Answers are grounded in this paper and cite their sources.
Finite observers do not merely update beliefs on a fixed stream of evidence. They must continuously reproduce the distinctions, historical anchors, and de-correlated challenges that make correction possible. Once an established belief influences the investment in those very structures, success and corrigibility can diverge: reasons continue to accumulate while the capacity to generate reasons for revision contracts. This reflexive divergence is epistemic hollowing.This paper proposes a dynamic framework for belief formation, justification maintenance, and epistemic hollowing in finite observers. Its point of departure is not probability, uncertainty, or error rates, but the more upstream condition of registration: a worldly difference must first be resolved by a finite interface and jointly sustained within a comparative window with retrievable anchors before it becomes a registrable event for the observer; observer-internal probability assignments only acquire an operational domain once this outcome space has been constituted. Consequently, historical identifiability is not treated as a background automatically provided by time, but as an achievement of registration that requires continuous resource investment.The paper further repositions JTB (Justified True Belief) as an entitlement architecture of 'continued taking-as-true,' rather than as a final definition of knowledge; shows that a substantial family of Gettier-like luck can be represented as saturation failure of predicates relative to registration equivalence relations; and argues that any belief-revision rule defined solely on the current registration space cannot recover task-relevant distinctions that have not entered its domain. Dynamically, the success history S and the challenge structure Γ can grow apart: success can be capitalized through memory into new justification, but only when the internal warrant-updating rule underestimates the dependency structure of the generating path; when it does not, repeated success does not inflate justification. Once a belief acquires entitlement, it in turn shapes future action, sampling, archives, classification, and challenge investment, producing a self-reinforcing trajectory in which success reproduction can crowd out failure-space reproduction. This gives rise to three hollowing mechanisms—distinction collapse, challenge collapse, and identity drift—corresponding to the three dimensions of corrigibility: registration capacity (Δ), de-correlated challenge capacity (Γ), and cross-temporal rule invariance (I).In dialogue with core literatures, the paper operationalizes Bateson's 'difference that makes a difference' as 'registered difference that makes a registered difference,' prefixes Ashby's requisite variety with 'requisite registrable variety,' shows how a substantial family of Gettier-like cases can be represented as registration saturation failure, pushes Hume's problem of induction from insufficient exogenous samples to endogenous sampling distortion, and engages the organizational decoupling literature to specify an endogenous persistence mechanism: success is capitalized into entitlement, entitlement reduces challenge reproduction, and the resulting evidential thinning is then read as further success. Institutions instantiate the same abstract architecture, but introduce additional parameters of delegation, conflict, and transferability.
No takes yet. Share an insight, caveat, or question.
Ruipeng Shi (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: