The Epistemic Stack: How Structural Tools Drift Toward Social Heuristics: From Cognition to Institutions to Machines

Human institutions begin as structural tools. Writing, records and written rules were invented to coordinate people beyond the scale where everyone knows each other. This paper proposes that, over time, these tools drift toward the social heuristics of the majority that runs them: precedent, status, consensus and in-group loyalty. The result is an environment we call the field, which people experience as reality rather than as one lens made permanent. We set out an eight-link causal chain. Every brain compresses data through learned predictions, and people weight social and structural data differently; the majority's weighting becomes the shared default. Fast social heuristics coordinate groups well where people know each other (Dunbar, 1992). Beyond that scale, groups invent structural tools, from the earliest accounting tablets to written rules (Nissen, Damerow & Englund, 1993; Weber, 1922/1978). Those tools then drift toward the social weighting of the people who run them, a pattern visible in creative compliance with law, in the devaluation of proposals from opponents, and in lower funding for interdisciplinary research. Rules accumulate rather than get removed (Adams et al., 2021), and fixes come from the same lens. Together these form an epistemic stack: layers of rules and knowledge in which each layer treats the one below as settled, so early assumptions pass upward unchecked. Institutions that build in a halt, a fixed point that forces a return to structural data, make fewer errors: the surgical safety checklist and crew resource management in aviation are examples. We predict that adding a structural halt to rule-making reduces drift in any institution, artificial intelligence included. Neither lens is a deficit; the problem is treating one lens as the only one. We present three case studies (bureaucracy, the mental health field, and artificial intelligence), show how the same event reads differently through each lens, explain why a lens shared by the majority is hard to detect, and propose a halt at three scales and five falsification protocols.