Mind Forge All tools
Working memory

Adaptive Dual N-Back

The classic Jaeggi-style working-memory task — with a per-modal accuracy engine that adapts the level to your current threshold, and breaks down hits, misses, and false alarms separately for the audio and visual streams.

What makes this one different

Most online n-back trainers log "max level reached in a session." That's misleading — a single lucky streak inflates the number. We adapt based on a rolling 10-trial window per modality, so you see your actual stable threshold.

Level 1 Init
Visual
Audio
—
Press J if same as n-back
Trial
0 / 20
Hits
0
False alarms
0
Press F when the visual cell matches the one n-back · Press J when the audio letter matches.
Ad slot — placed below the exercise, well clear of buttons & inputs.

Per-modal performance

The breakdown most online n-backs hide. Lots of people have asymmetric n-back — strong on visual, weak on audio, or vice versa.

Visual
—% accuracy
Audio
—% accuracy
Vis hits
0
Vis FA
0
Aud hits
0
Aud FA
0

Your last sessions

No sessions yet. Run one to start tracking.

What problem does it solve?

Working memory — the small "scratchpad" you use to hold a phone number, follow a multi-step instruction, or track where you are in a sentence — is one of the strongest cognitive predictors of fluid intelligence. The Dual N-Back task is the most-studied single exercise targeting it. The honest reason people keep coming back to it: it's the rare cognitive task where the difficulty scaling is measurable, not vibes-based.

The problem with most "brain training" apps is they don't tell you whether you got better at the task or just at recognizing the app. This one does.

Who is this for

It is not a clinical assessment. If you suspect a real working-memory problem, see a neuropsychologist.

How to interpret your results

Stable level (the highest level you can hold for ~10 trials at >70% accuracy) is the headline number. Anything else is noise.

Per-modal accuracy tells you whether your working memory is balanced or lopsided. Many people score 1-back on audio but 2-back on visual, or vice versa. That's a real cognitive signal, not a bug.

False-alarm rate matters more than people think. Pressing F or J every time you see something close is not "trying hard" — it's noise that lowers your effective threshold. Calmer performance usually unlocks higher levels.

Session-to-session, your stable level should move slowly (weeks, not days). If it jumps 3 levels in a day, you're probably chasing random fluctuation.

Limitations (read this before drawing conclusions)

The transfer problem

The largest, most rigorous study on this — the BBC "Brain Test Britain" study with 100,000+ participants (Gow et al., 2022 in Nature Human Behaviour) — found that practice on a working-memory task like n-back does not reliably transfer to broader cognitive ability. Improving your n-back number mostly means you got better at n-back.

Other limitations specific to this implementation:

  • It runs in a browser, so timing is slightly less precise than lab-grade software (expect ±15 ms jitter vs. ±2 ms in dedicated equipment).
  • We use the letters C, H, K, L, P, Q, R, S, T, X (the original Jaeggi set) but on a 3×3 grid. If you've trained on a 4×4 grid elsewhere, your numbers won't directly compare.
  • Sound quality depends on your speakers/headphones. We use short, distinct consonants, but on bad speakers they can blur together.
  • The adaptive algorithm is straightforward (rolling-window accuracy). A more rigorous staircase (QUEST, Psi) would be more efficient but harder to interpret.

Frequently asked questions

What's a "good" n-back level?

The original 2008 Jaeggi paper reported average untrained adults around 2-back to 3-back after a few weeks of practice. Reaching 4-back consistently is strong. 6-back is genuinely elite territory. But more important than the absolute number is your personal trajectory over weeks.

How long should I train?

Most research uses 20–25 minutes per day, 4–5 days a week. We cap sessions at 20 trials per level for that reason. Going longer tends to introduce fatigue that masks real learning.

Why does the audio one feel harder?

For most people, it is. Verbal working memory (audio) tends to be slightly weaker than visual-spatial working memory in untrained adults. The per-modal breakdown chart will show this clearly if it's true for you.

Can I use this instead of Lumosity/BrainHQ?

If your goal is "practice a well-defined cognitive task for free without giving my data to a corporation," yes. If your goal is a managed, multi-domain daily program with adaptive algorithms and progress reports, those services do more than this single page. We're not trying to replace them.