You weren't liquidated, but the system cut your position: insurance fund and ADL

Once the liquidation order hits the market, who fills the hole

Picking up where leverage vs liq price, the math left off: that piece takes the liquidation price all the way down to an actual number. But touching the liquidation price is only the moment the risk engine starts work — it is not the moment the position ends.

What happens after the trigger is this. The engine takes the position over and turns it into an order it sends into the market. That order needs someone on the other side, exactly like a close you place yourself. If the book is deep enough it fills in an instant, all you see on your statement is the word “liquidated”, and the stretch in between is something you never feel.

A thin book is an entirely different story. In a one-way cascade the bids get eaten layer by layer, so the liquidation orders behind have to reach down for worse and worse prices; in the more extreme case the order sits there unfilled, the position is still open, and the loss keeps growing. The 2024-08-05 BTC flash crash — 6 hours, 20% down had exactly this shape: every time a cluster of liquidation levels was punched through, fresh selling was released, the price dropped another step, and the next batch of liquidations was triggered. Liquidation at a moment like that is an accelerator, not a brake.

Partial liquidation is an earlier link in the same chain. When the position's margin ratio falls to a certain point, the system trims part of it instead of closing the whole thing, and lets what is left keep carrying. The intent behind the design is sound, but what goes out is still a market order, and when depth is short it goes out at the worst price all the same — partial liquidation does not let you dodge what this section is about, it only splits the problem into several events.

A word on scale while we are here. Exchange-wide daily liquidation volume is public on CoinGlass's liquidation data page, and anyone can pull the history themselves. Look at the days those spikes sit on and it is obvious that “the book could not absorb the liquidation orders” is not a theoretical corner case; it comes around several times a year.

So change one default assumption first: the liquidation price is a line in the ledger, not a promise about your fill. It tells you when the system takes over. It does not tell you what price you finally go out at. What happens inside that gap is the reason the insurance fund and ADL exist at all.

That also settles something else: your stop-loss should not sit right up against the liquidation price. Once that price is touched, whether you fill and at what price are no longer yours to control; the last price still under your control is the stop order you placed yourself.

Back to the gap itself. The liquidation price and the bankruptcy price are two separate lines with a stretch of buffer between them. The liquidation price is where your margin ratio falls to the maintenance margin requirement — at that point there is still something left in your account. The bankruptcy price is where equity lands exactly on zero. For how that first line resolves into a concrete number, leverage vs liq price, the math has a version you can plug figures straight into. The risk engine's job is to dispose of the position inside the empty stretch between those two lines.

There are only two outcomes. Either the fill lands on the near side of the bankruptcy price, the position closes with a little residual value left over, and that residue does not come back to you — it goes into the insurance fund. Or the fill slips past the bankruptcy price, account equity turns negative — what the glossary files under negative equity — and the insurance fund covers that negative number. So the insurance fund is not a separate pot of money the platform puts up; it is essentially the residual value of every historical liquidation, accumulated. And a note on names while we are at it: insurance fund, risk reserve, risk margin pool — different platforms and different documents use different words for the same layer of buffer. This site's glossary files it as “risk margin pool”.

Follow that logic and one more thing falls out. The reason the liquidation price sits ahead of the bankruptcy price in the first place is to leave the engine that stretch of time to get filled. Keep the buffer narrow and the platform's capital efficiency is better, but the engine has less room; keep it wide and you get closed out earlier. Tiering the maintenance margin rate by notional value is, at bottom, a balance struck between those two ends — the larger the position, the longer the stretch it needs, and the higher the margin ratio you are asked to hold.

One more piece of money that is easy to overlook: most platforms also charge a liquidation-related fee when they close you out, and that usually goes to the same pool. Whether it is charged and how it is worked out differs from venue to venue, and it is written on the rules page in force at the time.

This also explains the shape the fund has. It grows slowly while markets are calm, and a single one-way cascade can drain a large slice of it in a day. The right way to read the number is as a trend, not as an absolute on any one day — a balance that visibly steps down after a run means a meaningful share of that run's liquidations filled past the bankruptcy price, which is to say depth really did fall short.

