Checks whether an LLM response is grounded in the source context. Flags contradictory sentences and generates a corrective prompt.