Algorithmic Res Judicata: Generative AI Summarization, Unwritten Precedent, and the Administrative Record

Abstract: Automated summarization converts software configurations into unwritten administrative law. Public benefits and immigration agencies increasingly place machine-generated summaries between claimant case files and deciding officials. Under production quotas and mounting backlogs, caseworkers read the summary rather than the underlying record. Machine characterizations that travel forward into subsequent renewals function as algorithmic res judicata, establishing preclusive factual findings that no adjudicator ever resolved. Systemic filtering patterns operate across caseloads as automated stare decisis, generating unacknowledged precedent without agency explanation. Established administrative doctrines already prohibit this governance by proxy. The Chenery doctrine confines judicial review to the agency's contemporaneous reasoning. Notice statutes bar reliance on unpublished staff instructions, while procedural due process guarantees inspectable evidence. Treating summarization configurations as agency policy restores the statutory obligations of authorship, explanation, and disclosure. Keywords: Algorithmic Res Judicata, Automated Stare Decisis, Administrative Law, Reasoned Adjudication, Procedural Due Process, Technological Due Process, Chenery Doctrine, Agency Precedent, Case Summarization, Public Benefits, Immigration Adjudication, Street-Level Bureaucracy, Administrative Record, Production Quotas, Automation Bias, Moral Crumple Zone, Freedom of Information Act, OMB M-25-21, Administrative Procedure Act, Unacknowledged Policy Change