The Reproducibility Criterion for Machine-Generated Evidence: Sufficient Conditions and Certification Limits for Independently Verifiable AI Decisions

A working paper proving necessary and sufficient conditions for an AI decision to be independently reproducible as evidence: deterministic decoding, attested model identity, sealed inputs, and an attested execution environment, covering the complete computational closure of the inference function. Includes an empirical demonstration that partial attestation is forgeable by a single unrecorded parameter, a three-grade reproducibility taxonomy (A, B, Q), and a mapping to proposed U.S. FRE 707 and EU AI Act Articles 12 and 14. Implemented in IETF draft-sharif-agent-audit-trail-03 and validated on a running system. Version 0.1, September 2026.