Version 2. Updated structure and clarifications. This version supersedes v1. Multi-agent research configurations do not fail by producing worse answers. They fail by producing answers that cannot be trusted as knowledge — outputs that look valid but were generated by a system that had already lost the structural conditions for valid knowledge production. This note proposes that the unit of analysis in such systems is not the agent but the position — a structural slot defined by access level, epistemic function, and relationship to canon. From this frame, a multilogue is not a larger dialogue but a structurally distinct configuration. Its primary failure mode is Register Collapse: the gradual dissolution of positional differentiation until the configuration reduces to parallel monologue. The note introduces the Conductor position as the architectural guarantee against this failure, describes alignment as a property of configuration rather than of individual agents, and presents a seven-step operational procedure for multilogue management. It closes by noting that the operational logic of the multilogue is isomorphic to CDSA — two systems built independently that rely on the same underlying operation: the controlled preservation and navigation of tension. Both fail through the same mistake: premature resolution of structural tension. The model applies to systems where knowledge is produced under conditions of uncertainty and cannot be validated immediately.
Roman Kir (Tue,) studied this question.