Adaptive Interpreter is a mechanism-first variant classification framework that independently scores loss-of-function (LOF), dominant-negative (DN), and gain-of-function (GOF) mechanisms to predict not only pathogenicity but inheritance pattern. Through validation across 4,487 variants in 8 genes, this system demonstrates 97.3% sensitivity, 82% accuracy for predicting semi-dominant inheritance from DN scores, and 42% resolution of VUS (variants of uncertain significance). This study introduces two novel biological insights: The Semi-Dominant Hypothesis (“The DN IS the LOF”) — Homozygous DN variants lose the substrate required for poisoning, resulting in complete functional loss. This unifies the longstanding paradox of variants classified as both autosomal dominant and autosomal recessive. The CASCADE Phenomenon — In dimeric transcription factors, DN-mediated interface disruption can produce gain-of-function effects by shifting conformational equilibria toward constitutive activation. This work demonstrates how mechanism-aware scoring—developed collaboratively by human and AI researchers—can resolve ambiguous clinical classifications, reveal underlying molecular logic, and generate testable biological predictions.