Liquidity sweeps: what they are, and what the order flow shows

A liquidity sweep is a quick move through an obvious high or low that triggers the orders waiting beyond it, followed by a close back on the other side of the level. Traders also call it a liquidity grab or a stop hunt. In ICT terms it is a raid on buy-side liquidity (above a high) or sell-side liquidity (below a low).

On a candle chart a sweep is one wick. In the order flow it is a sequence you can watch: the push, the forced orders, the turn and the reclaim.

What gets swept

Below an obvious low sit orders that only exist as triggers: stop losses on long positions and, on perpetual futures, the liquidation prices of leveraged longs. While price stays above the low they do nothing. When price trades through it, they become market sell orders at once.

Above an obvious high it is the mirror image: stop losses on shorts and short liquidations become market buys.

That burst of forced orders is the liquidity. A large buyer can fill against it without chasing price higher, which is the usual story for why sweeps reverse. Whether they actually do is a separate question, answered further down.

The exact definition we use

“Obvious low” and “quick move” are too loose to count, so EdgeDepth’s record uses a fixed rule, measured on completed 15-minute and 1-hour candles:

1. The level. Take the 12 completed candles before the latest one: 3 hours of context on 15-minute candles, 12 hours on 1-hour candles. The level is their lowest low.
2. The sweep. The latest completed candle trades strictly below that level.
3. The reclaim. The same candle closes above the level.

At highs the rule is mirrored: the level is the highest high, the candle trades above it and closes back below. A candle that sweeps both sides counts as neither. A candle that only touches the level does not count. Only completed candles count, so a sweep never appears and then disappears while a candle is still forming, and a gap in the data means the candle is not evaluated rather than guessed.

In the research record this is candle_sweep_reclaim, on 15-minute and 1-hour candles. Many traders also call a sweep any move that reverses a few candles later; this definition is deliberately the strict, single-candle version, because that is the one a chart pattern can be tested on.

What the order flow shows during a sweep

The candle shows where price went. The order book and tape show who was trading while it happened.

Before the sweep. Price leans on the low. You can see bids resting near it and sellers hitting them on the tape. You cannot see the stops: stop orders are not in the public order book until they trigger. On perpetuals, the liquidation heatmap estimates where leveraged positions would be forced out, which is the closest thing to a map of that hidden side.

The push through. The triggered orders arrive together: a burst of aggressive selling on the tape, large trade bubbles on the chart and, on perpetuals, reported long liquidations. Treat reported liquidations as a floor: the exchange feed does not publish every one.

The turn. This is where a sweep and a breakdown separate. In a sweep, the selling is absorbed: sell volume stays high but price stops falling because large bids keep filling it, then buyers start lifting the offer. In a breakdown, bids pull back from the level, the book tilts toward the sellers and price keeps going.

The reclaim. Price closes back above the level. That close is all a candle chart records. A footprint of the same candle shows how much volume traded below the level before it came back.

To practise reading these pieces on a recording, open a replay with no sign-up and switch to Real-time. It is a recorded move rather than a sweep chosen for this page; a linked sweep example will be added here.

Sweep, breakout or failed breakout?

What price does at the level	What it is called
Trades through it, closes back inside on the same candle	Liquidity sweep (this definition)
Closes beyond it and holds there	Breakout or breakdown
Closes beyond it, then comes back on later candles	Failed breakout, a looser kind of sweep

Does a liquidity sweep reverse?

The story says the swept liquidity fuels a reversal. A chart can show you one that did. Counting all of them is the only way to see whether the story holds.

We counted every 1-hour sweep on every Binance USDⓈ-M perpetual in EdgeDepth’s record (up to 603 markets at a time, none of them TradFi-linked) from 15 July 2025 to 22 September 2026 (UTC): 331,288 sweeps of a low and 311,491 sweeps of a high. To see whether a sweep changes anything, each column is compared with how often any hour on the same markets moved that far.

24 hours later	After a sweep of a low	After a sweep of a high	Any hour
Up 2% or more	29.4%	26.5%	27.4%
Down 2% or more	32.0%	35.2%	33.1%
Up 5% or more	12.5%	12.5%	12.0%
Down 5% or more	13.5%	15.5%	14.0%

Shares are of the sweeps with a full 24 hours of data after the candle: 320,883 lows and 302,631 highs.

