The Minimal Decision Record: What a Consequential AI Decision Record Must Contain

Abstract This applied paper translates the Identity-Persistence Program into a decision-record schema for consequential automated decisions. It asks what a record must contain so an independent reviewer can determine what happened without interviewing the team or reconstructing context after the fact. The answer is a minimal decision record: a scope block with at most twelve fields, one structured operating envelope, an event block, and a completeness audit that either certifies the record or exhibits a concrete dispute its gaps cannot resolve. The paper distinguishes logs from decision records. Logs report observed surface behavior; records declare what governed the decision, what counted as the same case, what changes were tolerated, who had authority, what reviewers may inspect, and what can be replayed. The schema inherits its field structure from prior formal results but makes no new formal claim. Its operating rule is practical: simplicity is valid only when earned by structure, not by nesting fields into opaque blobs. The result is an adjudication-grade artifact: a record that separates what the system produced from what the declared scope demanded. The remaining claim is empirical rather than formal—whether such records reduce dispute cost, review time, and leakage at portfolio scale.