Split test

How it works: the lander assigns each visitor A or B (50/50, sticky per browser; ?v=a / ?v=b forces one for testing). The variant is stamped on the GHL contact as tai-var-a (TAI 1) / tai-var-b (TAI 2) at opt-in. Winning metric is a qualified booking: opt-in → finished the audit application → booked the audit calendar, not low-spend. Same definition as the Meta Lead. Verdict uses a two-proportion z-test at p<0.05; ignore it under ~100 opt-ins per arm.