If a platform publishes the pool's balance and its historical curve, the shape is worth a glance once a month. What you are looking for is not whether the number is big — detached from the platform's size that tells you nothing — but how far it dropped in the last few violent moves and how long it took to come back. Slow to recover means the buffer is still in its recovery period, and if a shock of the same size arrives inside that window, the odds of the next layer being called on are higher.

And one thing that is easy to get wrong, said plainly: the insurance fund protects the balance of the ledger, not you. What it covers is the money owed by the accounts that blew up. Your own capital was gone the moment the bankruptcy price printed. However thick the fund's balance is, it carries no compensation of any kind for someone who has already been liquidated.

Whether the fund is pooled per currency or shared across the whole platform, and how negative equity from isolated and cross positions is collected, differ by venue and do get changed. The current version lives in OKX's risk and rules documentation in its help center — don't paste another venue's wording on top of it. For the difference between the two margin modes themselves, see isolated or cross margin? how to choose on OKX perpetuals.

When the pool runs short, the system turns to the side that's making money

When a run drains the insurance fund below what it takes to cover the negative equity appearing, the system needs another way to square the books: pick positions on the opposing side and force part of them closed. That is ADL — auto-deleveraging — and it has its own entry in the glossary.

Here is the counter-intuitive part: the side that gets picked is the side that is winning. Somebody has to own the money the blown-up side owes, and when the fund cannot own it, the only option left is to settle part of the profit on the opposing book early so the two sides balance out. Your margin is sufficient, your direction was right, you did nothing wrong, and your position shrinks anyway.

The underlying reason is the structure of perpetuals. There is no delivery date, every long is matched by a short, and the platform is not your counterparty. Negative equity that appears in the market does not disappear because the bookkeeping is rearranged; it can only come out of three places. The liquidated trader's own margin is the first layer, the insurance fund is the second, and the unrealised profit of the opposing side is the third. ADL is the moment that third layer gets switched on.

The actual damage it does to you should be stated precisely too. The part that gets taken off is usually closed in profit, so this does not lose you money. The real loss is exposure: you read the direction right, the market is still moving your way, and the car has put you down at the side of the road. Getting back to the position you had means opening again, the price is no longer the price it was, and you pay the opening cost a second time, plus the funding rate from there on — across a stretch of trend those two together are not small change.

Something else worth being clear about: the last step in the chain is not identical everywhere. Some platforms keep a socialized loss layer alongside auto-deleveraging, spreading the shortfall proportionally across a set of profitable accounts rather than closing out a few people's positions outright. Whether that layer exists, where it sits in the order and how wide it reaches are all platform rules — go by the rules page in force at the time, and don't reason from one venue's mechanism to another's.

Read through the design intent and ADL is a way of stopping the bleeding, not a revenue line. The platform does not make money at this step; all it wants is no hole in the books, because once a hole is left, what gets affected is everyone's withdrawals. Understanding that helps you judge how it behaves: it closes the amount it has to, as fast as it can, and then stops. It does not trim extra while it is there.

So there is not much point being angry at ADL. It is more useful to treat it as a cost that comes built into the product. It is close in nature to the funding rate — both are terms you accept in exchange for holding long-dated exposure through a perpetual — the difference being that funding is deducted every day where you can see it, while this one does not show itself at all most of the time.

One more angle makes it clearer: liquidation is your margin running out; ADL is your counterparty's margin running out. Two ends of the same liquidation chain, one cleared out by the market and the other held down by the system. The difference is that the second end was never on your risk checklist, because it is not decided by the state of your position.

The two factors behind the queue: profit ratio and effective leverage

The queue is not random, and it is not ordered by position size. What the major platforms use is broadly profit ratio multiplied by effective leverage: the larger the product, the further forward you sit, and losing positions don't enter the queue at all.

