DUAL N-BACK · PRACTICE GUIDEDual N-Back Guide
Dual N-Back is a continuous updating task: compare each current position and tone with what appeared N trials earlier, then mark only the channels that truly match. This guide explains the exact browser implementation, how to read hits, misses and false alarms, how deterministic Practice works, and what the numbers do and do not mean.
The core rule
N-back asks you to compare the cue you are seeing or hearing now with the cue that appeared a fixed number of trials earlier. In 2-back, for example, the current cue is compared with the cue two steps behind it, not with the immediately previous cue. The comparison window moves on every trial, so the task is continuous rather than a set of isolated flash-card questions. This implementation keeps position and sound as separate channels. A position match can occur without a sound match, a sound match can occur without a position match, both can occur together, or neither can match. That is why the two response buttons are independent. The rule is simple to state, but reliable play depends on updating the recent sequence instead of remembering only the last item.
Visual, Audio and Dual modes
Visual 30 isolates the spatial stream. A square lights on a three-by-three grid and you mark Position match only when that square equals the square N trials earlier. Audio 30 isolates the sound stream and uses short tones synthesized by the browser, so there is no downloaded voice pack or external audio service. Dual 30 combines the two streams and scores them independently. Keeping these modes separate matters because a person can learn the N-back rule in one channel before attempting two channels at once. Raw percentages from different modes should not be treated as interchangeable difficulty ratings. The purpose of the mode labels is to make the exercise understandable and repeatable, not to imply a universal ladder of mental ability.
How matches are generated
Practice sequences are deterministic from the URL seed, the selected mode and the selected N. The generator first chooses a controlled set of eligible trials that should be matches. It then creates the remaining cues while explicitly preventing accidental N-back matches on trials that were intended to be non-matches. That detail avoids a common generator problem where random values silently change the match density. The same seed therefore recreates the same cue sequence and the same match locations. The platform does not need a server database or a licensed puzzle corpus for this specialist. This also makes the implementation independently testable: a verifier can recompute the N-back relationship from the generated sequence and compare it with every stored match flag.
Scoring without invented rankings
Each active channel on each trial becomes one of four outcomes. A hit means you marked a real match. A miss means a real match occurred and you did not mark it. A false alarm means you marked a match when the cues were different. A correct rejection means you correctly left a non-match unmarked. Response accuracy is the share of these channel decisions that were correct. The site also stores a local progress score so Practice history and My Mix have one compact value, with a small false-alarm penalty. That site score is not a population percentile, age norm, school rank, clinical score, driving assessment or employment measure. It is only an internal summary of performance on this browser task.
Choosing an N level
One-back is the easiest rule to learn because each cue is compared with the immediately previous trial. Two-back introduces a larger moving window and is the default starting level for Dual mode. Higher N values increase the amount of sequence context you must keep updated, but there is no reason to force a high number when accuracy collapses or false alarms rise. A useful practice session is one in which you can follow the rule consistently enough to learn from the outcome categories. The page supports 1-back through 5-back because that range provides progression without pretending that a higher number is automatically better for every user, device or session.
Adaptive mode
Adaptive 3×18 uses three shorter dual-channel blocks. It begins at the selected or locally recommended N. After a completed block, the next N can move up when accuracy is strong and false alarms remain controlled, stay the same when performance is in the middle range, or move down when accuracy is low or false alarms are frequent. The adjustment is based on the actual completed block rather than on a hidden difficulty label. The recommendation saved for later Practice is local to this browser. Adaptive mode is intended to reduce the friction of manually selecting an N every time, not to produce an intelligence score or a permanent classification.
Timing and interruptions
The cue pace is part of the session contract, so a background tab or a browser window that loses timing continuity should not quietly produce a normal record. If the page becomes hidden during a run, the session is invalidated and must be restarted for a clean result. Browser scheduling, power-saving behavior and audio policy can differ across devices, so this rule protects consistency rather than claiming laboratory-grade timing. Audio playback is generated through the Web Audio API after user interaction. If you cannot hear the Sound check, use Visual mode or fix device audio before starting an Audio or Dual session instead of guessing at unheard cues.
What research claims are not made
N-back is widely used as a working-memory task, but a web game should not turn that fact into claims it cannot support. A score here does not diagnose ADHD, memory impairment, neurological disease or any other medical condition. It is not an IQ estimate and it does not prove broad improvement in intelligence, school performance or unrelated abilities. Research on transfer from working-memory training to other skills has produced mixed findings, so the product language stays task-specific. The useful interpretation is narrower: you can observe how accurately you detect N-back matches under a repeatable rule and compare your own sessions when the setup is similar.
Local-first records
Completed Practice sessions can store recent results, total hits, misses, false alarms, personal bests for a mode and N combination, and a locally recommended N. These records stay in browser storage and do not require an account. Clearing site data can remove them, so they should not be treated as permanent cloud records. This design keeps the specialist usable without login friction, subscription APIs or personal profile data. It also means a result is primarily for self-comparison. If you move to a different device, browser, keyboard, speaker setup or input environment, differences in the setup should be considered before comparing raw numbers.
A practical routine
For a first session, start with Visual 1-back or Visual 2-back and focus on understanding the moving comparison window. Then try Audio at the same N and use Sound check first. When both single-channel modes feel clear, use Dual 2-back and respond to the two streams independently. Review misses and false alarms rather than looking only at the headline percentage. If false alarms are high, slowing your decision process mentally can be more useful than moving to a higher N. Adaptive can be used after the rule feels familiar. A small number of attentive sessions is more informative than repeatedly clicking through sessions while distracted.
Accessibility and controls
The main responses are large buttons designed for mouse, trackpad and touch, and the keyboard provides A for Position match and L for Sound match. Visual cues use a clear three-by-three grid with a highlighted cell, while Audio mode visually mutes the grid so the interface does not suggest that position should be answered. The page reports the current trial count and current N in text as well as visually. Audio remains optional because hearing conditions and device policies vary. The goal is to make the rule and result categories explicit, with no need to infer hidden scoring behavior from animation alone.
Reading a completed result
Suppose a Dual 2-back session reports high overall accuracy but several false alarms. That pattern is different from a session with the same percentage and several misses, because the first reflects over-calling matches while the second reflects failing to respond to real matches. Looking at the categories helps you decide what to practice next. Hits alone are also incomplete because a person could press both match buttons on every trial and collect hits while creating many false alarms. The scoring therefore includes all channel decisions, not just successful match clicks. Personal best logic also considers false alarms when percentages tie.
Why this is distinct from sequence memory
A traditional sequence-memory game usually shows an expanding sequence and asks you to reproduce it after presentation. Dual N-Back never asks you to replay a whole list. Instead, the stream keeps moving and the question on every trial is whether the current cue matches the cue N positions earlier. Older cues fall out of the comparison window as new cues arrive. The two-channel form can also require separate decisions about spatial and auditory information on the same trial. That continuous updating loop is why this specialist sits beside, rather than inside, the platform’s existing Memory Grid sequence and number-memory exercises.
Scope: This guide describes the actual V0.106 browser implementation. The specialist is local-first, uses deterministic cue generation and synthesized tones, and remains outside Today’s 6.