Parkinson was writing about civil servants and committee agendas, but the version that turns up in products is smaller and stranger: tell somebody roughly how long a thing will take, and they will take roughly that long, whether or not anything makes them.
Six identical rounds below. Every one asks you to find the same number of marked cells in the same size of grid. Before each round you are told what it usually takes — three of them say about twenty seconds and three say about six — and which kind you get alternates, so the practice you pick up along the way lands evenly on both.
Nothing is enforced. There is no countdown, no timer running down on screen, and nothing happens if you take longer. The estimate is only ever a sentence you read before you start, which is precisely the point: this is a test of whether a number somebody else wrote changes what your hands do.
Why there is no countdown
A visible timer would settle the question in the wrong direction. Watching a clock run out makes people hurry, which is not Parkinson's claim — his claim is that the allowance alone does the work, quietly, without pressure and without anyone noticing they are pacing to it.
It also has to be an estimate rather than a deadline. A deadline that nothing enforces gets discovered as toothless within two rounds and stops mattering. "This usually takes about twenty seconds" is what a real product says, and it never has teeth in the first place.
Three rounds of each is a very small comparison, and one round where you lost your place can swallow the whole result. The verdict uses the middle of each set and refuses to interpret a gap smaller than the noise three rounds can carry.
What it costs you
Every estimate you show is an instruction. "About five minutes to set up" does not describe the work, it shapes it. If your onboarding says ten minutes, it will take ten minutes, and it would have taken six if you had said six.
Which makes an honest generous estimate expensive. Padding a number to be safe is sensible engineering and it costs real time out of the person's day, every single time.
The lever also works the useful way round. Showing that something takes thirty seconds, when it truly does, gets it started — most abandonment before step one is a guess about length, not a judgement about value.
And it explains why removing steps often fails to speed anything up. If the expectation stays where it was, the remaining steps quietly grow to fill the gap the removed ones left.