Cascades and the liquidation map

Section 12 gave us the key point: a liquidation is not “the exchange took the money” somewhere in its books, it is a market trade. When a position runs out of margin, the exchange closes it itself, with a market order. Liquidating a long is a forced sell; liquidating a short is a forced buy. Such an order has no limit price and no patience: it takes whatever rests in the book, level by level, until the whole size is filled.

That is where the chain comes from. The price falls → the most leveraged longs run out of margin → their forced sells eat the bids → the price drops further → the next longs, with slightly less leverage or a slightly lower entry, reach their liquidation price. Each step sets up the next one. This is a liquidation cascade. The mirror case is an upward cascade of shorts: forced buys lift the price into the next shorts' liquidation prices (often called a short squeeze).

It matters how a cascade differs from an ordinary decline. In an ordinary decline every seller decides: sell now or wait, place a limit order or hit the market. In a cascade the decision was made in advance — when the trader chose the leverage. The exchange carries it out mechanically and at the worst moment: when the price is already moving against the position and the book below is thinning.

What a cascade looks like in EVA. The “Order book · Flow” panel has a liquidations feed: time, coin, side, price and amount. A single line is noise. A cascade is a run of lines of one side in a row, often at close prices, within seconds or minutes. The rest is usually visible too: a candle with a long body, a burst of one-sided aggression in CVD (section 7), open interest that is not rising but falling (section 10): a liquidation closes a position, so fewer contracts remain in the market. “Price sharply down + a run of long liquidations + OI falling” is the typical portrait of a downward cascade. The mirror: “price up + a run of short liquidations + OI falling” is an upward cascade.

Compare another picture: the price falls, OI rises, almost no liquidations. That is not a cascade but new positions — someone is opening shorts. Liquidations and OI together answer the question “who is moving the price — new participants or forced ones”.

A cascade cannot go on forever, and it usually stops for one of several reasons.

• The fuel ran out. Each liquidation removes a position from the market. Once all positions with a nearby liquidation price are gone, there is nothing left for the next step. Leverage has been “burnt out” — visible as a sharp drop in OI.
• It hit thick liquidity. Large limit orders (liquidity walls, section 4) can absorb forced sells without moving the price further. This is the absorption from section 8: aggression keeps coming, the price stands.
• Opposite buyers arrived. A sharp drop makes the price interesting to those who were waiting lower. Their market buys extinguish the remaining pressure.
• The side changed. After a long cascade, shorts opened on the way down often remain. If the price bounces, they are the ones at risk — and sometimes a cascade starts in the opposite direction.

None of these causes is visible in advance to the exact price. A wall can be pulled (section 4), opposite buyers may not come, and exchanges do not publish how many positions sit lower. So the honest question is not “where will the cascade end” but “are there signs the fuel is running out”: the liquidations feed thins, OI stops falling, hits on the bids no longer move the price.

The example's figure shows the first case of each outcome for BTC, not the prettiest ones. On 02.09.2026 at 08:50 UTC, $1.47M of longs were wiped out in five minutes (20 liquidations); the minute price went from 76,823.1 to 76,770.0, and an hour after the burst it was 76,640.4 — further down. The same day at 10:50 UTC, $5.75M of longs were wiped out (155 liquidations) — almost four times more — yet by the end of that five-minute bar the minute price was already 76,477.6 against 76,426.6 at its start, and in the next bar $1.07M of shorts were wiped out. An hour later — 76,665.0, higher. A bigger burst did not mean a longer fall.

The “Liquidations” chart layer puts five different things on one field, and the first rule is not to mix them up:

• fact — history bands: where positions have already been wiped out and for how much over the day; hovering names the amount, side and the exchange the number comes from;
• the estimated map — a model: where, by our estimate, other people's liquidation prices might be;
• levels of large Hyperliquid addresses — real liquidation prices of specific positions: Hyperliquid discloses positions publicly;
• option strikes with max pain and unfilled CME gaps — separate layers with their own tooltips.

The estimated map answers: “if positions were opened here, with usual leverage, where would their liquidation prices be?” The logic is simple. Exchanges do not say who holds what leverage, but one can see where trading happened and where open interest grew — that is, where positions were probably opened. From such an entry price the liquidation price sits at a distance that depends on leverage (section 12). The model spreads the presumed positions across several typical leverages and draws where their liquidations would cluster.

What the map shows. Where liquidation prices would cluster under typical leverage. It is useful as a question: “if the price gets here, could a forced wave happen here?” Next to the fact, the map lets you compare where things were already wiped out and where there may still be something to wipe out.

What the map does not show.

• Real orders. There is not a single order from the book on the map — orders are shown by the book and the heatmap (section 4). A map band does not “rest” anywhere; it cannot be pulled or filled.
• Anyone's real positions — except large Hyperliquid addresses, whose positions are public. Their levels are the hardest numbers on the layer: not an estimate but an actual figure from the exchange.
• Price direction. The popular idea that “price moves to liquidation clusters like a magnet” is a hypothesis. A check on the 04–24.09.2026 archive (against random minutes of the same coin and hour, on two halves of the sample) showed the price did not go more often to where the map had more untouched liquidations. The map did not behave like a magnet.
• Completeness when empty. An empty map does not mean “there are and will be no liquidations”: the day's wiped-out turnover always stands next to it, to tell silence from lost data.

Any liquidation total is the total of what exchanges reported, and they report differently. On the “Money” tab the “Liquidations” block states under the total how many coins and which exchanges it covers. The differences between exchanges are:

• Bybit, Gate and Kraken Futures send every liquidation;
• OKX — the exchange's liquidation history; it is counted while the sample is fresh, and if OKX is silent, the caption says so;
• Binance and Bitget — at most one liquidation per second per coin (Bitget: the largest for longs and for shorts). In a second of a cascade, when dozens of positions close, we see one line from them;
• Hyperliquid is not in the total: its public data does not mark liquidations. Its large addresses' positions are visible instead — a different source, shown separately.

So EVA's total and another service's total can differ several-fold, and neither has to be wrong. The liquidation map for OKX and for Bybit differs roughly two-fold in turnover — the venue is always named next to it. The “incomplete” mark on a coin means less data is in memory than the window — not everything is counted there.