A liquidation heatmap is a 2D chart with time on the x-axis and price on the y-axis. Darker zones mark price levels where larger clusters of potential liquidation orders sit. Plainly: it shows how many positions would be wiped if price reached that level.
Third-party aggregators like CoinGlass combine open-interest data from Binance, OKX, Bybit, and others. They back-solve for each leverage tier's liquidation price to estimate how much USD notional would be liquidated at each price. These are estimates — only exchanges know exact positions and leverage.
A magnet level is one of the darkest price clusters on the heatmap. Markets tend to drift toward concentrations of liquidity — when price reaches a magnet, the triggered liquidations create market-buy or market-sell pressure that pushes price further through the cluster.
No. It tells you 'if price moves to X, this many liquidations fire', not 'price will move to X'. But it gives a useful read: which nearby levels are magnets that might attract price, and where momentum can accelerate after a break.
Three steps: (1) Identify the largest liquidation clusters within 5-10% of current price; (2) Decide whether your entry treats those clusters as targets or risk zones; (3) Never place your stop-loss inside a heavy cluster — it gets swept with the liquidations and slippage is brutal.
What a liquidation heatmap actually shows
A liquidation heatmap is a 2D chart with price on the y-axis and time on the x-axis. At each price level, the colour encodes the USD-equivalent notional that would be liquidated if price touched there. Darker colour means a bigger cluster of forced exits waiting.
Underneath, the heatmap is built from estimated open interest, leverage distributions and the maintenance margin formula. The provider does not have your account, but they can model what fraction of OI is at what implied liq price given typical leverage stats. The output is good enough to be useful, but it is an estimate, not exchange-confirmed truth.
The shorthand: dark band = lots of forced exits waiting at that price. Light band = sparse, easy passage. A market moving toward a dark band is moving toward a place where price action accelerates.
Where the data comes from, and what CoinGlass is really estimating
The standard free reference is CoinGlass's liquidation heatmap. They aggregate position data from Binance, OKX and Bybit, then model the per-leverage-tier liq prices using each exchange's MMR table. They produce two views: a price-time heatmap and a stacked OI / leverage breakdown.
Glassnode has a more academic version on the futures liquidation volume metric — see their derivatives liquidation volume dashboard. Free tier limits the granularity, but it is a good cross-check.
For the academic definition of a liquidation event, Wikipedia's liquidation entry drops it into the broader finance context.
What actually gives a magnet price its pull
"Magnet price" is jargon for a price level that attracts the spot rate. The mechanism is feedback: when price gets close to a big liquidation cluster, even small price moves trigger some of those liquidations. Those liquidations print as forced market orders. Market orders consume the order book on the trigger side, pushing price further into the cluster, triggering more liquidations.
That is why a dark cluster is rarely a "support" level. From the long side, the cluster is everyone else's stop-out. Once price gets close, the forced selling accelerates the descent through the cluster, not the bounce off it.
The phrase "they're going to hunt the liquidations" is a folk version of this. There is no centralised "they". The mechanic is automatic: the higher the OI sitting at a level, the bigger the gravitational pull of that level on the market.
A bright cluster only tells you how much potential liquidation is stacked at a level. It says nothing about whether price will get there, and nothing about what happens if it does. Clusters get skipped, price turns halfway, clusters get cleared and reverse on the spot — all of that is routine. Trading the map as a signal means reading "what happens if the condition holds" as "the condition will hold". The usable read is conditional: if price trades through X, expect acceleration; if price never reaches X, the map says nothing about your position today.
Whether the map deserves any weight in the current regime is something you can settle without anyone's hit rate — keep your own log. Every time a clear cluster shows up, write down the price band, the spot price at that moment, and the timestamp of the snapshot; come back 24-48 hours later and mark whether price entered the band, whether it cut straight through or bounced, and whether volume expanded. A dozen entries in, you have your own read on when this chart is worth anything, which beats any number someone else hands you. Two cross-checks worth logging next to it: which way funding is leaning, and whether the cluster is decaying (positions behind an old cluster may have been closed long ago while the map still lights it up).
