Treat the first-run sequence as a permanent test surface

Superwall’s case study of Cal AI reports 424 variants across 123 experiments, 160 of them paywall designs, for a 31% relative lift in trial-to-paid over 12 months. It is a growth engine, not a screen you finish.