FlowAlpha › Chart patterns

Crypto chart patterns, with the hit rates we measured

Most pattern guides tell you what a shape looks like and stop there. These pages also tell you how often it worked for us — on 277 calls we published in public before the move, and on 2,140 more replayed over history. Winners and failures are both counted.

43%of our 277 resolved live calls worked (as of 6 Sep 2026)
277calls published before the move, then graded (as of 6 Sep 2026)
47%of 2,140 backtested (as of 6 Sep 2026)
12pattern families detected, 15m to 1D

Live and backtested are separate populations and are never averaged together.

Every pattern we detect

A call counts as a win if price closed 5% in favour before it wicked 5% against, measured from the confirming close over the following 30 days. "Live" means we posted it on the board at the time. "Backtested" means the same detector replayed over past data — a separate population, and not part of our published record.

The full table

PatternBiasLive callsLive hit rate (as of 6 Sep 2026)BacktestedBacktested rate (as of 6 Sep 2026)
Rectangle / Range Either way 6545% 10544%
Double Bottom Bullish 5549% 10347%
Double Top Bearish 4333% 7649%
Descending Triangle Bearish 3954% 10745%
Ascending Triangle Bullish 3540% 9646%
Falling Wedge Bullish 2157% 7748%
Inverse Head and Shoulders Bullish 8too few to say 1136%
Head and Shoulders Bearish 7too few to say 1856%
Rising Wedge Bearish 4too few to say 2425%
Bull Flag Bullish not published live yet 61048%
Bear Flag Bearish not published live yet 63448%
Cup and Handle Bullish not published live yet 27948%

Across every family, 43% of our 277 resolved live calls worked — which is what an unfiltered detector that publishes its own failures looks like — against 47% of 2,140 backtested ones. Families with fewer than 10 resolved calls show "too few to say" rather than a percentage, because a rate on four calls is noise, not a finding.

Why these numbers are lower than you may expect

A hit rate near half is what an honest, unfiltered pattern detector looks like. We publish every detection that clears the geometry test, not the handful that happened to work, and we grade them on a fixed rule rather than by eye. Sites quoting 70% and 80% are usually either selecting after the fact or counting a touch of the target rather than an ordered outcome.

What makes a pattern tradeable is not the raw hit rate anyway — it is the size of the winners against the size of the losers, and where the invalidation sits. That is why every call on our board keeps its peak move and its drawdown, both visible on the full record.