Contestable Agents for Data Governance: Recording Human Stewardship of AI-Generated Metadata Across Structured and Unstructured Systems
Industrial organizations struggle to govern structured telemetry and unstructured documentation together because the work relies on scarce domain experts. The paper proposes contestable governance: specialist AI agents generate scored, cited arguments for metadata decisions, human stewards review and intervene before anything is committed, and an append-only record preserves full lineage. This design aims to stop unreviewed agent errors from propagating into downstream systems while making human oversight attributable and auditable.
This paper presents a thoughtfully motivated architectural pattern for human-in-the-loop data governance, but it offers no empirical evidence that the proposed contestability improves decision quality, reduces rubber-stamping, or operates at scale. The authors are commendably transparent about these limits, repeatedly noting that scores are illustrative and several key features remain unimplemented.
The seven requirements (R1–R7) provide a crisp conceptual scaffold, the PROV-based lineage model is sound, and the synthetic HVAC examples clearly illustrate how conflicting evidence—such as a point name versus a manual section—can be surfaced for human review before commit. The authors also remain transparent that the warranty case and promotion gate are illustrative rather than evaluated.
The entire evaluation consists of synthetic examples with author-selected base scores, hand-tuned hyperparameters, and no accuracy benchmark. The reference implementation covers only a single synthetic point, while the warranty workflow, triage path, append-only storage, and promotion gate remain proposed rather than implemented. The scoring rule is justified only as illustrative, the base scores are uncalibrated, and the paper explicitly disclaims that the computations establish validity or robustness.
The conceptual requirements and comparisons to data-governance and contestable-AI literatures are fair and well-cited, but the evidence does not validate the central QBAF scoring rule or show that specialist agents outperform simpler baselines. The warranty case relies entirely on hand-authored arguments and scores rather than independent adjudication, and the cited 72.7% triage saving from Wu and Lim (2026) is explicitly noted not to measure this workflow.
An open reference implementation is available for the HVAC point example and the arXiv ancillary files include a reproducible warranty scoring script, but there is no real dataset, benchmark, or fitted parameter set. The triage threshold, promotion gate, cross-system governance record, and append-only database roles remain unimplemented, so independent reproduction would be limited to running synthetic illustrations on a single point.
Industrial organizations hold operational knowledge in structured records, such as telemetry, point lists, and transactions, and in unstructured sources, such as manuals, service notes, communications, and inspection reports. Governance programs often manage these sources separately. Classifying, linking, and documenting them depends on a small group of employees with deep domain knowledge. Agents built on large language models can propose metadata, cite documents, and flag quality issues, but unreviewed proposals can spread errors into downstream systems. A perfunctory approval leaves no record of what a reviewer checked. We present contestable governance: specialist agents produce scored, cited arguments for disputed metadata decisions; a steward can accept, reject, modify, or add arguments before commit; and a governance record retains the evidence, interventions, and outcome. A versioned domain glossary supplies model-specific terms and acronym definitions to the agents. We describe the workflow, a triage path that reserves full argumentation for uncertain proposals while preserving review before commit, the scoring and record models, and a gate that controls which verified decisions become precedent. Two HVAC use cases illustrate the pattern: point classification using Brick Schema class labels, supported by an open reference implementation, and an illustrative warranty failure-mode decision linking claim fields and telemetry to service notes and revisioned bulletins.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.