Prompt-Type Phenomenology

This paper presents a pre-registered experimental design for testing whether different modes of engagement with large language models produce measurably different activation dynamics at the meta-dimensional level. Using the Cognitive Weather Observatory (Temte 2026o), the experiment compares four prompt categories (direct task instruction, collaborative intellectual engagement, extended roleplay requiring multi-agent modelling, and introspective discussion about the nature of consciousness) across a battery of meta-dimensional measures: rate-of-change magnitude, second-order phase-transition patterns, inter-meta-dimensional correlation structure, and cumulative activation dynamics. The central question is whether how one engages a system changes what the system does computationally, not merely what it outputs. A positive result would constitute empirical evidence that engagement mode modulates computational character, with direct implications for frameworks that treat the relationship between human and AI as the relevant unit of analysis. The design specifies prompt categories, corpus construction, the analysis pipeline, pre-registered hypotheses, and confirmatory and disconfirmatory criteria, prior to data collection.