Sweeps did lean the textbook way, but only by about two points. After a sweep of a low, a 2% rise the next day was a little more common than after any hour, and a 2% fall a little less common; sweeps of highs mirrored it. Split into 6 periods, the lean held in 5 for lows and 5 for highs, and it flipped in the strongest trending stretch, September 2026, when swept highs kept rising.

The market’s own direction mattered more than the sweep. In falling periods, sweeps of lows were still followed by more 2% falls than 2% rises. A sweep nudged the odds; it did not set the direction.

These figures count how far price moved, not what trading the sweep would have paid. That needs a separate test.

Does trading it pay?

We tested that too. Over one month, 13 September to 13 October 2026, we took 100 markets drawn at random (with a fixed seed, before any result was seen) from the 520 recorded at the start, and traded every hourly sweep on them: long from the minute after a sweep of a low, short from the minute after a sweep of a high, with fees and slippage counted.

Every version lost money after costs. In both directions and with both exits, the share of winning trades stayed below what the average win and loss needed to break even. Shorts did worst. The month overlaps the September stretch in which swept highs kept going up.

This is one month on 100 markets. Funding is not included. Signals whose minute prices were missing before the exit were left out rather than counted (19% to 24% of each test), and while a market had a trade open, further sweeps on it were skipped. A fixed-rule test like this says the bare pattern did not pay. It does not rule out sweeps with better context doing better: whether liquidations actually fired, whether the book absorbed them, how crowded positioning was. That is what research lets you add to the question.

How it was counted

Condition: feature.candle_sweep_reclaim_1h = bullish (a sweep of a low) or feature.candle_sweep_reclaim_1h = bearish (a sweep of a high), feature_defs.v1. Outcome: the close 24 hours later. Comparison: every market-minute in the same window, sweeps included. The full range was too large for one scan, so it ran as 6 windows and the counts were added up.

Window (UTC)	Markets	Sweeps of lows (query)	Sweeps of highs (query)
15 July 2025 to 1 October 2025	552	57,589 (2b32b58b)	57,812 (6061e869)
1 October 2025 to 1 January 2026	599	74,436 (b86f46fd)	65,520 (cfb191cb)
1 January 2026 to 1 April 2026	603	74,416 (3dab3317)	63,407 (049a2881)
1 April 2026 to 1 July 2026	595	68,336 (71337688)	65,292 (287623e5)
1 July 2026 to 1 September 2026	529	43,815 (72282999)	42,704 (304a75a7)
1 September 2026 to 22 September 2026	523	12,696 (e039c5f8)	16,756 (47e7e780)

times.anchor_time between is inclusive at both ends, so occurrences at exactly 00:00 UTC on an inner boundary are counted in both adjacent windows: 78 sweeps of lows and 72 of highs (under 0.03%). 22 Sep to 14 Oct 2026 was not included: those scans exceeded the execution budget and the reference returned AUTH_REQUIRED on 15 Oct 2026.

The guided lesson walks through an earlier, shorter scan of the same pattern.

Liquidity sweep questions

Is a liquidity sweep the same as a stop hunt?

Mostly yes. “Stop hunt” describes the same move from the point of view of the traders whose stops were hit, and implies someone did it on purpose. The definition above only describes what price did. It does not assume intent.

What are buy-side and sell-side liquidity?

ICT terms for the resting orders on each side of the market. Buy-side liquidity sits above highs: short stops and short liquidations, which become buy orders when triggered. Sell-side liquidity sits below lows: long stops and long liquidations, which become sell orders.

Do liquidity sweeps work in crypto?

They happen constantly in crypto, especially on perpetuals, where leverage puts liquidation prices just beyond obvious highs and lows. Whether a sweep reverses is a different question: in the count above, 1-hour sweeps leaned toward a reversal by about two points over the next day, less than the market’s own direction moved things, and traded with fixed stops over a month every version lost money after costs. The pattern alone is not an advantage; the context around it is what to test.

What timeframe should I use?

The record measures 15-minute and 1-hour candles. The logic works on any timeframe, but a lower timeframe produces many more sweeps of levels almost nobody was watching.

How is a sweep different from a liquidation cascade?

In a liquidation cascade, forced selling triggers more forced selling and price keeps falling. A sweep is the case where the forced orders are absorbed and price closes back above the level within the same candle.