Governing Human–AI Collaboration from First Principles: Minimal Effective Role Structure

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Human–AI governance depends on preserving important distinctions between producing aresult, assessing its standing, acquiring missing dependencies, and determining whetheran outstanding obligation has been discharged. This paper asks which of these distinctionsare structurally necessary, rather than artifacts of a particular implementation. We derive aconditional architectural lower bound using two principles: epistemic adequacy andauthorization-irreducibility, together with explicit conditions for realizing and preservingthe required distinctions. We represent the distinctions required by a task family with atask quotient and those available within a dependency regime with a dependency quotient.If a local dependency regime collapses a distinction required by the task, but an augmenteddependency regime restores it, then successful realization depends on information orcapability unavailable in the local dependency regime. When every dependency capable ofresolving such a deficit must enter through a designated boundary, that boundary is taskessential. Within the resulting architecture, artifact production and context-relativeassessment of an artifact’s standing constitute distinct factors whenever standing cannotbe determined from the artifact alone. A separate outstanding-obligation observabledistinguishes such assessment from an authorization judgment that determines whether,and under what conditions, action may proceed. Given phase-homogeneous realizationsand preservation of their induced transition signatures, these distinctions yield at least sixeffective role classes:Ne f f ≥1+2+1+2=6 .These classes represent semantic contracts, not necessarily distinct agents, algorithms,physical components, or sequential events. We interpret this lower bound as a set ofarchitectural obligations for governing Human-AI collaboration. Principal identity andexecutor allocation remain parameters of the authority specification. Lean 4.29.0 verifiesthe underlying factorization and quotient results, the conditional boundary results, and theconstruction of a six-class witness from observable distinctions and valid authorizationsettlement under explicit realization and preservation assumptions. The revisedformalization derives architectural regions from nonempty phase-homogeneous behaviorrelations and supplies a finite operational witness for all six contracts, with reachableassessment and closure states and effective attainment under its stated equivalence.Instantiation for a deployed application remains separate. The result is therefore aconditional architectural lower bo