A harmonic pattern is a ratio rule applied to five chosen pivots

Five points, four legs, and a table

Every harmonic in common use is labelled X, A, B, C and D. The legs are XA, AB, BC and CD. The pattern is defined by where B sits as a proportion of XA, where C sits as a proportion of AB, where D sits as a proportion of XA, and sometimes by how far CD extends relative to BC. Nothing else is involved. There is no volume term, no trend filter and no requirement about time.

That means a harmonic is a geometric constraint on a sequence of swings, and the constraint is either satisfied or it is not. The important consequence is that a harmonic cannot be a reason for anything by itself. A proportion is a description of what has already been measured. Whether the market cares about the proportion is a separate question, and the ratios themselves do not answer it. Treat the named patterns as a shorthand for a specific combination of retracement and extension, which is genuinely useful for communication, and keep the question of significance apart from the question of definition.

The named patterns and the numbers that define them

The named shapes differ only in their permitted bands, which is worth saying before the list because the list can look like five different ideas. It is one idea with five sets of numbers. Each definition fixes where B may sit relative to XA, where C may sit relative to AB, and where D completes.

• Gartley: B retraces about 0.618 of XA, C sits between roughly 0.382 and 0.886 of AB, and D completes near 0.786 of XA, inside the XA leg.
• Bat: B is shallower, roughly 0.382 to 0.5 of XA, and D completes nearer 0.886 of XA, with a deeper BC projection.
• Butterfly: B retraces about 0.786 of XA and D extends beyond X, commonly around 1.27 of XA.
• Crab: the extreme version, with D projected much further beyond X and a large BC projection.
• AB=CD: the simplest form, two legs of similar length, often with similar duration.

Notice that the families are separated by where D falls relative to X. Patterns completing inside the XA leg are retracement patterns and patterns completing beyond X are extension patterns, which is a meaningful distinction because the two put you into a move at very different points and with very different distances behind you. The retracement levels involved are the same ones used everywhere else in charting, so there is no separate mathematics here. Harmonics simply require several of those levels to agree before granting the shape a name, and the name is a label rather than a reason.

The completion zone is a cluster, not a line

In practice D is never a single price. The pattern gives you several projections that should land near each other: a retracement or extension of XA, a projection of BC, and the point where CD equals AB. Those three rarely coincide exactly, so what you actually get is a band of prices, usually called the potential reversal zone. The tighter the three projections cluster, the narrower the band.

This is where a lot of confusion starts. Price reaching the zone is the pattern completing, which is a statement about measurement. It is not a statement that anything will happen. Every harmonic method that has survived contact with real trading adds a requirement after the zone is reached, whether that is a candle closing in a particular way, a shift in structure on a lower timeframe, or a hold of a nearby level. The zone narrows down where to look. It does not supply the reason to act. Treating zone entry as the trigger is the single most expensive habit in this area.

Tolerance decides how many patterns exist

Every pattern definition needs a tolerance, because exact ratios almost never print. If a Gartley requires B near 0.618 of XA, somebody has to decide how near. Widen that allowance and the number of qualifying patterns on any chart rises sharply. Narrow it and most candidates are rejected. The count of harmonics visible on a gold chart is therefore a function of a setting, not a property of gold.

Two consequences follow. First, two traders scanning the same chart will honestly disagree about whether a pattern is present, and neither is wrong, because they are applying different definitions. Second, any claim about how often these patterns appear is meaningless without the tolerance attached, which is why you should treat unqualified frequency claims with suspicion. The useful discipline is to fix your tolerance in advance and write it down, so that you are not quietly loosening it when you want a pattern to exist. A definition you adjust after looking at the chart is not a definition.

The hindsight problem: choosing X after the fact

Pivot selection is the deeper issue and it gets far less attention than the ratios. Before you can measure anything you must decide which swing high is X and which low is A, and that decision depends on how significant a swing has to be to count. Change that threshold and you get a different X, a different XA length, and therefore different projections and a different pattern.

On a completed move this is easy, because you can see which choice produces a clean fit. That is exactly the trap. The pattern that looks inevitable on a historical chart was selected, consciously or not, from several possible pivot sets, and the one that worked is the one you kept. In real time you do not have that luxury, because the swing that will turn out to be the relevant extreme has not finished forming. Anyone reviewing harmonics on old charts should assume a portion of the apparent reliability comes from this effect, in the same way it does with wave counting and with any other method that requires you to nominate the important points yourself.

How gold volatility treats a projected zone

Two features of this market deserve specific mention. The first is overshoot. Gold routinely trades through a projected level and comes back, so a stop placed just beyond D is placed inside the range of ordinary noise. If the zone is tight and the market is active, the arithmetic of the trade can be unworkable even when the read is reasonable. Sizing against current volatility rather than against the zone width is the honest adjustment.

The second is scheduled repricing. A completion zone reached during a data release is reached for reasons that have nothing to do with the measurement, and the reaction that follows is driven by the release. There is no sensible way to attribute the outcome to the pattern in that situation. The practical response is to know when the zone is likely to be reached and to treat releases as a reason to stand aside rather than a coincidence. If you want to work with these shapes, mark the zone in advance on the live chart and decide what reaction you require, because deciding afterwards is how a measurement turns into a justification.

Making the idea falsifiable

A harmonic can be made testable, and doing so strips away most of the mystique without losing the structure. Fix the tolerance. Fix the swing threshold that defines a pivot. State the completion zone as a price range before price arrives. State the reaction you require inside the zone and the price that invalidates the idea. Record the outcome whether or not it worked.

What you are left with is a location rule plus a confirmation rule, which is how most location based methods end up looking once the branding is removed. That is not a criticism of harmonics. A method that reliably points you to areas where several measurements agree has done something useful, because those areas are where other traders are also watching. The honest caveat is that the agreement of three projections derived from the same four legs is less independent than it appears, in the same way that several indicators built from the same prices are less independent than their number suggests. The zone is one observation, carefully measured, and it should be weighted like one.