Liquidation heatmaps: how to read one, and how accurate they are
How to read a liquidation heatmap: what the bands and colours mean, how the levels are estimated, estimated levels against real positions and reported liquidations, and a public score of how often the levels meet real liquidations.
A liquidation heatmap is a chart of the prices where leveraged futures positions would be forced closed, drawn over the price chart. Bright bands mark prices where a lot of money would be liquidated at once. Bands above price are short positions that would be forced to buy; bands below are long positions that would be forced to sell.
Most heatmaps are estimates. Exchanges do not publish where every position would be liquidated, so the levels come from a model. Before reading a bright band as fact, check which of the three kinds of liquidation data you are looking at: estimated levels, real positions, or liquidations that already happened.
Open a live map: BTC liquidation heatmap · ETH · SOL · Hyperliquid BTC, from real positions · every coin
What a liquidation is, and why the levels matter
On perpetual futures a leveraged trader puts up margin. If price moves far enough against the position that the margin no longer covers it, the exchange closes the position with a market order. That forced close is a liquidation. A long liquidation is a forced sell; a short liquidation is a forced buy.
Forced orders are market orders, so they push price further the way it was already going. When many positions share a liquidation price, reaching it can set off a chain: one batch of forced orders moves price into the next batch. That chain is a liquidation cascade, and it is why traders want to know where the levels sit.
The popular idea that price is drawn to liquidation levels like a magnet is a claim, not a rule. A large level is also a place where price can stall, or where nothing happens for days. The public score further down is one way to check how often levels and real liquidations actually meet.
How to read a liquidation heatmap
- Find the price. Everything on the chart is read from where price is now.
- Above price: short liquidations. If price rises into a band, the shorts there are forced to buy, which can push price higher still.
- Below price: long liquidations. If price falls into a band, the longs there are forced to sell.
- Brightness is size. A brighter band means more dollars at that price. In many tools the colours are relative to what is on screen, so the same yellow can mean very different dollars on a quiet day and a busy one; read the dollar labels where the tool gives them. In EdgeDepth each level is a rail: a fixed price, coloured by the liquidations expected there within 24 hours, or by its dollars.
- The profile at the right edge adds up the levels still standing at each price, so the largest clusters stand out at a glance.
- A touched level is treated as closed. The model takes the positions behind it to be closed. A saved level end can lag the first price crossing, as the example below shows. Ending does not prove anyone was liquidated there; only reported liquidations record that.
- Distance and time matter. A large level 15% away says little about the next hour. A small one just above price can matter more.
Not to be confused with an order book heatmap, which draws the limit orders resting in the book, not the prices where positions would be forced out. A search for “BTC heatmap” can mean either.
A real example: BTC before and after a level was crossed
What did the map show before price moved, and what happened afterward? This recorded example follows two estimated long-liquidation levels below BTC’s price on 15 October 2026. It was selected afterward to teach the distinction between an estimate, a price crossing and an exchange report, not to demonstrate predictive accuracy.
Before: read the estimate without the outcome
At 02:20 UTC, the last complete candle closed at $81,769. Below it, level A was $80,890 with $24.36M of estimated long liquidations. Level B was $80,326 with $26.39M. These were model estimates, not known positions or orders waiting in the book.
The candles end at the first panel’s clock. The blank area is not missing price data: it deliberately withholds the later minutes. The lower bars show only reports timestamped within the displayed complete minutes: red for longs, green for shorts. Both panels use the same axes and keep the pre-move estimates fixed.
Pause here: does the larger estimate at B mean price must reach B first? Can either line tell you when a liquidation will occur?
Reveal what happened next
The 02:24 candle crossed A. B was not reached before this window ended at 02:27. Binance separately reported $13.01M of long liquidations and $1.50M of short liquidations during the crossing minute. Those are market-wide reports for that minute, not a measurement of the exact positions estimated at A.
The answer: no. A was nearer, and price reached it while the larger B remained untouched during this window. Neither estimate specified an arrival time. The reported dollars do not have to equal the estimate: they cover a whole market-minute, cannot identify every position at A, and Binance’s published feed is incomplete.
The candle first crossed A at 02:24, while the stored rail record marks its end at 02:32. We keep those timestamps separate. A chart crossing is not proof that the model published an update at that instant.
Takeaway: a level is an estimate to inspect, a crossing is a price observation, and a liquidation report is separate evidence. One selected example cannot tell you how often a map succeeds or fails. Read the score and its limits, and compare other dates before drawing a conclusion.
