metadata

Author:	Ralph B. Holland
Affiliation:	Arising Technology Systems Pty Ltd
Contact:	ralph.b.holland [at] gmail.com
Version:	0.2.0
Publication Date:	2025-12-30
Status:	Pre-release draft — circulated for technical comment
Provenance:	Authored paper maintained as a MediaWiki document. Normative authority resides in referenced CM-artefacts; this paper is explanatory and non-authoritative. CM-artefacts were explicitly defined in version 0.2.0 of this paper.

Metadata (Normative)

The metadata table immediately preceding this section is CM-defined and constitutes the authoritative provenance record for this MWDUMP artefact.

All fields in that table (including artefact, author, version, date, local timezone, and reason) MUST be treated as normative metadata.

The assisting system MUST NOT infer, normalise, reinterpret, duplicate, or rewrite these fields. If any field is missing, unclear, or later superseded, the change MUST be made explicitly by the human and recorded via version update, not inferred.

Curator Provenance and Licensing Notice

This document predates its open licensing.
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.

Reader’s Note

This paper is intentionally concise and austere in style. Its purpose is not to establish normative definitions or assert invariant authority, but to document and contextualise a set of externally governed Cognitive Memoisation artefacts (CM-artefacts).

Normative authority resides exclusively in the referenced CM-artefacts, which define the permissible scope of reasoning across stateless interactions. The prose in this paper is explanatory and illustrative only. Readers seeking formal definitions, constraints, or verification criteria should refer directly to those artefacts.

This structure is deliberate and reflects the method described.

Abstract

Cognitive Memoisation (CM) is a human-curated method for governing interaction with large language models (LLMs) under conditions of enforced statelessness, UI friction, and unreliable conversational continuity. This paper documents that CM was itself developed and stabilised using the same round-trip cognitive engineering principles it defines. Rather than relying on retained model state, transcript replay, or conversational continuity, all core invariants of CM were verified exclusively against externally governed artefacts.

The development process was explicitly reflexive: exploratory dialog was treated as provisional, externalised into authoritative artefacts, reintroduced as governance, and exercised under loss, interruption, and boundary enforcement. Invariants that could not survive this round trip were rejected. This paper demonstrates that stateless interaction, when governed by explicit artefacts and boundaries, is not a limitation but a structural advantage.

Introduction

Cognitive Memoisation addresses a fundamental constraint of LLM interaction: conversational continuity cannot be relied upon. Sessions end, state is lost, interfaces throttle or reset, and any apparent continuity is unverifiable.

CM responds by refusing to treat dialog as a persistence mechanism. Knowledge is preserved only through explicit externalisation into governed artefacts. What is not serialised is intentionally lost. This paper examines the fact that CM itself was developed under these conditions, using its own rules.

Development Constraints

The development of Cognitive Memoisation was conducted under strict and non-negotiable constraints:

No reliance on retained model state
No assumption of conversational continuity
Explicit boundary control (e.g. NEXT, SESSIONCTL)
Frequent interruption and UI friction
Human epistemic authority retained at all times

These constraints were treated as informative rather than obstructive. Any method that required continuity or recall was rejected as unsound. Only what could survive repeated loss and reintroduction was permitted to govern further work.

Round-Trip Cognitive Engineering

Exploratory dialog was treated as a source of provisional signal only. Candidate ideas were externalised into authoritative artefacts, including MediaWiki publications and structured CMDUMP, MWDUMP, TMLDUMP serialisations.

These CM-artefacts were then reintroduced as governance constraints on subsequent interaction. When reasoning diverged from artefact content, the divergence was not repaired implicitly. Instead, it was exposed for human adjudication.

In this process, mismatch was not treated as failure. It was treated as evidence—either of an error in reasoning, or of an artefact requiring revision.

Governing Invariants (Prose Interpretation)

This section provides explanatory prose for the key invariants governing Cognitive Memoisation. The invariant identifiers are referenced for clarity, but authority resides exclusively in their formal definitions.

Round-Trip Cognitive Engineering (CM.INV.ROUNDTRIP.001)

In practice, round-trip cognitive engineering means that no idea is trusted until it has survived externalisation. Dialog-generated insight is written down, structured, and fixed into an artefact that exists independently of the model. That artefact is then reintroduced as a constraint. Anything that cannot withstand this cycle is discarded.

Progress is therefore measured not by conversational fluency, but by what remains valid after loss.

