Bluma Zeigarnik noticed that waiters in a Vienna café could recall the details of an unsettled order and lost them the moment the bill was paid. She took it into a laboratory, interrupted people in the middle of small tasks, and found they remembered the interrupted ones roughly twice as well as the ones they were allowed to finish.
Below are ten small jobs. Each one carries a name and a bar with four segments, and you fill the bar by clicking. Five of them will not let you finish. They stop part-way and move on without asking, and there is nothing you can do about it.
Then there is a memory test. You are being told about it now, which is a real weakness and I would rather name it than pretend — Zeigarnik's participants had no idea a test was coming, and the effect is about what stays with you when you are not trying to keep it. Knowing makes you rehearse, and rehearsal is the thing most likely to flatten the result.
What the test cannot be gamed on is which names stay. You are not told which of the ten will be interrupted, and you will not know until it happens.
Why the test is recognition and not recall
Asking someone to type the names they remember would produce two or three answers and no usable comparison. Instead the test shows twenty names — the ten you saw and ten you did not — and you pick the ones you recognise. The ten you never saw are the control: picking them is a false alarm, they are counted, and a run with a lot of them is a run where somebody was clicking generously rather than remembering.
Five against five is a very small comparison. A gap of one name is nothing at all, and the verdict will not read anything into it.
What it costs you
An unfinished task is not free. It occupies something. Every half-completed flow, saved draft and abandoned form a product leaves open is a small ongoing cost to the person who left it, not a neutral state waiting patiently for them.
Which is exactly why the progress indicator works so well as a hook — and why using it that way deserves care. Showing someone a profile that is 60% complete creates a real pull toward finishing it. That is genuinely useful when the remaining 40% is worth their time, and manipulative when it exists to serve a completion metric.
The pull disappears the moment the thing is done. Anything you wanted a person to remember should not be attached to the end of a task, because completion is the signal to let go. The confirmation screen is the worst place on earth to put something important.
Interruption is not always damage. A serial that ends mid-scene, a tutorial that leaves one step for later, a form that saves and resumes — these use the same mechanism, and they are the reason a cliffhanger works at all.