Late on a Tuesday night, I sat in my study with three browser tabs open — Coinglass, CoinAnk, and a raw CME order book — trying to reconcile three different numbers for the same phenomenon. All three claimed to know how much money would be forcibly liquidated if ETH crossed a given price. None of them agreed. The gap between the highest and lowest estimate was wide enough to drive a leveraged position through.
That night reminded me of something I learned the hard way during my 2017 audit work: a number is not a fact merely because it is precise. Precision is a style of presentation. Truth is a property of the world. The two are not the same, and confusing them is how careful people lose their capital.
This week, a Coinglass snapshot circulated widely: break above $2,480 and roughly $1.622 billion in short liquidations would trigger; drop below $2,245 and about $580 million in longs would go under. The asymmetry — shorts outnumbering longs by roughly 2.8 to 1 — became the headline. It is a genuinely interesting structural signal. But before anyone builds a trade on it, we owe ourselves the unglamorous work of asking how those numbers exist at all.
To understand what a liquidation heatmap is, you have to understand what it is not. It is not a ledger. It is not a record of real orders resting on an exchange. It is an inference — a reconstruction assembled from three ingredients: open interest, an assumed distribution of leverage, and a price-bucket model. Coinglass does not see individual positions. No third party does. Mainstream centralized exchanges do not publish per-position liquidation prices; that data lives inside their matching engines and stays there. What the heatmap offers is an educated guess, dressed in the visual authority of a heat gradient.
The mechanics being modeled are real enough, and I want to be fair to them. When a leveraged position's margin ratio falls below the maintenance margin requirement, the exchange's liquidation engine force-closes it. That forced close becomes a market order in the opposite direction — a liquidated long sells, a liquidated short buys. Cluster enough of these at a single price and you get a cascade: forced selling begets more forced selling, or forced buying begets more forced buying, until liquidity thins and the move accelerates beyond what any fundamental catalyst would justify. This mechanism is not fiction. I have watched it unfold in 2021, in 2024, and in dozens of smaller episodes that never made headlines. The cascade is a permanent feature of leveraged markets, not a passing phase. In a bear market, it is often the only engine still running.
So the question is never whether liquidation cascades exist. The question is whether any specific heatmap accurately predicts where they will occur — and here the answer is more uncomfortable than the colorful charts suggest. A heatmap is a weather forecast, not a photograph of the storm. It tells you where lightning is likely, based on patterns it has seen before. It cannot tell you when the strike lands, or whether the clouds will part instead.
Let me be concrete about the methodology, because this is where twenty-five years of watching markets and six months of auditing consensus code take over.
Coinglass estimates liquidation density by multiplying open interest by an assumed leverage distribution and mapping the result against price levels. Three assumptions do the heavy lifting, and each is a potential source of error. First, the leverage distribution — how much open interest sits at 5x, 10x, 25x, 50x, 100x — is not publicly disclosed. It is modeled. Second, isolated-margin and cross-margin positions behave very differently, and the model often cannot distinguish between them. A cross-margin trader with a large collateral buffer will not be liquidated at the price a naive calculation predicts. Third, hidden orders and market-maker hedging positions are invisible to the estimate entirely.
Stack those together and the honest error band on a figure like $1.622 billion is not 5%. It is plausibly 20% to 30%. The number you see is the center of a distribution, not a point. A trader who treats $1.622 billion as gospel and sizes a position around it is not trading on information; they are trading on a model's midpoint, which is a far more fragile thing to lean on.
None of this makes Coinglass malicious. It makes the product human — an approximation of a system that is deliberately opaque. And that opacity is the deeper story. The reason we depend on third-party estimators at all is that centralized exchanges refuse to disclose real liquidation data. Transparency, in the derivatives market, is something you rent from an aggregator precisely because the venues themselves will not provide it. That is a structural flaw, not a temporary inconvenience. When a single data provider becomes the market's shared nervous system, we have swapped one form of centralization for another — and called it insight. I find that troubling in the same way I find a single oracle feeding a billion dollars of DeFi collateral troubling: concentration is concentration, whether it wears the mask of a bank or the mask of a dashboard.
Now, the asymmetry itself. The 2.8-to-1 ratio between short and long liquidation pressure is the one signal worth taking seriously, and it deserves more nuance than the headline allows.
Read one way, it is a coiled spring. If ETH pushes through $2,480 on real volume, the forced buying of liquidated shorts can feed on itself — each liquidation lifts the price, which triggers the next, a positive feedback loop that amplifies volatility. In a thin order book, that can move price faster than any fundamental catalyst could dream of.
