Artifact-to-Finding Promotion: Evidentiary Requirements for Harm from Correctly Functioning AI Systems

This paper specifies the records required to investigate harm arising from a propositionally accurate AI output whose attributed warrant exceeded its established warrant. It uses artifact-to-finding promotion to name the incident pattern: an AI system performs within specification and a consequential decision assigns its output more evidentiary force than the system established. Version 1.1 distinguishes the ordinary technical and domain evidence required to prove input integrity, system operation, and output validity from five promotion-specific records: the output as rendered, qualifiers presented or omitted, the linked decision record, the system's documented warrant at deployment, and contemporaneous model and configuration provenance. It operationalizes attributed warrant through the documented rationale and observable decision, distinguishes representation-, decision-, and compound-layer promotion, and adds a controlled evidence-to-finding reconstruction pathway. The revision also makes source independence, representation completeness, the decision-support boundary, and proportionate control provenance explicit. Worked successful and failed reconstructions demonstrate the method and the consequence of missing records. The central forensic problem is that the decisive promotion record is the output as rendered to the decision-maker, not merely the underlying transaction log, and conventional logging architectures are not necessarily designed to preserve it. This paper is issued as a working paper. The incident pattern is offered for testing and criticism rather than as a settled result. Correspondence and challenge are welcome.