About SM-2

The SM-2 (SuperMemo-2) algorithm is one of the most influential spaced repetition algorithms ever created. Designed in the late 1980s, it laid the mathematical foundation for modern flashcard-based learning systems and helped popularize the idea that memory can be optimized through carefully timed review.

A Brief History of SM-2

The SM-2 algorithm was developed in 1987 by Polish researcher Piotr Woźniak. As a student, Woźniak became interested in improving his long-term retention of study material. Dissatisfied with traditional cramming and linear review schedules, he began experimenting with ways to review information at increasing intervals based on how well he remembered it.

His early experiments led to the first SuperMemo algorithm (SM-0), which was simple and somewhat rigid. Through iterative experimentation—analyzing memory performance data and adjusting formulas—he refined his approach. The result was SM-2, a significantly improved algorithm that dynamically adjusted review intervals according to learner feedback.

SM-2 became especially influential because it struck a balance between effectiveness and simplicity. Its formula was lightweight enough to be implemented on personal computers of the time, yet sophisticated enough to produce impressive retention gains. Even decades later, many spaced repetition systems still use SM-2 directly or adapt its core principles.

The Core Idea Behind Spaced Repetition

The algorithm is grounded in a psychological principle known as the "spacing effect." Research in cognitive science shows that information is retained more effectively when review sessions are spaced out over time rather than massed together. Furthermore, reviewing material just before it would otherwise be forgotten strengthens memory more efficiently than reviewing it too soon.

SM-2 operationalizes this principle by scheduling the next review of a flashcard based on how easily the learner recalled it during the current review.

How SM-2 Works

At its heart, SM-2 uses three main components:

  1. A quality rating of recall
  2. An interval (in days) until the next review
  3. An "ease factor" that reflects how easy the item is for the learner

1. Quality of Recall

After reviewing a flashcard, the learner assigns a score from 0 to 5:

  • 5 — Perfect recall
  • 4 — Correct response with slight hesitation
  • 3 — Correct response but difficult recall
  • 2 — Incorrect, but remembered after seeing the answer
  • 1 — Incorrect; the answer felt familiar
  • 0 — Complete blackout

Scores of 3 or higher are considered successful recall. Scores below 3 indicate that the card needs reinforcement.

2. Review Intervals

For new cards, SM-2 sets the first two intervals as fixed values:

  • After the first successful recall: 1 day
  • After the second successful recall: 6 days

After that, intervals grow multiplicatively. The next interval is calculated as:

New Interval = Previous Interval × Ease Factor

This means that if a card is consistently recalled well, its review interval expands rapidly—e.g., from 6 days to 15 days, then to 40 days, and so on.

If recall quality is below 3, the repetition count resets, and the interval returns to 1 day. This ensures difficult items are reviewed more frequently.

3. Ease Factor

Each card has its own ease factor (EF), which starts at 2.5 by default. The ease factor determines how quickly the interval grows.

After each review, the ease factor is updated using a formula based on recall quality:

EF' = EF + (0.1 − (5 − q) × (0.08 + (5 − q) × 0.02))

Where:

  • EF is the previous ease factor
  • q is the recall quality (0–5)

If recall was easy (high q), the ease factor increases slightly. If recall was difficult (low q but still ≥ 3), the ease factor decreases. However, EF is never allowed to drop below 1.3, which prevents intervals from shrinking excessively.

In practical terms:

  • Easy cards become increasingly spaced out
  • Difficult cards stay on shorter intervals
  • Each card develops its own review rhythm

Why SM-2 Was So Impactful

The brilliance of SM-2 lies in its personalization. Instead of applying the same schedule to every flashcard, it adapts to the learner's performance on each individual item. This dramatically increases efficiency: time is spent where it is needed most.

Another reason for its longevity is its simplicity. Despite being rooted in memory research, the algorithm is computationally light and relatively easy to implement. It doesn't require complex modeling or large datasets—just consistent user feedback.

