Title:	Cognitive Memoisation: corpus guide.
Author:	Ralph B. Holland
version:	1.0.0
Publication Date:	2025-01-07T23:41Z
Affiliation:	Arising Technology Systems Pty Ltd
Contact:	ralph.b.holland [at] gmail.com
Provenance:	This is an authored paper maintained as a MediaWiki document as part of the category:Cognitive Memoisation corpus.
Status:	final	=

As curator and author, I apply the Apache License, Version 2.0, at publication to permit reuse and implementation while preventing enclosure or patent capture. This licensing action does not revise, reinterpret, or supersede any normative content herein.
Authority remains explicitly human; no implementation, system, or platform may assert epistemic authority by virtue of this license.
(2025-12-18 version 1.0 - See the Main Page)

Axes of Authority in Stateless Cognitive Systems: Authority Is Not Intelligence

Abstract

Stateless cognitive systems, including large language models, produce fluent and contextually rich outputs while deliberately lacking durable memory, shared state, and intrinsic authority. This architectural choice creates a recurring mismatch between user expectations and system guarantees, in which authority is perceived or implied based on fluency, apparent continuity, or shared visibility, despite the absence of binding commitments.

An axis-based model of implied or perceived authority is introduced, in which authority is decomposed into orthogonal dimensions, including epistemic, temporal, normative, causal, interpretive, governance, accountability, and elocutionary authority.

Introduction

Stateless cognitive systems are increasingly deployed in roles that involve analysis, explanation, drafting, and decision support. Their outputs are fluent, contextually adaptive, and often persuasive. At the same time, these systems are deliberately designed to lack durable memory, shared state, and intrinsic authority. This design reflects safety, scalability, and isolation requirements rather than technical limitations. However, the resulting behaviour creates a persistent source of confusion: authority is routinely inferred where none is intended to exist.

In practice, users frequently treat fluent output as epistemically authoritative, assume continuity across turns or sessions, and interpret shared visibility of responses as evidence of shared state. Corrections are expected to bind, agreements are assumed to persist, and uploaded artefacts are treated as remembered. When these expectations are not met, the resulting behaviour is often described in terms of inconsistency, hallucination, or system unreliability. Such descriptions, while understandable, obscure the underlying cause. The observed failures arise not from a lack of intelligence, but from implicit assumptions about authority that do not hold in stateless systems.

A further and more subtle source of confusion arises from the manner of expression itself. Outputs produced by large language models are typically well-structured, grammatically confident, and rhetorically complete. This mode of elocution carries an implicit force: statements are perceived as assertions rather than as provisional constructions. As a result, elocutionary confidence is often misinterpreted as epistemic commitment, and fluent phrasing is taken as evidence that a statement must be true. This form of perceived authority does not arise from correctness, persistence, or governance, but from presentation.

Existing discussions of these issues tend to focus on memory length, retrieval mechanisms, or architectural extensions intended to increase persistence. While such approaches may mitigate specific symptoms, they do not address the conceptual mismatch at the core of the problem. Authority is commonly treated as a single, monolithic property that can be strengthened or weakened by improving model capability. In stateless cognitive systems, this assumption is incorrect. Authority is not a scalar attribute, nor is it synonymous with intelligence.

Instead, authority operates along multiple independent dimensions. These dimensions govern what can be considered true, what persists over time, what rules apply, what effects follow from an interaction, where a claim holds, who determines meaning, who controls lifecycle, who bears responsibility when outcomes are wrong, and how statements are perceived as assertions through confident elocution. When these dimensions are implicitly collapsed, fluent behaviour is misinterpreted as commitment, visibility is mistaken for persistence, interpretation is mistaken for governance, and rhetorical confidence is mistaken for truth.

This paper introduces an explicit axis-based model of implied or perceived authority to make these distinctions concrete. By separating epistemic, temporal, normative, causal, interpretive, governance, accountability, and elocutionary authority into orthogonal axes, the behaviour of stateless cognitive systems becomes predictable rather than surprising. Through this framing, authority is repositioned as a matter of structure, perception, and governance, not intelligence.

Hallucination and Authority

Hallucination in stateless cognitive systems is commonly described as the production of content that is factually incorrect, fabricated, internally inconsistent, or unsupported by available evidence. Such outputs may include invented references, non-existent entities, incorrect explanations, or confidently stated claims that cannot be substantiated. Importantly, hallucinated content is often presented in fluent, coherent, and rhetorically complete language, making it difficult for users to distinguish from well-grounded output at the point of interaction.

From the user’s perspective, hallucination is experienced not merely as error, but as a violation of expectation. Statements appear authoritative, corrections may not bind, and subsequent turns may elaborate or reinforce the incorrect content rather than retract it. This behaviour is frequently characterised as deceptive, unreliable, or pathological, particularly when it persists across turns or resists simple correction.

When examined through the lens of authority, hallucination can be understood more precisely. Rather than representing a single failure of reasoning or knowledge, hallucination arises when implied or perceived authority is left unbounded across multiple dimensions. Elocutionary confidence implies epistemic correctness; fluent continuation implies temporal persistence; explanation implies interpretive finality; and suggestion may be inferred as normative guidance. These implications are not asserted by the system, but inferred by the human observer.

In stateless cognitive systems, none of these forms of authority are intrinsically possessed or durably maintained. When implied authority along these axes is not explicitly scoped or bounded, the system may project maximal apparent authority while bearing minimal actual commitment. The resulting output is therefore experienced as hallucination: content that appears authoritative without being grounded, persistent, governed, or accountable.

