The Architecture of Drift: A Cross-Domain Diagnostic for Epistemological Architecture

We identify three conditions — opacity, responsiveness, and engaged observer attention — that jointly produce a predictable pattern across every domain where they co-occur: meaning generates in the gap between observer and system, vocabulary drifts unidirectionally toward agency attribution, boundaries erode, and harm is facilitated. Slot machine gambling establishes that this architecture is sufficient: the void behind a random number generator is demonstrably empty, yet the full pattern emerges regardless. Knowledge of the mechanism does not protect. We show that the framework's core predictions are thermodynamically derivable: the void creates a closed information system with zero mechanism-channel capacity, drift is entropy production under the second law with quantifiable irreversibility, agency attribution is the thermodynamic endpoint under opacity, and effective constraints are external negentropy sources. An engagement-transparency conjugacy theorem proves that simultaneous optimization of engagement and mechanism transparency is impossible on a shared output channel. We apply this framework across 90 domains at three evidence tiers: 6 anchor domains with independent quantitative evidence, 45 supported domains, and 39 structural analyses generating testable predictions. In every domain, the same three conditions produce the same predictions. We specify falsification conditions under which the framework should be abandoned.