Profit ratio is that position's unrealised PnL relative to the margin it occupies. Effective leverage is notional value divided by the position's current equity, with unrealised PnL counted inside equity. The two factors partly cancel each other out: a growing unrealised profit lifts the profit ratio, but it also lifts equity and pushes effective leverage down. So simply making a lot does not necessarily put you near the front. What actually gets pushed to the front is a combination — margin kept thin, a large unrealised profit, and leverage that was high from the moment the position was opened.

Why order it this way rather than by position size or time of entry? Two reasons. First, the hole should be filled with the smallest amount of closing possible, and a position with a high profit ratio releases more profit for every unit closed. Second, an account running high effective leverage contributes more risk to the system as a whole, so having it give up a slice first is consistent with how risk control works everywhere else. First come, first served looks fair, but it is neither efficient nor aligned with risk.

A comparison without numbers. Two people are both long and both winning. A opened at low leverage, so as the unrealised profit came in, the position's equity thickened along with it. B opened at high leverage with margin pared back to just enough to carry the position, leaving the whole unrealised profit sitting inside the position as a cushion. Their profit ratios might be similar, but B's effective leverage never came down, and in the queue B sits ahead of A.

Margin mode affects the reading here. Under isolated margin, the position's equity is simply its own slice of margin plus unrealised PnL, which is clean to work out. Under cross margin, the account balance and the unrealised profits and losses of your other positions all feed in, and the same notional value can produce an effective leverage some distance from your intuition. If you are on cross, estimate your place on an account basis; don't stare at that one position alone.

This is also why “I haven't added to the position, so I'm not taking on new risk” is an illusion. You do nothing, the unrealised profit rises, and the profit ratio rises with it. As long as effective leverage hasn't come down in step, your place in the queue keeps moving forward — a process in which you never made a single decision.

That bar of a few segments on the positions screen is reading this rank. Note that what it gives you is a relative position, not a probability: if nobody in the venue is being liquidated, nothing happens even with the bar full; conversely, in extreme conditions a bar that is only half lit does not mean you are entirely safe. The right way to use it is to watch which direction your own segment count moves over a few days.

Formula details, how the bar's tiers are cut, and whether losing positions are excluded outright all differ between platforms' implementations and get adjusted between versions — go by the rules page OKX publishes at the time. What this section gives you is the judgement logic, not a formula you can plug numbers straight into.

Spotting, in your trade history, the fill you never placed

You can identify it, and it rests on three features holding at once: you never placed this order, what got trimmed was a position on the winning side, and the fill type or source field on this record differs from an ordinary close. The exact field names change between versions — don't memorise names, memorise the combination.

How to screen for them in practice. Export the trade history by time, ring-fence the window around the last violent move, then inside it look for the fills that are closes, positive on PnL, and don't match anything in your own order records. What's left is basically it; go back and check the fill type field once at the end to confirm.

The first use of flagging them is reconciliation. You will know this movement of money wasn't something you clicked, you won't spend time trying to remember what you did that day, and you won't put it down to some automation tool going wrong. A position shrinking without your knowledge is enough on its own to make you doubt your account's security, and being able to attribute it straight away is worth something.

The second use is review. You don't choose an ADL fill's price and you don't choose its timing; mixed into your own trading record, it puts a little dirt on your win rate, your average holding time and your profit-to-loss ratio. If you log trades in R-multiples, the way the stop-loss vs position-size paradox sets out, be especially careful here — this trade's R was not produced by your strategy.

One more practical question on top: how the fee on this fill is worked out, and which price exactly is taken, differ from platform to platform. Don't assume it works like a market close of your own; check the wording in force on the rules page before you reconcile.

When the queue gets crowded, and how to move yourself back

ADL is a low-frequency event on major contracts in calm markets. It doesn't appear on the list in the 5 most common perpetual futures rookie mistakes either — beginners rarely survive to the point where they need to worry about it. It clusters under a handful of conditions, and every one of those conditions can be seen coming.

A one-way market with the depth drained. This is the main one. Liquidation orders don't fill, the scale of negative equity grows, the insurance fund is consumed fast, and the queue starts up behind it. The kind of move where a whole row of clusters on the liquidation heatmap is punched through level by level is this process made visible; the 2024-08-05 BTC flash crash is a complete sample of the shape from start to finish.

