Skip to main content

Spaced Repetition

Spaced repetition is a learning technique that involves reviewing information at increasing intervals over time to combat the forgetting curve and optimize long-term retention. It's one of the most evidence-based and effective methods for memorizing large amounts of information.

Core Principles​

The Forgetting Curve​

Hermann Ebbinghaus (1885) discovered that memory retention follows a predictable exponential decay pattern:

  • Without reinforcement, we forget approximately 50% of new information within 24 hours
  • After 7 days, retention drops to around 10-20% without review
  • Each review session "resets" the forgetting curve, but at a slower decay rate

The Spacing Effect​

The spacing effect demonstrates that distributed practice (spreading reviews over time) is far more effective than massed practice (cramming). Key findings:

  • Information reviewed at intervals is retained 200-300% better than information studied in one session
  • Optimal spacing intervals increase exponentially after each successful recall
  • The difficulty of retrieval strengthens memory (desirable difficulty)

Active Recall​

Spaced repetition relies on active recall rather than passive review:

  • Testing yourself (retrieving from memory) is more effective than re-reading
  • The act of retrieval itself strengthens neural pathways
  • Failed recall attempts still provide learning benefits by identifying knowledge gaps

Spaced Repetition Systems (SRS)​

What is an SRS?​

A Spaced Repetition System is software that automates the scheduling of review sessions based on algorithmic predictions of when you're about to forget information. SRS platforms track your performance and adjust intervals accordingly.

Common SRS Algorithms​

SM-2 Algorithm (SuperMemo 2)​

The foundational algorithm used by Anki and many other SRS platforms:

  • Uses an "easiness factor" (EF) that adjusts based on recall quality
  • Default intervals: 1 day → 6 days → then multiplied by EF
  • EF starts at 2.5 and adjusts based on performance (1.3 to 2.5 range)

Formula:

New Interval = Previous Interval × Easiness Factor

SM-17/18 (SuperMemo)​

Modern SuperMemo versions use more sophisticated algorithms:

  • Considers retrievability (probability of recall)
  • Optimizes for long-term retention vs. review workload
  • Uses two-component model of memory (stability and retrievability)

FSRS (Free Spaced Repetition Scheduler)​

Newer algorithm developed for Anki (2023+):

  • Machine learning-based approach trained on real user data
  • More accurate predictions than SM-2
  • Considers: card difficulty, memory stability, retrievability, and scheduling history
  • Available as an option in Anki 23.10+

Review Quality Ratings​

Most SRS platforms use a quality scale to adjust scheduling:

Anki's 4-button system:

  • Again (1): Complete failure - resets card to beginning of learning queue
  • Hard (2): Correct but difficult - shorter than normal interval
  • Good (3): Correct with moderate effort - standard interval
  • Easy (4): Trivial recall - much longer interval

SuperMemo's 6-point scale:

  • 0: Complete blackout
  • 1: Incorrect, but recognized answer
  • 2: Incorrect, but seemed easy
  • 3: Correct, but required significant effort
  • 4: Correct with some hesitation
  • 5: Perfect recall

Key SRS Terminology​

Card States​

New: Cards that have never been studied

Learning: Cards currently being learned (in short-term review cycles)

  • Multiple steps (e.g., 1m, 10m, 1d) before graduation
  • Failed reviews send card back to first learning step

Review/Young: Recently graduated cards (interval < 21 days typically)

Mature: Cards with intervals > 21 days (well-learned material)

Relearning: Previously learned cards that were forgotten, going through learning steps again

Interval Management​

Interval: Time between reviews (e.g., 3 days, 2 months, 1 year)

Graduating Interval: First interval when a card moves from Learning to Review state (typically 1 day)

Easy Interval: Interval assigned when "Easy" is pressed on a new card (typically 4 days)

Maximum Interval: Upper limit on review intervals (default: 36,500 days / 100 years)

Interval Modifier: Global multiplier applied to all calculated intervals (default: 100%)

  • Increase to > 100% for easier material or to reduce workload
  • Decrease to < 100% for harder material or higher retention targets

Fuzz Factor: Random variation added to intervals (±5%) to prevent cards from always coming due together

Card Difficulty​

Ease Factor / Easiness: Multiplier that determines how quickly intervals grow (typically 130-250%)

  • Starts at 250% for new cards
  • Decreases when "Hard" or "Again" is pressed
  • Increases when "Easy" is pressed
  • Lower ease = more frequent reviews

Leeches: Cards that are repeatedly forgotten (e.g., failed 8+ times)

