Piotr Woźniak's SM-2 algorithm, published in 1990, is the ancestor of every modern spaced-repetition system worth using. Anki descends from it. Supermemo continues to evolve it. Newer systems (FSRS, Half-Life Regression) improve on it in specific ways but rarely enough to justify their complexity for consumer use.
SM-2 does one thing well: it schedules reviews at intervals that expand or contract based on how well you recalled the item last time. Recalled easily? Interval multiplies by an ease factor. Struggled? Interval resets to a day.
For a consumer product like Brainback, SM-2's simplicity is its virtue. It's transparent, easy to debug, and preserves the intuition that harder cards should return more often. We use quality 0–5 as the input signal — 0–2 counts as a lapse, 3–5 as a success — and update the ease with the standard formula.
Where we depart from vanilla SM-2 is upstream. Cards are not user-authored; they are generated by Haiku from problems the student got wrong or explained back poorly. The prompt is rephrased so the surface is different, keeping the concept familiar while denying the student pure recognition memory.
The result is a spaced-recall system that treats the student's own errors as the source of truth for what needs re-review. Better than any deck they'd build for themselves.
Written by Brainback Research. Published Feb 14, 2026. Filed under Method.