⚠ Cold-start problem: The ML model trains on real outcomes (hit/miss) from brain findings. From a fresh install it takes
50-100 rated outcomes (about a week of brain activity) before predictions are reliable. This page lets you seed the model with
synthetic but realistic training data based on documented setup edge.
Each generated row uses real feature distributions plus a label drawn from the historical hit-rate of that setup type. This gives the model a "warm start" — predictions will be approximately calibrated immediately, then continue to refine from real outcomes.