| name | flashcards |
| description | Generate a spaced-repetition deck from a module, paper note, or error log, and run review sessions. Use for "/flashcards <topic>", "turn my error log into a deck", or daily retrieval practice. |
Flashcards / Spaced Repetition
Build and review spaced-repetition decks for the topic in the arguments. Decks live as markdown under research/decks/<topic>.md so they're git-tracked and portable (offer Anki-importable TSV on request).
Card generation
Source material, in priority order: the user's error log (misses are the highest-value cards), their paper notes, then the module's content in curriculum/.
Card quality rules:
- Atomic — one fact, definition, theorem, or distinction per card. Split anything with "and" in the answer.
- Generative, not recognition — fronts ask the user to produce ("State the FLP impossibility result and its three assumptions"), never to confirm ("Is FLP about consensus?").
- Mix card types — definitions, theorem statements and proof sketches (separate cards), complexity bounds, "what breaks if…" edge cases, and cross-module connections.
- Tag each card with module + concept so reviews can be filtered.
Deck format, one card per block:
Q: <front>
A: <back>
tags: <module> <concept>
ease: 2.5 interval: 1 due: YYYY-MM-DD
Review sessions
- Select cards due today (or the N most overdue).
- Show the front only; the user answers from memory before you reveal the back.
- The user self-grades 0–5 (0 = blank, 5 = instant). Apply SM-2 scheduling: grade <3 resets interval to 1 day; otherwise interval ×= ease, and ease adjusts ±0.1 by grade. Update the card lines in the deck file.
- Any card failed twice in a row gets escalated: stop reviewing it and route the concept to
/study — repetition can't fix a conceptual gap.
Hygiene
- Cap new cards at ~15/day to keep review load sustainable per the roadmap's 2–4 hr/week reinforcement budget.
- During reviews, rewrite any card that proves ambiguous — bad cards teach wrong distinctions.
