🔒

Unauthorized Origin

This build is not authorized to run on this domain.

Current origin:

🧠 N-Back
N=2 Best: --

Ready to Train?

Difficulty

Mode

Custom Configuration

Minimum: N + 10

Starting N-Level

2

~5 minutes per block

3
Get ready...
N 2
Trial 1/20
Block 1
Audio: -
Shape:
Color:

Block Complete!

Position

Hit Rate --
False Alarms --

Audio

Hit Rate --
False Alarms --

Shape

Hit Rate --
False Alarms --

Color

Hit Rate --
False Alarms --

Session Complete! 🎉

Score --
Peak N --
Accuracy --
Trials --
Duration --

Your Progress

Storage
0 / 30 MB
Sessions 0
Total Trials 0
Best Score --
Best N --

Recent Sessions

No sessions yet. Start training!

Charts (Score + N over time)

Data Management

Settings

Profile

This name will appear on shared score tokens.

Timing (Advanced)

Audio

Hybrid tone + speech synthesis. Higher quality = more realistic sound but larger memory usage.

Display

Keyboard Shortcuts

A or ← Position match
L or → Audio match
S or ↓ Shape match (triple/quad/custom)
C or ↑ Color match (quad/custom)
Esc Pause

The Science of N-Back Training

What is N-Back?

The N-back task is a continuous performance task used in cognitive neuroscience to measure working memory capacity. In the dual N-back variant, you must track two independent streams (typically visual position and audio) and identify when the current stimulus matches the one from N trials ago.

Does It Work?

The evidence on N-back training is mixed. While early studies showed promising results for improving fluid intelligence, subsequent research with better controls has produced less consistent findings.

What We Know Works

  • Training improves N-back performance itself
  • Some near-transfer to similar working memory tasks
  • Improvements in task-switching ability (some studies)

What's Less Clear

  • Far transfer to general intelligence (Gf)
  • Long-term retention of benefits
  • Transfer to real-world cognitive tasks

Important Considerations

Studies showing the largest effects often lacked active control groups. When participants in control groups expect improvement (expectation effects), the differences between training and control groups shrink considerably.

Key References

  • Jaeggi, S. M., et al. (2008). Improving fluid intelligence with training on working memory. PNAS, 105(19), 6829-6833. doi:10.1073/pnas.0801268105 Initial study showing Gf improvements
  • Redick, T. S., et al. (2013). No evidence of intelligence improvement after working memory training. Journal of Experimental Psychology: General, 142(2), 359. doi:10.1037/a0029082 Well-controlled replication found no Gf transfer
  • Owen, A. M., et al. (2010). Putting brain training to the test. Nature, 465(7299), 775-778. doi:10.1038/nature09042 Large-scale online study found limited transfer
  • Melby-Lervåg, M., & Hulme, C. (2016). There is no convincing evidence that working memory training is effective. Psychonomic Bulletin & Review, 23(1), 324-330. doi:10.3758/s13423-015-0862-z Meta-analysis emphasizing limited far transfer
  • Parong, J., et al. (2022). Expectation effects in working memory training. PNAS, 119(28). doi:10.1073/pnas.2036188119 Demonstrates role of expectation effects
  • Zhong, Chen, et al. (2025). Near and far transfer effects of working memory training. Acta Psychologia. doi:10.1016/j.actpsy.2025.105499 Recent study on longer visuospatial N-Back improving training

The Bottom Line

N-back training is a challenging cognitive exercise that will definitely improve your N-back performance. Whether benefits transfer to other cognitive abilities remains an open question. If you enjoy the challenge and find it engaging, there's no harm in practicing—just maintain realistic expectations about broader cognitive benefits.

Disclaimer

This application is for educational and entertainment purposes only. N-Back Trainer is not a medical device and is not intended to diagnose, treat, cure, or prevent any disease or medical condition. The cognitive training provided by this application should not be considered a substitute for professional medical advice, diagnosis, or treatment. If you have concerns about your cognitive health, memory, attention, or any other medical condition, please consult a qualified healthcare professional. Individual results may vary, and the research on cognitive transfer effects remains inconclusive. By using this application, you acknowledge that you do so at your own discretion and risk.

About N-Back Trainer

A free working memory training tool featuring adaptive difficulty with Bayesian skill tracking.

How to Play N-Back

N-Back is a working memory task where you must remember stimuli from N turns ago and respond when the current stimulus matches.

The Basic Concept

If you're playing 2-back, you need to remember what happened 2 turns ago. When the current position or sound matches what appeared 2 turns earlier, press the corresponding button.

Step-by-Step

  1. Watch and listen: Each trial shows a square lighting up in a 3×3 grid while playing a letter sound.
  2. Remember: Keep track of the last N positions and sounds in your mind.
  3. Match: When the current position matches the position from N trials ago, press the A key (or ←). When the current sound matches, press L (or →).
  4. Both can match: Sometimes both position AND sound match—press both keys!
  5. Don't guess: Only respond when you're confident there's a match. False alarms hurt your score.

Example (2-Back)

Imagine this sequence of positions: Center → Top-Left → Bottom-Right → Center → Top-Left

On trial 5 (Top-Left), you should press the position match key because Top-Left also appeared on trial 3 (exactly 2 back).

