Author: CBB (@Cbb0fe). Translation credit in the source: TechFlow.
CBB has published a detailed review of a 10-month arbitrage operation he ran with his brother between Hyperliquid’s HIP-3 markets and Interactive Brokers, or IBKR. According to his account, the two built a bot-driven setup from scratch after HIP-3 went live and ultimately generated about $10 million in profit by September 2026.
He wrote that the pair had spent the previous eight months running what he described as the top arbitrage bot on HyperEVM, but returns there had tightened after months of competition with Wintermute and the arrival of another participant. Their edge, he said, was never about fighting institutions over the long term. It was about deploying a strategy quickly, then extracting what they could before larger players showed up.
That search for the next trade led them to stock perpetuals on Hyperliquid.
From HyperEVM arbitrage to HIP-3 stock perpetuals
CBB placed the starting point in October 2025. HIP-3 launched on Hyperliquid on Oct. 13, and three days later Unit/TradeXYZ introduced its first stock perpetual market, XYZ100. Because more than 40% of Hyperliquid’s supply was still set aside for community distribution, the pair believed generating volume on HIP-3 could be worth pursuing.
He said the logic was similar to what had previously taken them toward opportunities on HyperEVM: they wanted to trade where volume on Hyperliquid’s ecosystem could matter. This time, they decided to test whether they could build and run a stock-perpetual arbitrage bot between HIP-3 and traditional finance rails.
At that point, CBB said, they had almost no background in traditional markets. He wrote that he had never traded a stock in his life and barely understood what futures were. What they did know was that IBKR looked highly competitive for the type of hedging they needed, so they decided to learn it as they went.
In the first few days, he was mostly trying to figure out how to use the IBKR platform at all. He said he was taking screenshots of almost everything and sending them to Claude with questions such as what a screen meant, what a field did, what he was supposed to do next, and how they were supposed to hedge XYZ100. His brother, at the same time, was digging through IBKR’s API and mapping out what was possible and what was not.
Coming from crypto exchanges, where APIs are often straightforward to connect and automate, IBKR felt like a very different system. Market data subscriptions, contract specifications, order types, permissions, API limits, Trader Workstation, and IB Gateway all had to be learned. Early on, they were not even sure the whole idea would be viable.
After about a week of working through IBKR, though, they began to see a path.
How the arbitrage bot was structured
CBB described the strategy in simple terms. The team treated IBKR quotes as the real price and continuously checked whether HIP-3 was trading away from that reference.
- If a market on HIP-3 traded at a discount to IBKR, they would go long on Hyperliquid. Only after the Hyperliquid leg filled would they open the matching short on IBKR.
- If a market on HIP-3 traded at a premium to IBKR, they would do the reverse: short on Hyperliquid first, then buy on IBKR after the crypto leg executed.
On paper, that logic was straightforward. In practice, it required a long list of parameters for every market.
For the IBKR leg, he gave the example ["NVDA", 55, 400, { maxDelta: 800, slippage: 0.1 }]. In his explanation, 55 was the minimum hedge size. Because IBKR charges a minimum $1 commission, they did not want to constantly execute tiny hedges, so the bot would let delta build until it reached 55 NVDA shares before hedging. The number 400 was the maximum size for one IBKR hedge order, meant to limit slippage. The maxDelta value of 800 was a safety valve: if IBKR trades kept failing and the mismatch between the two legs reached 800 shares of NVDA, the bot would stop trading that market. Slippage of 0.1 was the maximum slippage they were willing to tolerate when hedging on IBKR.
The HIP-3 side had its own settings. CBB shared the line NVDA: pair("NVDA", "xyz:NVDA", {makerSize: 400, makerOffsetBuy: 0.12, makerOffsetSell: 0.12, cancelDelta: 0.02, takerRatioBuy: 0.05, takerRatioSell: 0.1, takerMin: 1, takerMax: 2000, limit: 110000, makerEnabled: true, preMarketOffset: 0.04 }).
