Credit Decision Explainability Flow

From a scored applicant to reason codes and a fairness report

Credit Decision Explainability Flow From a scored applicant to reason codes and a fairness report 01 / Sources 02 / Attribute 03 / Rank 04 / Map 05 / Consume Applicant Record · feature values · 01 / Sources · one row Applicant Record feature values one row Scoring Model · gradient boosting · 01 / Sources · probability Scoring Model gradient boosting probability SHAP Attribution · local, per feature · 02 / Attribute · additive SHAP Attribution local, per feature additive Fairness Monitor · outcomes by group · 02 / Attribute · population Fairness Monitor outcomes by group population Rank Contributions · adverse first · 03 / Rank · signed Rank Contributions adverse first signed Reason-Code Map · feature to text · 04 / Map · top drivers Reason-Code Map feature to text top drivers Adverse-Action Notice · applicant letter · 05 / Consume · required Adverse-Action Notice applicant letter required Fairness Report · adverse-impact ratio · 05 / Consume · audited Fairness Report adverse-impact ratio audited feature values one applicant prediction score scored outcomes population contributions signed values top adverse ordered reason codes human text impact ratio by group Legend primary data policy / PII async batch data store

Explanation Path

  • • One scored applicant plus their feature row feed local SHAP attribution
  • • Signed contributions are ranked so the most adverse features surface first
  • • Top drivers map to plain-language reason codes on the applicant notice

Fairness Path

  • • The same model's population outcomes feed a separate fairness monitor
  • • Group approval rates yield an adverse-impact ratio for the audit report
  • • Explanation and fairness stay distinct so neither dilutes the other

Data-flow diagram • Built with Archify • Create yours ↗ • Hover to trace • R route • Click to focus • +/− zoom • M radar • [/] views • P play story • T theme • E export