Audit-Stable Meaning: Why a Changed Definition Can Break a Past Decision

Some audit failures get blamed on bad execution. A large class of them are something else entirely: the meaning moved. Audit-stable meaning is the property that keeps a past decision provable after the definitions change.

The Problem

A compliance decision is only as durable as the meaning behind it. A loan is approved because an applicant met the definition of "qualified." A claim is paid because a procedure matched the definition of "medically necessary." A transaction clears because a counterparty fell outside the definition of "high risk." Months later, an auditor asks a simple question: was that decision correct under the rules that governed it?

In systems without decision-bound semantics, that question often cannot be answered. The ontologies, definitions, and classification rules have moved on. The term that justified the decision no longer means what it meant when the decision was made. The decision may have been sound at the time, but the system can no longer prove it, because it never bound the decision to the meaning in force at that moment.

This is not an execution failure. The code ran correctly. It is a semantic failure: the meaning was unstable, and stability was never treated as a requirement.

Definition Laundering

When meaning can move freely, it can also be moved deliberately. A definition is quietly revised, and a whole population of past decisions silently changes compliance status without any decision being re-made or re-examined. We call this definition laundering: rewriting the present meaning of a term so the historical record reads differently, with no trace that anything was reinterpreted.

Version control does not prevent this. Versioning a document or a schema records that meaning changed, and how. It does not bind any individual decision to the specific meaning that governed it. Knowing a definition was edited on a given date tells you nothing about which definition a particular decision was evaluated against. The edit history of the dictionary is not a receipt for the word you actually used.

The Property: Audit-Stable Meaning

Audit-stable meaning is the property that every governed decision remains evaluable, for as long as it must be retained, under the exact semantics that governed it at the moment it was made.

Audit-stable meaning is not a feature you bolt on. It is a property a system either has or lacks, and it can be tested. A system has it only if four invariants hold:

• Decision-bound semantics. Each decision is bound to exactly one immutable semantic snapshot, and that binding is fixed at decision time.
• Non-retroactivity. Later changes to meaning cannot alter the authoritative compliance status of decisions already made. Only the bound snapshot is authoritative for a given decision.
• Reproducibility. Given the recorded inputs and the bound snapshot, a deterministic replay reproduces the original determination.
• Drift visibility. Every semantic change is explicit, enumerable, and documented as a diff between snapshots.

A system that fails any one of these does not provide audit-stable meaning, no matter how sophisticated its other governance capabilities are.

How It Works, in Plain Terms

The architecture separates two things that are usually conflated. Living semantics is the domain knowledge that legitimately evolves: definitions get refined, edge cases get added, rules improve. Enforcement semantics is the frozen snapshot bound to a specific decision. The two coexist. Meaning stays free to evolve going forward, while every past decision stays anchored to the snapshot that governed it.

Each governed decision produces an Evidence Package, also called a Proof-Carrying Decision: a replayable evidence record that binds the decision to its semantic snapshot and its supporting inputs. Anyone holding the Evidence Package, the bound snapshot, the recorded inputs, and the deterministic replay logic can re-run the determination and get the same result. That is what makes the decision provable later, and it is what turns definition laundering from a silent rewrite into a detectable one. The snapshot is fixed, so a changed present-day definition cannot quietly restate the past.

Where This Sits

Audit-stable meaning is not the enforcement layer. In the FERZ model, authorization is the enforcement layer of AI governance: the non-bypassable boundary that decides what is allowed to happen. Audit-stable meaning is the substrate underneath it. It is what makes authorization replayable, what gives evidence its integrity over time, and what lets a governance decision be re-examined years later under the meaning that produced it. Enforcement decides. Audit-stable meaning keeps the decision provable.

Why AI Raises the Stakes

Probabilistic systems make meaning harder to hold still. Embeddings shift between model versions, fine-tuning moves decision boundaries, prompt templates encode their own implicit ontologies, and retrieval-augmented generation pulls in context that changes underneath a query. The reproducibility that matters here is specific and achievable: the deterministic governance evaluation is replayed against the recorded model output, not by re-executing the model. The goal is not to make a model deterministic. It is to make the governance decision about that model's output provable.

The full treatment, including the formal definitions of the four invariants, the disqualifying tests that operationalize them, and the reference architecture, is in the paper.

Read the Paper

Versioned Meaning: How to Make Ontologies Audit-Stable
Edward Meyman, FERZ, Inc. Technical Note, Version 1.5.
DOI: 10.5281/zenodo.18328587 · Read and download on Zenodo

@techreport{meyman2026versioned,
  author      = {Meyman, Edward},
  title       = {Versioned Meaning: How to Make Ontologies Audit-Stable},
  institution = {FERZ, Inc.},
  year        = {2026},
  type        = {Technical Note},
  number      = {Version 1.5},
  doi         = {10.5281/zenodo.18328587},
  url         = {https://doi.org/10.5281/zenodo.18328587}
}

Frequently Asked Questions

What is audit-stable meaning?

Audit-stable meaning is the property that a governed decision stays evaluable under the exact semantics that governed it when it was made, even after the underlying definitions change.

How is it different from ontology versioning?

Versioning records how definitions changed over time. It does not bind a specific decision to the meaning that governed it. Audit-stable meaning adds that binding, so a past decision can be replayed under its original semantics.

What are the four invariants?

Decision-bound semantics, non-retroactivity, reproducibility, and drift visibility. A system that fails any one of them does not provide audit-stable meaning.

What is definition laundering?

Quietly revising a definition so that past decisions read differently without being re-made or re-examined. Audit-stable meaning prevents it by binding each decision to a fixed semantic snapshot.

How does it apply to AI systems?

In probabilistic systems, the governance evaluation is replayed against the recorded model output rather than by re-running the model, so the governance decision stays provable even as models drift.

Related: FERZ Governance · The Four Tests Standard (4TS) · Versioned Meaning (paper)

CITE THIS PAGE

FERZ, Inc. (2026). Audit-Stable Meaning: Why a Changed Definition Can Break a Past Decision. https://ferz.ai/articles/audit-stable-meaning

@misc{ferz2026auditstablemeaning,
  author       = {{FERZ, Inc.}},
  title        = {Audit-Stable Meaning: Why a Changed Definition Can Break a Past Decision},
  year         = {2026},
  organization = {FERZ, Inc.},
  url          = {https://ferz.ai/articles/audit-stable-meaning}
}