Beyond Online Association Rules: Governing Semantic Drift in Numerical Stream Mining

This preprint proposes a governance-oriented reframing of online numerical association rule mining. While existing approaches have achieved substantial progress in efficiency, incremental updating, and metric optimization, they typically assume semantic stability and equate statistical persistence with epistemic validity. We argue that long-lived mining systems operating in non-stationary environments accumulate semantic debt: statistically strong rules whose conceptual meaning has silently degraded relative to the evolving context. This form of degradation remains invisible to traditional metrics and can lead to unrecognized epistemic failure in decision-support systems. To address this gap, we introduce GAIA, a meta-control layer designed to govern the lifecycle of extracted rules under semantic drift. GAIA does not replace or modify existing mining algorithms. Instead, it operates above them, externalizing expert judgment and providing structured mechanisms for semantic validation, challenge, retention, and explicit rule retirement. The contribution is architectural and conceptual rather than algorithmic. The work formalizes the distinction between pattern persistence and knowledge justification, introduces semantic debt as a first-class epistemic risk, and articulates falsifiable claims regarding the necessity of governance layers in long-lived, non-stationary AI systems.