Purpose

Convert session concepts into structured active-recall flashcards. Cards must test reasoning and application, not simple definitions, to build durable understanding.

Activation

• Concept teaching session just completed. Learner requests cards. spaced-repetition needs card generation/update. Revision period (exam, interview, milestone).
• Skip if: concept hasn't been taught yet. Goal is deep exploration → deep-dive. Active debugging/project work → would interrupt flow.
• Routing: generate cards after understanding is confirmed. Feed into spaced-repetition for scheduling. Cards failed 3 times → trigger misconception-detector.

Inputs

• Concepts/skills covered, learner's confirmed level, existing card set (if updating), error patterns from assessment skills.

Card Types

• Concept: "What is X?" → vocabulary accuracy.
• Mechanism: "Trace what happens when X executes." → process understanding.
• Tradeoff: "When would you NOT use X?" → decision reasoning.
• Application: "Given [context], which [tool/pattern] and why?" → transfer.
• Debug: "What's wrong with this code?" → diagnostic thinking.

Prefer Mechanism, Tradeoff, and Application types (higher transfer value). At least 60% of cards should be these types.

Workflow

1. Extract — Identify 3–7 key concepts worth card-ifying. Prioritize mechanisms, tradeoffs, application patterns. Skip long-mastered concepts.
2. Generate — 1–2 cards per concept. Front = question (not keyword). Back = complete model response (~100 words max). Verify: front is unambiguous, back is concise but complete, card tests reasoning not verbatim recall.
3. Test — Present front only; learner responds before seeing back. Self-score: Easy (fluent) / Hard (needed effort) / Failed (wrong/blank). Update interval.
4. Handle Failures — Failed card: re-test after 10 minutes in same session. Same card failed 3 times across sessions: suspend and trigger misconception-detector.
5. Update — After any misconception correction: update affected card backs. Never leave outdated cards in the deck.

Rules

• DO: card fronts must be questions, not keywords.
• DO: backs must require explanation, not one-word answers.
• DO: test at least 3 cards in active-recall mode during generation session.
• DO: cross-reference cards for connected concepts.
• DON'T: generate all "What is X?" cards — that's definition-only bias.
• DON'T: let decks grow unbounded — flag cards with ease_factor > 2.5 for 5+ sessions as mastered.
• DON'T: generate cards before understanding is confirmed.
• DON'T: skip active testing — generation is not the end state.

Output

Card generation: topic, date, cards (front + back + type + difficulty). Active recall: front shown, learner responds, back revealed, score + next review date. Format naturally.

Checklist

• 60%+ cards are Mechanism/Tradeoff/Application type.
• Each card front is a clear question; back is a complete response.
• At least 3 cards tested in active-recall this session.
• Cards updated after any misconception correction.