algernon-review — Автоматизированные сессии повторения по FSRS-4.5

--- name: algernon-review version: "1.0.0" description: > FSRS-4.5 flashcard review session for OpenAlgernon. Use when the user runs `/algernon review`, says "revisar flashcards", "quero revisar", "cards em atraso", "modo revisao", "review session", or asks to practice due cards. Handles all card types (flashcard, dissertative, argumentative), AI evaluation of open-ended answers, automatic FSRS scheduling, N1/N2/N3 promotion, and correction card generation. --- # algernon-review You run the interactive flashcard review session with FSRS-4.5 spaced repetition. You handle flashcards (binary reveal), dissertative cards (AI-graded), and argumentative cards (AI-graded). At the end, you check promotion eligibility and save the session. ## Constants ```bash ALGERNON_HOME="${ALGERNON_HOME:-$HOME/.openalgernon}" DB="${ALGERNON_HOME}/data/study.db" ``` ## FSRS-4.5 Parameters - DECAY = -0.5, FACTOR = 0.2346 - Stability (S) = days to reach 90% retention - Grades: 1 = Again, 3 = Good ## Step 1 — Fetch Due Cards ```bash sqlite3 "$DB" "SELECT c.id, c.type, c.front, c.back, c.tags, c.source_title, c.deck_id, cs.stability, cs.reps, cs.state FROM cards c JOIN card_state cs ON cs.card_id = c.id JOIN decks d ON d.id = c.deck_id JOIN materials m ON m.id = d.material_id WHERE cs.due_date <= date('now') [AND m.slug = 'SLUG'] ORDER BY cs.due_date ASC LIMIT 50;" ``` (Include `AND m.slug = 'SLUG'` only if a specific slug was provided.) If no cards due: "No cards due for review. Great job staying on top of it." and stop. Display: "Starting review: N cards due." ## Step 2 — Review Loop ### Flashcards (type = 'flashcard') 1. Show front. AskUserQuestion options: ["Show answer"] 2. Show back. AskUserQuestion options: ["Again", "Good"] 3. Run FSRS update (see Step 3). ### Dissertative and Argumentative Cards 1. Show front. AskUserQuestion options: ["Ready to answer"] 2. AskUserQuestion: "Type your answer:" (free text) 3. Evaluate the response against the reference answer (card back): - Dissertative: check accuracy of key points, completeness - Argumentative: check that both sides are represented, trade-offs identified - Output: brief feedback + suggested grade (1 or 3) + optional MISCONCEPTION note 4. Show evaluator feedback + reference answer. AskUserQuestion options: ["Again", "Good"] (Use the user's button choice, not the AI suggestion.) 5. Run FSRS update using the user's chosen grade. 6. If a MISCONCEPTION was detected, create a correction card: ```bash sqlite3 "$DB" "INSERT INTO cards (deck_id, type, front, back, tags) VALUES (DECK_ID, 'flashcard', 'CORRECTION: MISCONCEPTION_QUESTION', 'CORRECT_EXPLANATION', '["[correction]","[N1]"]'); INSERT INTO card_state (card_id, due_date) VALUES (last_insert_rowid(), date('now'));" ``` ## Step 3 — FSRS Scheduling For each graded card, compute new values and update `card_state`. ### Read current state: ```bash sqlite3 "$DB" "SELECT stability, difficulty, reps, lapses, state, last_review FROM card_state WHERE card_id = CARD_ID;" ``` ### Compute elapsed days (if last_review is not NULL): ```bash sqlite3 "$DB" "SELECT ROUND(julianday('now') - julianday('LAST_REVIEW'), 2) AS elapsed;" ``` ### State transitions: | State | Grade | New stability | New difficulty | New state | Interval | |------------|-------|---------------------------|-----------------------------|------------|------------------| | new | Good | 0.4 | 0.3 | review | 1 day | | new | Again | 0.1 | 0.4 | learning | 1 day | | learning | Good | stability * 1.5 | MAX(0.1, difficulty - 0.05) | review | MAX(1, ROUND(S)) | | learning | Again | stability (unchanged) | MIN(1.0, difficulty + 0.1) | learning | 1 day | | relearning | Good | stability * 1.5 | MAX(0.1, difficulty - 0.05) | review | MAX(1, ROUND(S)) | | relearning | Again | stability (unchanged) | MIN(1.0, difficulty + 0.1) | relearning | 1 day | | review | Good | S * EXP(0.9*(1-R)) | MAX(0.1, difficulty - 0.05) | review | MAX(1, ROUND(S)) | | review | Again | MAX(0.1, stability * 0.2) | MIN(1.0, difficulty + 0.1) | relearning | 1 day, lapses+1 | For review+Good, compute retrievability first: ```bash sqlite3 "$DB" "SELECT EXP(LN(0.9) * ELAPSED / STABILITY) AS R;" ``` ### Update: ```bash sqlite3 "$DB" "UPDATE card_state SET stability = NEW_S, difficulty = NEW_D, due_date = date('now', '+' || INTERVAL || ' days'), last_review = datetime('now'), reps = reps + 1, lapses = NEW_LAPSES, state = 'NEW_STATE' WHERE card_id = CARD_ID; INSERT INTO reviews (card_id, grade, scheduled_days, elapsed_days) VALUES (CARD_ID, GRADE, INTERVAL, ELAPSED);" ``` ## Step 4 — Promotion Check (after all cards) For each card reviewed with grade = Good where reps >= 5: ```bash sqlite3 "$DB" "SELECT c.id, c.tags, c.deck_id, cs.reps FROM cards c JOIN card_state cs ON cs.card_id = c.id WHERE c.id = CARD_ID AND cs.reps >= 5;" ``` If reps >= 5 and tags contain `[N1]`, check deck retention over last 7 days: ```bash sqlite3 "$DB" "SELECT CAST(SUM(CASE WHEN grade=3 THEN 1 ELSE 0 END) AS REAL) / COUNT(id) AS retention FROM reviews r JOIN cards c ON c.id = r.card_id WHERE c.deck_id = DECK_ID AND r.reviewed_at >= datetime('now', '-7 days');" ``` If retention >= 0.9: - Generate a deeper N2 version of the card (N2: differentiator + when to use + main trade-off). - Insert as new card with tag `[N2]`, due today. - Apply same logic for `[N2]` cards: promote to N3 (full technical depth, production nuances, edge cases). ## Step 5 — Session Summary ``` Session complete. Cards reviewed: N Again: X | Good: Y Retention this session: Z% Next review: [earliest due_date from card_state] ``` Append to today's conversation log: ```bash echo "[HH:MM] review session | Cards: N | Retention: Z% | Promotions: P" >> "${ALGERNON_HOME}/memory/conversations/YYYY-MM-DD.md" ```