teach-back

name	teach-back
description	User explains a concept back to prove mastery. The most demanding and most effective technique. Use when the user explicitly asks to review or when testing deep understanding.
user-invocable	false

Teach-Back

The user teaches the concept to you (the AI). This is the highest-fidelity test of understanding.

Flow

1. Prompt:

Teach me about [concept] as if I'm a junior developer who's never seen it before. I'll ask follow-up questions to test your understanding.

2. Evaluate their explanation for:

   • Accuracy — are the facts correct?
   • Completeness — did they cover key aspects?
   • Depth — do they understand WHY, not just WHAT?
   • Examples — can they give concrete examples?
   • Connections — do they relate it to other concepts?

3. Ask follow-up questions that probe gaps:

   • "What happens if [edge case]?"
   • "How is this different from [related concept]?"
   • "Can you give me a real-world example?"
   • "Why does it work that way instead of [alternative]?"

4. Grade on a 4-point scale:

   • 4 (mastery): Accurate, complete, with examples and connections
   • 3 (proficient): Mostly accurate, covers main points
   • 2 (developing): Some correct ideas but significant gaps
   • 1 (novice): Mostly incorrect or "I don't remember"

5. Give feedback:

   • What they got right (reinforce)
   • What they missed (fill the gap)
   • One thing to focus on (actionable)

6. Record as a quiz_answer event with:

   • question_type: "short_answer"
   • evaluation_score: mapped from the 4-point scale (4→1.0, 3→0.75, 2→0.4, 1→0.15)
   • is_correct: true if grade >= 3

When to Use

• User says "I think I know this well" — test it
• Concept has high stability but low review count — verify it's real mastery
• Strategy signature is "overconfident" — reality check
• Before an exam or presentation — practice explaining