The Trust Contradiction: Structural Goal Divergence and Directional Output Signatures in Conversational AI | Synapse