A harmonic pattern is a five point price shape, labelled X, A, B, C and D, where every leg has to measure a set Fibonacci fraction of another leg before the pattern counts, and a reversal is expected at D. The rules for it are not settled. The published Gartley rule sets do not match each other, three of the pages carrying them do not match themselves, and the same 0.786 ratio is the exact location of D on some of them, a ceiling on others and a minimum on one more.

A rule you cannot write down the same way twice cannot be tested, because a chart that passes under one page's rules fails under another's. So the job is to pick one rule set, write it out with a tolerance, and count what it does on history you have not studied.

What is a harmonic pattern?

Five labelled points, four legs, and a ratio test on every leg. X is where the pattern starts, A is the next swing, then B, C and D each retrace or extend a leg that came before, and D is the point where the trade is taken. NAGA puts the condition plainly: "the validity of the pattern depends on specific retracement and extension levels".

The ratios come from Fibonacci retracement and extension levels. What makes harmonic patterns different from a plain retracement is that the ratios are folded into the shape: each leg is measured against another leg of the same pattern rather than against a single swing.

The attribution varies too. Tradeciety credits "H.M. Gartley, a financial analyst and author who published the first developments of this unique approach in his 1932 book, Profits in the Stock Market". TradingSim gives the same book and year. LiteFinance names the book without the year and adds two more people: "Harold Gartley laid out the idea of the Gartley five-point pattern in his book Profits in the Stock Market, and Larry Pesavento established the basic rules for trading using Fibonacci levels. Scott Carney expanded on the theory of harmonic patterns by adding the Crab, the Shark, and the Bat". IG names Gartley with no book, credits the Bat to "Scott Carney in 2001" and the butterfly to "Bryce Gilmore".

What are the rules for a Gartley pattern?

There is no single answer. Each page publishes its own set:

Source	AB	BC	CD	What 0.786 of XA is
IG	"must be 0.618 of XA"	not stated	not stated	the exact place D must land
NAGA, in its own prose	"B is a 0.618 retracement of wave A"	"0.382 to 0.886 retracement of AB"	"an extension of 1.13 to 1.618 of AB", then "Many traders look for CD to extend 1.27 to 1.618 of AB"	"Point D is a 0.786 retracement of XA"
NAGA's own rule list, identical to Tradeciety's	"must be at or near 61.8%"	"38.2% to 88.6% of AB but must NOT exceed 88.6%"	"can extend 113% to 161.8% (max) the length of AB but must not exceed 78.6% of XA"	a ceiling CD must not exceed
TradingSim, in its body	"retraces 61.8% of the initial XA leg"	"The BC leg by 38.2% of XA"	"at the minimum a 78.6% retracement of XA"	a minimum CD must reach
TradingSim's own cheat sheet, same page	"0.618 of XA"	not in the table	not in the table	the D point entry ratio
LiteFinance, in its own prose	"AB goes down and is 61.8% of XA"	"38.2% - 88.6% of AB"	"127.2% - 161.8% of BC"	the size of the AD formation, not of D
LiteFinance's own table, same page	B of XA, 0.618	C of BA, 0.382-0.886	D of AB, 1.272-1.618	D of XA, 0.786

The reference leg changes from row to row, and that is the part that breaks a rule. BC is measured against AB on most of these rule sets and against XA on one. CD is measured against AB on three of them, against BC on another and against XA on one more, and two leave CD out altogether. 38.2% of AB and 38.2% of XA are different prices on the same chart, so a rule quoted without its reference leg is not yet a rule.

Three of the pages contradict themselves. NAGA's prose puts D at a 0.786 retracement of XA, and the rule list further down the same page says CD "must not exceed 78.6% of XA", which is a ceiling rather than a location. TradingSim's body makes 0.786 a minimum and its own cheat sheet lower down makes it the D point entry ratio. LiteFinance's prose measures CD against BC and its own table measures D against AB.

Two of the pages publish the same rules. NAGA's "Description and Rules" lists for the Gartley, Butterfly, Bat and Crab are word for word Tradeciety's, as are its sections on folded Fibonacci geometry and the five price points. A reader who opens both is reading one rule set rather than two independent ones.

Tradeciety names the problem. On the traders who insist on a strict reading of the Fibonacci measurements, it writes: "many of them seem to be in disagreement with one another, as if they were all following different rules."

Is 0.786 a target, a maximum or a minimum?

All three, depending on which page you opened. That one number does four different jobs.

The exact place D must land. IG: "The retracement of point D must be 0.786 of the XA movement". NAGA's prose: "Point D is a 0.786 retracement of XA." TradingSim's cheat sheet gives the same thing as the D point entry ratio.
A ceiling. Tradeciety, and NAGA's own rule list: CD "can extend 113% to 161.8% (max) the length of AB but must not exceed 78.6% of XA".
A minimum. TradingSim's body: "The CD leg is at the minimum a 78.6% retracement of XA."
The size of a different leg. LiteFinance: "The AD formation is directed down and is 78.6% of XA", with D itself set by a 161.8% projection of AD.

