From Derivation to Instrument

This preprint presents the development of a semiotic/hermeneutic framework for evaluating large language model (LLM) outputs through four successive rubric designs. Beginning with a theory-derived instrument grounded in semiotics and hermeneutics, the paper documents the evolution toward an empirically curated evaluation framework featuring atomic scoring, independent semiotic and hermeneutic assessment, four hard-gate criteria, a selectable criterion library, and a pre-registered holistic "Gut Check" designed to operationalize the hermeneutic circle. The work is intended as a contribution to LLM evaluation methodology, computational hermeneutics, and human-AI communication research.