A reason code explains a credit decision, such as why an application was denied or offered less favorable terms. Lenders may provide specific reason codes in an Adverse Action Notice; interpretable models make the explanation easier to defend. section 4.4.4
Reason codes can link model results to understandable factors. With WoE transformations, an auditor can trace how a particular bin affects a score. LIME and SHAP can also break down individual decisions, while global interpretation methods can reveal broader patterns. section 4.4.4 section 6.0
Credit-scoring workflows use reason codes to support:
- Connecting model scores to reasons or partial explanations at the scoring and inference stage. section 4.5.3
- Making decisions more transparent to borrowers and regulators through interpretable outputs. section 6.0
- Providing readable explanations through an LLM interface only when it translates deterministic model outputs, with privacy and compliance safeguards. section 6.6
Tracing a score effect
A bin for three to five delinquencies can have a negative WoE value that corresponds to a lower credit score, giving an auditor a traceable explanation of the result. section 4.4.4
