Contestable Agents for Data Governance: Recording Human Stewardship of AI-Generated Metadata Across Structured and Unstructured Systems

The paper presents "contestable governance," a framework addressing the challenge of managing operational knowledge that industrial organizations hold across structured records (telemetry, point lists, transactions) and unstructured sources (manuals, service notes, communications, inspection reports). Since governance programs typically manage these sources separately and rely on a small group of domain experts, the authors propose specialist agents built on large language models that produce scored, cited arguments for disputed metadata decisions. Human stewards can accept, reject, modify, or add arguments before committing decisions, and a governance record retains the evidence, interventions, and outcomes, avoiding the problem where unreviewed proposals spread errors or perfunctory approvals leave no record of what was checked. A versioned domain glossary supplies model-specific terms and acronym definitions to the agents. The paper describes the workflow, a triage path reserving full argumentation for uncertain proposals while preserving review before commit, the scoring and record models, and a gate controlling which verified decisions become precedent. Two HVAC use cases illustrate the pattern: point classification using Brick Schema class labels with an open reference implementation, and an illustrative warranty failure-mode decision linking claim fields and telemetry to service notes and revisioned bulletins.

(Source：arXiv.org)