When credit policy and production code diverge
Lending platforms often discover rule drift only after vintage curves bend. How sample testing surfaces silent exceptions.
Credit committees approve a policy. Engineering ships a decision service. Six months later, override codes and feature flags have rewritten the effective underwriting without a formal change record.
In risk management audits for lending platforms, we reconcile policy text against decision logs and code configuration. Sampling originated loans by segment often reveals cohorts that should have been declined—or approved with tighter limits—under the published framework.
Close the loop with a living rule inventory: each decision node named, owned, versioned, and linked to the committee paper that authorised it. Without that map, portfolio risk reporting describes intentions rather than operations.