There is a chart on Hyperliquid, on Coinglass, and on roughly every analytics dashboard a perp trader has on their second monitor, that gets called the liquidation map. It is a histogram, usually heatmap-styled, with notional dollars on one axis and price on the other. Tall bars cluster around a few specific prices. The implication, when you stare at it long enough, is that you are looking at the load-bearing levels of the market — the prices where, if touched, a wave of forced selling or buying will follow.
It is not nothing. The map is genuinely showing you something real. But the gap between what the map shows and what most traders read into it is wide enough that the dashboard ends up doing harm more often than good — especially on Hyperliquid, where the structural reasons it deceives you are slightly different from the binance-pattern CEXes most public maps were designed against.
This is a field guide. What the bars actually represent, what you can fairly conclude from them, and the two structural blind spots the map quietly carries — the ones that turn a useful chart into a cautionary tale when you trust it more than you should.
What the bars actually are.
A liquidation map is a forward-looking estimate, not a record of fills. Most public maps work by combining two inputs:
- Open-interest snapshots, scraped or polled from the venue.
- An estimated leverage distribution, fit to recent funding rates, basis, and historical fills.
A bar at $3,840 of height $14M means: roughly $14M of inferred long-perp exposure has its estimated liquidation price near $3,840, given the leverage the model thinks those positions are running.
A few things to internalize about that:
- It is an estimate of where margin runs out, not a guarantee that any specific account will be liquidated. The position could be partially hedged on spot. It could be in a portfolio-margin account where the liquidation price is computed against an entirely different basket. The notional could be wrong by a wide margin.
- It refreshes slowly. Hyperliquid's OI updates by funding epoch; the maps you see on most dashboards are minutes-stale at best. By the time a cluster prints on your screen, the trade that put it there is hours old.
- It collapses dimensions. You are looking at a single underlying. Most non-trivial perp positions are part of a basket — a long ETH offsetting a short BTC, a delta-neutral basis trade, a vol-target overlay. The map prints both legs as if they were standalone bets.
So the bar is not a count of liquidatable accounts. It is a model's best guess at the price floor (or ceiling) below which margin tension starts to bind. Useful, but inferential.
The three things the map does tell you.
When you have spent enough screen time with liquidation maps across regimes, three readouts come through reliably. Treat these as the upper bound on what you can fairly extract from the chart.
Where the magnet is, roughly. When one side of the book has built up materially more inferred liquidations than the other within 2–4% of spot, price tends to drift toward that side over the following hours. The mechanism is not mystical: market makers see the cluster too, and they price the optionality. If $40M of long liquidations sit at $3,720 and the next short cluster is twice as far away, a flat-tape afternoon is going to lean toward $3,720. Not always. But often enough that you can quantify the bias.
Where the cascade boundary is. A cluster of $80M or more inferred liquidations stacked within 0.5% of price is a real risk surface. Once the first 10% of those positions get hit, the resulting market sell pressure tends to walk price into the next 10%, and so on, until either the cluster is mostly cleared or the book thickens. This is the only place the map points at a self-fulfilling dynamic. Treat it as a vol-regime indicator, not a trade trigger.
Whether the funding regime is consistent. If funding has been positive for a week and the map nonetheless shows the heavier cluster on the short side, something is off — either the funding was not actually a strong signal of one-sided positioning, or the leverage being used is lower than the model assumes. A consistent map confirms a regime. A contradictory map is a flag to be more careful, not to fade harder.
That is the list. Three things. None of them are predictions; they are frame-of-reference reads.
The two things the map doesn't show.
This is where most traders get into trouble. The blind spots are not subtle once you know them, but they are rarely visible on the chart itself.
