Governance Simulation in Algorithmic Underwriting Exception Handling: A Decision Integrity Protocol Assessment of Earnest Operations LLC

This audit does not assess whether Earnest Operations LLC's underwriting model produced biased outcomes. That question, whether the Cohort Default Rate variable functioned as a proxy for race, is the subject of the Massachusetts Attorney General's allegations and is not re-litigated here. This audit examines a narrower, structurally distinct question: whether Earnest could reconstruct, for an individual underwriting decision, which of two governance categories actually produced it, an algorithmic knockout determination, or a documented human exception with the senior oversight that Earnest's own policy required. According to the Massachusetts Attorney General's findings, published in the settlement announcement of 10 July 2025, underwriters at Earnest "frequently bypassed the models without clear standards or documentation, and often favored applicants based on assumptions", despite a company policy requiring senior oversight for exactly this kind of model exception. Earnest has denied the allegations and states it settled to avoid prolonged litigation, without conceding a violation of state or federal law. Both positions are represented in this audit. If the Attorney General's finding is accurate, the structural condition it describes is not primarily a fairness failure. It is a governance simulation: a documented oversight mechanism existed on paper, but its operational presence, whether specific overrides were actually reviewed and approved as the policy required, cannot be established after the fact. This is the second audit in the DIP series assigned to the Governance Simulation category, alongside TD Bank AML, and the first to examine a human-process rather than automated-system observability gap.