Dataset for contextualized divergence detection

Dataset for contextualized divergence detection containing the following. 1. Authoritative Medication GuidelinesThe MedDocs/ directory contains the authoritative medication guideline documents. These documents serve as the primary clinical reference source from which advice is extracted and synthesized.2. Train and Test SplitsThe dataset is divided into two primary splits: train — 2,110 samplestest — 8,435 samples Each split contains structured samples used for model development and evaluation.3. User Profile DefinitionWithin each data file, the following fields together define the user profile: AgeGenderMedical ConditionCurrent Medications The Current Medications field directly references the relevant medication documents in MedDocs, grounding each user profile in specific authoritative sources.4. Raw Extracted AdviceEach sample also includes the raw advice text extracted from the medication document. This raw advice is used during dataset construction to control the generation of the synthetic Reference Advice. It is provided for transparency and traceability. It is not used during inference.5. Synthetic Reference AdviceThe Synthetic Reference Advice represents the controlled, generated advice instance associated with the persona. It is derived from the authoritative medication documents.6. Divergence LabelThe Synth Type field serves as the label indicating the type of divergence present in the synthetic advice. This functions as the primary classification target for downstream modeling tasks.