Read another way, it is a warning about what we cannot see. A large short book does not necessarily mean the market is bearish. It can mean the market is hedged. If a meaningful share of those shorts are hedges — spot longs paired with perpetual shorts, or option market-makers delta-hedging their books — then those positions will not be force-liquidated on a rally, because the spot leg is profitable and offsets the loss on the derivative. Distinguishing speculative shorts from hedged shorts is the difference between a genuine squeeze setup and a mirage. The heatmap cannot tell you which is which. That distinction lives in the funding rate and the basis, not in the heat gradient. When funding is deeply negative, shorts are paying to stay short — that is conviction, or at least crowding. When funding is flat while open interest is high, you may simply be looking at insurance, not a bet.
Which brings me to the missing variable, and it is not minor. The snapshot never tells you the current price of ETH. Without knowing how far spot sits from $2,480 and $2,245, the entire map loses its operational value. A threshold 12% away is a very different proposition from one 1% away — one is a distant possibility, the other an imminent probability. Omitting the reference price turns a risk radar into a decoration. Any honest presentation of liquidation data begins with where price stands right now, because liquidation risk is always a function of distance, not of level.
And the risk does not stop at the centralized exchange. My long-running concern about DeFi lending protocols is that they share the same collateral — ETH — with the CEX derivatives market, but they run on different clocks. When ETH falls fast enough to trigger CEX long liquidations, the drop propagates into on-chain money markets. Aave and Compound positions near their liquidation thresholds get called, and their liquidations push yet more ETH onto the market. Worse, on-chain liquidations depend on price feeds, and price feeds carry latency. An oracle that reports a stale or lagging price during a violent move can either trigger liquidations that should not happen or delay the ones that should — a failure mode that has nothing to do with the heatmap and everything to do with infrastructure. This is the part of the system the derivatives charts almost never show. The CEX cascade and the DeFi cascade are not two events. They are one event wearing two costumes, and most liquidation maps render only half of the body.
I want to pause on that, because it is where the real education lies. In 2022, when the algorithmic stability of a certain protocol dissolved in front of everyone, the post-mortems focused on the model that failed. They paid less attention to the plumbing — the oracles, the bridges, the liquidators — that turned a bad day into a systemic one. The lesson was not that leverage is dangerous. Everyone knows leverage is dangerous. The lesson was that risk travels along paths no one draws on a chart. A heatmap of CEX liquidations is a map of the visible plumbing. The invisible plumbing is where the flood begins.
Here is the counter-intuitive part, and the one that keeps me up at night.
The liquidation heatmap is not a neutral observation. It is a participant. When a widely-followed data point names $2,480 and $2,245 as the levels where fortunes turn, it does not merely describe the market — it teaches the market where to look. Traders cluster their stop-losses and take-profits around published thresholds. Market-makers anticipate liquidity at those prices and position ahead of it. The map, in other words, helps draw the very terrain it claims to survey. This is not conspiracy; it is reflexivity, and it is the ordinary behavior of any market that runs on shared data.
The practical consequence is uncomfortable for anyone who treats these levels as natural boundaries. They are partly self-fulfilling and partly self-defeating. Self-fulfilling, because attention concentrates liquidity there. Self-defeating, because sophisticated players know the crowd is watching, and the crowd's visible intentions are the easiest thing in the world to trade against. A level everyone can see is a level everyone can prepare for. The cascade may come — but not necessarily where the map insists.
There is a second blind spot worth naming. The heatmap captures leverage on centralized venues, where positions are visible to the model. It says nothing about the leverage that matters just as much — the recursive leverage in restaking ecosystems, where the same ETH is pledged, re-pledged, and lent against in layers that no heatmap models and few regulators understand. If a genuine cascade arrives, the damage will not be confined to the positions the chart shows. It will surface in the positions the chart has never heard of. The visible map is the comfortable one. The real risk is almost always the part of the system that no one bothered to draw.
So what do we do with a number like $1.622 billion? We hold it the way we should hold all estimates: as a hypothesis, not a verdict. We cross-check it against at least one independent source. We watch the funding rate to learn whether the shorts are conviction or insurance. We track open interest for the leverage quietly accumulating underneath. We remember that the venues generating this data are the same venues that profit from every liquidation, and we price that incentive into our trust accordingly. And we treat the missing reference price as a reminder that context is not decoration — it is the difference between information and noise.
Truth is immutable, unlike the price action. The price will do what it does, and the cascade will come when it comes. Our job is not to predict it, but to stop mistaking a model's midpoint for a fact — and to survive long enough that the distinction still matters.