How crypto liquidations actually work: mechanics, cascades, and the math of survival

Every day, tens of millions of dollars in crypto positions are closed by force. Not by their owners. By the exchange, at market, at the worst possible moment for the account holder, and by design. Liquidation is the load-bearing mechanism of leveraged crypto trading, and almost nobody using leverage can explain what actually happens when it fires. We record the liquidation stream across 636 perpetual contracts in real time, so here is the full picture: the mechanics inside the engine, why cascades form, what the data shows, and the arithmetic that decides whether an account survives its first genuinely bad week.

The mechanics, step by step

When you open a leveraged futures position, you post margin. A $1,000 position at 10x requires $100 of your money; the exchange effectively fronts the rest of the exposure. That loan is not charity, and the exchange protects itself with a simple rule: your losses must never exceed your margin. The enforcement of that rule is liquidation.

Two numbers govern everything. Initial margin is what you post to open. Maintenance margin is the minimum equity the exchange requires to keep the position alive, typically a fraction of a percent to a few percent of position value depending on size and tier. As price moves against you, your unrealized loss eats your margin. The moment remaining equity touches the maintenance requirement, you no longer own the position. The liquidation engine does.

What the engine does next matters for everyone in the market, not just the account being closed. On Binance Futures the position is closed by a forced order into the live order book. The engine is not patient and it is not price sensitive; its job is to end the exposure before losses exceed margin, and it will sell into whatever bids exist. If the book is thin, the forced close itself moves price. Any margin remaining after the close and fees is returned to the trader. Usually there is little. If the close cannot be completed above bankruptcy price, the venue's insurance fund absorbs the difference, and in extreme scenarios exchanges resort to auto-deleveraging, closing positions of profitable traders on the opposite side. That last mechanism is rare, but its existence tells you how seriously the system takes the rule that losses stop at margin.

One practical detail worth knowing: liquidation is triggered by mark price, a composite index designed to resist manipulation, rather than the last traded price alone. This protects you from being liquidated by a single rogue wick on one venue, and it also means your liquidation level is not exactly where a naive calculation from the chart suggests.

Why cascades form

A single liquidation is a private tragedy. A cascade is a market event, and it emerges from three ingredients that are always present.

First, traders cluster. Entries concentrate around obvious levels: breakouts, round numbers, support zones everyone watches. Leverage choices concentrate too, because the interface offers the same presets to everyone. Similar entries plus similar leverage equals similar liquidation prices. The liquidation map of a popular coin is not a smooth distribution; it is a series of dense pockets sitting under and over the market like charges.

Second, forced orders move price. When price touches the first pocket, the engine sells those positions at market. That selling pushes price lower, into the next pocket, whose positions are then also sold at market. Each wave finances the next. This is why liquidation moves have a signature look on the chart: acceleration through levels that held for days, on enormous volume, in minutes, followed by a sharp partial snap-back once the pockets are empty and forced selling simply stops.

Third, leverage makes the pockets shallow. At 20x, a 5% move reaches the cluster. In altcoin perpetuals, 5% is not an event. It is Tuesday. Our own logs show routine days where a mid-cap alt travels its entire 20x liquidation distance intraday, in both directions, without any news at all.

The important reframe: cascades are not malfunctions and they are not conspiracies. They are the deterministic consequence of crowded leverage meeting forced execution. The market is not hunting you personally. It does not need to. The mechanism finds the crowd automatically, because the crowd built the trigger itself. Whether specific large players position around visible liquidation pockets is a separate question, but here is the honest part: they do not need to do anything illegal for the pocket to get hit. Gravity does the work.

What the data actually shows

Exchanges publish the liquidation stream in real time but do not archive it; once an event scrolls past, the public record is gone. Binance provides no historical liquidation endpoint at all. We have been recording the stream continuously across 636 perpetuals since May 2026, which lets us say a few things from data rather than folklore.

Liquidations are wildly lopsided in time. The stream spends most hours nearly silent, then prints double-digit percentages of a day's total volume in bursts of minutes. Averages are meaningless here; the distribution is all tail. Second, long liquidations dominate in this market's structure. Retail crypto skews long chronically, so the ammunition on the downside is almost always larger, which is one reason drops are faster than rallies of the same size. Third, the aftermath of a large cascade is measurably calmer: once positioning has been forcibly cleared, the same coin often needs days to rebuild enough open interest to produce another violent move. Positioning is the fuel gauge, and after a cascade the tank reads empty. You can inspect per-coin liquidation history, long versus short, on any coin's page in our Data Terminal, alongside the open interest that feeds it.

What we could not find in the data is equally worth reporting. We tested liquidation-driven entry strategies against random-entry controls, the same honest benchmark we apply to every popular strategy, and the simple formulations retail is sold, buy the cascade, fade the wick, did not survive testing as standalone rules. Cascades mark where forced sellers finished. They do not, by themselves, mark where buyers begin.

The arithmetic of survival

Everything above is mechanism. This section is the part that decides outcomes, and it is nothing but division.

Your liquidation distance is roughly 100 divided by your leverage, in percent, minus fees and funding. At 5x you can survive an adverse move approaching 20%. At 10x, about 10%. At 25x, about 4%. At 50x, about 2%. Now put those numbers against the market's actual behavior: the median liquid altcoin perpetual routinely travels 5-10% within a day, and 3-4% within an hour during active sessions. At 25x and above, you are not trading a view on the market anymore. You are betting that the next few hours contain no noise, and noise is the one thing crypto reliably supplies.

The deeper error is confusing per-trade survival with account survival. A 4% liquidation distance does not mean you lose occasionally; compounded over dozens of trades, it makes eventual ruin close to certain, because a single routine fluctuation ends each position regardless of whether the idea was right. Professional desks run effective leverage of low single digits not out of timidity but because the arithmetic of ruin dominates the arithmetic of returns at any horizon that matters. High leverage does not increase your expected return. It increases the speed at which your account reaches its expected value, and for most retail strategies that value, measured honestly after fees against random controls, is negative to begin with.

If you take one action from this guide, make it this: compute your liquidation distance before entry, compare it to the coin's ordinary daily range on its data page, and if the routine day can reach your liquidation level, your problem is not entry timing. It is size.