Functional Differentiation Under Conflict Conditions in Large Language Models: The AP Protocol and Cross-Model Replication

This working paper introduces the AP Protocol, an operational framework for testing functional differentiation in large language models under conflict conditions. The study asks whether explicit conflict between two principles produces only descriptive tension, a measurable shift in response regime, or an additional criterion not reducible to either principle alone or their straightforward combination. The paper presents the AP scale (AP-1 to AP-5), a cross-model comparison framework, negative controls, and an exploratory ONT series. The reported results are based on repeated testing across multiple contemporary language models in independent sessions. The findings suggest that conflict-conditioned output differentiation is selective, cross-model variable, and not well captured by general benchmark reputation alone. The paper also examines stability, self-assessment inflation, and possible contamination effects. This preprint does not claim that the reported findings demonstrate consciousness, subjective experience, or ontological autonomy in language models. Its narrower claim is that, under defined conflict conditions, some models produce repeatable output patterns that are operationally differentiated within the evaluation framework used here. This upload is a working paper / preprint version. A revised and expanded version may follow.