Open BTC’s public map and daily record. Its live values change; the two teaching panels above are pinned to the dated records. They are a reconstruction, not screenshots of today’s terminal. The saved step times are 01:40 UTC for A and 01:45 UTC for B. The stored record does not prove the exact time a subscriber received each update.
Optional: what trade bubbles add
Look inside a different move: GTC's trade bubbles
This is a different market and date, shown only to distinguish executed trading from resting orders. It is not the BTC liquidation example above.

The circles are grouped recorded trades: green for taker-led buying, red for taker-led selling, larger for larger grouped values. The coloured blocks behind them are resting order-book depth. Neither a trade bubble nor a bright block identifies a liquidation. Use the separately reported liquidation layer for that evidence.
Open the GTC event record or read the order-book heatmap guide.
Three kinds of liquidation data
| Estimated levels | Real positions | Reported liquidations | |
|---|---|---|---|
| What it is | A model’s estimate of where positions would be forced out | Open positions read from the exchange, at its own liquidation prices | Forced closes the exchange published after they happened |
| Built from | Price, volume and open interest, plus assumptions about leverage and entry | Hyperliquid, which exposes each account’s open positions | The exchange’s liquidation feed |
| Tells you | Where forced orders might sit | Where they do sit, for the positions covered | Where they did happen |
| Cannot tell you | Whether the positions exist | Positions it cannot locate | Every liquidation: the feeds are incomplete |
| In EdgeDepth | Liquidation rails on Binance and Bybit | The Hyperliquid layer | Reported-liquidation markers and pane |
Estimated levels: how most heatmaps are calculated
A liquidation price follows from the entry price and the leverage. A long opened at $100 with 20x leverage is liquidated roughly 1 ÷ 20 = 5% lower, a little closer once the exchange’s maintenance margin is taken off, so near $95.5. At 50x the same long is liquidated within 2%; at 100x, within 1%.
No exchange publishes every position’s entry and leverage, so an estimated heatmap works backwards. It watches where open interest grows, assumes the new positions were opened near that price with a spread of leverage, computes their liquidation prices and, in most tools, removes them once price trades through. Every tool makes its own assumptions about leverage, entries and when positions close, which is why two heatmaps of the same coin disagree.
EdgeDepth’s rails are this kind of estimate. Rails come from a model calibrated on Bybit’s liquidation record, not from known positions. In development they beat no-fuel and even-split baselines, and the 24 h touch chance was 27.5% predicted against 28.6% observed. The forward test on live data reports from 16 Nov. A rail is not a forecast that price will reach it. Read the forward test registration.
Real positions: Hyperliquid
Hyperliquid is different. It publishes each account’s open positions with the exchange’s own liquidation price and real leverage, so its levels can be read rather than guessed. EdgeDepth reads them from its own Hyperliquid node every minute and draws the significant levels as rails.
Two limits come with that. Hyperliquid is one venue, so these levels describe Hyperliquid traders, not Binance’s or Bybit’s. And a cross-margin position shares its account’s margin, so its liquidation price moves as the account’s other positions gain or lose.

The node reads every wallet it has seen trade, which on 15 Oct came to 98.7% of BTC open interest, but a level needs a usable liquidation price near the mark. The chart’s key shows the located share; at the time of the BTC chart above it read $1.43B of $3.20B (44.8%). Cross-margin positions with no usable price, and levels far from the mark, are left out rather than painted as false precision. More on the method on the Hyperliquid liquidation heatmap page.
Reported liquidations
Reported liquidations are the forced closes an exchange published after they happened. They are the only one of the three that records what took place, and they are incomplete. Binance’s public stream publishes one selected liquidation per symbol in each 1000 ms window, so when several land together most go unpublished, and its totals are a floor (why a liquidation count is not a count). Missing reports are never filled in with estimates. Read them beside price and the tape as a lower bound, and keep them separate from the estimated levels.
Are liquidation heatmaps accurate?
We checked 11 liquidation map and heatmap tools on 4 October 2026. None published how often its levels met real liquidations. A heatmap that never shows its misses cannot be told apart from one that only looks convincing.
EdgeDepth scores its own map every day, misses included. For BTC on Binance Futures, over 8 scored days (5 to 12 Oct 2026):
- Price spent 9% of the time at the map’s levels, about 15 of 172 hours.
- 64% of the liquidated dollars Binance reported came in those minutes: $197.76M of $308.58M.
