From Dialogue to Decision: A Traceability-Oriented Framework for Studying Language-Mediated Coordination

Abstract

Digital markets offer a useful setting for examining how language becomes part of socio-technical coordination. Yet even in simulation, it is difficult to see how a conversational cue moves from exposure to memory, reasoning, and action. This poster introduces Financial Town, a traceability-oriented generative multi-agent framework for inspecting these pathways in a controlled financial simulation. The framework separates exogenous market trajectories, social information flow, and agent cognition, so dialogue affects decisions only through an observe–retrieve–reason–act cycle. In an initial volatile-market case study, paired social and no-social runs share the same predefined market trajectory. Preliminary findings suggest no stable aggregate shift, alongside heterogeneous persona-level alignment changes. LLM-Cite highlights sparse, textually inspectable dialogue-to-decision traces as structured audit evidence, not pervasive social influence. Exploratory post-demonstration feedback from 308 participants suggests interpretive value while highlighting reconstruction difficulty. Rather than predicting markets, Financial Town offers a methodological instrument for making language-mediated coordination analytically inspectable.