The trade the vault had to take.

A liquidation engine assumes the order book can take the position it closes. In the six hours before the build, everything resting within 25 price levels of Hyperliquid’s ETH book averaged about 7,000 ETH. The position liquidated that morning was 160,234 ETH.

Order books refill, so a better yardstick is how long the market takes to trade that much. At the overnight pace, matching the position would have taken more than eight hours of Hyperliquid’s ETH trading. On Binance, under an hour. Same coin, same morning, on a venue with about a tenth of Binance’s ETH volume.

Sold into the book in one go, a position that size would have walked the price through every resting bid. That is not what the data shows: open interest came down steadily, over five hours, in the afternoon.

At 06:55 UTC open interest on Hyperliquid’s ETH perpetual jumped 28,000 ETH in a single minute. By 07:00 the bids resting within 25 levels of its book had swelled from $7M to $114M, 19 times their overnight average. By 07:30, 167,000 ETH of new positions were open, close to the 175,179 ETH Lookonchain later attributed to one wallet.

It didn’t stop there. Open interest kept climbing in waves until 09:31, when it peaked at 748,903 ETH. Open interest counts contracts, and every contract has a long and a short side: it measures how much was at stake, not who held which side.

On Binance and OKX, ETH open interest barely noticed. Whatever was happening was happening on one venue.

The same contract trades on every venue, and arbitrageurs keep the prices within a few basis points of each other. Overnight, Hyperliquid’s ETH and Binance’s closed each minute within about 5 bps below and 9 bps above one another.

During the build, that leash snapped. At 09:06 ETH touched $1,970 on Hyperliquid while Binance never got past $1,955.90, and at the 09:31 peak in open interest it swung the other way, 37 bps below. Buying that size on one venue pulled that venue’s price with it.

The unwind looked nothing like it. From 12:00 to 17:00, while open interest fell by half, Hyperliquid stayed level with Binance: on average within 2 bps. ETH did fall 4.7% that afternoon, but Binance fell just as far, to its low at the same minute. If the selling on Hyperliquid pushed ETH down, it pushed every venue down together; the data can’t tell that apart from a market that was falling anyway.

A backstop liquidation doesn’t go through the order book: the position and its margin are transferred to the exchange’s liquidator vault. Nothing trades, so open interest doesn’t move. When Hyperliquid posted its statement at 10:26, open interest was still within 10% of its peak.

At noon it started to fall, and kept falling for five hours, a little less each hour. Lookonchain sized the liquidated part at 160,234 ETH; open interest fell about twice that, so other positions were closing too. By 17:00 it was back within 7% of where it stood at 06:00.

The afternoon’s market sells outweighed the buys by about a tenth. HLP ended the day about $4M down while the trader kept about $1.8M, and Hyperliquid said ETH leverage would drop from 50× to 25×.

ETH spent that spring falling, and the whale’s day doesn’t stand out in it. On Binance’s daily candles it has a smaller range than the day before.

Everything that made 12 March remarkable happened inside one venue and inside a few hours: a book holding under 5% of the position, open interest that more than doubled and came back, a price that ran ahead of Binance’s and then fell back in line. None of it shows up in a daily close, or in a price feed that averages across venues. You only see it in minute data, venue by venue, with the order book and positioning next to the price.

Every chart here comes from Aperiodic’s 1-minute microstructure metrics: order-book depth, order flow, open interest and prices, computed point-in-time from raw exchange feeds for every perpetual on Binance, OKX and Hyperliquid. The same few API calls replay any day, on any of them, without touching a raw L2 update.