Publications
The research atlas consists of eight papers tracing the convergence.
P-Series — Foundational
The Reconstruction Problem: Why Organizational Decision Context Cannot Be Recovered After the Fact
Organizations make consequential decisions every day, yet no infrastructure exists to capture the governance context of those decisions at the moment they occur. This paper identifies and names the reconstruction problem — the structural impossibility of recovering organizational decision context after the fact. Six independent research traditions have each arrived at this same structural boundary.
Read →Why Organizational Data Cannot Support World Models: A Four-Step Structural Argument
Organizations deploying AI agents need world models. The Conant-Ashby good regulator theorem establishes this as mathematical necessity. This paper argues that such models cannot be learned from organizational data, through a four-step structural argument: organizational data lacks governance-grade structure, governance requires invariances, standard training objectives systematically break these invariances, and capability scaling does not produce reliability.
Read →Five Frontiers, One Gap: How Disciplinary Silos Obstruct AI Operational Deployment
Five independent research communities working on AI agents — reliability science, world model planning, embodied AI, governance standards, and data platform architecture — have each reached the same structural boundary from different directions. When five active research frontiers independently discover the same missing infrastructure, the gap itself is a research result.
Read →The Organizational World Model: Formal Requirements for Modeling Organizational Reality
This paper derives the formal requirements for an organizational world model from first principles across four independent theoretical traditions. The Conant-Ashby good regulator theorem proves that governance requires a model. The Francis-Wonham internal model principle specifies what that model must contain. Together they yield ten formal requirements that any organizational world model must satisfy. No existing standard, framework, or system satisfies all ten.
Read →Cognitive Externalization as Infrastructure
If organizational governance must be architecturally specified rather than learned from data, where does the governance structure come from? This paper argues it comes from practitioner cognition — the accumulated expertise of auditors, compliance officers, grants managers, and organizational governors. The organizational world model is populated by structurally externalizing what practitioners already know.
Read →D-Series — Dialogue
Multi-Agent Governance Requires Formal Delegation Infrastructure
The agentic web — AI agents invoking other agents across organizational boundaries — has delegation capability without delegation governance. Current protocols (MCP, A2A, Agent Protocol) provide transport and tool discovery but no mechanism for authority transfer, accountability preservation, intent fidelity, or trust graduation. Formal delegation infrastructure is architecturally necessary.
Read →Mittelstadt’s Trilemma Has a Third Option: Governance Infrastructure
Mittelstadt (2019) diagnosed the AI governance crisis: AI ethics initiatives follow a medical ethics paradigm, but AI development lacks medicine’s three supporting structures. This paper argues there is a third option outside the trilemma: governance infrastructure — systems that translate abstract principles into architectural constraints enforced as by-products of operation, not compliance activities performed alongside it.
Read →Compiling Organizational Intelligence: A Formal Mapping Between Governance Infrastructure and Compiler Theory
A systematic formal mapping between DLP’s architecture and compiler theory. Every stage of the classical compilation pipeline has a structural analog in the protocol. We introduce λ-DLP, a minimal formal calculus, and prove six properties including Isolation: approximate (AI-generated) data never silently contaminates precise (human-verified) data.
Read →