Availability Without a Model: On the Semantic Fragmentation of Operational Metrics

This paper examines the epistemic status of availability as it is used in contemporary operational and distributed information systems. While availability is commonly treated as a unified system property, in practice it is derived from heterogeneous observational layers, including infrastructure state, service behavior, and business-level functionality. The paper argues that such usage produces a form of semantic fragmentation: scalar indicators that are numerically stable but lack declared structural equivalence. A minimal formalization shows that equality of availability values does not imply informational isomorphism across observational domains. Interpreting monitoring subsystems as observational localities within the framework of Coherent Observational Epistemology (COE), the paper characterizes model-free availability as a case of inter-local inference performed without verified structural alignment. The Unified Availability Model (UAM) is presented as an example of model-based availability that restores cross-layer coherence. The analysis contributes to the philosophy of measurement and observation in complex technical systems, demonstrating how engineering indicators may function as epistemic constructs rather than direct system properties.