The Gartley candidate D lies inside XA; the Butterfly variant compared here projects D beyond X. This changes the measurement and rejection rules, not the probability of profit. We compare definitions and hypothetical coordinates, and show why a failed Gartley cannot simply be renamed Butterfly after price passes X.

Definitions and a common measurement rule

Use ordered X–A–B–C–D points and linear-scale price differences. AB/XA means |B−A|/|A−X|; AD/XA means |D−A|/|A−X|. Screen-line length and D/A are different quantities. A ratio needs a named numerator and denominator.

This comparison uses Carney’s public Gartley definition with B at 0.618 XA and D at 0.786, and his Butterfly variant with B at 0.786 and D at 1.27 XA. C, AB/CD relationships, BC projection and convergence still require checking. The source page discusses more than one Butterfly relationship; choose and record a complete variant before measuring candidates.

The signed formula P(r) = A − r × (A−X) works for either direction. P(0)=A and P(1)=X; r greater than 1 places P beyond X. For the same XA, Gartley D=P(0.786) coincides in price with Butterfly B=P(0.786), but they occur at different stages: B precedes C and the final leg.

The five points of the XABCD structure
Point Xthe origin of the pattern — the starting high or low
Point Athe first extreme in the opposite direction
Point Bthe retracement of leg XA — its depth is one condition, not a complete classification
Point Cthe retracement of leg AB — record its ratio and projection under the selected definition
Point Da candidate completion price; the XA-based projection here is measured from A

Gartley — the blueprint for the whole family

For a hypothetical EUR/USD frame X=1.25000 and A=1.28000, XA=300 pips at 0.0001. Gartley candidate B=1.26146 and D=1.25642; Butterfly 1.27 gives B=1.25642 and D=1.24190. Gartley D is 64.2 pips above X; Butterfly D is 81 pips below X. The two D prices differ by 145.2 pips. C is unspecified, so these are preliminary coordinate checks, not complete patterns or entry signals.

In the mirror frame X=1.28000 and A=1.25000, the signed XA is −0.03000. Gartley B=1.26854 and D=1.27358; Butterfly B=1.27358 and D=1.28810. Gartley D remains 64.2 pips below X, while Butterfly D is 81 pips above X. These two frames are independent hypothetical directions, not sequential trades. Neither difference is an expected profit.

Price levels are not leg ratios

Compare the lengths of specified legs, not two labelled price levels. Shifting the whole synthetic structure illustrates the difference without forecasting a market.

Scroll sideways to see the full diagram

Ratios depend on price differencesTwo X, A, D series differ by 1000 units. In both, A is 100 units above X and D is 27 below X. AD/XA is 1.27 in each.PRICE DIFFERENCES, NOT D / ASeries IX 100A 200D 73Series IIX 1100A 1200D 1073I: XA = 100; AD = 127; AD/XA = 1.27.II: XA = 100; AD = 127; AD/XA = 1.27.Adding 1000 to every price preservesleg lengths and their ratios.

Two independent models in arbitrary units. I: X = 100, A = 200, D = 73; II adds 1000 to each. Panels use the same distance scale relative to their own base. B, C, time and candle history are unspecified; only three-point geometry is tested. Scott Carney — Butterfly Pattern defines the 1.27 projection relative to XA length.

Adding 1000 to X, A and D preserves XA = 100 and AD = 127. Both series give AD/XA = 1.27 despite different D and A prices.

D/A is 0.365 in I and about 0.8942 in II. Neither is AD/XA. That quotient cannot test the specified 1.27 ratio despite using the same point names. A complete Butterfly still requires its other conditions.

Butterfly — the extension pattern, when price runs past X

The selected Carney Butterfly uses 1.27 XA. TradingView’s XABCD drawing documentation describes alternatives of 1.272 or 1.618 XA; “or” does not mean any intermediate value is an equivalent rule. On the 300-pip bullish frame, 1.272 gives D=1.24184, only 0.6 pip below the 1.27 price. Using 1.618 gives D=1.23146, 104.4 pips below it. That is a different model assumption, not rounding, and does not automatically classify a Crab either.

“Beyond X” has different meaning for protection. For bullish Gartley, X is below candidate D; for bullish Butterfly, X is already above D and does not bound further decline from D. Define geometry rejection, exit logic and the actual protective order separately. A chosen stop cannot guarantee its fill or limit every possible loss.

Do the two ratios require the same X?

Take another hypothetical set A=1.2800, B=1.2600, D=1.2400. The Butterfly B condition requires XB = A − (A−B)/0.786 ≈ 1.254555. The D condition requires XD = A − (A−D)/1.27 ≈ 1.248504. These two required anchor prices differ by about 60.51 pips, so no one exact X satisfies both conditions.

With A fixed and separate ±2-pip tolerances on B and D, the allowed X intervals are approximately 1.254300–1.254809 and 1.248346–1.248661. The half-widths are 0.0002/0.786 and 0.0002/1.27. The intervals do not overlap. Shifting X to repair one ratio cannot repair the other; even overlapping intervals would not validate C or the remaining requirements.

