Driving Trusted Subsurface Data for AI at Scale: Enforcing Data Quality Thresholds and Governance in ADNOC's Data Foundation and ENERGYai

Abstract

Abstract As subsurface data platforms increasingly power artificial intelligence (AI), machine learning (ML), and agentic workflows, advisory or post-hoc quality control (QC) is insufficient to prevent non-compliant data from propagating at scale. This paper presents ADNOC's enforceable governance and quality framework within its Upstream Data Foundation (powered by OSDU®), which standardizes, enriches, validates, and controls access to key subsurface data classes so that ENERGYai, ADNOC's agentic AI platform, consumes only trusted, AI-ready data. Data providers submit datasets through standardized portals that trigger automated ingestion pipelines. A smart governance pipeline transforms incoming datasets into ADNOC standard templates, harmonizing parameter names, units of measurement and metadata structures. The pipeline supports both historical data enrichment (OSDU-resident data validated and enriched using agreed sources, target attributes, and quality rules) and controlled new ingestion via predefined templates and governed submission workflows. A quality engine evaluates datasets against industry-standard dimensions (e.g., completeness, uniqueness, consistency, conformance, validity, and accuracy), executes domain QC rules, and persists auditable results with lineage and timestamps. The implementation demonstrates a scalable and highly automated approach to governing large volumes of upstream subsurface data across ADNOC operating companies, assets, and data domains. End-to-end automation significantly reduces manual validation effort and accelerates time to availability while improving data consistency, standardization, and discoverability across enterprise platforms. Richer, standardized metadata, augmented where applicable by ML-based fact extraction from reports, improves searchability, contextual understanding, and user confidence at the point of data consumption. Centralized quality dashboards provide management with transparent visibility into data quality scores, trends, and remediation progress. Enforceable QC-score thresholds act as a platform-level quality gate: datasets that fail defined requirements are automatically quarantined and excluded from platform services, while datasets that pass are automatically published and tagged as AI-ready, enabling governed consumption by agentic workflows. The key differentiator is governance that actively restricts data access based on measurable QC thresholds rather than treating quality scores as advisory, delivering a production-grade blueprint for operationalizing trusted, AI-ready subsurface data within an enterprise-scale, OSDU-aligned data foundation.