Hedwig von Restorff put a list of similar items in front of people, made one of them look different, and found that the odd one came back far more often than its neighbours. It is the reason a primary button is a different colour from everything around it, and the reason a page with nine primary buttons has none.
The demonstration has an obvious trap, and it is the same trap the serial position page is built around: if the odd item sat at the start or the end of the list, position alone would explain the result. You keep the beginnings and endings of lists regardless of what they look like.
So below, across five lists of nine words each, the boxed word never appears in the first two or last two places. It lands somewhere in the middle five, in a different place each time. What its recall gets compared against is the other middle words in the same lists — same region, same disadvantage, same everything except the box.
The box is the only difference. The word is the same size, the same typeface and the same colour as its neighbours, sitting in the same place on screen for the same length of time. It simply has a line drawn around it.
What it is measured against
Five lists give five boxed words and twenty unboxed ones from the same middle region. The comparison is those two rates, and nothing else — the first two and last two words of every list are shown, but they are excluded from the scoring entirely, because they carry an advantage the box has nothing to do with.
There is a third bar in the results, and it is the one to look at first. Nine of the names in each recognition round were never on screen. Picking those is a false alarm, and a run with several of them is a run where somebody was clicking generously rather than remembering — in which case the other two numbers mean nothing and the verdict says so instead of interpreting them.
What it costs you
The effect is a budget, not a technique. One distinct thing on a screen is remembered. Two are competing. Six are a pattern, and a pattern is the thing the odd one was supposed to stand out from. Every element you emphasise spends from the same pot, and the pot is small.
Emphasis is relative to its surroundings and nothing else. There is no such thing as an inherently prominent component. A coral button is loud on a quiet page and invisible on a busy one, which is why a component library can hand you a "primary" style and still not give you emphasis.
Distinctiveness gets remembered, which is not the same as understood. The odd item is what comes back later — that is all this law promises. It says nothing about whether the person knew what it did, and an emphasised control that is unclear is simply a memorable confusion.
The most expensive place to spend it is a place nobody needed help with. Highlighting something the eye was going to reach anyway costs the same budget as highlighting the thing people miss, and buys nothing.