Toward Reproducible Machine Learning in Enterprise Platforms: A Governance Framework for MLOps Standardization Across Distributed Data Pipelines

Reproducibility is one of the biggest challenges and consistently under-addressed problems in enterprise AI deployments. Even as organizations make great investments in data infrastructure and model development tooling, they often experience difficulty in reproducing previous experimental results, validation of models deployed in production, and data provenance of predictions at scale. To tackle the enterprise distributed data pipeline issue of reproducibility in a systematic manner, this study proposes a five-layer MLOps governance framework for standardized data pipelines. Structured survey data from 250 practitioners across six industry sectors was used for an empirical investigation and then a comparative case study benchmarking analysis was conducted. Data lineage adoption (β = 0.31, p < .001), model version control maturity (β = 0.28, p < .001), and monitoring coverage (β = 0.27, p < .001) were the most significant factors when using mixed methods analysis, which included Pearson correlation analysis, multiple linear regression, one-way ANOVA, and chi-square contingency testing. Results from a one-way ANOVA showed that the adoption of structured governance was among the most significant organizational factors that influence the reliability of the ML pipeline with highly significant differences across the MLOps maturity levels (F(4, 245) = 147.3, p < .001, η² = 0.706). Post-implementation validation showed 69.4% increase in the rate of experiment reproducibility and 74.5% decrease in latency of model drift detection. The proposed framework outlines actionable governance level-by-level specifications that can help organizations transition from disjointed, ad hoc MLOps, to standardized, auditable, and cross-team reproducible workflows.