What a score measures
Textbook-ness: how many of the pattern's defining conditions co-occurred, weighted by rules written in advance. That has real uses, ranking attention, enforcing minimum standards, making detection auditable. It is a description of the setup's anatomy, produced entirely from the left side of the chart.
Why it cannot be probability
Outcomes depend on things no anatomy score contains: the regime, who is positioned where, what data lands tomorrow. Conditions can align beautifully into a trap; markets punish exactly the configurations everyone learned to love, adaptively. Our validation made this concrete: score deciles and outcome ranks simply failed to correlate out-of-sample, while a dumb geometric filter, reward distance, passed. Anatomy and prognosis are different sciences.
Honest usage
Use scores as gates and rank-orderers: ignore sub-threshold detections, review the day's highest-scoring candidates first, audit what the engine saw. Then judge the trade on the things scores cannot hold, context, location quality, risk geometry. And distrust any tool anywhere that prints setup probabilities without publishing the validation, because that number is marketing wearing mathematics.
FAQ
Is a higher-scoring setup at least somewhat better?
As anatomy, yes, more conditions aligned. As prognosis, our out-of-sample data says do not assume it; treat score as a quality floor, not a forecast.
Why publish scores at all then?
Auditability and ranking: scores make detection transparent and reviewable. The label rule keeps them honest: counts, never odds.
What number should drive the trade decision?
Risk geometry and context: reward-distance, invalidation clarity, higher-frame agreement. Those are checkable facts, not predictions.
ⓘ See these ideas on real price: open the free XAUUSD live chart.