Surrogate Model

The surrogate top-1 prediction was usually too strict, but top-10 predictions often contained strong candidates. The feature map uses fewer dimensions than candidates to preserve budget efficiency, accepting some misspecification. DOPP fits a local linear surrogate to approximate the composite PPA score from engineered features. The method uses a linear surrogate because the candidate set is finite and local to one search run, despite nonlinear true PPA. As candidate sets become larger and more diverse, the fixed feature map becomes less expressive and surrogate ranking degrades.