Download PDF View Markdown source
Current AI-mediated coordination work accumulates reasoning, decisions, and rationale across sessions and participants, but most of that coordination state is lost at session boundaries, fragmented across documents, tickets, and chat logs, or embedded implicitly in vendor-specific product schemas. This paper defines the Coordination Knowledge Substrate (CKS) — a design pattern in which coordination knowledge, governance semantics, and conflict-handling logic live in a persistent, human-governed substrate outside the LLM — and defends it through six architectural commitments: hybrid substrate-LLM division of labor, conflict preservation as first-class substrate state, human-governed authority, AI as substrate mediator, tool-agnosticism at the substrate layer, and linear-cost scaling inherited from database-backed external memory.
The core theory is defended through five claims drawing on systematic review of 2024–2026 literature across governance-first AI architectures, epistemically-structured substrates, and execution-substrate research. A sixth claim extends the core theory to human-AI-human coordination: the same substrate that supports a single user coordinating across sessions supports multiple roles coordinating across sessions, with AI mediating over the substrate rather than in per-session interactions. The paper demonstrates both the core theory and the extension through one proof-of-concept in a regulated industry, instantiated in commodity spreadsheet infrastructure with a commercially available LLM integration.
The contribution is a design pattern, not an empirical evaluation. CKS articulates a recurring structure of coordination problem and architectural response — named so that subsequent work can adopt it, vary it, compose it with adjacent patterns, or argue against it. The paper positions CKS as one contribution to an emerging research direction in CSCW-AI, governance-first AI architectures, and epistemically-structured substrates.
This paper is part of a trilogy on the Coordination Knowledge Substrate (CKS) pattern. See the main paper index for the full corpus, including foundational concept papers and operational deep-dive treatments.
Cite as:
Li, W. (2026). Coordination Outside the Model: A Human-Governed Substrate Pattern for AI Systems. Coordination Architecture Theory Papers.
https://liwenxinxin.github.io/coordination-architecture-theory-papers/papers/paper-1.html