Trying to classify trades across all market types just didn’t cut it. A setup that works in a quiet, sideways regime falls apart in high-volatility trends. So I added a regime classifier upstream. Gaussian Mixture Models over rolling return/volatility clusters. Not perfect, but good enough to route data into bull/bear/range buckets. Now each downstream model is trained on what it’s actually meant to see. Accuracy jumped. So did my confidence in the predictions. [[ML]] [[Serendipity]]