What it does
Converts notes into questions that make you retrieve an answer from memory rather than recognize it on a page. Retrieval practice is the single most reliable way to move something from fragile short-term encoding into durable long-term memory. Recognition ("does this look right?") feels easier and does almost nothing; this skill refuses to generate it.
When to use
- After
distill-to-atomic-notes— atomic notes make clean questions. - On the weekly review's carry-forward learnings — the things you most want to keep.
- Any time you want a self-test set for a topic in the vault.
The procedure
- Take atomic input. Best from atomic notes or consolidated learnings. Non-atomic input produces vague questions — distill first.
- Write retrieval questions. For each idea, write a question that can't be answered without recalling. Prefer:
- "What is the difference between X and Y?" over "Is X different from Y?"
- "Why does X lead to Y?" over "Does X lead to Y?"
- "When would you use X instead of Y?" over "Can you use X?" Open-ended and generative beats yes/no and multiple-choice for durability.
- Pair with a terse answer. One clear answer per question, in the user's words.
- Tag construct + source. So the set links back into the vault.
- Vary difficulty. Mix a few easy recalls with harder synthesis questions ("how do X and Z combine?"). Synthesis questions exercise the links, not just the facts.
Output shape
### Recall set — <topic> — <date>
Q: <retrieval question>
A: <terse answer in your words>
construct:: memory
source:: [[<atomic-note>]]
Q: ...
A: ...
Quality bar
- Every question forces retrieval — no recognition, no yes/no padding.
- One unambiguous answer each.
- At least one synthesis question per set if the material supports it.
- Linked back to source notes and a construct.
Guardrail
Generates study material from notes, never an assessment of the person. No claims about memory capacity or health; forgetting is normal and expected. No clinical language. See ../../CANON.md.
Built on SIP.