Perp DEX vs CEX liquidations: what each venue shows

On an on-chain perp DEX such as Hyperliquid, every position, its liquidation price and every liquidation are public, so liquidations can be checked wallet by wallet. Centralised exchanges publish liquidations only as a stream of force orders without the account behind them, and some streams leave orders out, so their totals are a lower bound.

This guide compares what each kind of venue makes visible, how the liquidation itself works on Hyperliquid, and what that means when you read liquidation numbers across venues.

The short version

The difference is not that one kind of venue liquidates more fairly than the other. It is what an outsider can see.

What is public on Hyperliquid

Hyperliquid runs its perpetual order books on its own chain. In its documentation, every order, cancel, trade and liquidation happens transparently, with one-block finality.¹ Anyone can read an account's open positions, their size, their margin and their liquidation price through the public API.

That makes two things possible that centralised data does not allow:

• Liquidation levels before they happen. Because every position carries its own liquidation price, we can sum the size per price level for the wallets we track and draw it as a liquidation heatmap. No leverage has to be assumed.
• Liquidations confirmed per wallet. When a tracked position disappears between two polls while the mark price reached its liquidation price, we check the wallet's fills. Only a fill marked as a liquidation of that wallet counts, with its time, price and size. The full method is on our liquidations metric page.

How a liquidation works on Hyperliquid

Hyperliquid's documentation describes the process in stages:²

1. Book liquidation. When account equity falls below the maintenance margin, the position is closed with market orders sent to the order book. If the book closes enough of it, any remaining collateral stays with the trader.
2. Partial liquidation for large positions. For positions above 100k USDC, only 20% goes to the book at first, followed by a 30-second cooldown in which liquidation orders cover the whole position.²
3. Backstop. If equity falls below two thirds of the maintenance margin and the book has not done the job, the liquidator vault takes the position over. The liquidator vault is a strategy within HLP, the protocol vault, so its profit and loss from backstop liquidations goes to HLP's depositors.
4. Auto-deleveraging. Only if an account's value or an isolated position's value turns negative are profitable positions on the other side closed against it, ranked by profit and leverage.³

Two details follow from the docs. Hyperliquid charges no separate clearance fee on liquidations, and because liquidation orders go to the public book, all users can compete for that flow.² In a backstop liquidation, however, the maintenance margin is not returned to the trader.²

What centralised exchanges publish

Centralised exchanges know every position, but they do not publish them. What they publish is a stream of liquidation (force) orders, and the streams differ:

• Binance pushes force-order snapshots on its all-market liquidation stream once per second, and for each symbol only the latest liquidation order within that second is pushed.⁴ In a fast market, several liquidations of the same symbol in one second appear as one.
• Bybit describes its all-liquidation topic as pushing all liquidations that occur on Bybit, every 500 milliseconds, per symbol.⁵ The stream is subscribed symbol by symbol, and we cover the top 100 symbols by turnover.

Neither stream says which account was liquidated, so there is no way to follow a single trader's liquidations, and no public liquidation prices for open positions.

Why exchange totals are a lower bound

Put the two together and the direction of the error is always the same: published exchange totals can only miss liquidations, never invent them.

• On Binance, the one-per-second snapshot means a cascade of many liquidations in the same symbol is reported as a fraction of its real size.
• On Bybit, symbols outside the subscribed set are not seen.
• On both streams, liquidations that happened while a collector was disconnected are not published again later.

Hyperliquid numbers have their own limit. In real time we confirm liquidations only for the wallets we track, and a partial liquidation of a tracked position looks like a reduced position; liquidations of every other account come in later, from the hourly files of Hyperliquid's official archive. Until an hour is imported, its Hyperliquid total is also a lower bound, but for a different reason: coverage of wallets, not sampling of the stream.

Reading liquidation numbers across venues

Three habits avoid most misreadings:

• Compare a venue with itself. A comparison of the same venue across periods of the same coverage is far more reliable than a comparison of venues with different publication rules.
• Read the split as reported, not as market share. If one exchange shows most of the liquidations in an hour, part of that can be how much its stream publishes.
• Use side and timing, not only size. Which side was liquidated, and when, survives sampling better than the exact total.

Our liquidations page shows totals per venue and side, with the coverage of each stream stated. Where the data comes from, and since when each stream is collected, is documented in our methodology.

Limits of this comparison

• Docs and streams change. Publication rules are taken from each venue's official documentation as read on the access dates below.
• No insight into exchange internals. How a centralised exchange runs its own liquidation engine and insurance arrangements is outside what its public streams show, so we do not describe it here.
• Lighter. We also collect liquidations on Lighter, another perp DEX. This guide uses Hyperliquid as the example because its liquidation process is documented in detail.