What is a liquidity sweep?
A liquidity sweep is a price move that pushes beyond a level where resting orders are likely to sit — the high of a range, a prior swing low, an obvious round number — triggers those orders, and then reverses back through the level it broke. The sweep is the round trip, not the poke.
The mechanism behind the idea is uncontroversial. Stop-loss orders cluster just beyond obvious levels because that is where a large number of traders place them, and a cluster of stops is a pool of guaranteed market orders. Any participant who needs size filled has a reason to prefer trading where that fuel exists.
Where the idea gets slippery is in the definition. If a sweep is only a sweep once price comes back, then the label can only be applied after the reversal has already happened — which is precisely when it stops being tradeable. Most educational material glosses over this, and most charts you have seen marked with sweeps were marked in hindsight.
Why do most charts mark sweeps that were never sweeps?
Most marked sweeps are ordinary wicks selected after the fact, because the eye can only find the ones that worked. Scroll back through any chart and you will see clean sweeps everywhere; the ones that pierced a level and simply carried on are invisible, because nobody circles those.
This is selection at its most seductive, since the selection happens automatically and feels like observation. A wick through a level that reversed looks like a sweep. A wick through a level that continued looks like a breakout. Both are the same event at the moment it forms, and they are only separable afterwards.
The consequence is that the pattern's apparent reliability is close to 100% in any hand-picked sample and close to meaningless out of sample. If you want to know what sweeps are actually worth, the only honest procedure is to define the trigger mechanically, apply it to every occurrence in a long history, and count what followed — including the ones that would have embarrassed the thesis.
Can you tell a real sweep from a fake one in advance?
In our own testing, the honest answer is that advance separation is much harder than the material suggests, and we could not find features that reliably told the two apart at the moment of the poke. We tried the usual candidates — how fresh the level was, how many times it had been touched, the size of the pool, the speed of the approach, the shape of the piercing candle.
What we found instead was that the confirmation rate stayed stubbornly near chance regardless of which features we added. That is a negative result about forecasting, and it is worth stating plainly rather than burying: the moment of the poke does not appear to carry much information about which way the next few hours go.
What changed the picture was reframing the question. Instead of asking which pokes will reverse, we measure what each poke actually did over a fixed window afterwards and grade it as confirmed, failed or expired. That measurement is trivially accurate, because it is a record of what happened rather than a claim about what would. It also makes the platform's sweep labels honest: a level is marked as a confirmed sweep only once the reversal is in the data, and marked as failed when it is not.
The practical implication for a trader is uncomfortable but useful. If a sweep only becomes identifiable after the reversal, then a sweep strategy is really a strategy about what to do after a confirmed reversal — entries, stops and targets measured from the reversal, not from the poke. That is a testable strategy. 'Enter when you see a sweep' is not.
Where do stops actually cluster?
Stops cluster where a large number of traders would independently choose the same invalidation point, which in practice means just beyond structurally obvious prices rather than at arbitrary levels.
The most reliable clusters sit beyond swing highs and swing lows that are visible on a higher timeframe, because those are the levels almost everyone can see. A swing point that required a custom indicator to find is not a pool; a swing point that is the most obvious feature on the daily chart probably is.
Session extremes matter for the same reason. The high and low of the Asian range, the previous day's high and low, and the weekly open are all reference points that a wide range of participants use, which is what makes the area just beyond them crowded.
Round numbers deserve a mention with a caveat. Whole figures attract resting orders, but they also attract limit orders and option barriers, which push in the opposite direction. The cleanest pools are the ones where a structural level and a round number coincide — and the messiest analysis is the kind that treats every round number as a pool because it is easy to find.
How should you trade around liquidity sweeps?
Trade the confirmed reversal rather than the poke, and size the trade from the structure that confirmed it, because the poke itself is the part that carries no reliable information.
Concretely, that means waiting for price to close back inside the level it broke, using the extreme of the sweep as the invalidation point, and measuring the target from the entry rather than from the level. Measuring reward from the level instead of from the entry is one of the most common ways a sweep backtest flatters itself, because the level is always a better price than the one you actually got.
Be honest about the cost of waiting. Confirmation removes the trades that never reverse, which is the point, but it also gives up the best part of the move on the ones that do. Whether that trade-off is worth making is an empirical question about your instrument and timeframe, and it is answerable — run both versions over the same history and compare, rather than assuming the earlier entry is better because it looks better on the chart you already know the outcome of.
Finally, count your variants. Sweep strategies have a lot of knobs: which levels qualify, how far past the level counts as a poke, how long the reversal has the right to take, what closing back inside means on which timeframe. Every combination you try is another trial, and the best of forty combinations needs a much higher bar than the first one.