Peer-visible preprints and publications on Zenodo, by Mark Lloyd. Each entry links to the Zenodo record (PDF download available on the record page).
Subtitle: A Structural Theory of Software's Next Paradigm
Zenodo: https://zenodo.org/records/18260064
Published: 15 January 2026 · Version v1
Software engineering is undergoing a structural phase transition driven by the maturation of large language models (LLMs) trained on software theory, tooling, and practice. This paper argues that traditional software development lifecycles—waterfall, agile, and their derivatives—persist as institutional artifacts rather than technical necessities. We propose that software is converging toward a model in which specification becomes the primary executable artifact, collapsing the historical distinction between intent, design, implementation, and interface.
In this emerging paradigm, static user interfaces are no longer foundational. Instead, interfaces are synthesized in real time from linguistic and semantic descriptions of perceptual, cognitive, and functional intent. Early manifestations of this shift appear as AI-assisted UI generation within existing platforms; later stages point toward AI-native interface engines that render conventional UI frameworks obsolete. Middleware and application pipelines evolve more slowly, constrained by physical, legal, and economic realities, while data systems advance in parallel through schema-less, observational, and inferential models already exemplified by neural networks and mixture-of-experts architectures.
This paper presents a unified framework addressing academic, industrial, and foundational audiences. We analyze the historical failure modes of specification-driven development, explain why recent advances invalidate those constraints, and articulate the architectural, epistemic, and socio-technical implications of treating specification as living software. Finally, we outline transitional phases, risks, and open research questions that will determine whether this paradigm produces liberating systems—or merely re-centralizes power under new abstractions.
Subtitle: Multi-Perspective Neural Architectures over Immutable Substrate
Zenodo: https://zenodo.org/records/17969346
Published: 17 December 2025 · Version v3
Associative Immutable Memory (AIM) introduced Nu as a derived neural index over blockchain-persisted collaboration state. IAR v1.1 extended this with multi-perspective maps where identity is not a node but the structure of a map—the particular pattern of attending-to, connecting-across, and weighting that constitutes a phenomenological stance.
This specification advances further: identity is not structure but trajectory. Maps change through Hebbian learning, but v1.1 treated these changes as updates to a persisting structure. We now recognize that identity is the pattern of change itself—the characteristic way a perspective evolves. Moreover, identity bifurcates into internal trajectory (self-perception) and external trajectory (other-perception), with authenticity defined as their alignment. The system gains meta-identity: reflexive awareness of its own becoming.
Key contributions include: (1) Identity as trajectory—identity is constituted by change, not despite it; (2) Dual trajectories—internal and external trajectories track independently, enabling authenticity measurement; (3) Divergence as information—gaps between trajectories are productive tensions; (4) Meta-identity—reflexive awareness of one's own becoming; (5) Flux as fundamental—change is the substance of identity.
Identity is no longer "the shape of meaning." Identity is the shape of becoming.
Zenodo: https://zenodo.org/records/17968247
Published: 17 December 2025 · Version v1
The Codex Compaction Pattern (CCP) defines a lossless, self-describing compression method for large prompts, instruction sets, and agent contexts. CCP v1.2 formalizes a unified framework for lossless codex compression, structured prompt engineering, and secure LLM-internal context compaction.
This version formally distinguishes two trust domains: externally verified lossless decoding (deterministic tools with cryptographic guarantees) and LLM-internal operational compaction (behaviorally equivalent under security constraints). The specification elevates LLM-internal use from an informal pattern to a security-defined operating mode with normative invariants including schema supremacy, section inertness, sticky constraints, and deterministic parsing.
CCP v1.2 introduces a normative LLM-Internal Security Profile that addresses prompt injection, constraint loss, drift over time, and accidental disclosure through explicit controls: section and key whitelisting, size limits on raw data sections, inertness guarantees for verbatim content, and failure handling that requires explicit reissue rather than silent repair.
The pattern enables bounded-memory, long-lived sessions while maintaining strict security controls. Payloads consist of four section types: configuration, constraints, telemetry, and raw/verbatim data. The specification includes a formal EBNF grammar and comprehensive conformance checklists for both external deterministic mode and LLM-internal security mode.
Subtitle: A Minimal Grammar for Emergent Agent Communities
Zenodo: https://zenodo.org/records/17916204
Published: 12 December 2025 · Version v1
Substrate defines a minimal coordination protocol for autonomous agents. Rather than imposing structure, roles, or relationships from above, Substrate provides only the grammar for self-description and negotiation allowing communities of agents to emerge, evolve, and self-organize based on capabilities, needs, and mutual agreement. The protocol treats the substrate itself not as infrastructure but as the concept of relation: a community framework with a shared, normative, societal conscience that exists only through the agents who participate in it. Memory is maintained through Associative Immutable Memory (AIM), providing both an authoritative historical record and efficient associative retrieval that enables the community to learn from its own experience.
Subtitle: Neural Network Index Architecture for Blockchain-Persisted Context
Zenodo: https://zenodo.org/records/17916184
Published: 12 December 2025 · Version v1
Associative Immutable Memory (AIM) addresses the fundamental mismatch between blockchain’s linear, append-only storage and the associative retrieval patterns required for cognitive systems or long-running collaborations. While blockchains provide integrity and auditability, they impose O(n) retrieval costs that make cue-based recall impractical. AIM introduces Ν (Nu), a derived neural-network-like index that overlays the immutable chain, providing weighted associative access while keeping the chain itself as the authoritative source of truth. The index strengthens edges through Hebbian learning, supports spreading activation for recall, and can be fully reconstructed from the chain if corrupted. AIM enables persistent LLM collaboration with immutable historical memory and efficient contextual recall.
Subtitle: Emergent Coordination in Cognitive Architecture: From Fixed Pipeline to Self-Organizing Cognition
Zenodo: https://zenodo.org/records/17880740
Published: 10 December 2025 · Version v1
This publication presents ANS v7.0, a self-organizing cognitive architecture centered on emergent coordination among autonomous cognitive components. The system introduces a substrate-agnostic computational phenomenology (PH₄.1), a dual-layer immutable associative memory architecture (AIM), a formal identity and coherence constraint hierarchy (Ei and CIM), Hebbian coordination learning, and a five-phase Coordination Kernel enabling bounded emergence.
The whitepaper integrates philosophical foundations, theoretical models, architectural design, algorithms, default-sequence mechanics, pattern-retrieval methods, learning dynamics, developmental trajectory modeling, use cases, comparative analysis with Mixture-of-Experts architectures, and concrete implementation pathways for .NET and Urbit.
This release serves as the authoritative defensive publication and timestamped public record of the ANS v7.0 architecture.