We’d rather over-explain than overstate. Here is exactly how the numbers across CryptoPatterns.ai are produced — and where their limits are.
Each chart pattern carries a historical follow-through rate. To produce it, the engine scans real historical candles for the most liquid pairs across 20+ exchanges and detects every occurrence of the pattern. For each occurrence it then checks what price did over several horizons — 1 hour, 4 hours, 24 hours and 7 days — and counts it as a follow-through if price moved in the pattern’s predicted direction by a meaningful threshold within that horizon (a larger threshold for longer horizons). The win-rate is simply the share of occurrences that followed through, and it is always shown alongside the sample size it is based on.
This is deliberately a follow-through measurement, not a full trade simulation. It does not assume a specific entry fill, stop or target, and it does not deduct trading fees or slippage. That makes it a clean, comparable measure of how a pattern has historically behaved — but it is not the same as the live, fee-and-stop reality of an actual trade. Patterns with small sample sizes are noisier; we flag low-sample patterns rather than presenting their numbers with false confidence.
Browse historical pattern follow-through rates in the pattern library.
It is a historical follow-through rate: across thousands of past occurrences of that pattern, the share where price moved in the pattern’s predicted direction by a meaningful threshold within a set window (1h, 4h, 24h or 7d). It is computed from real historical candles and shown with its sample size. It is not a full trade simulation and does not include fees, slippage or a specific stop, so treat it as evidence of historical tendency, not a guarantee.
Yes — always. Pattern win-rates are computed over all detected occurrences, winners and losers alike, and are shown alongside the sample size they are based on. We never cherry-pick the occurrences that worked.