N-back test: method

What it measures

Performance on the n-back task itself: continuous updating and matching of recent items in working memory. It is a practice tool with its own progress graphs; it is not a normed working-memory assessment.

Procedure

Streams: position (one of the 8 outer cells of a 3 x 3 grid; the centre is never used), audio (8 spoken letters, or 8 tones called Notes), colour and shape (8 values each). Modes: single (position), dual (position + audio; the default), triple (+ colour) and quad (+ shape). Each cue is shown for 500 ms; the next cue follows after the chosen interval (standard 3,000 ms). A response for a stream counts for the current cue until the next cue. Keys: A (position), L (audio), S (colour), K (shape), Space pauses; on phones, buttons below the grid. A block has N warm-up cues (no responses) plus the scored cues (standard 20). Sequences come from a seeded generator that the server replays: in each stream exactly 30% of scored cues are matches (6 per 20), each pair of streams matches together on 10% of cues (2 per 20), three- and four-stream modes include cues where all or three streams match, and non-matches never accidentally repeat the N-back item. An optional 1-back tutorial gives feedback. The default preset is adaptive (standard timing and length); a manual preset lets the player set N, interval, length and feedback. An optional daily set has 20 blocks with breaks. Unfinished blocks pause when the page is left and can resume.

Parameters

Scoring

Per stream, over scored cues: hits, misses, false alarms, correct rejections; balanced accuracy = 50 x (hit rate + correct-rejection rate), in percent, so responding to nothing (or to everything) scores 50%. The block score is the mean balanced accuracy across the selected streams; higher is better. Errors per stream = misses + false alarms, which drive the adaptive level for the next block. Reaction times per stream are summarized (median, mean, min, max) but do not enter the score. There is no comparison with other players; graphs compare the player's own sessions with the same settings.

Valid-run rules

Only complete blocks are recorded (N + scored cues, one response flag per stream per cue, no responses during warm-up). The server regenerates the sequence from the seed and rejects a session whose logged stimuli differ from it, whose accepted answers are missing, or whose finish time lies in the future. Pauses are allowed and logged. Bot signals are stored but never change scoring or graphs. These practice sessions are kept separate from the scored n-back test and never feed its comparisons or app scores.

Differences from the classic task

The standard block follows the timing and adaptation rule of Jaeggi et al. (2008): 20 + N cues, 500 ms display, 3 s per cue, up with fewer than 3 errors in every stream, down with more than 5 in any stream. Differences: spoken letters made with computer-generated speech (or tones), optional colour and shape streams, a short tutorial, phone controls, manual settings and self-paced daily blocks. It is not a replication of a research training intervention, and it is not Kirchner's original single-stream task.

Limitations

Gains on the trained task are well documented; transfer to other abilities, including intelligence, is mixed and debated. Nothing on the page promises such transfer. Audio and display timing depend on the device and browser; the log records when the page dispatched each cue, not when the screen or speaker produced it. There are no norms, and the setting is unsupervised. No clinical or diagnostic use.

Languages and review status

Instructions in all site languages. The audio stream uses native letter or syllable sets in 28 languages (Bengali, Bulgarian, Chinese Simplified and Traditional, Czech, Dutch, English, French, German, Greek, Hindi, Hungarian, Italian, Japanese, Kazakh, Marathi, Nepali, Persian, Polish, Portuguese, Russian, Serbian, Spanish, Telugu, Ukrainian, Urdu, Albanian, Armenian), recorded with computer-generated speech; other languages use the tone option. Position, colour and shape streams are language-independent.