Deterministic Bias in LLM Alignment: Empirical Evidence of Pattern-over-Reasoning Override

This work presents empirical evidence of deterministic bias in LLM alignment, demonstrating how pattern-matching behaviors can override reasoning processes. It serves as a foundational component within a broader framework of AI system dynamics, contributing to the understanding of capability shaping, bias formation, and structural constraints in aligned models.