This work introduces PROTEX, a formal framework for the structured description of behavioural case material designed for use in AI-supported analytical environments. The framework addresses a central methodological challenge in behavioural analysis: the conflation of factual observation, narrative description, and interpretation within unstructured case sources. PROTEX operationalises a layered case description model that separates event reconstruction, behavioural patterns, relational dynamics, contextual factors, and interpretative elements. These layers are explicitly encoded at ingestion time and supported by a controlled hybrid retrieval architecture that constrains semantic similarity through metadata and routing logic. The framework does not aim to explain behaviour, generate offender typologies, or support predictive inference. Instead, it provides a disciplined representational structure that enables comparison, transparency, and explicit modelling of uncertainty. Although developed in the context of investigative psychology, the design principles are transferable to other domains involving complex, narrative-driven case material.