From Control to Interpretability: Autonomy, Structural Inheritance, and the Reversal of Evaluative Direction in Advanced AI Systems | Synapse