Versioned Meaning: How to Make Ontologies Audit-Stable

This preprint presents a formal framework for audit-stable meaning in regulated AI and decision systems. It argues that many compliance failures stem not from incorrect execution, but from semantic instability: when ontologies and definitions evolve, past decisions become unverifiable because auditors cannot determine what a term meant at the time a decision was made. The paper defines four invariants required for audit-stable meaning—decision-bound semantics, non-retroactivity, reproducibility, and drift visibility—and distinguishes enforcement semantics from living semantics. It introduces a reference architecture for semantic snapshotting, version-scoped enforcement, and cryptographic binding of decisions to their governing semantic contexts, preventing retroactive reinterpretation (“definition laundering”) and enabling deterministic audit replay. This version (v1.4) is a specification-grade preprint that supersedes earlier conceptual drafts. It expands the formal treatment of semantic drift, replay requirements, and governance failure modes across probabilistic AI components, including embedding evolution, fine-tuning shifts, and retrieval-augmented generation (RAG) context drift. The work is intended for researchers, regulators, auditors, and system architects working in high-stakes domains such as healthcare, financial services, and government. It is published as a preprint to support open dissemination and prior-art disclosure while remaining subject to future revision.