The Random Forest's AUC advantage over Logistic Regression (typically 0.02-0.04) comes primarily from capturing non-linear DTI interactions in the 680-740 FICO band. For regulatory reporting under Reg B, the LogReg model remains the compliant choice — but the RF's feature importance rankings should inform which variables to add to the next scorecard revision. The transition matrix shows Grade B concentrating 42% of total default volume despite only 18% of originations — this is the portfolio's hidden risk.