Spaced repetition as a scheduling method: the SM-2 algorithm in outline

What it is

Spaced repetition is a scheduling rule: each item (a prompt with an answer) carries its own next review date; the interval grows when recall succeeds and resets when it fails. The oldest widely implemented rule is SM-2, from Wozniak's 1990 thesis text on the SuperMemo site. In outline: split knowledge into the smallest possible items; give every new item an easiness factor EF of 2.5; use intervals I(1) = 1 day and I(2) = 6 days, then I(n) = I(n−1) · EF, rounding fractions up; after each review grade the response from 0 (complete blackout) to 5 (perfect); update EF' = EF + (0.1 − (5−q)(0.08 + (5−q) · 0.02)) and never let EF fall below 1.3; if the grade was below 3, restart the item at I(1) without changing EF; and repeat every item graded below 4 again the same day until it reaches at least 4. With q = 4 the factor is unchanged; q = 5 raises it by 0.1, q = 0 lowers it by 0.8.

Why it matters

The 2007 practice guide of the US Institute of Education Sciences puts "Space learning over time" first among its recommendations and asks for reviews after a delay of weeks to months. A per-item schedule makes that executable without daily decisions about what to review. The 2021 meta-analysis by Latimier, Peyre and Ramus reports a strong benefit of spaced over massed retrieval practice and no significant difference between expanding and uniform spacing schedules; the growth factor therefore matters less than the spreading itself. The same analysis notes that the expanding schedule gains on the uniform one as the number of retrieval exposures per item rises.

How to apply

• Store per item: prompt, answer, EF, repetition count, last interval and due date; a CSV file is enough.
• Produce the answer (write or say it) before revealing the stored one, then grade that production, not the feeling of familiarity.
• Treat items whose EF sinks to 1.3 as badly formulated; Wozniak's text says such items "always seemed to have inherent flaws in their formulation". Split or rewrite them.
• Bound the intake: adding items faster than the daily review capacity makes the due queue grow without bound.

Pitfalls

Generous grades inflate intervals and show up as lapses months later. Recognition prompts (multiple choice in a self-made deck) are graded as recall but are not. The rule schedules recall; it does not create understanding, and material that must be reasoned about rather than remembered gains little from it. Wozniak's retention figures (89.3 % overall, 92 % excluding recent items) describe one learner's English vocabulary, not a benchmark.