---
name: "teach_back_method"
description: "Run the teach-back learning loop: explain a concept from memory out loud, find the exact sentence where you stumble, re-study only that gap, and log each pass so progress is trackable. Triggers when a learner wants to test or deepen understanding, is stuck in rereading notes, or says 'I think I get it but can't explain it'."
metadata: { "includeInPrompt": true }
---

# Teach-Back Method

## Purpose
A study protocol for one concept at a time. The learner teaches the concept out loud, from memory, in plain words. The point where the explanation breaks is the point where understanding is missing. Fill that gap, teach again, and log each pass to watch understanding grow over weeks. For the research evidence behind each step, see `references/evidence.md`.

## Workflow

1. **Bound one concept.** Not a chapter. Name the concept and name the audience you will teach (a friend, a smart 12-year-old, a colleague). Narrow is what makes the gap visible.

2. **Explain from memory, out loud, closed book.** No notes, no peeking. Plain words, logical order. If you read from notes, that is copying, not learning; the closed-book pass is what forces retrieval.

3. **Mark the stumbles.** Note each place you:
   - stall on a sentence you cannot finish,
   - use a term you cannot define in plain words,
   - skip a step with "it just follows", or
   - lean on jargon to sound right.
   Stop at the first undefined term and unpack it before moving on. These marks are the gaps.

4. **Re-study only the gaps.** Go back to the source material for the marked points, nothing else. Then explain from memory again.

5. **Give an analogy and break it.** State the analogy, then state one place it fails. A useful analogy has to preserve the real relationship; finding its boundary proves you understand the structure, not the story.

6. **Transfer check.** Apply the concept to a new case with different surface details. Explain which features matter and which do not.

7. **Log the pass.** Record concept, audience, pass number, gaps found, hints or re-study used, and a 1-5 confidence score before and after. Schedule the next teach-back. `bin/teachback.py` handles the log and the review schedule:

```sh
teachback.py start "photosynthesis" --audience "smart 12-year-old"
teachback.py log "photosynthesis" --gaps "can't explain why water splitting happens first" --before 2 --after 4
teachback.py due
```

## Output Contract
- A recorded explanation pass (audio transcript or written text), in the learner's own words.
- A dated gap list: the exact stumbles, what re-study filled each, what remains shaky.
- A confidence delta (before/after 1-5) per pass, and the next review date.

## Operating Rules
1. Closed book during the explain step. Open-book explanation is a copying task, not a learning task.
2. One concept per pass. Whole chapters hide gaps instead of exposing them.
3. At least two to three passes. The first pass exposes surface gaps; deeper passes expose missing causal links. One pass is never enough.
4. Never accept jargon as an answer. "Mitochondria produce energy" is a label, not an explanation; ask what that means in plain words.
5. Record only what the learner actually produced. Do not summarize away their stumbles.
6. Spaced reviews, increasing depth: day 1 micro (one-sentence explain), day 3 structured pass, day 7 new-angle pass, day 14 transfer to a novel scenario, day 30 full teach-back.
7. Know when this is the wrong tool: procedural/motor skills need deliberate practice; syntax and vocabulary need spaced repetition; trivial concepts and genuinely time-constrained situations do not repay the cost.
8. No confident unverified claims about effectiveness. Cite the evidence in `references/evidence.md` honestly: the named "Feynman Technique" is folklore, but its mechanisms (self-explanation, retrieval practice, the expectation of teaching) are well replicated.