He said makerSize determined how much size to post, makerOffsetBuy and makerOffsetSell set the spread they wanted relative to fair value, and cancelDelta told the bot when price movement was large enough to cancel and repost. For taker flow, takerRatioBuy and takerRatioSell set the spread threshold before crossing the spread, while takerMin and takerMax controlled the size they were willing to execute. The limit parameter capped total exposure in that market, makerEnabled switched passive quoting on or off, and preMarketOffset added extra spread during pre-market hours because liquidity on the traditional side was much thinner then.
First month: about $850 million in November volume
By late October 2025, the system was ready for a live test. The first few days were messy. The IBKR API caused trouble, connections sometimes dropped, and his brother had to keep finding ways to hold the whole setup together.
Even so, they quickly realized the opportunity set was large. CBB wrote that the experience felt like picking money up off the ground.
In November, they executed roughly $850 million in volume on HIP-3 and made more than $500,000 in profit. He described that as a strong start.
December was quieter. Volume came in at about $550 million for the month, and profit was still solid, but they were already wondering whether their time might be better spent elsewhere. The trade was good, in his telling, but not yet a gold mine. They chose to stay with it anyway because, as he put it, they usually find it hard to stop as long as there is still something left to extract.
January 2026: metals boom, then a $1.1 million loss
The major acceleration arrived in January 2026. Gold and silver started ripping higher, and demand on Hyperliquid for those markets turned intense. CBB said the previous two months had effectively prepared them for exactly this kind of move.
That environment exposed a new operational constraint: liquidity. Almost everyone wanted to be long commodities on Hyperliquid, which meant the pair constantly needed to commit more capital on the IBKR side for hedging. They kept adding money to IBKR, but moving that amount of capital triggered banking friction. CBB singled out EtherFi as being particularly helpful in allowing them to withdraw large sums quickly.
In January alone, they traded about $1.7 billion in volume on Hyperliquid, and funding income by itself exceeded $600,000.
Then came the failure that defined the month.
On Jan. 27, CBB said, he had just landed in Dubai and was about to get coffee with his brother to talk about the bot when he received a liquidation warning from IBKR. He could not initially understand how that was possible because it was still early and nothing major seemed to be happening in the market.
When he logged in, he found that they were net short $120 million in gold futures while gold was in the middle of a sharp rally. They shut the bot off immediately. CBB wrote that he was shaking and genuinely scared of being liquidated. Because all trading on the IBKR side had been handled by the bot, he was not deeply familiar with manually managing those positions there.
Over the next 15 to 30 minutes, he manually closed the $120 million gold short. Later that afternoon, once the market opened and they could fully calculate the damage, the loss came to $1.1 million.
The cause, he said, was simple and painfully stupid. Data from the IBKR interface had stopped refreshing properly. The bot believed there was a position mismatch between Hyperliquid and IBKR and kept selling gold futures on IBKR to correct a difference that did not actually exist. It repeated the action over and over until the gold short reached $120 million and margin warnings finally surfaced.
The incident forced them to add more controls. They needed to verify that data coming from IBKR was fresh, and they needed additional checks before allowing the bot to keep increasing positions. More broadly, CBB said the bot was never originally designed for the level of volume and opportunity that had emerged by then.
The two spent a full day fixing the system and deployed a new version the next day.
Even after the fix, confidence was shaken. CBB wrote that they briefly questioned whether they really knew what they were doing and whether the risk-reward profile still made sense after losing $1.1 million for such an avoidable reason. It was the first time since launch that they seriously considered stopping.
Around $600,000 recovered on the next day’s silver dislocation
They kept going.
CBB said the pair’s pattern was always the same: understand what broke, repair it, and move on. Nearly every bot they had built together had suffered an outsized loss at some point, and somehow they had always managed to recover.
The day after the new version went live, silver sold off sharply after touching an all-time high, and a spread of about 3% opened between Hyperliquid and IBKR at one stage. They captured the move and made about $600,000. It did not erase the previous $1.1 million loss, but in CBB’s telling, it was the first sign that the strategy was functioning again.
