged position blows up so badly that the trader’s margin and the exchange’s insurance fund cannot cover the loss, someone has to absorb it. The exchange reaches across the book and force-closes positions on the winning side. Traders who were right get taken out of their trades to pay for traders who were wrong.
Most of the time this is invisible. On October 10, 2025 it was not. The crypto market recorded its largest single-day liquidation event, roughly $19 billion market-wide, with about $10.3 billion of that on Hyperliquid according to CoinGlass data. Hyperliquid’s auto-deleveraging closed $2.1 billion of positions in twelve minutes.
“Most exchanges do it by ranking traders and picking up the biggest winners first,” Feng said.
She is describing an algorithm the industry calls the Queue, used by Binance, Hyperliquid and Lighter. It sorts traders by profit and leverage and closes the top of the list until the shortfall is covered.
Nobody can agree how much it cost
Tarun Chitra, the founder and chief executive of the risk-modeling firm Gauntlet, published a paper on the October event in November 2025. His summary of it on X was not subtle. “Did @HyperliquidX autodeleverage (ADL) $650m of PNL that it didn’t have to? Was this 28x more than the minimal necessary? Did almost every exchange (incl. @binance) copy-pasta a Huobi heuristic from 2015?”
The last question is the one to sit with. The Queue is not a modern piece of engineering. It traces back to a 2015 design and has been copied across the industry since.
The paper sets out an impossibility result: no policy can simultaneously satisfy exchange solvency, revenue, and fairness to traders. Under repeated auto-deleveraging in extreme conditions, Chitra wrote, “the greedy Queue strategy completely fails,” concentrating losses on a small set of winners and closing more than it needs to. He put the excess profits lost by winning traders at between $45.0 million and $51.7 million, corresponding to roughly $653.6 million of positions closed.
Dan Robinson, a research partner at the crypto investment firm Paradigm, went after both the model and the number in public.
“This paper is simply wrong about its central topic: how Hyperliquid’s ADL works,” he wrote. “Tarun is describing a different (much crazier) algorithm, which also might explain how he calculated that traders somehow paid $653m to cover a $23m deficit.”
Hyperliquid allocates in contracts, Robinson argued, not by taking the full equity of each winner in order, so a count of positions closed is not a count of money destroyed. He was blunt about the tone of the exchange too. “I don’t like obscurantism or gatekeeping,” he wrote, before working through the paper section by section.
Chitra gave ground. He later pointed readers to “the mistakes made plus the things that are still correct,” and listed the fixes he was working on, crediting Robinson and a co-author by name. A follow-up paper in February 2026 found that Hyperliquid’s production queue achieved about 50% of an upper bound on regret during the October episode, against roughly 2.6% for an optimized alternative. That work also notes Binance overutilized auto-deleveraging considerably more than Hyperliquid did.
What survived the argument is the part that matters to traders. Robinson, defending Hyperliquid, still wrote that “I agree that Hyperliquid’s ADL is not Sybil-resistant” and that this “is a decent reason to change to a pro-rata rule,” applied in proportion to notional rather than equity value. The critic and the defender both want the Queue gone. They disagree about what it cost.
Fairness as a feature
“We spread it pro rata across the entire opposing side,” Feng said. “No one gets singled out. Everyone takes a tiny proportional slice.”
“The winners are never the single target that’s being punished,” she said. Her shorthand for it: “never punished for winning.”
Chitra’s own recommended fix is a risk-aware pro-rata scheme that allocates by position leverage rather than ranking winners. Margin Trade shipped its version while the argument was still going on.
The caveat is size. Margin Trade reached mainnet on June 9, 2026, after a testnet that opened on May 28. Hyperliquid processed about $172 billion over one thirty-day stretch this year, which one tracker put at roughly 32% of onchain perp volume; another count this year had its share at 44%. Margin Trade’s own promotional post in July claimed $1.36 million in weekly volume on its flagship market. No pro-rata rule has been tested by a crash on a book that small.
Traders have heard versions of this promise before, in a nearby industry. Chris Hunter, CEO of Maven Trading, said on the On The Margin podcast that proprietary trading firms build rulebooks with the same asymmetry. Firms “will have a lot of rules that will kind of catch you so that you can’t get a payout,” he said, describing an adversarial posture where the house is “trying to catch the trader on mistakes so that they could not pay them in the same way that an insurance company is going to try to catch a car crash for any mistakes that they make.”
Auto-deleveraging exists to keep an exchange solvent, not to avoid paying anyone. The trader still ends up closed out of a winning position.
Going where nobody else has listed
“We are going where there isn’t competition,” Feng said.
That is the second front. In practice it meant listing perpetuals on Pearl, an AI-compute token trading under the ticker PRL, before larger venues did. “It’s AI related, which is a very sexy narrative right now,” Feng said.
Pearl mines with matrix multiplication instead of conventional hashing. That is the operation behind AI training and inference. A GPU doing machine-learning work earns the coin on the side. Mainnet went live on April 27, 2026, and Together AI announced a partnership three weeks after that.
Feng checked the team first. Pearl Research Labs was co-founded by Omri Weinstein, a complexity theorist with a Princeton doctorate who has held posts at Columbia and the Hebrew University. “It’s not, like, a fake team,” Feng said. “It’s a very transparent team.”
The maths turned fast. Renters piled in on RunPod and Vast.ai, difficulty climbed, and daily revenue for an RTX 5090 fell from about $33.80 to $17.19. That is a 49% cut in a matter of weeks, and block rewards only shrink from here. PRL traded mainly on SafeTrade and MEXC, where liquidity is thin. The people running those rigs had almost no way to lock in a price. It is the squeeze that pushed an earlier generation of bitcoin miners into hedging.
Then Feng described the problem with her own plan.
The miners will not trade. They mine, they sell, and they treat derivatives as something other people do. “This is a big gap that we’re experiencing right now,” she said. Her team is producing explainers about hedging aimed at an audience that has so far shown little interest in it. Feng came to crypto from Bitmain, the mining hardware manufacturer, so she is pitching a group she used to sell to.
The unglamorous layer
None of this is where the attention goes. Capital keeps flowing to venues promising speed and to challengers hunting Hyperliquid’s lead, including funded newcomers like GRVT. Onchain perpetuals still account for only around a tenth of the volume that centralized exchanges handle, so there is room for all of them to keep growing while the underlying question goes unresolved.
Kaledora Kiernan-Linn, cofounder and CEO of Ostium, said on the On The Margin podcast that her original bet was on this category becoming the main event. The thesis that “TradFi or Real World Asset Perpetuals would be the second blockbuster product to come out of crypto after stable coins, has been identical, really, since day one,” she said, and the growth has followed: “we’ve seen sort of parabolic growth over the last year and change.”
Growth of that kind arrives with more leverage, more correlated positions and more days like October 10. On those days the difference between venues is not the interface or the fee schedule. It is a rule buried in the risk engine that decides whose winning trade gets closed.
Feng, for her part, does not oversell the moment. “It is very challenging I have to say,” she said of building into this market. “It’s not a great market.”
