Credit Scorecard: Weight-of-Evidence & ROC Discrimination (2D)

This 2D scorecard mechanic works the way many real credit bureaus build a scorecard by hand: income, debt-to-income ratio and payment history are each cut into five bins against a synthetic labelled population, and every bin gets a weight-of-evidence (WOE) value computed directly from that population's good/bad split. Instead of training a classifier automatically, you set each feature's weight yourself — at 1.0× for every feature the combined score is the statistically optimal sum of WOE, but detuning any weight away from 1.0× throws away real separating power even though nothing about the underlying population changed. A live test applicant panel shows the WOE contribution from each of your three features and the resulting default probability updating instantly, while a full ROC curve swept across the whole population — plus its AUC and derived Gini coefficient — shows exactly how much discrimination power your hand-tuned weights are keeping or losing.