  • Indicates ineffective card design or prerequisite knowledge gaps
  • Should be: rewritten, split into multiple cards, or suspended until prerequisites are learned

Review Scheduling​

Due Date: When a card is scheduled for review

Overdue: Cards past their due date (accumulated backlog)

  • Large backlogs can demotivate and reduce retention
  • Better to reduce daily new cards than accumulate overdues

Daily Limits:

  • New cards/day: How many unseen cards to introduce (e.g., 20/day)
  • Review cards/day: Maximum reviews per day (e.g., 200/day)
  • Reviews should typically be unlimited; new cards control workload

Card Actions​

Bury: Temporarily hide a card until the next day

  • Useful when a related card just appeared
  • Automatically unburies at midnight
  • Manual bury: right-click → Bury

Suspend: Indefinitely pause a card from appearing in reviews

  • Intervals are preserved (unlike deleting)
  • Used for: leeches, outdated information, cards needing revision
  • Must be manually unsuspended to resume reviews

Reset: Erase all scheduling history, returning card to "new" status

  • Use sparingly - loses valuable algorithm data
  • Alternative: use "Reposition" to move new cards in queue

Reschedule: Manually set a new due date or interval

  • Useful after long breaks or importing cards from another source
  • Can specify: new interval, place in review queue, or reset completely

Deck Organization​

**Deck:** Container for cards (e.g., "Spanish Vocabulary", "Medical School::Anatomy")

**Subdeck:** Nested deck structure using `::` separator

- Example: `Languages::Spanish::Verbs`
- Subdecks inherit parent deck settings unless overridden

Filtered Deck: Temporary deck created by search query

  • Used for: cramming before exams, reviewing specific tags, catching up on overdue cards
  • Cards return to original deck after session

Parent Limit: Whether daily limits apply separately to each deck or are shared with parent

  • Enabled: each deck has independent limits
  • Disabled: reviewing subdeck counts toward parent's limit

Advanced SRS Concepts​

Retention Rate: Percentage of reviews answered correctly

  • Typical targets: 80-90% for general knowledge, 90-95% for critical material
  • < 80%: material too difficult or intervals too long
  • > 95%: reviewing too frequently (inefficient)

True Retention: Actual measured retention from review performance

Desired Retention: Target retention rate set in FSRS or other algorithms (e.g., 0.9 = 90%)

Stability: How long a memory can last before it becomes unretrievable (FSRS concept)

Retrievability: Current probability of successful recall (FSRS concept)

Load Balancing: Distributing reviews evenly across days to avoid spikes

  • Fuzz factor helps with this automatically
  • Advanced: use add-ons like "Load Balancer" for better distribution

Spaced Repetition Techniques​

Effective Card Creation​

Minimum Information Principle: Break complex information into atomic facts

❌ Poor: "What were the causes of World War I?" ✅ Better: Multiple cards for each specific cause (assassination, alliances, militarism, etc.)

Cloze Deletions: Fill-in-the-blank format

The {{c1::mitochondria}} is the {{c2::powerhouse}} of the cell.
  • Creates focused questions
  • Reduces card creation time
  • Avoids "recognition" vs "recall" trap

Bidirectional Cards: Create reverse cards when appropriate

  • Front: Spanish word → English translation
  • Back: English word → Spanish word
  • Use for: vocabulary, definitions, equations

Image Occlusion: Hide parts of images for visual learning

  • Anatomy diagrams (hide labels)
  • Architecture/engineering drawings
  • Musical notation

Optimal Spacing Strategies​

Leitner System: Physical flashcard method with boxes

  • Box 1: Daily review
  • Box 2: Every 2 days
  • Box 3: Every 4 days
  • Box 4: Every 8 days
  • Box 5: Every 16 days
  • Cards move forward on success, back to Box 1 on failure

Pimsleur Method: Audio-based language learning with specific intervals

  • 5 seconds → 25 seconds → 2 minutes → 10 minutes → 1 hour → 5 hours → 1 day → 5 days → 25 days → 4 months → 2 years

Expanding Retrieval Practice: Start with short intervals, rapidly expand

  • Useful for initial learning phase
  • Example: 1 minute → 10 minutes → 1 hour → 1 day → then standard SRS

Review Strategies​

Daily Consistency: Review every day to prevent backlog

  • Better to do 15 cards/day consistently than 100 cards sporadically
  • Set realistic daily limits based on long-term sustainability

Time-Boxing: Allocate fixed time (e.g., 30 minutes) rather than card count

  • Prevents burnout from large review queues
  • Prioritize: overdue reviews → due reviews → new cards

