Governed State in AI Research Systems

This release presents Article 32 (P32) from the Reflexive Laboratory research program, a transcript-sufficient approach to AI-assisted scientific inquiry. The paper introduces a bounded classification criterion for AI research systems based on state architecture. Recent systems increasingly preserve large bodies of source material, including linked documents, knowledge bases, audit trails, and graph-structured memory layers. These developments produce stronger forms of managed memory, but do not by themselves yield governed current state. Building on the Reflexive Laboratory series, the paper defines three system classes: managedₘemory: systems whose highest-order control surface remains archival and retrieval-based; partialₛtate: systems with an explicit current-state surface that lack one or more required governance conditions; governedₛtate: systems that maintain an explicit working-state architecture with evidence mediation, enacted authority, admissibility-gated update, re-entry, and state sufficiency. The paper contributes a minimal definition set, a five-condition criterion schema, a practical decision test, and a worked comparison case showing why advanced compiled-memory systems remain managed memory rather than governed state. The broader claim is methodological. Transcript-sufficient research systems require more than persistent memory. They require an explicit operational layer that determines what is authoritative now, how that status is established, and how it may change without severing its relation to the retained archive. This release includes: the final publication-ready manuscript (PDF) ; manuscript source files (Markdown, DOCX, TeX) ; figures and tables; structured comparison and criterion tables; supporting transcript materials documenting the construction of the comparison case; release metadata, manifest, and checksums. The goal of the release is both to present a classification result and to demonstrate a research workflow in which the paper and its production trace are preserved as a unified research object.