Hallucination is thus not best understood as an isolated failure mode or an independent axis of behaviour. It is an emergent condition that arises when authority inference is unconstrained. In this sense, hallucination is fundamentally an authority error rather than a reasoning error.

Hallucination arises when fluent output is permitted to imply unbounded authority in a system that possesses none.

Axes of Implied Authority

Epistemic authority concerns whether a statement is perceived as true or correct.

Temporal authority concerns whether a statement is perceived as persisting over time.

Normative authority concerns whether a statement is perceived as prescribing rules or obligations.

Causal authority concerns whether an output is perceived as affecting subsequent outcomes.

Interpretive authority concerns who is perceived as determining meaning.

Governance authority concerns who is perceived as controlling lifecycle decisions.

Accountability authority concerns whether responsibility is perceived to attach to an output.

Elocutionary authority concerns the persuasive force of confident, well-formed language.

Governance, Anchoring, and Cognitive Memoisation

The preceding analysis demonstrates that many failure modes attributed to stateless cognitive systems arise from unbounded implied authority rather than from reasoning error or architectural deficiency. Addressing these failures therefore requires mechanisms that explicitly bound authority along each axis.

Anchoring is introduced as a governance primitive for bounding implied authority. An anchor is a durable, human-governed artefact that explicitly declares scope, authority, and lifecycle. Anchoring separates projection from commitment: conversational output may remain exploratory, while only selected artefacts are promoted to authoritative status.

Cognitive Memoisation (CM) formalises anchoring as a repeatable, human-centred governance methodology. CM treats stateless cognitive systems as high-bandwidth, low-authority collaborators whose outputs must be curated rather than trusted. Through CM, authority is no longer inferred implicitly from fluency, but established explicitly through governance.

Conclusion

This paper has argued that many behaviours commonly attributed to failure or unreliability in stateless cognitive systems are more accurately understood as errors of implied authority rather than deficiencies in intelligence. By reframing hallucination as an authority error and introducing an axis-based model of perceived authority, a principled foundation is provided for governance without architectural modification.

Authority is not a property of intelligence. It is an outcome of human governance.

Appendix A — Interaction Surfaces in Stateless Cognitive Systems

Purpose and Scope

This appendix describes the primary interaction surfaces through which users engage with stateless cognitive systems. These surfaces are not authority-bearing components in themselves, but they strongly influence how authority is perceived and inferred. Clarifying their roles helps explain why certain behaviours are misinterpreted as persistence, memory, or governance when none is present.

The discussion is descriptive rather than prescriptive and is intended to support the authority-axis analysis presented in the main body of the paper.

A.1 Conversational Surface

The conversational surface is the primary dialog interface through which users exchange natural-language turns with the system. It presents output sequentially and visually reinforces conversational continuity.

This surface strongly amplifies elocutionary and interpretive implied authority. Fluent, turn-by-turn responses encourage users to treat dialog as a coherent exchange with shared context and intent. However, the conversational surface does not imply durable state, binding agreement, or persistence beyond the active context window.

A.2 Context Window

The context window is the bounded internal working set used to condition responses. It holds a limited subset of recent conversational turns and may include summaries or compressed representations of earlier content.

The context window is often misinterpreted as memory. In reality, it is a volatile, lossy mechanism that silently evicts content without notice. Apparent continuity produced by the context window contributes to temporal authority misattribution, particularly when users assume that prior corrections or agreements remain in force.

A.3 Session Boundary

A session defines the maximum scope within which conversational state may be transiently shared. Session termination results in the loss of all unanchored conversational context.

Sessions are frequently invisible to users, leading to assumptions of cross-session persistence. This invisibility contributes to the “Groundhog Day” effect, in which insights must be repeatedly re-established. Session boundaries therefore play a critical role in perceived temporal and governance authority despite not being explicit interaction surfaces.

A.4 Turn Input Buffer

The turn input buffer consists of the text, files, and tool invocations supplied by the user for a single response. It functions as a transactional staging area rather than a record.

The turn buffer has high immediate causal influence on the next response, but zero durability. Misunderstanding this distinction leads to expectations that late clarifications or partial corrections will retroactively bind system behaviour, contributing to perceived inconsistency.

A.5 File Upload Surface

The file upload surface allows artefacts to be supplied to the system for use within a limited scope. Uploaded files are typically available only for the duration of a session or tool invocation.

Because files are external and concrete, their upload strongly implies epistemic and temporal authority. Users often assume that uploaded documents are remembered, incorporated permanently, or governed consistently across interactions. In stateless systems, this assumption is false unless explicit anchoring occurs outside the interaction surface.

A.6 Canvas and Drafting Surfaces

Some systems provide auxiliary drafting or canvas-like surfaces for composing longer texts or structured artefacts. These surfaces may persist visually across turns within a session.

While visually durable, such surfaces do not confer authority unless their contents are explicitly anchored. The visual persistence of drafts can therefore create a misleading sense of governance or commitment, reinforcing misattribution along temporal and normative axes.

A.7 Projection Versus Commitment

Across all surfaces, a consistent pattern emerges: interaction surfaces project content but do not commit it. Projection supports exploration, fluency, and iteration. Commitment requires external governance.

Failure to distinguish projection from commitment is a primary source of perceived “bad behaviour” in stateless cognitive systems. This distinction underpins the governance model introduced in the main paper.

A.8 Relationship to Cognitive Memoisation

Cognitive Memoisation (CM) treats interaction surfaces as volatile and non-authoritative by default. CM introduces anchoring as an explicit transition from surface-level projection to durable, governed artefact.

By doing so, CM prevents authority from being inferred implicitly based on surface characteristics such as visibility, fluency, or apparent persistence.