RSTM: A Formal Model of Coercion- and Runtime-Mediated Semantic Transitions in Application Systems
Abstract
Modern application systems depend on dynamic typing, object binding, deserialization, reflection, symbolic dispatch, plugin discovery, and runtime module loading. These mechanisms are useful because they translate flexible values and metadata into executable behavior, but they also create security failures in which an input remains syntactically admissible while its policy meaning or selected execution target changes. Existing weakness taxonomies, secure-coding guidance, static analysis, and runtime hardening address important symptoms, yet they do not provide one compact representation that connects accepted values, semantic transformations, execution selection, context, and protected state transitions. This article introduces the Runtime-Mediated Semantic Transition Model (RSTM), a formal and operational framework for analyzing coercion- and runtime-driven application failures. RSTM represents application states, admissible values, semantic domains, transformation boundaries, runtime operators, execution targets, contexts, and protected invariants. It defines two principal divergence classes—semantic-value drift and execution-binding drift—and derives a general attack condition and seven safety properties. An eight-step review workflow and a normalized transition card operationalize the representation for architecture review, testing, and incident analysis. A retrospective qualitative evaluation maps seven documented case families. In a repeatability check, three master’s students independently coded the same five fixed case packets under one codebook, yielding 86.7% macro-average pairwise agreement across semantic domain, runtime operator, target role, and earliest actionable control boundary. In a frozen same-case comparison, RSTM matched a structured weakness-taxonomy and threat-modeling baseline on information capture (46/50 versus 46/50) and produced no earlier-boundary win. These findings support RSTM as a consistent cross-mechanism representation on the evaluated cases, while not establishing improved vulnerability discovery, expert-analyst reliability, reduced analyst effort, or practical superiority over the baseline.