Chemistry is the worst predictor of long-term relationship success and dating apps spent 15 years optimizing for it. John Gottman ran a 40-year couples research program out of the University of Washington. The original study tracked 130 newlyweds and predicted divorce with 93.6% accuracy from observing a single 15-minute disagreement. The variable doing the predicting is contempt frequency. Eye rolls. Sarcastic corrections. Dismissive sighs during conflict. Stable couples maintain a 5:1 ratio of positive to negative interactions during disagreement. Below 1:1, divorce becomes statistically near-certain. Six years after the wedding, couples who stayed together turned toward each other's bids for attention 86% of the time. Couples who divorced did it 33% of the time. The behavior is tiny: asking a question, sharing a thought, looking up when the other person walks in. The signal is enormous. Chemistry measures novelty response and dopamine. It captures how someone behaves when both of you are excited and the lighting is good. By month 18 neither condition holds. Then the predictive behaviors finally surface: how they treat you when they're tired, sick, or losing an argument. Dating apps optimize for the metric with the lowest predictive validity in the entire field. Attraction is what gets you to swipe. Contempt patterns don't show up on a profile.