Liquidity management and speed became the next priorities
From there, they started strengthening the operation in a more systematic way. CBB’s view was clear: if that much money was still available, more participants and more staffed-up teams would come in. If they wanted to stay competitive, they had to improve quickly.
The first issue was capital management. This trade was very different from pure crypto arbitrage, where moving funds between venues can take less than five minutes. Here, they had to send bank wires into IBKR and move money back out through the banking system.
To handle that, they designed a dynamic system tied to available liquidity at IBKR.
- When liquidity at IBKR was tight, they were willing to lose some money unwinding existing positions in order to free capital, and they demanded wider spreads before opening new trades.
- When liquidity at IBKR was abundant, they did the opposite, accepting smaller spreads to initiate fresh positions and deploying capital more aggressively.
The second area was speed. Up to that point, they had treated IBKR quotes as the reference price. That worked, but the feed was relatively slow. As more traders entered the same game, they knew it would become a latency contest, and relying only on IBKR data would stop being enough.
They started looking for alternatives and found Databento. With Databento and a Nasdaq license, CBB said, they were able to obtain much faster direct market data. They submitted the application in January and got approval at the end of the month.
February: metals stayed hot and oil volatility added more fuel
Metals remained active in February, and they traded roughly $1.5 billion in volume during the month.
CBB also wrote that, toward the end of February, Trump decided to bomb Iran, sending market volatility sharply higher and pushing oil above $100. By that stage, the pair was making roughly $60,000 to $120,000 a day from arbitrage spreads and funding, with weekends feeling dull because traditional markets were closed.
He said that if they made “only” $40,000 over the previous 24 hours, they would start wondering what had gone wrong. They would inspect the bot, adjust parameters, and try to understand where performance was leaking and what else could be improved.
Claude became part of that process as well. According to CBB, they fed their full Hyperliquid and IBKR execution history into Claude so it could analyze where they were losing the most money, what was breaking, and what could be optimized. He said this was the first time they had used AI for trading analysis and that it made a meaningful difference.
That obsession with iteration did not fade even when things were going well. CBB said he and his brother talked about the bot all day. His brother pushed code updates almost daily, while he kept tuning parameters based on market conditions.
After metals and oil, semiconductors took over
By late April 2026, as the Iran conflict began to cool, they thought the period of outsized profitability might finally be ending. For the previous several months they had been making about $500,000 a week, and CBB could not see what would keep driving opportunities at that level.
Then semiconductors and other bottleneck trades took off. He named tickers such as SNDK and MU and said they started trading like pure memecoins.
That sequence stood out to him. When they began building the bot in October 2025, almost nothing was happening. From there, they moved through metals, then oil, and then semiconductors, with each phase behaving in a far more extreme way than they initially expected.
CBB said luck clearly played a role. They happened to be in the right place at the right time with a product suited for exactly that kind of market. At the same time, he argued that committing early to HIP-3, to stock perpetuals, and more specifically to TradeXYZ also reflected some real foresight.
Across May, June, and July, monthly volume stayed between $1.5 billion and $2.5 billion, while weekly profit held around $400,000 to $500,000.
By September, the pair said the trade may be nearing its end
The write-up was dated to early September 2026. By then, CBB said, a number of institutions had entered the trade, and Ethena had also announced plans to move into stock basis trading in the coming weeks.
For him, that likely meant the opportunity was close to running out. Still, he described the ride as extraordinary. Over 10 months, he said, the pair traded a combined $32 billion across HIP-3 and IBKR, accounted for 1.5% of TradeXYZ’s total volume, and made $10 million in profit.
He added that the result depended on having substantial liquidity available to deploy. Even so, the annualized return on actual invested capital still came out to roughly 35% to 45%, depending on the period.
What mattered most to him, though, was not just the profit. CBB wrote that the experience gave the pair their first real exposure to traditional finance and how it works. Ten months earlier, they had never traded a stock and barely knew what futures were. Now, by his account, they had traded $32 billion.
He ended the post by saying the only thing left was to pray for Hyperliquid season three.