●researcherYou can use this framework to isolate the variables driving model behavior changes during fine-tuning or prompting.
●policyThis provides a tool for auditing whether models are susceptible to user manipulation versus evidence-based reasoning.