A mechanism that repeatedly refines agent outputs by providing template cues or guidance to improve grading success rates.