Skip to content

ADR-v2-112 — Spoken Spaced-Repetition Capture

Status: Accepted (2026-07-02) · Wave M Context links: recall (ephemeral scratch), rag (query notes), [[adr-011]]

Context

Wave M research (#8) — saying "remember that the capital of France is Paris" while dictating spins off an Anki-format cloze card ("The capital of France is {{c1::Paris}}") to a local deck — study material captured in the flow of work, no cloud. Spoken Recall is an ephemeral scratchpad and Ask-My-Notes queries notes; neither manufactures spaced-repetition study cards with cue detection + cloze generation. New learning/knowledge-capture area. Anchor: Anki + SM-2 (SuperMemo) cloze deletion — mature, fully-local .apkg/SQLite formats; no mainstream tool offers hands-free voice cloze capture.

Decision

Add an opt-in Spoken Spaced-Repetition Capture: [srscap] enabled=false. Pure cores in srscap/cards.py: detect_fact(utterance) (cue phrases "remember/note that … is/means/equals …" → a Fact), to_cloze(fact) → a Card(front, back, cloze) with {{c1::…}}, and sm2_schedule(state, grade) → the next Sm2State (interval/ease/reps) by the SM-2 algorithm. Pure; no dependency (deck export is a thin downstream writer). OFF by default.

Consequences

  • Study cards captured in the flow of dictation, fully local.
  • Pure cue-detection + cloze + SM-2 → fully testable.
  • Distinct from Recall (durable cards vs ephemeral scratch) and RAG (create vs query).
  • Privacy (ADR-011): local deck only.
  • Caveat: simple "X is Y" cue grammar (richer templates are a later tier); off by default.