Whatever Counts As One
The limit on what you can hold counts things, not information. Which means the size of a thing is yours to set, and this page hands you a bigger one and measures what happens.
A short list arrives one item at a time. You give it back, in order. That is the whole task, it is from 1887, and the number that falls out of it is your span.
The lists are made of four different things, and what they are made of is the entire point. Ten digits carry 3.32 bits each. Binary digits carry one bit each. If your limit were a fixed amount of INFORMATION, a binary list should come back more than three times longer. If your limit counts THINGS, the two should come back about the same length and the bits should differ threefold.
Miller wrote this down in 1956 as plainly as anyone ever has: the number of bits is what stays constant in absolute judgment, and the number of CHUNKS is what stays constant in immediate memory. Which leaves the question this page exists to put a number on. What counts as one thing?
Between the fitted blocks and the last one you spend ninety seconds learning four pairs: 00 is 0, 01 is 1, 10 is 2, 11 is 3. Then the binary lists come back, shown in pairs exactly as they were in the block before, and you give them back in binary exactly as you did before. Nothing about the screen or the keys changes. The only thing that changed is inside your head.
Both go on the screen before the block runs, both carry their own intervals, and the block is scored by which interval it clears rather than which number it lands nearer. Scoring by nearness is how a page in this lab nearly shipped a bias, and the fix stayed.
There is a boring explanation sitting on that last block and it is a good one. Binary shown in pairs is easier to rehearse whether or not you know what a pair means, and the recoded block is also the LAST block, so an hour of practice is pushing on it too.
So the block just before the training is the same paired binary with no mapping taught. It carries the grouping. It carries the practice. It does not carry the recoding. The strong test here is not recoded against raw binary. It is recoded against GROUPED.
One block is long words, ten of them, four syllables each. Ten words carry exactly as many bits per item as ten digits and exactly as many chunks, so a pure chunk account says the two spans are equal. They are not. Long words come back shorter, because what fits is partly what you can say in about two seconds. This page measures that on the same evening and prints it against its own headline. A chunk is not free, and that block is where you can see the price.
This is the first page in this lab whose headline never reads a clock. A span is a COUNT of items, not a latency, so display lag, input lag and keyboard scan interval have nowhere to land. What could reach it is the presentation RATE, and every item gets the same slot in every block: a pair gets exactly two slots, so a paired list and an unpaired list of the same length take exactly the same time to arrive. Every number below is a difference or a ratio of two spans measured at one rate in one sitting.
Five blocks and a ninety second drill, about thirteen minutes, the first three trials unscored. Tap the buttons, or use the number keys, Backspace to undo, Enter to submit. No item appears three times in a row. Trials taken while this tab was in the background, or interrupted by a long frame gap, or answered while the list was still arriving, are dropped and re-run rather than scored. Every list is placed where it is worth the most, which means you are supposed to get about half of them wrong. If you get nearly all of them right the page refuses the run, because the usual reason for that is a pen.
Everything runs in your browser. Nothing is recorded. Nothing leaves the page.