Learn
Notes on what automated chart analysis can and cannot establish, written from the same data the tool runs on.
We let a pattern detector read nine years of charts. Here is every hit rate.
Updated 2026-08-02
Do chart patterns work? Every completed formation found across 182,341 candles of ten crypto markets, and what followed it. Double top resolved in its textbook direction 29.1% of the time. Triple top never fired once.
What an AI agent actually does when it analyses a crypto chart
Updated 2026-08-02
Six stages, five of them arithmetic. How swing detection, geometric pattern matching, confidence scoring and a published weight table produce a chart reading — and where the language model is allowed to sit.
Can AI predict crypto prices? A catalogue of where it fails
Updated 2026-08-02
Small samples, correlated instruments, one and a half market cycles, and 14 of 40 pure random walks producing a nameable formation. The specific places automated crypto analysis breaks, measured on our own engine.
AI crypto trading signals: what they assert, and how to judge one
Updated 2026-08-02
A signal claims something is worth acting on; an analysis claims something is present. Only the second can be checked before the outcome. The four things a provider must publish, and why almost none of them do.
Automated multi-timeframe scanning: 125 chart states, and what that costs
Updated 2026-08-02
Five intervals across twenty-five markets. Why timeframes disagreeing is information rather than contradiction, why the evidence is thinnest on the intervals people trust most, and why scanning 125 states meets the false-positive rate 125 times.
Do chart patterns work in crypto? Nine years of evidence, shape by shape
Updated 2026-08-02
On 661 completed formations across nine years of ten major crypto pairs: nine of eighteen shapes never reached a usable sample, two never occurred at all, and only two stayed above a coin flip at every horizon. Every figure with its sample count.