A Conditional Framework for Policy Collapse under Information Limits

Fuente: Zenodo
Salvato in:
Dettagli Bibliografici
Autore principale: Pérez Contreras, Benjamín Felipe
Natura: Recurso digital
Pubblicazione: Zenodo 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866901992552529920
author Pérez Contreras, Benjamín Felipe
author_facet Pérez Contreras, Benjamín Felipe
contents <p dir="ltr">This document provides the formal theoretical foundation for the Hybrid Meta-Controller (HMC) architecture introduced in the main paper Entropy-Induced Collapse in Learned Decision Policies under Partial Observability and Architectural Mitigation. While the primary work focuses on empirical validation and system design, this supplement establishes the ontological and informational constraints governing epistemic stability in decision-making agents.</p> <p dir="ltr">We show that policy collapse under partial observability is structurally unavoidable under finite informational constraints, and cannot be eliminated by improved training procedures, optimization methods, or increased model capacity. Under a minimal set of decision-theoretic axioms, we prove the existence of a critical informational bound, denoted αcrit, below which no purely inductive policy can guarantee long-term viability. This bound arises necessarily from the interaction between finite observation channels and viability-preserving state distinctions, rather than from empirical or algorithmic limitations.</p> <p dir="ltr">We further show that asymptotic stability under partial observability requires a qualitative architectural transition, in which control shifts from inductive inference to non-inductive (structural) regulation once informational conditions fall below this bound. This necessity is formalized as the αcrit Principle of Epistemic Viability (PEV), which characterizes a fundamental constraint on the persistence of decision-making agents operating under informational limits.</p> <p dir="ltr">Together, these results establish a non-normative, information-theoretic account of epistemic stability grounded in agent viability, providing necessary structural conditions for the design of decision architectures capable of operating safely under partial observability.</p> <p dir="ltr"> </p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19076073
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle A Conditional Framework for Policy Collapse under Information Limits
Pérez Contreras, Benjamín Felipe
Information Theory
Epistemic Stability
Partial Observability
Decision Theory
Policy Collapse
Hybrid Decision Architectures
<p dir="ltr">This document provides the formal theoretical foundation for the Hybrid Meta-Controller (HMC) architecture introduced in the main paper Entropy-Induced Collapse in Learned Decision Policies under Partial Observability and Architectural Mitigation. While the primary work focuses on empirical validation and system design, this supplement establishes the ontological and informational constraints governing epistemic stability in decision-making agents.</p> <p dir="ltr">We show that policy collapse under partial observability is structurally unavoidable under finite informational constraints, and cannot be eliminated by improved training procedures, optimization methods, or increased model capacity. Under a minimal set of decision-theoretic axioms, we prove the existence of a critical informational bound, denoted αcrit, below which no purely inductive policy can guarantee long-term viability. This bound arises necessarily from the interaction between finite observation channels and viability-preserving state distinctions, rather than from empirical or algorithmic limitations.</p> <p dir="ltr">We further show that asymptotic stability under partial observability requires a qualitative architectural transition, in which control shifts from inductive inference to non-inductive (structural) regulation once informational conditions fall below this bound. This necessity is formalized as the αcrit Principle of Epistemic Viability (PEV), which characterizes a fundamental constraint on the persistence of decision-making agents operating under informational limits.</p> <p dir="ltr">Together, these results establish a non-normative, information-theoretic account of epistemic stability grounded in agent viability, providing necessary structural conditions for the design of decision architectures capable of operating safely under partial observability.</p> <p dir="ltr"> </p>
title A Conditional Framework for Policy Collapse under Information Limits
topic Information Theory
Epistemic Stability
Partial Observability
Decision Theory
Policy Collapse
Hybrid Decision Architectures
url https://doi.org/10.5281/zenodo.19076073