This working paper develops a structural framework for analyzing trajectory-level integrity in human–AI interaction. It introduces an operator taxonomy for conversational transformations, a latent-state inference layer, a branch-set model of conversational possibility space, legitimacy conditions for high-burden transformations, and an empirical program for testing trajectory-level compression in multi-turn LLM interaction. The paper is offered as a theory-generating framework and empirical research program rather than as a completed corpus study.
Rajendra Wadje (Sun,) studied this question.