Every other page in this series measures something you did. This one measures something you did and something you believe, and then puts them next to each other. That is the whole law: the gap between the two, and which one you would have sworn by.
Below is one panel of nine words. You will be told a word and you click it, twelve times. The panel is wearing two different sets of clothes, and it alternates between them every round. Everything that could affect how fast you find a word is identical in both — the nine words, the grid, the position and size of every cell, the type size, the label colour and the colour behind the label. What changes is corner radius, border weight, a shadow, and the tone of the panel behind the cells. Taste, in other words, and nothing that a stopwatch should care about.
At the end you get two questions, and you answer them before you are shown any numbers. Then the numbers.
The honest caveat, up front: you have read the title, so you know what this page is hoping to catch you doing. That is the single biggest weakness in this demo and I cannot design it away. It is why the preference half is a demonstration and the timing half is the measurement — you can decide to answer a question contrarily, but it is much harder to decide to be faster.
Why the rounds alternate
You get faster at this as it goes — you learn the nine words, you learn where to look. If all six rounds of one look came first, that practice would land entirely on the other one and the result would be an artefact of the running order. So the looks strictly alternate, which spreads practice evenly across both by construction rather than by hoping.
The rounds are also matched in pairs: the two rounds of a pair put the target in the same cell of the grid, so within a pair the only difference is the surface and the particular word. That pairing is what the comparison at the end is built on — six pairs, and for each one, which look you were faster in.
Six is a cruel number here, and deliberately so. If you flipped a coin six times, getting the same face every time happens about once in thirty-two runs. So you would have to be faster in all six pairs for the result to mean anything at all, and anything less than that is a coin. Most people's times come out looking exactly like a coin, which is the point.
What it costs you
Your users' praise is not usability data. "It feels so much easier now" after a visual refresh is a report about the refresh, and it will arrive whether or not a single task got shorter. If it is the only evidence you have, you have no evidence.
It also runs the other way, and that is the expensive direction. A genuinely faster flow in an ugly shell gets described as clunky, and the team spends the next quarter fixing something that was never broken. Anyone who has watched a rewrite get rejected on feel knows this one.
Beauty buys patience, and patience is real. The effect is not purely an illusion to be debunked — people forgive more, try longer, and blame themselves rather than the product when it looks considered. That tolerance is worth having. It is just not the same thing as working, and the trouble starts when a team spends it without knowing it has been spent.
So measure the two separately, and never let one stand in for the other. Time on task and error rate answer one question. How it felt answers a different one. Both are worth knowing; a team that only collects the second will ship something admired and slow.
The argument I keep having is about redesign reviews, where the new thing is always shown finished and the old thing is always shown as it is. Of course the new one wins. Comparing a polished mock to a live screen is not a comparison, and by the time anyone has numbers the decision was made months earlier, on a slide.