The reversal zone and price confirmation — where you actually enter

PRZ means potential reversal zone: record the selected XA, BC and AB/CD projection prices and their differences under one definition. A tight geometric cluster does not guarantee a reversal, a short executable stop or a profitable result. Do not silently change the projection convention to rescue a failed candidate.

An entry condition may wait for a recorded candle or oscillator event, but it is another rule, not guaranteed removal of expensive failures. A confirmed pivot or closed candle may be available later than projected D. Record that delay and the subsequent actual hypothetical fill instead of assigning entry retrospectively to the ideal geometric level.

Practical takeaway: Measure price-leg differences and their ratios. Dividing absolute price levels changes the meaning of a harmonic ratio.

A separate GBP/USD candidate and execution model

Hypothetical example — bullish Bat on GBP/USD, daily chart (illustrative values)
Point X1.2515 — the low from which leg XA starts
Point A1.2670 — the high of leg XA; the whole leg measures 155 pips
Point B1.2600 — AB/XA ≈ 0.4516; one Bat coordinate condition
Point C1.2656 — a bounce of 0.800 of leg AB, below the high at A
Computed point Dexact XA projection 1.253267; display 1.2533; CD/BC ≈ 2.2023
Reversal zonechosen interval 1.2525–1.2540, width 15 pips; not demonstrated projection convergence
Confirmationassumed later close-condition; complete OHLC candles not supplied
Entryhypothetical actual Ask fill at 1.2540 after the condition
Stop lossassumed Bid fill at 1.2495, below X; planned 45-pip price distance
First targetexact geometric 1.258513006; assumed Bid fill rounded to 1.2585 gives 45 pips
Second targetexact geometric 1.261753994; assumed Bid fill rounded to 1.2618 gives 78 pips and 1.7333 reward/risk

The exact candidate AD is 137.33 pips. Equal AB=CD would project D=1.2586 and alternate 1.27 AB=CD gives 1.25671: respectively 53.33 and 34.43 pips from the XA-based 1.253267. The chosen 15-pip interval is therefore not evidence that these projections converge. This is an incomplete candidate and a separate arithmetic trade model, not a verified Bat. Under the assumed fills, reward of 45 or 78 pips versus planned risk of 45 pips gives 1 or 1.7333 before commission and funding. The spread is already represented by Ask entry and Bid exits. Neither the ratios nor a positive hit rate alone establishes net expectancy.

Measurement and test checks

Several checks prevent contradictions in the definition and chronology. Passing them establishes consistent arithmetic; it does not establish profitable performance.

  • Changing a tolerance after the result. Record the selected complete convention and your allowed deviation. AB/XA=0.67 differs from 0.618 by 0.052; whether a detector accepts it depends on a previously stated tolerance, not a convenient label.
  • Ignoring decision delay. A closed-candle or confirmed-pivot variant is knowable later than an early touch. Its later entry cannot be filled retrospectively at the earlier ideal D; a confirmation filter does not guarantee reduced losses.
  • Treating additional context as proof. Record higher-timeframe and indicator conditions as explicit variants. Price-derived filters are not automatically independent evidence and do not guarantee the direction of the next trade.
  • Inferring expectancy from reward/risk. A 1:1 price payoff may have positive or negative expectancy depending on actual probabilities and costs. Partial exits, break-even stops and full exits need separate calculations; none is superior by definition.
  • Mixing timeframes or feeds. Changing candles changes selected extrema and available information. Save the source, price side and timeframe, and compare rules on a fixed sample before making an effectiveness claim.
  • Confusing geometry rejection and protection. State the cancellation event, protective price side and fill assumption. A stop beyond X is not the same instruction for a candidate that already places D beyond X. Ordinary stops can slip.

What to do next with harmonic patterns

Compare methods defined by all their conditions, not just B and D. Keep the original rejected candidate when testing a new name or new anchor. Projected D, recognized local reversal and actual fill are three different stages.

  1. Check signed coordinates. For X=1.30000 and A=1.34000, Gartley candidate D at 0.786 is 1.30856, or 85.6 pips above X. Butterfly 1.27 gives 1.28920, or 108 pips below X. Without B and C these are only projections.
  2. Use a fixed sample. Record all selected pivots, ratios, tolerance failures and no-entry decisions. Check C, BC and AB/CD convergence as well as B and D; do not choose only completed reversals.
  3. Record execution separately. Write the actual hypothetical entry side, protection, exit variants and costs. Potential reward divided by planned price risk is not a success probability or a sufficient profitability rule.
  4. Set an alert instead of watching the screen. Hang a notification on the calculated level of point D. When price gets there you can calmly judge whether a reversal candle is forming inside the zone or the market is walking straight through the level.
  5. Evaluate without a magic sample size. Simulation or demo practice can test whether the recorded rules are reproducible. Twenty chosen trades cannot certify a strategy; include unseen periods and report limits, costs and parameter changes.

Related reading: Fibonacci retracements — the basics for the mathematics behind these patterns; Fibonacci extensions — setting targets as a companion to the BC projection topic; Elliott wave theory for the historical context of market geometry and its links to Gartley's idea.