Fig. 1: An illustrative patient-level evaluation workflow for LLM-supported MAT.

Here, a Spanish discharge summary is followed by five open-ended questions covering key clinical concepts that will be answered by the patient. Each question has a “gold-standard” answer generated from the source English note, allowing evaluators to compare the patient’s responses with the intended information. This approach measures both the understandability and actionability of the translation by assessing how well patients grasp critical details from the LLM-generated text.