Contracts that were thin to begin with. Perpetuals on small-cap coins, the book in the small hours and across long holidays — the same notional size causes far more slippage here than on a major contract, and negative equity shows up more readily.

The heavily crowded side of the book. A funding rate parked at an extreme value in one direction for a long stretch means one side is packed. Once that side starts blowing up in a chain, the scale is concentrated — and the other side, which is you if you're on the sparse side, happens to be sitting on unrealised profit.

An insurance fund that has just been drained. A balance that visibly steps down after one run means there is less buffer to draw on when a shock of the same size comes next. This is also why you read this number as a trend.

Contracts that have only just listed. Resting depth hasn't spread out yet and quotes are wide, so one medium-sized liquidation is enough to punch through the book. Opening at high leverage in a new contract's first days is nothing like the same multiple on a major contract.

Cross-instrument synchronisation in macro events. One headline hits dozens of contracts at once, and if the insurance fund is a single shared pool, the pressure stacks rather than being counted separately. Watching only the book depth of the contract in your own hands at a time like that will understate how tight things are overall.

The state of your own side. A large unrealised profit, margin kept thin, leverage that was high to start with — the earlier items decide whether the queue starts at all; this one decides where you are standing when it does.

Three of those you can watch ahead of time: whether the funding rate has been leaning one way for a long stretch, how fast open interest drops in a sharp fall, and whether the pool's balance has just taken a step down. All three point at the same thing — how much buffer is left in the market. They don't forecast the market; they tell you how tight the liquidation chain will be if a move does come.

Put together, they add up to a conclusion that isn't very comfortable: the moment ADL is most likely to come looking for you is in the few days when you have read the market best and made the most. That is not a coincidence — it is a direct corollary of the queueing rule.

So what can you do? Only two directions actually move your rank: bring effective leverage down, or bank the unrealised profit. Everything below is operational detail around those two.

Add margin once the unrealised profit grows, instead of letting it carry the position. When an isolated position is well up, plenty of people feel that with unrealised profit underneath there is nothing left to manage. From the queueing rule it is exactly the opposite: thin margin plus a big unrealised profit is the textbook front-of-queue configuration. Put a little margin into the position and the profit ratio doesn't change, effective leverage comes down, and the rank comes down with it.

Scale out. Turning part of the unrealised profit into realised cuts the profit ratio factor directly. This is something you should be doing anyway; now there's one more reason.

Set leverage at a level you can hold long-term, at the moment you open. Open high and worry about it once you're up — that setting is one we generally advise against; it scores you points on the liquidation side and the ADL side at the same time. Work out the liquidation price and effective leverage together before you open; the liquidation price calculator and the position calculator will run it for you.

Put the ADL indicator into your daily check. It costs the same glance as the liquidation price and the funding settlement countdown, and you are already looking at those two.

Don't size up on obscure contracts. This holds for liquidation and for ADL alike, for the same reason: the book can't carry it.

If you want to hold exposure through extreme conditions, consider moving part of it to spot. Spot has no liquidation and no ADL. This is not strategy advice, only a reminder that both of these risks belong to leveraged perpetuals; change the instrument and they don't exist.

With several positions open, look at the most extreme one first. The queue is built per position, not per account. You might have three longs at once with two of them at very low leverage, but as long as one is carrying a big unrealised profit on thin margin, that's the one that gets pulled out. Go through them one by one; don't just read the account summary line.

Compress all of this into a checklist you can use at three moments. Before you open, fix the leverage, and don't count on unrealised profit diluting it later. While you hold, glance at the indicator alongside the liquidation price every day, and add margin or scale out as the unrealised profit thickens. Afterwards, reconcile, and flag separately the fills you didn't initiate. None of the three is difficult; the hard part is that the second one has to be done on the days you feel most comfortable.

ADL is a low-probability event, but its probability distribution is skewed — it picks out precisely the days when the liquidation chain is tightest, and on those days you are most likely holding a good-looking unrealised profit. The cost of leverage is not only that one line at the liquidation price; beyond the line there is this stretch as well.