Guide

How SM-2 spaced repetition actually works

SM-2 is the scheduling algorithm introduced in 1987 by Piotr Woźniak for SuperMemo 2 and later made famous by Anki. It decides one thing: when you should see a flashcard again. This guide explains the math behind it — no marketing, just how it works and why it works.

The problem: the forgetting curve

In the 1880s Hermann Ebbinghaus measured how quickly memorized material decays. His finding, replicated many times since: retention drops steeply within the first day, then levels off. Without review you typically lose the majority of newly learned material within days.

The counterintuitive part: each time you successfully recall something, the decay slows down. Review at the right moment — just before you would forget — and the memory becomes more durable than before. Spaced repetition automates the search for that "right moment" so you spend the fewest possible reviews keeping material in memory.

What SM-2 tracks per card

SM-2 keeps three numbers for every flashcard:

• Repetition count (n) — how many successful reviews in a row.
• Interval (I) — days until the card appears again.
• Ease factor (EF) — how easy this card is for you, starting at 2.5 and never dropping below 1.3.

The rating scale: 0 to 5

Internally, each review produces a quality score q from 0 (total blackout) to 5 (perfect, instant recall). IntelMemorAI's four buttons map onto that scale:

What a review does to the interval

First, the rule for scheduling: the first two successful reviews get fixed intervals, after that each interval multiplies by the ease factor:

Second, the review updates the ease factor itself:

A "Good" rating (q=4) leaves EF unchanged. "Easy" (q=5) adds +0.1, so the card's intervals grow faster forever. "Hard" (q=3) subtracts 0.14, permanently slowing that card's growth. And if you fail (q < 4 in the original formulation; IntelMemorAI's "Again"), the repetition count resets to zero — the card goes back to day 1 and must be relearned, and the ease factor drops.

A worked 30-day example

Take one card, EF starting at 2.5, and assume you rate it "Good" every time. The schedule unfolds like this:

In 30 days you reviewed the card exactly 4 times, and the next review isn't due until day 61. Now the same card if you had rated it "Easy" (q=5) at review 2: EF becomes 2.6, so review 3 lands at round(6 × 2.6) = 16 days instead of 15, and the gaps keep compounding. And if you press "Again" on day 8: the count resets, the card reappears the next day, and its EF drops by 0.2 — future intervals will grow more slowly for the life of that card.

Why this beats re-reading and cramming

Two well-established findings from memory research do the heavy lifting. The testing effect: retrieving an answer strengthens memory far more than re-reading it, even when retrieval feels harder. And the spacing effect: the same reviews spread over weeks beat the same reviews crammed into one evening. SM-2 combines both — every card is an active recall attempt, scheduled at the longest interval you can still survive.

How IntelMemorAI applies it

IntelMemorAI implements SM-2 exactly as described above and removes the two chores that make people quit Anki: writing cards and planning reviews. Upload a PDF, drop a YouTube link (captions not required), or type a topic — the AI generates the cards, a quiz, and a 7-day study plan, and SM-2 takes over scheduling from your first review. Your mastery per card is tracked so you can see which material is actually sticking.