The TradingView pattern library.

• Elliott translation: LS–Head is wave 5 with divergence; RS is the B-wave bounce failing under the old high. Same structure, two languages.
• Entry: neckline break close, or (higher R:R) the retest of the neckline from the other side. Target = head height from the break.
• Trap: sloping necklines invite premature entries — require a close beyond the neckline drawn through both armpits, and check volume expands on the break.

• Confirmation matters: the pattern doesn't exist until the neckline (the swing between the touches) breaks — trading the second touch is a range fade, not a pattern trade. Both are valid, but size and target them differently.
• Best filter: divergence on touch 2 (or 3) plus a failed-poke wick beyond the level — the same trapped-trader logic as the range strategy's failed-break entry.
• Intraday note: on US500/GER40 5m, "double tops" at yesterday's high that are actually expanded-flat B-waves will wick above touch 1 before reversing — give the second touch room or wait for the neckline.

• The one rule that filters most failures: volume must contract during the pause and expand on the break. A "flag" on rising volume is distribution, not rest.
• Elliott translation: flags/pennants are small wave-4s (or ii's); the measured move (pole ≈ pole) is the equality guideline in disguise. If the first pole was already a wave 5, the "flag" is a topping B-wave — check the higher-timeframe count before assuming continuation.
• Where they shine: the momentum-scalp entries of Strategy 6 are 1-minute bull flags by another name; WTI and US100 produce the cleanest ones during their high-energy windows.

• Ascending/descending: the flat side is accumulated orders; the break through it is the tradeable event (~87% for descending-triangle breakdowns in Bulkowski's data). The volume-breakout strategy is the execution template.
• Symmetrical: direction is near coin-flip pre-break — never pre-position. In wave-4 position, expect the break with trend and a final-wave thrust (see EW triangles).
• Wedges: the highest-information bilateral pattern because of the momentum story: five overlapping pushes with stacking divergence. Rising wedge after an extended rally = ending diagonal = expect the entire wedge retraced. Falling wedge into support = its bullish mirror.

TradingView ships a dedicated XABCD drawing tool (and the community publishes dozens of auto-scanners) for the harmonic family. All share the same anatomy — an impulse XA, retracement AB, minor BC, and a completing CD leg into the Potential Reversal Zone — differing only in their required ratios.

• Gartley/Bat within trends, Butterfly/Crab at extremes: match the pattern family to the regime, exactly as ADX gates the strategies in Section 3.
• Full trade plans: the site's harmonic combinations C5 (Bat + RSI divergence) and C6 (Gartley + fib cluster) give the checklist, stop and target logic — this table adds the sibling patterns to widen the map.
• Elliott bridge: a Butterfly/Crab D beyond X is usually an expanded-flat B-wave or wave-5 extension completing — harmonics and waves are two grammars describing the same crowd behaviour. When both name the same level, that's real confluence.

Why published success rates mislead

• Selection effect: studies count confirmed patterns (post-breakout). Your live problem includes all the almost-patterns that never confirmed — the denominator is different.
• Market & timeframe drift: most stats come from daily US stock data. 5m index-CFD behaviour (spreads, session opens, news) changes the numbers materially.
• Definition sensitivity: change the pivot threshold that defines "a swing" and half your double tops appear/disappear. Whatever detector you use, freeze its parameters before measuring.

Re-measuring on your market in ~40 lines

# Sketch: double-top detector + outcome measurement over ZigZag pivots
# pivots = DataFrame from the ZigZag algo on the Elliott page (time, price, kind)
TOL = 0.15   # peaks within 0.15×ATR are "equal"
results = []
for i in range(len(pivots) - 3):
    p1, t, p2, conf = pivots.iloc[i:i+4].itertuples(index=False)
    if p1.kind == p2.kind == 'H' and t.kind == 'L':
        if abs(p1.price - p2.price) <= TOL * atr_at(t.time):
            neck = t.price
            target = neck - (max(p1.price, p2.price) - neck)   # measured move
            hit = first_touch(after=conf.time, level=target)
            stop = first_touch(after=conf.time, level=max(p1.price, p2.price))
            results.append('win' if hit and (not stop or hit < stop) else 'loss')
win_rate = results.count('win') / len(results)   # YOUR number, for YOUR market

Run it per instrument and per session window. Typical findings worth knowing in advance: continuation patterns test far better with the prevailing regime (ADX > 25) than the headline rates suggest, and reversal patterns' edge concentrates at higher-timeframe levels — matching everything Section 3 preaches about stacking location with pattern.