Interleaving: Mix topics during review sessions

  • Prevents pattern recognition shortcuts
  • Strengthens discrimination between similar concepts

Before Bed Review: Take advantage of sleep consolidation

  • Reviewed material is preferentially consolidated during sleep
  • Avoid introducing completely new, complex information right before sleep

Anki​

Strengths:

  • Free and open-source
  • Highly customizable (add-ons, card templates, algorithms)
  • Cross-platform (Windows, Mac, Linux, iOS, Android, web)
  • Best for: serious learners, medical students, language learners

Key Features:

  • Supports SM-2 and FSRS algorithms
  • Rich media (audio, images, video)
  • LaTeX support for math/science
  • Shared decks (AnkiWeb)
  • Extensive plugin ecosystem

Limitations:

  • Steeper learning curve
  • iOS app costs $25 (Android free)
  • Default interface is utilitarian

SuperMemo​

Strengths:

  • Original SRS software (since 1987)
  • Most sophisticated algorithm (SM-17/18)
  • Incremental reading (integrate reading and flashcard creation)

Limitations:

  • Windows only
  • Expensive ($66 one-time)
  • Dated interface
  • Steeper learning curve than Anki

RemNote​

Strengths:

  • Combines note-taking with SRS (bidirectional linking like Roam/Obsidian)
  • Automatically creates flashcards from notes
  • Built-in spaced repetition without separate app

Use Case:

  • Students who want integrated notes + flashcards
  • Knowledge workers building a "second brain"

Quizlet​

Strengths:

  • User-friendly interface
  • Large library of pre-made sets
  • Social features (study groups, shared decks)

Limitations:

  • Limited true SRS functionality (more basic scheduling)
  • Premium features behind paywall
  • Less customization than Anki

Mnemosyne​

Strengths:

  • Free and open-source
  • Clean, simple interface
  • Good for beginners

Limitations:

  • Fewer features than Anki
  • Smaller community and plugin ecosystem

Evidence Base​

Research Findings​

Retention Gains:

  • Bahrick et al. (1993): Spacing increased retention from 35% to 80% over 8-year period
  • Cepeda et al. (2006): Meta-analysis found average effect size of d=0.42 for spacing effect

Optimal Intervals:

  • Cepeda et al. (2008): Optimal spacing is 10-20% of desired retention period
  • For 1-year retention: review after 36-73 days
  • For 5-year retention: review after 6-12 months

Desirable Difficulty:

  • Bjork & Bjork (2011): More difficult retrieval creates stronger learning
  • 80-85% success rate appears optimal (balance difficulty vs. frustration)

Limitations and Critiques​

Not Universal: Spaced repetition works best for:

  • Declarative knowledge (facts, vocabulary, definitions)
  • Recognition/recall tasks
  • Stable information (doesn't change)

Less Effective For:

  • Deep conceptual understanding (requires elaborative rehearsal)
  • Procedural skills (need deliberate practice, not just recall)
  • Creative application (transfer requires varied practice contexts)

Time Investment: Front-loaded effort

  • Creating effective cards takes time
  • Daily review commitment required
  • Benefits compound over months/years, not days/weeks

Integration with Other Learning Methods​

Complementary Techniques​

Elaborative Interrogation: Ask "why" questions before creating cards

  • Deepens understanding before memorization
  • Creates better retrieval cues

Dual Coding: Combine verbal and visual information

  • Add images to cards
  • Draw diagrams from memory
  • Use image occlusion

Interleaving: Mix subjects during study sessions

  • Don't review all Spanish cards, then all biology cards
  • Randomize or use Anki's default mixing

Testing Effect: Combine SRS with practice problems

  • SRS for foundational knowledge
  • Practice problems for application
  • Alternate between both

Learning Pipeline​

  1. Initial Learning: Lectures, reading, videos (acquire information)
  2. Note-Taking: Capture key concepts in organized notes
  3. Card Creation: Convert notes into atomic flashcards
  4. Initial Review: Short-interval repetition (1m, 10m, 1h)
  5. Spaced Review: Long-term SRS scheduling (days, weeks, months)
  6. Application: Use knowledge in practice problems, projects, teaching others

Best Practices​

Card Design​

  • One Fact Per Card: Avoid complex multi-part questions
  • Context Cues: Include enough context to avoid ambiguity
  • Mnemonic Hints: Add memory aids in extra field (revealed after answer)
  • Sources: Tag cards with source (textbook chapter, lecture number) for later reference
  • Personal Connection: Relate to personal experience when possible

