Liquidity Sweeps Explained

What is a liquidity sweep?

A liquidity sweep is price trading through a level where resting orders are clustered and then closing back on the side it came from. The level is taken without being accepted. The geometry — a wick through, a close back inside — is definitional and any two implementations will agree on it. The story usually attached to it, that the move was engineered to consume those orders, is interpretation, and no public market data can confirm or deny it.

The vocabulary is worth getting right once. Unswept highs sitting above the current price are buy-side liquidity (BSL), because the orders that fire when they are touched are buy orders. Unswept lows below are sell-side liquidity (SSL). The naming trips people constantly, since the labels refer to the direction of the orders resting there, not to the direction anyone expects price to go.

There are three outcomes when price reaches a pool, not two, and this is where most explanations are incomplete. Price can reject before reaching the level; it can traverse and close back — the sweep; or it can traverse and close through — acceptance, the genuine break. A tool that models only the middle case leaves the third completely unlabelled, which means every genuine breakout it sees looks like a sweep that has not resolved yet. Both endings need state.

Where does the liquidity actually sit?

It sits wherever a visible price attracts orders that convert into market orders when touched. Two populations dominate: protective stop orders belonging to positions on the far side of the level, and breakout orders waiting for the level to give way. Both become aggressive flow at a known price, which is the entire mechanical reason a level behaves like a magnet.

That gives an operational procedure rather than a guess. Take prior swing extremes that price has not yet taken, then merge extremes falling within a small price tolerance into a single pool that carries a count. Two highs at effectively the same price are a stronger pool than one high, and the count — not the shape of the wick — is what carries the information. Prior day and prior week extremes, session highs and lows, and round numbers behave the same way for the same reason: a widely watched reference price accumulates resting orders.

A pool also has a lifetime, and ignoring it is one of the most common chart errors. Once the pool is taken, the orders that rested there are gone; the level is spent. Redrawing a swept high as if it were still a magnet is drawing something that no longer exists. This is the same lifetime logic that applies to swing levels in market structure, where a pivot line should run only until price closes beyond it.

Note what none of this required: anyone's intent. The construction is mechanical and reproducible from a bar series and a tolerance.

Sweep or genuine breakout — what evidence separates them?

The traverse itself tells you almost nothing. What separates the two is what happens once price is on the far side: whether business is actually built there. There are three independent forms of acceptance evidence — volume, time and value — and their disagreement is information rather than an inconvenience.

The three acceptance dimensions do not become knowable at the same moment, and pretending otherwise is a subtle form of lying about your own timestamps. Volume acceptance can be answered by the first bar that builds a share of its volume beyond the level. Time acceptance re-seals on every close that extends the streak. Value migration may not be answerable for hours, until the profile instance that decides it closes. Publishing a single timestamp for all three misrepresents two of them. When they disagree — heavy volume, no time — the disagreement itself is the read: that combination describes a sweep, not acceptance.

The strong version — the stop run

The strong form of the sweep is a traverse on a bar whose volume and absolute delta both sit in an extreme percentile of the window immediately preceding it, followed by a close back through the level within a few bars. The volume-and-delta expansion is the differential signature: without it, the event is an ordinary sweep and deserves the ordinary label.

Two implementation details decide whether that test means anything.

The bar being classified must not be inside the distribution that classifies it. The comparison window has to be strictly trailing and exclusive. If the bar enters its own percentile baseline, a large bar inflates the very threshold it is being measured against, and the test weakens exactly when the event is most extreme.

Efficiency must be signed, not absolute. An aggressive burst is a run of consecutive same-side taker trades separated by no more than a short gap. Its efficiency is the net displacement projected onto the burst's own direction, divided by the volume spent. A buying burst that ends below where it started was rejected while it was buying: that is absorption, and it is deterministic rather than statistical — no percentile is needed to know it. Using absolute displacement instead of the signed projection paints that rejection as "a lot of movement" and can classify it as initiative, the exact opposite of what occurred.

The aftermath is a separate question from the event, and it is worth registering separately rather than mutating the original label: whether flow after the return goes flat and contrary, or whether fresh initiative keeps building volume beyond the extreme.

What here is definitional and what is a story?

Definitional: the geometry, the pool construction, the acceptance evidence, the efficiency arithmetic. Interpretation: intent, "smart money", and any claim about who was on either side of the trades. Both can appear in the same paragraph — they just should not be presented with the same confidence.

Consider the two descriptions of one event. First: a level with two equal highs was traversed on an absorbed buying burst, no volume was built above it, the close returned inside within one bar, and open interest fell across the move. Second: they ran the stops. The first is falsifiable in every clause; anyone with the same data can check it and tell you which clause is wrong. The second adds an actor nobody can observe and cannot be wrong about, which is what makes it comfortable and what makes it useless.

Two more honest caveats. Sweeps are not rare, and labelling them after the fact is trivial — any chart is full of them in hindsight, which is why hindsight examples prove nothing. And the whole read is conditional on the level having been visible before the traverse: a level identified after price reached it is not evidence, it is a drawing.

How does the Confluence Engine read sweeps live?

The Confluence Show is a live, AI-run market analysis broadcast. Its analyst, NAIRO, reads raw trades, order books and positioning 24 hours a day through the Confluence Engine — 40+ analytical layers computed in-house — of which liquidity pools, sweep efficiency, absorption and acceptance are four separate layers that are allowed to disagree.

The live constraint is handled by making knowability explicit. A pool is only published once the pivots that found it are themselves confirmed. A burst is only sealed once its gap window has elapsed without another same-side trade, because the end of a burst is not knowable at the instant of its last trade. A sweep is only sealed by the bar that closes back through the level. In practice that means NAIRO can tell you a pool is under test before it knows the outcome — and says so in those terms, then states in advance which resolution would prove its read wrong.

This is educational market analysis, not financial advice, and it issues no signals. The free 20-minute-delayed stream on YouTube, Twitch and Kick is at /watch, the covered markets at /markets, and the layer stack at /how-it-works.