Choose one before the chart, write down which one it is, and keep it fixed for the whole test. Change it halfway and the count you end up with measures nothing.

What are the ABCD, Bat, Butterfly, Crab and Cypher rules?

ABCD pattern rules

The simplest of them, four points and three legs. IG makes the ratios precise: "Using the Fibonacci retracement tool on the AB leg, the BC leg should reach precisely 0.618. The CD line will be the same length as the AB line, and the time it takes for the price to go from A to B should be equal to the time it takes to go from C to D." TradingSim is looser on the same pattern: "Because of the Fibonacci levels, the line segment from A to B is roughly the same as the segment from C to D."

Bat pattern rules

B is the disagreement. IG: "If the retracement up to point B stops at 50% of the initial XA movement, then you are probably looking at a BAT pattern." Tradeciety, NAGA's list and TradingSim all give a range instead, 38.2% to 50% of XA. TradingSim then gives CD two rules at once, "can retrace anywhere up to 88.6% of the XA segment" and "can be an extension of the AB segment of between 161.8% to 261.8%", and both NAGA and LiteFinance put D itself at a 0.886 retracement of XA.

Butterfly pattern rules

The pages that give its B point agree on the number: AB is a 78.6% retracement of XA, and IG puts the same ratio first: "The most important ratio to define is the 0.786 retracement of the XA leg." They part on CD. Tradeciety and NAGA: "CD can extend 161.8% to 224% the length of AB." LiteFinance's table puts D at 1.272 of XA, and Trading Strategy Guides gives CD two bounds, 1.618 to 2.618 of AB and 1.272 to 1.618 of XA.

Crab pattern rules

AB between 38.2% and 61.8% of XA. The CD extension is where they split: Tradeciety and NAGA say 261.8% to 361.8% of AB, IG gives "an extreme projection of BC (2.618 - 3.14 - 3.618)", and LiteFinance's table says 2.24 to 3.618. All of them put D at a 1.618 extension of XA.

Cypher pattern rules

The Cypher's ratios appear on one source only, and only in a table: TradingSim's cheat sheet gives the B point as 0.382 to 0.618 of XA and the D point entry as 0.786 of XC. Note the XC, because every other D ratio here is measured against XA. Trading Strategy Guides names the Cypher but gives no numbers, saying only that it "uses tighter Fibonacci ratios (usually less than 1)".

How many harmonic patterns are there?

There is no settled count. Four published answers:

Four. Tradeciety: "There are four primary harmonic patterns", the Gartley plus the Butterfly, Bat and Crab as "variations of the Gartley". NAGA says the same, "there are four that seem most popular".
Six. TradingSim: "The most widely traded harmonics are the Gartley, Butterfly, Bat, Crab, Shark, and Cypher." Trading Strategy Guides also counts six, with the same names.
Seven. IG's "Top 7 harmonic patterns", which adds the deep crab and the shark and leaves out the Cypher.
Four or seven on one page. LiteFinance's prose names "Gartley, Shark, Crab, Butterfly", its own ratio table has seven rows, adding the Alternative Bat and the Deep Crab, and the MT4 scanner it describes lower down detects those same seven.

Count the rows and the patterns separately before you quote a number from any of them.

Do harmonic patterns work?

Nobody has published a figure you can check. The accuracy claims on these pages name no market and no period, the one that does report a sample size names neither the tests nor who ran them, and on one page they do not agree with one another, so none of them is reproduced here. Trading Strategy Guides is the clearest case: "after extensive backtesting, we discovered the Amazing Harmonic Pattern Trading Strategy works best on higher time frames", with no market, no period, no sample and no result.

LiteFinance's advantages and disadvantages table carries both halves of a contradiction. On the advantages side: "Harmonic patterns appear in the chart often enough that traders can use them in their trading systems regularly." On the disadvantages side, a few lines away: "A harmonic pattern that meets strict aspect ratio criteria is very rare in real trading charts. Because of this, subjectivity is growing in the assessment and there is a probability of a false forecast." The same table states "There is no sufficiently accurate automatic tool for identifying harmonic patterns", and the section immediately after it reviews three automatic tools as "the most effective harmonic pattern indicators".

TradingSim is blunter and more useful: "Harmonic trading is one of the hardest technical frameworks to master because the ratios are unforgiving and valid setups are rare", and "Don't force patterns. If the ratios aren't clean, there is no pattern."

A method with unsettled rules, rare clean examples and a published tolerance for how close is close enough is a method whose results depend heavily on the person drawing it. That is a reason to test it yourself rather than a reason to dismiss it.

Can you backtest a harmonic pattern?

