Signal · Pattern Engine
Pattern Engine.
Closelook's pattern-hit feed, refreshed after every US close. Each signal fires when a name clears the multi-filter confluence gate — regression bands, pivot clusters, SMA context. Regime classification sits on top so you know whether a "triggered" signal is in a trend-following or mean-reverting environment.
How the scanner reads
Step 1 is the interpretability grid — every instrument is classified by Hurst exponent, 252-day R², autocorrelation, and a short-vs-mid-term stationarity score. That produces a regime label (🟢 trending / 🟡 mixed / 🔴 mean-reverting) which contextualises any pattern hit.
Step 2 is the confluence pattern: five filters evaluated per day per instrument — 126-day regression band, 252-day regression band, 504-day regression band, pivot-cluster proximity, and SMA-200 status. A signal fires when at least two filters are triggered or at least four are "near" — the count is the confluence confidence.
Scope is intentionally narrow: one pattern, one regime grid, ~50 names. Interpretability before breadth. More patterns follow once this one proves out.