Scheduling Discipline​

  • Review Before New: Always complete due reviews before adding new cards
  • Sustainable Pace: Set new card limits you can maintain during busy periods
  • Vacation Mode: Reduce new cards to zero before breaks, focus on reviews only
  • Gradual Ramp-Up: Start with 5-10 new cards/day, increase slowly

Deck Maintenance​

  • Regular Leech Review: Weekly check for cards failed > 4 times
  • Update Outdated Cards: Revise when information becomes obsolete
  • Delete Low-Value Cards: Remove trivial or non-essential cards after 6 months
  • Tag Ruthlessly: Use tags for exams, priority, subject, difficulty

Advanced Optimization​

  • Retention Target Tuning: Adjust desired retention based on importance (90% for critical, 80% for nice-to-know)
  • Load Balance: Use fuzz factor or load balancer add-ons
  • Sibling Spacing: Bury related cards to avoid pattern recognition
  • Optimize Ease Factor: Reset ease for old decks imported from other sources

Common Pitfalls​

Card Quality Issues​

Recognition vs. Recall: Avoid cards you can answer by pattern-matching

❌ "The capital of France is ____" (too easy to guess from context) ✅ "What is the capital of France?" (forces retrieval)

Orphan Facts: Isolated information without conceptual framework

  • Solution: Add "big picture" cards explaining relationships
  • Create "map" cards showing how facts connect

Overly Complex Cards: Multi-step reasoning required

  • Solution: Break into prerequisite cards
  • Use cloze deletions for step-by-step processes

Scheduling Problems​

Ease Hell: Repeatedly hitting "Hard" or "Again" creates very low ease factors

  • Cards get stuck in frequent review cycles
  • Solution: Use "Set Ease" add-on to reset to 250%, or rewrite card

Backlog Spiral: Missing reviews creates overwhelming due counts

  • Solution: Use filtered deck to gradually work through backlog
  • Reduce new cards to zero until caught up

Premature Optimization: Obsessing over algorithm settings instead of studying

  • Default settings work well for most learners
  • Focus on card quality and daily consistency first

Workflow Issues​

Passive Card Consumption: Using pre-made decks without understanding

  • Solution: Always edit or create your own cards from source material
  • Pre-made decks are good starting points, not final product

Review Procrastination: Skipping days due to large review count

  • Solution: Time-box reviews (30 minutes/day) rather than "complete all reviews"
  • Better to maintain streak with partial reviews than break habit

No Application: Only memorizing facts without using knowledge

  • Solution: Combine SRS with practice problems, projects, teaching
  • SRS is foundation, not complete learning system

Startup Implications​

Product Design Considerations​

Adaptive Algorithms:

  • Modern learners expect FSRS-level sophistication (not just SM-2)
  • Personalization based on individual forgetting curves
  • Differentiate difficulty by subject domain (languages vs. medical terminology)

User Experience:

  • Reduce friction in card creation (templates, AI-assisted generation)
  • Mobile-first design (reviews happen during commutes, waiting)
  • Gamification without sacrificing effectiveness (streaks, XP, but preserve spacing integrity)

Integration Opportunities:

  • Combine with note-taking tools (RemNote model)
  • Auto-generate cards from textbooks, lectures, videos (AI extraction)
  • Community decks with quality curation (Reddit-style upvoting)

Market Positioning​

Learner Segments:

  • Students (K-12, University): Exam preparation, language learning, STEM subjects
  • Professionals: Certification prep (medical boards, bar exam, professional licenses)
  • Lifelong Learners: Language enthusiasts, hobbyists, retirees

Competitive Advantages:

  • Superior algorithm (FSRS baseline, ML-enhanced personalization)
  • Seamless multi-platform sync
  • Content marketplace (high-quality pre-made decks)
  • Analytics and insights (retention stats, time-to-mastery predictions)

Technical Challenges​

Scheduling at Scale:

  • Efficient interval calculations for millions of cards
  • Load balancing across user base (server costs)
  • Offline-first with sync conflict resolution

Data Privacy:

  • Personal learning data is sensitive
  • Local-first architecture vs. cloud analytics trade-off

Algorithm Transparency:

  • Users want control over scheduling
  • Balance simplicity (hidden algorithm) vs. customization (exposed parameters)

References​

Foundational Research​

  • Ebbinghaus, H. (1885). Memory: A Contribution to Experimental Psychology
  • Bahrick, H. P., Bahrick, L. E., Bahrick, A. S., & Bahrick, P. E. (1993). Maintenance of foreign language vocabulary and the spacing effect. Psychological Science, 4(5), 316-321
  • Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354-380
  • Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. Psychology and the Real World, 2, 59-68

Algorithm Documentation​

Practical Guides​