What a Sweep Does Not Prove

Learning objective. By the end of this lesson you will be able to state precisely where observation ends and inference begins in the liquidity model, and explain why that boundary is a position-sizing decision rather than a philosophical one.

Prerequisite. The three-part test from lesson 3. This lesson is its other half.

Start with what is actually on your screen

A price chart contains four numbers per candle and a volume figure. That is the whole dataset. Open, high, low, close — and on spot gold, a volume figure that counts price updates rather than contracts traded.

Everything else you have ever read about a chart is inference laid on top of those numbers.

Some of that inference is reasonable. Some is decoration. The purpose of this lesson is to help you tell which is which on your own chart, because the industry that teaches this material has almost no incentive to draw the line for you — confident explanations sell better than accurate ones.

The claim, and the honest version of it

The standard claim runs roughly: banks need liquidity to fill large orders, so they drive price into obvious stop clusters to trigger them, then reverse into the real move.

Here is what can be supported and what cannot.

Supportable. Large orders need counterparties. A participant wanting to buy size cannot do it at a level where nobody is selling, and areas holding many resting orders are, by construction, areas where counterparties exist. This is close to a description of how any order-driven market works.

Also supportable. Traders cluster stops in visible places. You do it. The person teaching you does it. Beyond an obvious high is the most natural place to put one.

Not supportable from a price chart. That a specific move was caused by a specific participant acting with a specific intention. That anyone “engineered” a particular sweep. That the move which followed was “the real move” and the one before it was fake.

The distinction is not academic. The first two justify watching obvious levels. The third justifies certainty — and certainty is what gets sized wrong.

Three things people believe a sweep proves

“It proves institutions were there.” No. It proves price traded beyond a level and came back. Who traded, in what size, and why, is not in the four numbers. There may well have been large participants involved. There may equally have been a thin book at an awkward hour and a modest order clearing it. Both produce identical candles, and the chart cannot separate them.

“It proves the reversal is coming.” No. It is consistent with a reversal. It is also consistent with a pause, with a retest that fails, and with the range continuing for another six hours. Lesson 3’s part 3 exists precisely because the shape alone does not carry this information — and part 3 is the part that most often fails.

“It proves the level was significant.” Only circularly. The level was significant enough that price reacted; that is the extent of it. Levels price ignores entirely do not get written up afterwards, which is why the ones you remember all look meaningful. This is survivorship bias operating on your own chart history, and it is why the exercise in lesson 3 asks you to count failures.

Why this is a position-sizing lesson

Here is the part that makes this worth reading rather than merely worth agreeing with.

Certainty and size are linked, whether or not you link them deliberately. A trader who believes a sweep proves institutional intent takes a larger position than one who believes it raises the odds somewhat. Same setup, same chart, different size — and the difference came from a belief about causation that the chart never supported.

The belief also changes behaviour after entry. If a sweep proved something, then price moving against you is a temporary error to be waited out, and the story supplies an infinite supply of reasons to hold. If a sweep merely shifted the odds, then price moving against you is ordinary, and your invalidation level still means what it meant when you wrote it.

Epistemic humility is not a mood. On a trading account it has a numerical expression, and that expression is lot size.

Worked example — the same candle, two readings

Schematic.

Price reaches above EQH1, closes back below, and moves down and away. Textbook, all three parts passed.

Reading A. Institutions swept the highs to fill short orders and are now driving price down. Confident, causal, and unverifiable. It supports a larger position, an early add, and — when price retraces — holding through the invalidation level, because the story says the move is real and the retrace is noise.

Reading B. Price reached beyond an obvious pool and did not sustain above it. Structure is intact, and this is consistent with continuation lower. It is invalidated by a close back above EQH1. Modest, falsifiable, and identical in every observable respect.

Both readings come from the same candles. The second one survives contact with a bad outcome. The first one has to be abandoned or rationalised, and traders generally rationalise.

Notice too that Reading B is not less actionable. It contains everything needed to act — direction, condition, invalidation. What it lacks is the story, and the story was never the part doing the work.

Exercise — no trade required

Take five sweeps from your own chart history, including at least two that did not work.

For each, write two paragraphs: the causal version, and the observational version. Then mark every word in the causal version that describes something you could not have seen on the chart — every because, every they, every intended.

Most people find the causal version is shorter once the unverifiable words are removed, and that what remains is the observational version. That is the finding. The story adds length and confidence; it adds no information.

Common failure mode

Treating this lesson as a reason not to trade. It is not. Every method operates on incomplete information, and demanding proof before acting is simply a slower way of never acting. The argument is not know nothing, do nothing — it is know what you know, size accordingly, and write invalidation before entry.

The opposite failure: using humble language while trading on certainty. Writing “may” and “supports” in the journal while sizing as though the outcome is known is worse than plain overconfidence, because the record now looks disciplined and cannot correct you. The test is not the words. It is whether your invalidation level ever gets moved.

Invalidation condition. For the reading, not the trade: the observational read is invalidated when price closes back beyond the swept level and holds. That is a defined condition with a defined response. The causal read has no invalidation condition at all — which is the clearest evidence that it is not doing analytical work.

Observation checklist

1. My written read contains no claim about who traded or why.
2. Every clause could be checked against the chart by someone who disagrees with me.
3. My position size reflects uncertainty, not conviction about a story.
4. My invalidation level was written before entry and has not moved.
5. I have recorded the setups that produced nothing, not only the ones that resolved.

Knowledge check

1. What does a completed three-part sweep actually establish?

That price traded beyond a marked level, did not close beyond it on your chosen timeframe, and subsequently moved away. Nothing about participants, intent, or what happens next. It is a description of a shape in four numbers, and its usefulness comes from the shape recurring — not from any account of who produced it.

2. Two traders take the same setup. One reads institutional intent, the other reads a probabilistic condition. What differs?

Position size, and behaviour at the invalidation level. The first has a story that explains away adverse movement and therefore a reason to hold past the level. The second has a condition that either holds or does not. The entry may be identical; the outcomes diverge on the trades that go wrong, which is where account survival is decided.

3. Why does this lesson claim survivorship bias applies to your own chart history?

Because levels that price ignored are not memorable and do not get reviewed. Recall over-samples the levels that reacted, which makes reaction feel like the default. The correction is to count the levels that did nothing at the moment you mark them, before knowing the outcome — which is what lesson 3’s exercise builds.

Risk note

Educational only, not financial advice. This lesson makes no claim that any pattern produces any outcome, and explicitly argues against treating one as if it does.

Gold is traded with leverage; a position can lose more than the amount you intended to risk. Contract size and tick value are broker-specific — verify yours in your MT5 symbol specification.

A setup remains research until the MT5 Expert Advisor confirms execution. Manual trades stay outside the public record. The forward-test evidence published for this project is a record of what the system did, not a claim about what it will do.