What You Kept
A second after you look away, almost all of it is gone and a handful is left. This puts a number on the handful, and then asks the question nobody has settled: is it a handful of things, or one blanket stretched too thin?
Coloured squares flash for a fifth of a second. A second later, one of them comes back and you say whether it changed. Up to three or four squares this is trivially easy. Past that it collapses, and it collapses in the specific way that a fixed number of held items plus guessing would collapse. Luck and Vogel put that in Nature in 1997 and the number has barely moved since.
Then the harder round. Instead of asking whether a colour changed, the page asks which colour it was, on a continuous wheel, and reads the shape of your errors. That is where the twenty year argument lives: a few things held sharply and nothing for the rest, or everything held badly at once.
- 1️⃣ Round one. Two, four, six and eight squares. Flash, gap, one square returns. Same or changed.
- 2️⃣ Your capacity comes out as K = N x (hits − false alarms), which is Cowan's formula and the reason the false alarms matter as much as the hits.
- 3️⃣ Round two. Three squares or six, and instead of a yes or no you point at the colour on a wheel.
- 4️⃣ Those errors get fitted twice, once as slots and once as a stretched resource, three parameters each, and the page tells you which one your own data preferred.
- 🔒 The buttons are armed at probe onset, not merely locked. A press before the square exists is not an answer, and the number of them is printed at the end.
- ⚖️ Half of every set size really changes. Answering "changed" every time gives a capacity of exactly zero, and so does answering "same" every time.
- 🎨 One circle in CIELAB, not an RGB rainbow. A page that measures errors in degrees cannot use a ruler whose units change along its length.
- 🎡 The wheel is randomly rotated every trial, so a motor habit cannot look like a memory.
- 🙈 No feedback while anything is being measured. Feedback trains a criterion, and the formula assumes your criterion sat still.
- ⏱️ The flash is measured, not assumed. A browser shows you an array in whole screen refreshes, so the delivered duration is timed and printed with its spread.
The shorter run is a real choice, not a footnote, but it buys the time by making every interval on the results page wider. The mixture fit in particular is hungry, and it says so when it is underfed.
Set the number of squares yourself and find the point where it stops feeling like looking and starts feeling like remembering. For most people that switch happens between three and five and it is not gradual. Feedback is on here, because nothing here is measured.
Everything on this page runs in your browser. Every colour is computed on your machine and forgotten, nothing is recorded, nothing leaves the page.