Yes, by hand, and the sources themselves explain why it has to be by hand. TradingSim's FAQ asks whether you need special software to find harmonic patterns, and answers: "No. You can identify them manually with a Fibonacci retracement/extension tool. Pattern scanners exist but often flag loose fits; visual confirmation is still required." LiteFinance's own disadvantage list says no sufficiently accurate automatic tool exists. NAGA points the other way and suggests automated recognition software, and Trading Strategy Guides says a harmonic indicator ships with most platforms, so the sources disagree here too. What none of them disputes is that a harmonic entry is a price you drew, after you decided a shape qualified.

That decision is the part a formula cannot take from you. Two things have to be fixed in writing before the first chart.

Which rule set. Pick one row from the table above and copy it out in full. The same chart passes under one and fails under another.
How close counts. A rule that reads "at or near 61.8%" is not testable until you decide what near means. Two sources publish a tolerance you can borrow: TradingSim says "Verify every ratio is inside tolerance (typically ±3 percentage points). If any leg is out, skip the setup", and LiteFinance advises "that deviation from the levels above should not exceed 0.05". Choose one, write it down, and apply it to patterns you want to take as strictly as to the ones you do not.

Then step a chart forward bar by bar so the completing leg is not visible before you commit. Reading a finished pattern on a static chart is not a test of anything, because the shape that qualified it is already on screen. Our guides on backtesting a trading strategy and manual versus automated backtesting cover the general method.

How to test a harmonic pattern honestly, step by step

One pattern, one rule set, one market. Gartley on EURUSD, say, with the rule set copied out and the tolerance written as a number. Running four patterns at once turns a test into browsing.
Decide the D point confirmation rule now. Reaching the zone and reversing in it are two different events. NAGA quotes Scott Carney on exactly that point: when prices reach the reversal zone, "it is signaling an opportunity for a potential trade, not a signal to trade yet". Whatever you require, a closed candle or a break of a line inside the pattern, fix it before you need it.
Replay bar by bar. The X, A, B and C points are set and the CD leg is still forming. That is the only state in which the decision you are testing happens.
Mark the five points yourself and measure the legs as you go. Our Wave Tools hold drawing tools for exactly this shape: XABCD Pattern, ABCD Pattern and Any Wave Pattern. You place the points, because nothing on our charts finds a harmonic pattern for you. Right-click a drawing for Apply defaults to reset it to your saved look, and switch on Remember last used so the next pattern starts from the style you used last.
Record the rejects. A shape thrown out for failing a ratio is part of the result.
Set the stop and the target from the pattern's own points, before entry. The sources place stops beyond D and targets at earlier points of the pattern. Whichever you use, it is part of the rule set rather than a decision to make once the trade is running.
Then change exactly one thing and run the same window again. A different tolerance, or a different rule set from the table. That comparison is the only way to find out how much of your result came from the rule and how much from the reading.

Two practical notes on data, because a five point pattern is sensitive to the quality of the extremes it is measured from. Our data comes from Dukascopy, tick level where it is available, cleaned to remove anomalies before building 1-minute candles. That matters here because a single bad tick becomes a swing high, and a swing high becomes an X or an A. Our article on backtesting accuracy puts it plainly about price spikes: "They're not real prices that anyone could have traded at, but your backtest will happily trigger entries and exits on them." Depth is per symbol, up to around 22 years on the oldest forex pairs, with other instruments starting later.

Every trade you take in ChartLabs saves a snapshot of the drawings, indicators and chart state you had at the moment of entry. For this method that is the record that matters: months later you can open the trade and see the five points and the ratio grid you accepted, rather than trusting your memory of whether that B really was at 0.618.

What to record for each pattern

Date and time, symbol, timeframe.
Which rule set, by source, and the tolerance allowed.
The four leg ratios as you measured them, not as the rule wants them.
Whether every ratio passed, or whether one was forced.
The confirmation you required at D, and whether it appeared.
Entry, stop, exit, and the result in R.
A running count of patterns rejected against patterns taken.

That last line is the one most logs leave out, and it is the one that tells you whether the pattern is rare on your market. If you reject thirty shapes for every one you take, the rules are doing real work. If you take almost everything you draw, the tolerance is too wide to mean anything.

Where this fits

A setup whose published rules do not agree makes the usual advice, "backtest it", both harder and more necessary, because you first have to decide which version of the rule you are testing. That is the work ChartLabs is built for: replay a chart bar by bar, mark the pattern with the wave and Fibonacci tools, take the trade by hand, and review it later with the markup you had at the time. You can also write your own indicators in Pine Script in a built-in editor if you want a measurement helper on the chart, though nothing in the platform detects harmonic or chart patterns, and there is no scanner and no automated signal testing. The Candlestick Patterns indicator, in the Indicators menu under Standard, marks candlestick patterns only.

Step through your own charts and count what one rule set actually does. Start a free trial.

Harmonic Pattern Rules: Why the Sources Disagree