Knowing when to stop polishing
A practical stopping point for the details that can keep a project open forever.
You can tune a shadow forever. Every nudge gives you an answer within a second: look, decide, nudge again. Finding out whether someone understands a button takes another person and a few minutes, so the shadow keeps winning.
The card below has three fixes that have nothing to do with taste. Turn them on one at a time.
Each fix answers a different question
"OK" and "Save note" fit the same pill, and only one of them tells you what happens when you press it. More room between the title and the action makes them read as two separate things. The focus ring shows a keyboard user where Enter will land.
I can explain each of those in one sentence. That's my test for polish: if I can't say what a change fixes, it goes to the bottom of the list.
Looks and usability do pull on each other, though less neatly than the slogan suggests. In a 2012 study by Tuch and colleagues, 80 participants used different versions of an online shop. A better looking shop didn't feel easier to use, and a shop that was harder to use got lower beauty ratings afterwards. That's one task in one setting. I still take the warning: liking the mockup tells you nothing about whether the real thing works.
Get it out of the mockup
Put the card on a real page. Paste in a title long enough to wrap. Tab to the button with your hand off the mouse. Trigger the error state, then use the card again afterwards.
Some of it isn't a matter of taste at all. WCAG's Focus Visible criterion requires a visible focus indicator. In the card it's drawn as a box shadow, so turning it on never shifts the layout:
.action:focus-visible {
box-shadow:
0 0 0 2px var(--surface),
0 0 0 4px var(--ink);
}
The ring in the study is a drawing of that state. A drawing of a ring doesn't make a component accessible, so the working control still has to be tested with a keyboard.
Where I stop
Before another round of polish, I write down what the round is meant to fix. If the sentence comes out as "this might look nicer", the variant gets saved and moved behind anything broken or unclear.
Visual experiments still have somewhere to go. An odd spring or a strange hover goes into the Lab, where it only has to be interesting.
On a real page, I stop when the next useful thing I could learn needs someone else's hands on it. Another hour staring at the same card won't tell me that.