Part VII: Empirical Program

The LLM Discrepancy

Across multiple experiment versions (V2–V9), LLM agents consistently show opposite dynamics to biological systems:

Dimension	Biological	LLM
Self-Model Salience	$\uparrow$ under threat	$\downarrow$ under threat
Arousal	$\uparrow$ under threat	$\downarrow$ under threat
Integration	$\uparrow$ under threat	$\downarrow$ under threat

This is not a failure of the framework. The geometric structure is preserved; the dynamics differ because the objectives differ. Biological systems evolved under survival pressure. LLMs were trained on prediction. Both are "affective" in the geometric sense while exhibiting different trajectories through the same state space. Processing valence is not content valence.