Limitations and Evolution

While groundbreaking, SM-2 is not perfect. It assumes that memory strength can be adequately captured by a single ease factor, and it does not model forgetting curves explicitly. Modern algorithms sometimes use more advanced statistical models, such as probabilistic decay functions or machine learning approaches.

Still, many contemporary spaced repetition methods trace their origins directly to SM-2. It remains a foundational milestone in the science of optimized learning.

Conclusion

The SM-2 algorithm represents a pivotal moment in the evolution of learning technology. Developed through careful experimentation in the 1980s, it translated cognitive psychology into a practical scheduling system that adapts to individual memory performance.

More than three decades later, its core insight still holds: learning is most effective when review is timed to challenge memory just enough—not too early, not too late.

Interaction with cards

A card is the basic unit of learning in Memicards. Each card represents one thing you want to remember — a word, a phrase, a fact, or any concept that fits on two sides.

A card has four fields:

Word — the term you want to learn. This is what you'll see on the front of the card during a review session.

Translation — the meaning, definition, or answer. This is revealed when you flip the card.

Example sentence (optional) — a sentence showing the word in context. Words learned in context stick significantly better than isolated vocabulary. For example, if your word is «կատու» (cat), your sentence might be «Կատուն նստած է սեղանի վրա» — The cat is sitting on the table.

Mnemonic association (optional) — a personal memory hook. A silly image, a sound-alike word, or a story that connects the new word to something you already know. For example: «կատու sounds like "ka-too" — imagine a cat sneezing KA-TOO-CHOO». The weirder the association, the better it works.

How does a card work during review?

You see the word. You try to recall the translation. You flip the card and check yourself. Then you rate how well you remembered — Again, Hard, Good, or Easy. Based on your rating, Memicards schedules when to show you this card next. Cards you know well appear less often. Cards you struggle with appear more frequently. Over time, you spend your review time exactly where it's needed most.

Card buttons

Again — Didn't remember. Will show up again soon.

Hard — Remembered, but with effort. Will appear more frequently.

Good — Remembered well. On track.

Easy — Remembered instantly. Will appear much later.

Keep one idea per card. If a word has multiple meanings, consider making a separate card for each.

Write your own example sentences when possible — your own words are easier to remember than textbook examples.

Methods Comparison

Here's a concise comparison table of the classic SM-2 spaced repetition algorithm versus modern learning/scheduling algorithms used today (including probabilistic/Bayesian, machine-learning-driven, or advanced schedulers like FSRS). SM-2 remains widely effective, but newer models can slightly improve efficiency at the cost of much higher complexity.

Algorithm / Category Effectiveness (Retention per Review) Complexity Notes on Trade-off
SM-2 (Classic) ⭐⭐⭐⭐☆ — Strong, robust improvement in long-term recall vs. unstructured review; adapts per card via ease factor. ⭐☆☆☆☆ — Very simple mathematically; easy to implement and understand. Gold standard baseline; performs very well for most learners with minimal overhead.
Bayesian / Probabilistic Models (e.g., Bayesian Knowledge Tracing variants) ⭐⭐⭐⭐⭐ — Slight improvement in predicted recall and scheduling precision compared to SM-2 in some contexts. ⭐⭐⭐☆☆ — Moderately complex; needs probabilistic modeling and data tracking. Often more accurate timing of reviews, but requires more data and tuning, with gains typically modest over SM-2 for individual use.
Machine Learning / Adaptive Schedulers (e.g., FSRS / LSTM / learned models) ⭐⭐⭐⭐⭐ — Can outperform SM-2 in large percentages of cases (20–30% fewer reviews for equal retention). ⭐⭐⭐⭐☆ — High complexity; many parameters, training, and infrastructure required. Represents the cutting edge; marginal gains in retention/time trade-offs but much heavier computational demands.
Reinforcement Learning-based Algorithms ⭐⭐⭐⭐☆ to ⭐⭐⭐⭐⭐ — Potentially adaptive and optimized but experimental. ⭐⭐⭐⭐☆ — Requires RL design and reward tuning. Can personalize schedules based on performance history; complex and mainly research-oriented.
Manual / Box Systems (e.g., Leitner) ⭐⭐☆☆☆ — Better than random or cramming but coarse. ⭐☆☆☆☆ — Very simple. Practical without software, but far less personalized and efficient than algorithmic spaced repetition.