Tips for Success

  • Start with Single N-Back: If dual feels overwhelming, practice with position-only first.
  • Subvocalize: Quietly rehearse positions ("center, top-left, bottom-right...") to keep them in memory.
  • Focus on the rhythm: The timing is consistent—use it to your advantage.
  • Don't overthink: Trust your gut. Hesitation often means you've lost the memory trace.
  • Practice daily: Short, consistent sessions (15-20 minutes) work better than occasional long ones.

Features

  • Single (position), Dual (audio), Triple (shape), and Quad (color) N-Back modes
  • Bayesian adaptive difficulty targeting optimal challenge
  • Detailed progress tracking with 30MB local storage
  • Encrypted backup with Progress Restore Key
  • Shareable, cryptographically signed score tokens
  • Available on Web, and eventually... Windows, macOS, and Linux

Game Modes

  • Single (1-N): Track position only. Great for beginners or focusing on spatial memory.
  • Dual (2-N): Track position and audio simultaneously. The classic dual n-back paradigm used in cognitive research.
  • Triple (3-N): Track position, audio, and shape streams. Significantly more challenging with exponentially higher cognitive load.
  • Quad (4-N): Track position, audio, shape, and color streams. Maximum difficulty for experts and advanced practitioners.

How Adaptive Difficulty Works

The adaptive difficulty system uses a Bayesian particle filter to estimate your skill level in real-time and automatically adjust the N-level to keep you in an optimal challenge zone.

  • Skill Tracking: The system maintains 500 particles, each representing a possible skill level. After each trial, particles are weighted based on how accurately they predict your performance.
  • Signal Detection Model: Your responses are modeled using signal detection theory, which separates your ability to detect matches (sensitivity) from your tendency to respond conservatively or liberally (response bias).
  • Target Success Rate: The system aims for approximately 80% per-stream accuracy. Joint accuracy targets decrease with more streams due to multiplicative difficulty (64% for Dual, 51% for Triple, 41% for Quad).
  • N Selection: After each block, the system predicts your success rate at each potential N level and selects the one closest to the target. If performance is too high, N increases; if too low, N decreases.
  • Rapid Learning: The system learns quickly within 10-20 trials but continues refining its estimate throughout your session. Playing consistently helps it dial in your optimal difficulty.

How Scoring Works

Your score (0-10000) combines performance quality with configuration difficulty to ensure that harder achievements are properly rewarded. The scoring system is computed entirely in Rust/WebAssembly to prevent manipulation.

Performance Components

  • Joint Accuracy (50% weight): The percentage of trials where ALL enabled streams were correct simultaneously. This is the primary measure since N-back requires parallel tracking across multiple streams.
  • Balanced Accuracy (35% weight): Average of hit rate and correct rejection rate, smoothed using Jeffreys prior to handle rare matches fairly. Prevents gaming by never responding or always responding.
  • Sensitivity d-prime (10% weight): Signal detection theory measure of how well you discriminate matches from non-matches, independent of response bias. Computed using z-score transformations.
  • Speed Bonus (5% weight): Small bonus for faster correct responses within the response window. Normalized relative to window duration.
  • False Alarm Penalty (35% reduction): High false alarm rates reduce performance score even if other metrics appear strong. Encourages disciplined responding.

Difficulty Multiplier

  • N Level: Higher N contributes superlinearly using exponent 1.35 (N^1.35) since each additional level is disproportionately harder due to memory decay and interference.
  • Stream Count: More concurrent streams increase difficulty exponentially. Multiplier uses exponent 1.10 (streams^1.10).
  • Timing: Faster presets (Advanced, Expert) earn higher difficulty credit. Trial length affects difficulty with exponent 0.65.
  • Lures: Enabling N-1 lures (near-matches) adds 8% difficulty bonus due to increased interference.
  • Feedback: Immediate feedback reduces difficulty credit by 5% since it makes learning easier during sessions.
  • Match Probability: Lower match rates increase vigilance demands. Base rate of 30% is optimal; deviations affect difficulty.

Final Score Formula: Raw Score = 10000 × Performance^0.7 × (0.5 + 0.5 × Challenge), where Challenge is normalized difficulty relative to your configured maximum N. The 0.7 exponent ensures diminishing returns at extreme performance levels. This means you need BOTH high performance AND high difficulty configuration to achieve top scores.

Built With

  • Core Engine: Rust compiled to WebAssembly for performance and security
  • Frontend: Vanilla JavaScript
  • Desktop: Tauri for native Windows, macOS, and Linux builds
  • Storage: IndexedDB via Rust bindings for local progress persistence
  • Audio: Piper TTS

Support Development & Unlock Premium

Support development via Buy Me a Coffee to unlock premium features:

  • Triple and Quad N-Back modes
  • Custom mode configuration with flexible stream selection
  • N levels beyond 3 (up to N=30)
  • Shape and Color stream tracking

Leave your email or contact me at nback@jesse-anderson.net for your license key.

Already a supporter?

Enter your license key to unlock premium features:

Created by Jesse Anderson

Version loading...

Message

🔑 Activate Premium License

Enter the three-part key from your email