Build `drift_detector(baseline_mean, current_mean, std, threshold)` — returns `True` if the current feature mean has drifted significantly from baseline: - Compute z-score: `|current_mean - baseline_mean| / std`. - Return `True` if z-score `>= threshold` (drift), else `False` (stable). - `std == 0` → return `False` (no spread → can't compute meaningful drift; flagging would be alert noise). - This is the simplest production drift signal — runs nightly per feature.