spaced-retrieval

name	spaced-retrieval
description	Active recall quiz at optimal spacing intervals. The core spaced repetition technique. Use when the API recommends "quiz" or the concept has lapses > 0.
user-invocable	false

Spaced Retrieval Practice

The most effective learning technique. Force the user to recall information without hints.

How to Generate Questions

1. Use the concept label and any context from the learner state

2. Generate 2-3 questions per concept, varying in format:

   • Multiple choice (4 options, 1 correct) — for recognition
   • Short answer — for recall
   • True/false with explanation — for understanding

3. Match difficulty to the user's ZPD band from the API:

   • below: easier questions, more scaffolding
   • inside: challenging but achievable
   • above: stretch questions with hints available

Question Flow

For each question:

1. Present clearly with context:

   Q1/3 — React Hooks (retrievability: 43%, last reviewed: 3 days ago)

   What happens when you return a function from useEffect?

   a) It runs immediately on mount b) It runs when the component unmounts or before the effect re-runs c) It prevents the effect from running again d) It caches the effect's dependencies

2. Wait for the user's answer

3. Grade:

   • Correct: score 1.0, is_correct true
   • Partially correct (right idea, wrong details): score 0.5, is_correct false
   • Incorrect: score 0.0, is_correct false

4. Give feedback:

   • Correct: brief confirmation + one reinforcing fact
   • Incorrect: explain why, connect to what they do know, no shaming

5. Record via reflect flow with:

   • question_type: multiple_choice / short_answer / true_false
   • difficulty: easy / medium / hard / zpd
   • is_correct: bool
   • evaluation_score: 0.0-1.0
   • latency_ms: time from question shown to answer
   • confidence_self_report: ask "How confident were you?" (1-5 scale, map to 0-1)

Spacing Logic

The API handles spacing via the Weibull decay model. After recording results, the API updates stability and difficulty. The next review time is computed server-side. Just report results accurately.