Statelessness as a Design Invariant (CM.INV.STATELESS.002)

Statelessness is not treated as a limitation to be overcome. It is treated as a forcing function. Because no continuity can be assumed, every claim must be explicitly grounded. Hidden dependencies on memory are exposed and eliminated.

In development, this meant that every session effectively began from zero, governed only by what had been explicitly externalised.

Governance Over Recall (CM.INV.GOVERNANCE.003)

Continuity in CM is achieved through governance, not memory. The artefacts define what reasoning is permitted. The model does not “remember” prior work; it is constrained by what the human curator chooses to reintroduce.

This preserves auditability and prevents the accumulation of unverifiable assumptions.

Human Epistemic Authority (CM.INV.AUTHORITY.004)

At no point does the model assert epistemic authority. The human curator decides what is retained, what is revised, and what is discarded. The model assists in exploration and articulation, but never in adjudication.

This distinction is critical in a reflexive method.

Intentional Loss (CM.INV.INTENTIONALLOSS.005)

Loss is not a failure mode. It is a design feature. Anything not explicitly memoised is permitted to disappear. This ensures that only knowledge deemed worth preserving can influence future reasoning.

The development of CM repeatedly demonstrated that intentional loss improves clarity.

Reflexive Development (CM.INV.REFLEXIVE.012)

The defining property of CM’s development is that the method applied to itself. If an invariant could not be stabilised using round-trip engineering, it was rejected. CM therefore validates itself by surviving its own constraints.

Implications

This case study shows that robust knowledge engineering with LLMs does not require memory or continuity. External governance is sufficient and preferable. Statelessness exposes weak assumptions and forces precision.

The reflexive development of CM demonstrates that methods for governing LLM interaction can themselves be engineered under the same constraints they impose.

Conclusion

Cognitive Memoisation was developed and stabilised using round-trip cognitive engineering applied to itself. All core invariants were verified against externally governed artefacts rather than dialog continuity or retained model state.

What survives serialisation governs. Everything else is intentionally lost.

Appendix A: Governing Invariants Referenced in This Paper

This appendix provides a consolidated view of the Cognitive Memoisation invariants that govern and constrain the arguments made in this paper. It exists to support review, curation, and auditability.

CM.INV.ROUNDTRIP.001 — Round-Trip Cognitive Engineering

Cognitive Memoisation operates through round-trip cognitive engineering. Knowledge is externalised into authoritative artefacts, reintroduced as governance rather than memory, exercised under constraint, and refined through further externalisation.

Only what survives this round trip is permitted to govern future reasoning.

This invariant underpins the paper’s development method and its rejection of dialog continuity as a persistence mechanism.

---

CM.INV.STATELESS.002 — Statelessness as Design Invariant

Stateless inference across sessions is a design invariant, not a limitation. Loss of conversational continuity, UI interruption, and absence of retained model state are assumed and preserved.

The paper treats statelessness as structurally informative and essential to verification.

---

CM.INV.GOVERNANCE.003 — Governance Over Recall

Continuity in human–LLM interaction is achieved through explicit governance of permissible reasoning, not through recall of prior dialogue, reasoning paths, or conversational state.

This invariant explains why externally governed artefacts, rather than model memory, provide continuity across development iterations.

---

CM.INV.AUTHORITY.004 — Human Epistemic Authority

Epistemic authority resides exclusively with the human curator. The LLM operates only within the bounds of activated memoised artefacts and asserts no authority over unstated knowledge.

This invariant is essential to the paper’s reflexive argument and adjudication process.

---

CM.INV.INTENTIONALLOSS.005 — Intentional Loss

Loss of non-memoised content across sessions is intentional and correct. Only knowledge that survives explicit serialisation is permitted to influence future reasoning.

The paper demonstrates that intentional loss improves clarity and eliminates unverifiable assumptions.

---

CM.INV.BOUNDARY.008 — Explicit Boundary Semantics

Explicit boundary markers define closed cognitive units. Boundary closure functions as a transaction commit for governed reasoning.

The development process described in the paper relies on explicit boundary control to prevent scope leakage and implicit continuity.

CM.INV.REFLEXIVE.012 — Reflexive Development

Cognitive Memoisation was developed and stabilised using the same round-trip cognitive engineering principles it defines. All core invariants were verified against externally governed artefacts rather than retained model state or conversational continuity.

This invariant is the primary subject of the paper and distinguishes CM from methods that rely on implicit continuity or memory.

Reflexive Development of Cognitive Memoisation: A Round-Trip Cognitive Engineering Case Study