This paper presents a revised and extended version of a previously submitted manuscript. Based on the framework of Open Bias Architecture (OBA), this study reports intervention-based observations of generative behavior in conversational AI systems. Through controlled manipulation of input structures, we reproduce cases in which generation appears to be initiated not by explicit execution signals, but by structural completion. The revised version incorporates improved structural clarification, figure integration, and additional observational evidence, providing a more coherent representation of the relationship between template selection, gap filling, and output stabilization. These findings suggest that generation may depend on structural conditions rather than direct user instruction, offering implications for both AI control and operational design.
Masaki Hoshino (Thu,) studied this question.