The Last Ones Go First
The end of a list comes back best of all, and fourteen seconds of arithmetic is enough to take it away without touching the beginning. Two knobs, two humps, and a page that commits to a number before it turns the second knob.
Twelve unrelated words arrive one at a time. Then you type back as many as you can, in any order. Everyone produces the same shape: the first few come back, the last few come back best of all, and the middle sags. That curve is a hundred and forty years old and it is not the interesting part.
Two humps could be one memory with two good positions in it, or two memories. A curve on its own cannot say which. What can say which is a knob that moves one hump and leaves the other alone, and then a second knob that does the opposite.
Fourteen seconds of arithmetic between the last word and the first keystroke. The end of the list should go. The beginning should not move.
Each word on screen three times as long. The beginning should rise, because early words get rehearsed while the list is still short. The end should not move.
Eighteen lists are measured with one knob at a time. Then this page stops, fits two rival models to them, and prints what each one predicts for a condition you have not run yet: slow words AND the arithmetic, both knobs at once. One store says the end of the list is still the tall part. Two stores says it is gone. Both numbers go on the screen with their own intervals, and only then does that block run.
About twenty minutes if you answer and move on, twenty five if you use every second of every recall. It is a real one: two practice lists and then thirty, a keyboard, and a room where nobody is going to talk to you. Nothing is recorded and nothing leaves the page. Writing the words down breaks it, and the page can tell.