Reading this week's BTC map, panel by panel
Practical workflow: open CoinGlass, set BTC, set the 24h or 7d window. Look for the two or three darkest horizontal bands on the chart. Note the price levels.
From here, ask three questions. (a) Where is BTC trading right now relative to those bands? If price is sandwiched between two heavy clusters, expect a range. If price is above all the clusters with a heavy cluster below, the market has a downside attractor. (b) How fresh are the clusters? A cluster that has been building over weeks is more reliable than a cluster that built up in the last 24 hours. (c) How does the cluster shape your stop placement?
Repeat for ETH. The two assets often share clusters at similar percentage offsets from spot, but not always — when they diverge, the divergence itself tells you something about positioning.
Replaying the 2024-08-05 BTC flash crash, level by level
In the days before 2024-08-05, BTC was trading in the low $60,000s — $62,500 on 07-31, drifting to $59,800 by 23:00 UTC on 08-04. CoinGlass's heatmap showed the heaviest long-liquidation cluster sitting between $49,000 and $52,000, with thinner bands stacked above it around $54,000 and $58,000. Note what is and is not usable here: historical dollar totals for a cluster cannot be pulled back up after the fact, so the durable read is where the clusters sat and which one was thickest — not a figure in millions.
Once the carry-trade unwind kicked off on 2024-08-04 evening Asia time, BTC started shedding levels: $57,200 by 02:00 UTC, then the $54,000 shelf gave way at 04:30 and price covered the ground to $51,000 in about five minutes. Each leg lower fed the next band. The low print at 06:00 UTC on 2024-08-05 was $48,830 — the bottom edge of the heaviest cluster — before the market bounced.
The heatmap did not predict the news. What it framed in advance was the conditional: given a triggering event, the path of least resistance ran down through $54,000 into the $49,000 band. That is the value of the chart — it tells you where the floor is likely to sit, given that the move starts at all.
Clusters have lifespans, and the time axis shows it
Clusters have lifespans. A long-term cluster built up over 30 days is durable — even if the market chops past it briefly, the OI re-anchors there. A short-term cluster built up in 24 hours is fragile — it can dissipate as quickly as it appeared, especially if traders react to seeing it.
Practical use: trade off the long-term clusters for swing positioning. Use the short-term clusters for intraday flow context, but do not anchor your view on them.
The 2024-08-05 $49k cluster was a 7-day build, which is in the middle. The cluster predated the crash trigger, but it was not so old that the market had already adjusted around it.
Placing stops so you are not standing inside the cluster
The cluster becomes your reference, not your trigger.
The single most actionable rule out of all of this: do not park your stop inside a heavy cluster. Two reasons.
First, the cluster triggers in lockstep. Your stop fires at the same time as everyone else's, and the order book is empty on the trigger side because everyone is selling at once. Slippage is unpredictably bad — easily 0.5-1% beyond your stop level.
Second, the cluster magnet effect means the market is more likely to touch the price than the heatmap-free intuition suggests. You will get stopped out and watch the market reverse.
Better rule: place your stop a few percent past the cluster, on the assumption that if price walks through the cluster, you have legitimate trend confirmation. The cluster becomes your reference, not your trigger.
Three ways people misread this chart
One: treating clusters as walls. They are not walls, they are speed bumps. A cluster can be smashed through if the macro flow is strong enough. The 2024-08-05 case happens to be where the cluster did hold, but on a strong trend day they get vaporised.
Two: confusing implied with realised. The heatmap shows implied liq prices. The actual liquidations happen at market and can print anywhere within a wide band around the implied price, especially in fast markets.
Three: trading the heatmap as a primary signal. The heatmap is a context layer, not a trigger. Trade your setup, use the heatmap to size and to place stops. If you let the heatmap drive entries, you will be trading other people's stress.
Pair this with the leverage and liq math article so you can map your own positions onto the heatmap. And the 2024-08-05 replay ties the heatmap into the full event timeline.