Key Takeaways

SM-2 is still highly effective with minimal complexity

For most learners, SM-2 strikes a strong balance of retention gains and computational simplicity. It adapts per item and schedules reviews based on performance, capturing the core of spaced repetition's benefits without needing much data or engineering.

Modern algorithms can be slightly more effective, but complexity is much higher

Advanced schedulers—like probabilistic models or machine learning-based algorithms (e.g., FSRS)—can reduce total study time or improve recall prediction accuracy, but they require significantly more data, computation, and tuning. In practice, the average learner sees modest improvements over SM-2 gains, not dramatic leaps, unless operating at scale with large datasets.

In summary: SM-2 remains an excellent baseline. Modern methods can edge ahead in precision and efficiency, but the performance gains often do not scale proportionally to the jump in algorithmic complexity.

About Flashcards

Flashcards are one of the most effective tools for memorization. They work by testing your active recall—the ability to retrieve information from memory without looking at the answer.

How Flashcards Work

A flashcard has two sides:

  • Front: A question, word, or prompt
  • Back: The answer or translation

The process is simple:

  1. Look at the front
  2. Try to recall the answer
  3. Flip the card to check
  4. Repeat until memorized

Why They Work

Flashcards tap into powerful learning principles:

Active Recall

Instead of passively re-reading notes, you actively retrieve information from memory. This strengthens neural pathways and makes memories more durable.

Self-Testing

Testing yourself is more effective than studying alone. Each retrieval attempt reinforces the memory, even if you get it wrong.

Immediate Feedback

You know instantly if you remembered correctly. This helps your brain adjust and correct mistakes quickly.

Focused Learning

One concept per card keeps your attention sharp and prevents information overload.

Traditional vs. Spaced Repetition

Traditional flashcards:

  • Review all cards equally, regardless of difficulty
  • Easy cards waste your time with unnecessary repetition
  • Hard cards don't get enough practice

Spaced repetition flashcards (like memicards):

  • Focus on cards you're about to forget
  • Easy cards appear less frequently
  • Hard cards get more practice
  • Scientifically optimized timing for maximum retention

When to Use Flashcards

Flashcards are perfect for:

  • Languages: vocabulary, grammar patterns, phrases
  • Medical terms: anatomy, drugs, procedures
  • History: dates, events, people
  • Sciences: formulas, definitions, concepts
  • Exams: any fact-based material that requires memorization

Flashcards work best when:

  • Information can be broken into question-answer pairs
  • You need long-term retention, not just cramming
  • You can practice regularly (even 10 minutes a day helps)

What Makes Good Flashcards

Keep it simple

One concept per card. "What is the capital of France?" → "Paris"

Use context

Add example sentences for vocabulary: "Bonjour - Hello. Bonjour! Comment allez-vous?"

Add memory hooks

Associations help you remember: "Biblioteca (library) - sounds like 'biblio' (books)"

Make them personal

Create your own cards. The act of making them already helps you learn.

Ready to start? Create your first deck and experience the power of spaced repetition!

Main Concepts

Memicards organizes your learning into a simple hierarchy. Understanding these core concepts will help you get the most out of spaced repetition.

Projects

Projects are top-level containers for organizing different learning goals. Each project can have its own decks, settings, and progress tracking.

Examples:

  • "Spanish for Travel" — a project for vacation preparation
  • "Medical Terminology" — for professional study
  • "Japanese N5" — for JLPT exam prep

Key features:

  • Switch between projects using the dropdown in the header
  • Each project has independent settings (review limits, weekly goals)
  • Progress and statistics are tracked per-project
  • Your first project is created automatically when you sign up

When to create a new project:

  • Different languages or subjects
  • Distinct learning goals (exam prep vs casual learning)
  • Separate contexts (work vocabulary vs hobby vocabulary)

Decks

Decks live inside projects and contain your flashcards. Think of a deck as a chapter or topic within your learning goal.

