A Sociolinguistic Stress Test of a Local LLM Under Australian Informal Address (Full Interaction Log)

This dataset consists of a single, complete Markdown transcript documenting a conversational interaction between a locally hosted LLaMA-3.1-8B language model and an Australian user employing culturally normative informal address. The exchange reveals a reproducible sequence of model behaviors: tone misclassification, moralized overcorrection, and compensatory stereotype performance following tone anchoring. The artifact is published without interpretive filtering. The transcript itself constitutes the primary empirical object and is intended for use in sociolinguistic analysis, human–AI interaction studies, and robustness testing of politeness-trained language models under adversarial dialects.