Why do rational, risk‑aware individuals repeatedly take expressive or positional actions in digital environments where the expected value is negative?This paper introduces a minimal structural lattice explaining systematic miscalibration in persistent, identity‑bound digital systems without invoking addiction models, irrationality, manipulation, or behavioral pathology. The framework identifies three primitives that jointly generate compulsive initiation, heuristic drift, latent‑risk accumulation, and delayed activation panic: Temporal Asymmetry — decision‑time reinforcement is immediate and somatically encoded, while consequence‑time activation is delayed, conditional, and unobservable. Persistence Without Perceptual Decay — digital records preserve high‑fidelity state indefinitely, without contextual erosion, meaning loss, or embodied danger signals. Micro‑Contextual Coherence vs Global Invariance — humans act for local emotional/social coherence, but infrastructure preserves globally invariant records that can be reinterpreted under future norms, audiences, and power structures. These primitives interact to invert reinforcement gradients: short‑horizon cues trigger action; long‑horizon liabilities accumulate silently. Risk remains imperceptible until recontextualization collapses local meaning into global frames, producing sudden, discontinuous consequence realization. The model is structural, not psychological. It requires no assumptions of deception, weakness, intent, or adversarial design. It explains why calibration deteriorates even in honest systems and among high‑capacity, risk‑aware agents — and why warnings, self‑regulation, and platform‑level behavioral prescriptions fail. The lattice generalizes beyond social media to any persistent identity‑linked system with delayed or conditional consequences: workplace communication, credentialing, financial platforms, surveillance-linked services, and future infrastructures combining lifelogging, AR overlays, and biometric continuity. Scope: foundational substrate only — no empirical validation, diagnostics, or interventions. Implications extend to AI alignment under persistent memory, governance design, multi‑agent incentive geometry, and uncertainty modeling in delayed‑feedback environments.
LLC 3 Pilgrim (Thu,) studied this question.