the app is consitently gettting clinical details wrong. you need to have a seperate model checking that what it puts in the leter is actaully said in the transcript.