Play first
Practice starts with one obvious action and immediate correctness feedback. Daily and Sprint are short repeat-use runs; Benchmark keeps the longer fixed form when comparability matters.
V0.348 · GAME-FIRST VISUAL WORKING MEMORY
Change Detection
Remember the colors and locations, hold them through the blank delay, then judge one probe as Same or Different. Results keep accuracy, sensitivity and K evidence separate.
Practice
ProgressReady
Accuracy—
Median RT—
RunClean
Remember the colors and locations. Accuracy matters more than speed.
PausedVisual timing and response timing are frozen until you resume.
Remember the colors and locations.
A single square returns at one remembered location. Choose Same if its color matches, Different if it changed.
Keyboard: S / ← = Same · D / → = Different
18 scored · mixed 4 / 6 / 8 · 500 ms study · 900 ms delay
RUN COMPLETE
CleanPractice result
Accuracy—
Sensitivity d′—
K max—
Median correct RT—
Hit rate—
False-alarm rate—
Criterion c—
Scored trials—
| Set | N | Accuracy | Hit | False alarm | d′ | K |
|---|
K shown here uses K = N × (hit rate − false-alarm rate), clamped to 0…N, separately by set size. It is a model-dependent descriptive estimate, not an IQ score, diagnosis or population percentile.
Exact trial review
Finish a run to inspect every scored trial.
Recent runs, bests & Daily streak
Measurement notes & fixed benchmark forms
Benchmark A–E each preserve the V0.249 deterministic 60-trial single-probe stream: 20 trials at set sizes 4, 6 and 8 with balanced Same/Different truth. The main Benchmark button rotates forms locally; you can pin a form in Mode & options.
Accuracy, d′, hit rate, false-alarm rate, criterion c and K answer different questions. No single value is promoted as a universal “memory grade.”
Advanced V0.217 evidence & repeatability Lab
ADVANCED · SECONDARY LAYER
Change Detection Evidence Lab
Whole-array, single-probe, long-delay, fixed A–E, adaptive, custom, review and backup tools remain available without competing with the live game surface.
Evidence stays separated
Accuracy is readable, d′ accounts for hits and false alarms, criterion shows response tendency, and K remains broken out by set size rather than collapsed into an opaque grade.
Repeat exactly
Exact share/replay preserves the seed and setup, local history keeps provenance, and trial review plus CSV makes every completed run inspectable without an account.
Change-detection performance depends on procedure, device, attention, timing and scoring assumptions. Compare your own runs under similar conditions; this page is not a clinical assessment.