The first blind spot is hedged positions. A "long perp" on Hyperliquid is, materially, a different animal depending on what else the same wallet holds. A directional long is one risk surface. A long-perp leg of a basis trade against an offsetting short on a spot venue is another — and the latter is roughly half the universe of size in any normal funding regime. The basis-trade long has a "liquidation price" only in the most technical sense: the trader has no intention of letting it run there, because the spot leg is the actual P&L driver. They will add margin, they will close the leg, they will size down before they ever let it auto-liquidate. The map shows the basis-trade leg as if it were a directional bet. It is not.
The structural consequence is that the long side of the map almost always over-states the population of accounts that would actually let themselves get blown out. The cluster at $3,720 might represent $42M of inferred liquidations, but the unhedged portion is something closer to $12–18M. The cascade math the map invites you to do — $42M of forced selling will push price how far — is doing arithmetic on phantom dollars.
The second blind spot is cross-margin. Hyperliquid supports cross-margin accounts, and the public maps do not reliably know which accounts are crossed. A cross-margined long-ETH-perp does not get liquidated at the ETH-specific price the map prints; it gets liquidated when the whole portfolio's equity drops below the maintenance margin, which can be much higher or much lower than the per-position number, and depends on positions the map cannot see at all. A long BTC perp doing well at the same moment quietly buys the ETH leg another five percent of room.
Order-of-magnitude estimates from Hyperliquid's own published account-mode breakdowns put 30–50% of open size in crossed accounts in normal conditions. That is a third to half the chart that does not behave the way the chart says it does.
Neither of these is fixable from the outside. Better fitting will not recover the spot hedges. Smarter inference will not reveal the unrelated long-BTC keeping a cross-margined ETH-short alive. The map can only show you per-position math on a market that is mostly portfolio math.
How we use it, and the rule we use it under.
Engine ships a strategy in the marketplace that reads liquidation clusters on Hyperliquid as one of three confirming signals — not as a primary trigger. The strategy file makes this explicit:
## Signals
- Primary: funding_rate < -0.015% over 2 consecutive epochs
- Confirm A: OI delta > +$1.5M in last 30m
- Confirm B: inferred-liq cluster within 1.2% on the short side, > $20M
The primary rule has weight 1, the confirms have weight 0.5 each. The strategy needs the primary plus at least one confirm to fire. The liquidation map is a confirm — never a trigger.
The decision-log entries make the read auditable. When the agent fires, you see the funding number, the OI delta, the cluster size, and which one tipped it. When the agent does not fire on a setup that looked like it should fire, you also see why — usually because the cluster was real but on the wrong side, or because the cluster the map showed was upstream of a basis-trade leg and the rule discounts those by a configurable factor.
You do not need our strategy file to do the same thing. You need a rule, written down, that says: the liquidation map is a confirm, not a trigger. If you find yourself trading off the map's clusters alone, the map is doing the steering, and the map cannot see half of what is actually on the book.
A short checklist.
Before you let a liquidation map influence a decision:
- Confirm the cluster persists across at least two refreshes. A single-snapshot cluster is noise.
- Check whether the cluster sits inside or beyond the recent realized-vol cone. Inside the cone, it is likely already priced. Beyond it, it is a stress surface.
- Cross-reference with funding. A cluster on the same side as the funding regime is consistent; a cluster contradicting funding is more often a measurement artifact than alpha.
- Discount the long side by roughly a third for basis-trade hedging. Assume 30–50% of any bar is cross-margined and does not behave the way the bar implies.
- Decide before the trade whether the map is a confirm or a trigger. If it is the trigger, stop.
A note on the chart, finally.
Liquidation maps are useful instruments, but the chart is doing inference, not measurement, and what it cannot infer matters more than what it can. The traders who get the most out of the map are the ones who treat it as one of three or four signals, all in writing, all reviewable after the fact. The traders who get the least out of it are the ones who let the bars do the deciding for them.
If you are going to use the map, write down what you are going to do with it before you open the chart. That is the discipline that separates a useful signal from a confidence trick — on liquidation maps, and on every other piece of perp microstructure data you will read this cycle.