A Parametric Model of Cross-Domain Over Confidence Transfer in ExpertsThe Dunning–Kruger effect describes how confidence and competence reconverge as mastery increases — a symmetry that is restored by expertise. We argue that this symmetry *breaks* under cross-domain transfer: earned metacognitive calibration fails to generalise while the associated confidence does, pushing experts back toward the peak of ignorance in unfamiliar domains. We present a minimal model with nine parameters capturing five mechanisms: the baseline confidence–competence curve, non-monotonic domain proximity susceptibility, context-dependent humility that decays with domain distance, a unified amplification factor encoding both credential halo effects and error-dependent overconfidence, and temporal dynamics with reinforcement and expertise-gated correction resistance. The model predicts maximum vulnerability at intermediate domain proximity and identifies a phase transition between transient and persistent overconfidence at r=dr = d r=d. Six testable predictions are derived, including non-monotonic proximity dependence, asymmetric directional transfer, superlinear credential scaling, and error-specific amplification. An interactive companion is distributed as supplementary material.
Anonymous (Sat,) studied this question.