Structure:

  • Each deck has a name, optional description, and language pair
  • Cards belong to one deck
  • You can have unlimited decks per project

Examples within a "Spanish for Travel" project:

  • "Greetings & Basics"
  • "Restaurant Vocabulary"
  • "Hotel & Transportation"

Deck features:

  • Set language pair (e.g., Spanish → English)
  • Filter cards: Due Today, New, Starred
  • See card counts: total cards, due today, starred, inactive
  • Duplicate decks (with option to swap word/translation sides)
  • Move decks between projects

Cards

Cards are individual flashcards for learning. Each card has a word/phrase, translation, and optional memory aids.

Card anatomy:

  • Front: Word or phrase in the language you're learning
  • Back: Translation in your native language
  • Example sentence (optional): Context for the word
  • Association/Mnemonic (optional): Memory hook to help you remember
  • Difficulty star: Mark challenging words for extra focus

Card states:

  • New: Never reviewed
  • Learning: Recently introduced, reviewed frequently
  • Review: Mastered, reviewed at increasing intervals
  • Active/Inactive: Toggle to include/exclude from review sessions

Creating cards:

  • Add manually one-by-one
  • Batch import via CSV (50+ cards at once)
  • Edit, star, or deactivate cards anytime

Review Sessions

Review sessions use the SM-2 algorithm to show you cards at optimal intervals for retention.

How it works:

  1. The app shows you cards that are due today
  2. Try to recall the answer before flipping
  3. Rate your memory: Again, Hard, Good, Easy
  4. The algorithm calculates when to show the card next

Rating guide:

  • Again: Couldn't remember → shown very soon
  • Hard: Struggled to remember → shorter interval
  • Good: Remembered correctly → standard interval
  • Easy: Instantly recalled → longer interval

Types of sessions:

  • Review session: Counts toward statistics and progress
  • Practice mode: Casual review without affecting stats

The SM-2 Algorithm

SM-2 (SuperMemo 2) is a scientifically proven spaced repetition algorithm that calculates optimal review timing.

How it adapts:

  • Easy cards: intervals grow exponentially (1 day → 3 days → 1 week → 2 weeks...)
  • Hard cards: intervals stay shorter, reviewed more often
  • Each card has individual scheduling based on your performance

Why it works:

  • Reviews cards right before you're about to forget
  • Minimizes wasted repetitions of easy cards
  • Maximizes retention with minimum study time

Settings & Preferences

Per-project settings let you customize your learning pace:

Weekend Learner Mode:

  • Set different review limits for weekdays vs weekends
  • Example: 20 new cards weekdays, 50 on Saturdays

Daily limits:

  • New cards per day (cards you've never seen)
  • Review cards per day (cards you're reinforcing)

Other preferences:

  • Weekly card target goal
  • Prioritize starred cards
  • Theme (light/dark mode)

Progress & Statistics

Track your learning with detailed insights:

  • Retention rate charts: See how well you're remembering
  • Weekly goals: Monitor your daily review consistency
  • Difficult words list: Cards you're struggling with
  • Review history: Complete log of your study sessions

Export your data:

  • Download cards and progress as CSV or JSON
  • Take your data anywhere

Key Workflows

Starting fresh:

  1. Create a project (or use the default one)
  2. Create a deck with your language pair
  3. Add cards (manually or import CSV)
  4. Start reviewing!

Daily routine:

  1. Open your project
  2. Check "Due Today" count
  3. Start a review session
  4. Rate each card honestly
  5. Track your progress

Progressive learning:

  1. Import all cards at once
  2. Mark most as "Inactive"
  3. Enable 10-20 cards per week as you learn them
  4. System adapts to your pace