If liquidations arrived evenly over time, 9% of the time would hold about 9% of them. These levels held 64%, about 7.2× the average rate.
The caveats travel with the number. Liquidations bunch up in big moves, which also reach more levels, so the time bar flatters the map. Binance’s reported feed is a lower bound, so dollar figures are a floor. And it measures whether liquidations happened while price was at the levels, not whether price will reach a level next. The harder test is still running: we wrote down the pass and fail rules for our liquidation rails before 13 October, and publish the result on or after 16 November, pass or fail. Read the registered test.
Each coin’s page carries its own daily record: the BTC map and its record, or every coin. Ask the same of any heatmap you use: does it say which layers are estimates, and does it publish how often its levels met real liquidations?
Mistakes to avoid
- Treating a bright band as a price target. It marks where forced orders would sit, not where price has to go.
- Reading every heatmap as the same data. Estimated levels, real positions and reported liquidations answer different questions. Know which one is on screen.
- Adding up reported liquidations as a complete total. Exchange feeds are a floor. The daily crypto liquidations chart separates Binance and Bybit, shows long and short totals with report counts, and provides a dated PNG to cite. Hyperliquid totals remain unavailable until a verified daily venue series exists.
- Ignoring distance and time. Levels far from price, or ones that have stood for days, describe a different market from the next hour’s.
- Forgetting that levels move. Positions open and close all the time. A level that has been passed, or whose positions were closed, is gone even if a stale chart still shows it.
- Learning from one vivid example. The BTC example above was selected after the move. Look for the times price went near a level and nothing happened, too.
Watch it happen in replay
A heatmap shows where the levels were. A replay shows what happened when price got there: the resting orders thinning above price, the forced orders hitting the tape, the levels ending one by one. Open a recorded move from the event archive, or open the public replay with no sign-up and turn on Reported liquidations in the Liquidations menu. In Real-time, each reported liquidation lands as a diamond at its price and second. How to start a replay.
Liquidation heatmap questions
Liquidation map or liquidation heatmap: what is the difference?
Both mark the prices where leveraged positions would be liquidated. “Heatmap” usually means the levels drawn over the price chart through time, so you can see how they built up and where price went. “Map” often means a single snapshot: dollars at each price, drawn as bars. EdgeDepth’s per-coin pages draw the levels over recent candles, with the levels price reached this week and their dollars.
Do liquidation levels act like magnets?
A level does not establish that price will move toward it. Price can cross a level, stall short of it or move away. The BTC example shows a crossing and separate reports, not that the level caused the move. The score measures how much liquidation actually happened at the levels; it does not say price is pulled toward them.
Why do different heatmaps show different levels?
Most are estimates, and every tool makes its own assumptions about leverage, where positions were opened and when they closed. Venues differ too: a Binance map and a Hyperliquid map of the same coin describe different traders. Hyperliquid levels come from real positions, so they differ from any estimate by design.
Liquidation heatmap or liquidity heatmap?
They are different charts that are easy to mix up. A liquidation heatmap marks the prices where leveraged positions would be forced closed. A liquidity heatmap, also called an order book heatmap, draws the limit orders resting in the book: the bids and offers traders have placed on purpose. One shows trigger prices for forced orders, the other shows orders that are already there.
Which timeframe should I use?
Match the window to the question. For the next few hours, the levels near price on a short window matter most. Longer windows show the bigger clusters but pile up older estimates, and some of those positions will have closed without the model knowing. Whatever the window, read distance from price before brightness.
Is there a free liquidation heatmap?
Yes. Every coin’s liquidation map page is free to open, with the levels price reached this week, their dollars, and the map’s daily record. In the EdgeDepth terminal, liquidation rails are free on BTC; other coins need Pro.
What happens when price reaches a level?
The positions there are forced closed, if they still exist. The forced orders are market orders, so they add to the move; whether the move continues depends on what else is in the book and on the tape. The model treats a touched level as closed, but its saved end time can lag the first price crossing. The exchange may or may not publish the liquidations it triggered.
What does the Hyperliquid heatmap show that a Binance one cannot?
Binance does not publish positions, so a Binance heatmap is always an estimate. Hyperliquid publishes each account’s positions and liquidation prices, so its levels are read from real positions, within the coverage shown in the chart’s key. See the Hyperliquid liquidation heatmap.
Looking for a calculation? Open the Reading library. For recorded exercises, browse guided lessons.