How much better before anyone notices

Light one candle in a pitch-black room and the change is enormous — you go from blind to able to see. Light one more candle in a room already blazing with a hundred, and nobody notices a thing. Same physical addition: one candle, one packet of photons. Wildly different felt change. The brain didn’t measure how much you added. It measured how much you added relative to what was already there.

That ratio rule is one of the oldest quantified findings in all of psychology, and it carries the name of a nineteenth-century experimenter: Weber’s law, later extended by Fechner. Ernst Weber found that the smallest change a person can detect — the just-noticeable difference, or JND — isn’t a fixed amount. It’s a roughly constant fraction of the starting value. To notice a weight got heavier, you have to add about the same percentage whether you’re holding a light bag or a heavy one. Perception runs on proportions, not absolutes. The law is approximate, and it stretches and breaks at the extremes of any sense, but across the ordinary middle range it holds with unusual reliability.

Here’s why a builder should care: almost everything you tune is a change against a baseline, and the user’s baseline decides whether your work is even perceptible. Shave 40 milliseconds off a two-second load and you’ve done real engineering that no human will ever feel — 40ms against 2000ms sits under the JND. Shave the same 40ms off a 120ms interaction and it can be the difference between “laggy” and “instant.” Identical effort, opposite outcomes, because the ratio is what registers.

Once you hold the JND in your head, a set of scattered product decisions turn out to be the same decision.

Performance. Improvements that don’t clear the JND are, to the user, no improvement at all — you paid for them and got nothing perceptible back. That’s not an argument against them; they may matter for cost, or headroom, or crossing the line later. But it should change how you sequence work: either push an optimization far enough to cross the noticeable threshold, or batch several small ones so they land above the line together. A dozen invisible 15ms wins can be a single very visible one if you ship them at once.

Price. Weber’s law is exactly why a small percentage increase can slip by unremarked while the same absolute dollars on a cheaper plan cause revolt — the raise is judged against the old price, not in a vacuum. That’s a real lever, and one to hold with clean hands: fine for keeping pace with genuine cost, ugly when it’s a ratchet betting the user won’t do the math.

Quality. To make something feel meaningfully better — sharper images, faster search, cleaner copy — you usually have to clear a proportional gap, not just do slightly more than before. “A bit better than last version” often reads as “the same.” If you want the user to notice the upgrade, aim past the JND on purpose.

⁂

And here’s the edge that should keep you honest: the threshold cuts both ways. If small improvements go unnoticed, so do small degradations. A quality that erodes a few percent per release. A load time that creeps up a hair each quarter. A plan that gets a little worse each renewal. Each step sits under the JND, so no single one draws a complaint — and yet the user ends up somewhere they’d never have accepted in one jump. The same law that tells you when you’re wasting polish tells you how a good product rots without anyone raising a hand. Watch the cumulative drift, not just the per-release diff.

So before you spend a week making something better — or let something get a little worse — ask the ratio question first: better, or worse, compared to what the user is already standing in? And by enough of a fraction that a human will actually feel it?