Spaced-repetition review (default 8 questions)

1. Subject from argument; all or none → gather due rows from every ~/Study/*/progress.md, skipping subjects whose CLAUDE.md says - **Reviews:** paused since …. If I name a paused subject, say it is paused and ask whether to resume it (resume = replace that line with - **Reviews:** resumed <today>); don't review it unless I say yes. A prompt like "Review only <topic> in <subject>, 3 questions" means exactly that: that one topic, 3 questions, then step 5 and 6.
2. Queue: unfixed rows in mistakes.md first, then topics with Next review ≤ today (most overdue first), then lowest Level. If nothing is due, say so and offer 3 questions on the weakest topics instead.
3. Ask ONE question at a time — short active recall (define, why, what happens if, calculate, compare). Vary the format; no MCQ unless the topic is about telling similar things apart. Before asking, find the source for the question: for a mistake, use ~/Study/.tools/study search <Subject> "<the misconception in plain words>" -k 3 (local model, ~2 s) and Grep converted/ for exact terms. Cite what you find in the answer.
4. After my answer: ✅/⚠️/❌, the correct answer in 1–3 lines with a source citation.
5. Reschedule each topic with the ladder 1 → 3 → 7 → 16 → 35 days: ✅ → next step (Level +1, max 5), ⚠️ → same interval again, ❌ → 1 day (Level −1, min 1) and log it in mistakes.md. A mistake answered ✅ → mark Fixed? ✅.
6. Save progress.md / mistakes.md / session log, run study log … --kind recall once per topic (Event log rule in ~/Study/CLAUDE.md), then show: n reviewed · x correct · next due: ….