Spaced Repetition: The Science-Backed Guide to Long-Term Memory

In 1885, German psychologist Hermann Ebbinghaus performed one of the most influential experiments in cognitive psychology. He memorized lists of nonsense syllables (like 'DAX,' 'BUP,' 'ZOL'). Then he tested his recall at various time intervals.

His results revealed a stark truth about human memory. Without review, we forget roughly 50% of new information within one hour. We lose about 70% within 24 hours and 90% within a week.

The Forgetting Curve Is Exponential

The curve isn't linear. Forgetting happens rapidly at first, then gradually levels off. But Ebbinghaus discovered something equally important. Each time you review material at the right moment, the curve flattens.

Each Review Extends Retention

The first review might extend retention from 1 day to 3 days. The second might extend it to a week. The third to a month. Eventually, after enough well-timed reviews, information enters permanent long-term memory. It requires only occasional maintenance.

This principle forms the foundation of all spaced repetition systems.

The Pattern of Reviews

• Without review: about 50% forgotten in 1 hour, about 70% in 24 hours, about 90% in 1 week
• After 1st review (at about 24 hours): Memory strength resets. The next forgetting curve is shallower. Retention lasts about 3-4 days
• After 2nd review (at about 3 days): Retention extends to about 1-2 weeks
• After 3rd review (at about 1 week): Retention extends to about 1 month
• After 4th review (at about 1 month): Retention extends to about 3-6 months
• After 5+ reviews at optimal intervals: Information reaches near-permanent status. You need only occasional reviews every few months

Before software existed, German science journalist Sebastian Leitner created the first practical spaced repetition system in 1972. He used physical flashcard boxes to implement spacing. All new cards start in Box 1 (reviewed daily). When you answer a card correctly, it moves to the next box (reviewed less frequently). When you answer incorrectly, it returns to Box 1.

This system automatically adjusts review frequency. Cards you know well move to higher boxes and are reviewed rarely. Cards you struggle with stay in early boxes and are reviewed frequently.

How the Leitner System Works

The Leitner system was revolutionary because it automated the spacing decision. The boxes determine when you review each card, not your intuition. However, the system has limitations. Intervals are fixed per box (not per card). It doesn't account for individual differences in memory strength.

The Five-Box Structure

1. Box 1: Review every day (new and difficult cards)
2. Box 2: Review every 2 days
3. Box 3: Review every 4-5 days
4. Box 4: Review every 9-10 days
5. Box 5: Review every 2-4 weeks (cards you know well)

Card Movement and Concentration

Correct answer means the card moves up one box. Wrong answer means the card returns to Box 1. The system naturally concentrates your study time on the cards you find most difficult.

In 1987, Polish researcher Piotr Wozniak developed the SuperMemo algorithm (SM-2). This became the first computer-based spaced repetition algorithm. SM-2 assigns each card an 'easiness factor' (EF) that starts at 2.5. The factor adjusts based on your self-reported difficulty rating (typically 0-5 scale).

The interval between reviews is calculated by multiplying the previous interval by the EF. Cards you rate as easy get longer intervals. Cards you rate as hard get shorter ones. SM-2 was a breakthrough and remains the algorithm behind Anki, the most widely used spaced repetition app.

Limitations of SM-2

However, SM-2 has significant limitations that became apparent over decades of use. It relies heavily on subjective self-ratings. It doesn't model the forgetting curve mathematically. It uses the same initial intervals for all users regardless of their actual memory patterns. It doesn't learn from your review history. These limitations led to the development of FSRS.

The SM-2 Formula

• Interval(n) = Interval(n-1) × EF, where EF starts at 2.5
• User rates each review on a 0-5 scale. Ratings below 3 reset the card to the beginning
• EF adjusts with each review: EF' = EF + (0.1 - (5 - quality) × (0.08 + (5 - quality) × 0.02))
• Limitations: No mathematical model of forgetting, same starting parameters for all users, doesn't learn from historical data, relies on subjective ratings

FSRS (Free Spaced Repetition Scheduler) was developed by Jarrett Ye starting in 2022. It represents the most significant advance in spaced repetition algorithms in decades. Unlike SM-2, which uses simple multiplication rules, FSRS uses a mathematical model of memory.

The model has three key variables: Stability (how long until you'll forget), Difficulty (how inherently hard the card is for you), and Retrievability (the probability you can recall the card right now). The algorithm uses machine learning to optimize its 19 parameters based on your actual review history. It essentially learns how your specific memory works.

Superior Accuracy and Efficiency

In head-to-head testing, FSRS predicts whether you'll remember a card with over 30% greater accuracy than SM-2. This means fewer unnecessary reviews (saving time) and fewer forgotten cards (improving retention). FSRS was adopted by Anki as an optional algorithm in 2023 and is the default algorithm in FluentFlash.

The Practical Difference

FSRS users typically need 20-30% fewer daily reviews to maintain the same level of retention. The algorithm doesn't waste your time reviewing cards you already know well.

How FSRS Works

• Three core variables: Stability (S) = time until 90% forgetting probability. Difficulty (D) = inherent card difficulty for you. Retrievability (R) = current probability of recall
• FSRS models the forgetting curve mathematically: R(t) = (1 + t/(9·S))^(-1), where t is time since last review
• The algorithm optimizes 19 parameters using YOUR review history. It learns how YOU forget, not how an average person forgets
• Result: 30% plus more accurate predictions than SM-2. This means fewer wasted reviews and fewer forgotten cards
• FluentFlash uses FSRS as its default algorithm, automatically optimizing review schedules for each user

Getting started with spaced repetition is simple, but doing it effectively requires understanding a few key principles. The most common mistake is creating too many new cards too quickly. This leads to an overwhelming daily review load that causes people to quit. Start small, be consistent, and let the system work.

Start Small and Build Gradually

Start with 10-20 new cards per day maximum. Each new card generates reviews for weeks to come. A small daily addition creates a sustainable review load. You can always increase later.

Write Atomic Cards

Write cards that test one fact at a time. "What is the capital of France?" is better than "List all European capitals." Small, atomic cards form the foundation of effective SRS.

Consistency Is Essential

Review every day. Consistency matters more than session length. A 10-minute daily session is far more effective than a 70-minute weekly marathon. The algorithm assumes daily reviews when calculating intervals.

Be Honest With Your Ratings

If you hesitated or had to think hard, rate the card as 'hard' even if you eventually got it right. Accurate ratings lead to better scheduling.

Clear Backlogs First

Don't skip days if possible. Missed days create a backlog. If you do miss a day, do your reviews before adding new cards to clear the backlog first.

Trust the Algorithm

The algorithm will show you cards that feel too easy or too hard at times. Over time, the FSRS algorithm in FluentFlash learns your patterns. The scheduling becomes highly personalized.

While spaced repetition was popularized by language learners, the technique works for any knowledge domain. You need it when retaining large amounts of factual information over time. Medical students use it to memorize thousands of drug interactions, anatomy terms, and diagnostic criteria. Law students use it for case law and statutory elements. Software engineers use it for API syntax, system design patterns, and interview preparation.

Match Your Cards to Real Use Cases

The key is creating cards that match how you'll need to use the knowledge. For language learning, create cards that test recognition and production separately. Test reading the word versus producing it. For medical study, create cards that mirror how you'll encounter information clinically. For programming, create cards that present a problem. Test whether you can recall the solution approach.

Transfer to Real Performance

The more your flashcards resemble real-world retrieval situations, the more effectively spaced repetition transfers to actual performance. This principle applies across all domains.