---
name: spaced_repetition_systems
description: Build a spaced-repetition deck that survives past week two. Atomic card design, cloze deletions, scheduler setup (FSRS/SM-2/Leitner), interval tuning, avoiding ease hell, and a 15-minute daily review habit. Triggers when someone asks to memorize anything with flashcards, build a study deck, tune Anki settings, or fix a review backlog.
---

# Spaced Repetition Systems

## Purpose
Turn raw material into a deck that actually gets reviewed. Most decks die in week two for three fixable reasons: cards are badly written (two ideas per card, wordy fronts), the scheduler is misconfigured (ease hell), and the daily load is unsustainable. This skill fixes all three in order: card design first, scheduler second, habit third.

## Workflow
1. **Scope the deck.** Ask what the material is and why it must be recallable. If the user doesn't understand the material yet, do not make cards: summarize and teach first. Rule one of Wozniak's 20 rules: never learn what you don't understand. See `references/card-formulation.md`.
2. **Draft cards atomic.** One idea per card. Minimum information principle: each card tests exactly one fact, one mechanism, one name. Front = one sharp question; back = the smallest complete answer. See `references/card-formulation.md` for the cloze-deletion and interference rules.
3. **Work a sample out loud.** Show 5-10 sample cards before making the full deck. Wait for approval on the format, then produce the rest. Never hand over a 200-card deck sight unseen.
4. **Set up the scheduler.** Pick one: FSRS for real decks (built into current Anki; or implement from the spec), SM-2 for simplicity, Leitner boxes for tiny/short-term decks. Defaults and tuning in `references/scheduling.md`. The correct scheduler choice beats any per-card tweak.
5. **Fix ease hell if it's already happening.** If the deck runs SM-2 (Anki's classic mode) and reviews are piling up at the 130% ease floor, migrate to FSRS or apply the pass/fail discipline in `references/scheduling.md`. Do not tell the user to press Easy to fix scheduling; that just trades one scheduling error for another.
6. **Cap the daily habit.** New cards per day: start at 5-10, raise only when reviews stay under 15 minutes. Reviews before new cards, every day, same slot. Suspending leeches beats grinding them. Full protocol in `references/habit.md`.
7. **Check weekly.** Retention on mature cards should land near the target (default 90%). Consistently below: cards are badly formulated, reformulate before touching scheduler settings. Consistently above with ballooning reviews: lower the retention target, don't delete cards.

## Output Contract
Deliver the deck as plain text in a portable card format (one card per block: `Front:`, `Back:`, `Tags:`), plus a one-page setup note: scheduler choice, the 3 settings you changed from defaults and why, new-cards-per-day cap, leech policy. Everything in the note must be traceable to `references/` or a primary source.

## Operating Rules
- One idea per card, always. A card that regularly gets "half right" is two cards; split it.
- Never invent facts for cards. Cards come only from source material the user provided or that you verified. Unverifiable = no card.
- Cloze deletion for facts and names; full Q/A for mechanisms, contrasts, and "when to use" decisions.
- Near-duplicate cards are the main source of confusion: give every card a context cue (subject prefix or tag) and a front that differs sharply from its neighbors.
- Don't touch the scheduler's advanced parameters until card formulation is clean. Ninety percent of "the algorithm isn't working" is bad cards.
- New cards are a budget, not a goal. Missing a day's new cards is fine; missing reviews two days in a row is a backlog event: deal with it via `references/habit.md`, not by doubling up.
- No em dashes in anything this skill outputs. Plain sentences.