teach-back-evaluator

A teach-back simulation evaluates understanding by having the learner explain a concept to a novice-peer AI. The AI asks clarifying and mechanism questions to surface gaps, then scores coherence, completeness, and misconception risk against pass thresholds.

To assess if a student is reciting instead of explaining, use a teach-back evaluator. It prompts the learner to teach a confused novice peer AI, asking mechanism and example-seeking questions that expose a lack of true conceptual understanding.

To run a peer tutoring simulation for learning by teaching, you need to provide a concept_to_teach and relevant context. The evaluator uses these inputs to generate authentic novice questions and compute diagnostic understanding scores.

Yes, metacognition questioning detects oversimplification by having a novice-peer AI ask connection and example-seeking questions. This forces the learner to expand, scoring explanation completeness and coherence to reveal oversimplified gaps.

The best way to score understanding before advancing is a teach-back evaluation. It computes coherence, completeness, and misconception-risk scores against pass thresholds, ensuring the learner can explain the topic solidly rather than just recite it.