"The Magical Number Seven, Plus or Minus Two" is the most cited paper in interface design and one of the most misread. It gets quoted to cap navigation at seven items, or seven tabs, or seven anything, and none of that is in it. Miller opened by complaining that he was being persecuted by an integer — he was half joking about how often the number seven turned up in unrelated experiments, and he was not confident it meant one thing.
What the paper actually argues is the interesting half, and it's the half that gets dropped. Capacity isn't measured in items. It's measured in chunks — and a chunk can be as big as your familiarity with the material lets it be. One letter is a chunk. So is one word. So is a phone number you've dialled for twenty years. Recoding items into bigger chunks is, in Miller's words, the lifeblood of the thought process, and it is the only lever that actually moves the limit.
So the test below runs twice on the same kind of material, and changes only whether it's bundled for you.
Why the number is soft
Digit span really does come out near seven for most people, which is where the famous number came from. But digits are the most practised symbols you own. You have been bundling them into pairs and triples since you learned a phone number, and you do it without noticing, which means a digit-span test measures your chunking as much as your capacity.
Change the material to something you can't bundle — unfamiliar shapes, colours, nonsense syllables, or anything tested while you're prevented from rehearsing — and the number falls. Nelson Cowan's review of that work puts pure capacity closer to four. Seven is what you get when the material is doing half the work for you.
Which is why the number is the wrong thing to take from the paper. If capacity moved between four and seven depending only on how familiar the material was, then familiarity — not item count — is the thing you have any control over.
What it costs you
Seven is not a menu length. Nothing in the research says a navigation bar should have seven items, because nobody is holding your navigation bar in working memory. They can see it. The limit applies to what people must carry, not to what's on screen.
Format the thing so it chunks itself. A sixteen-digit card number in four groups of four is not a nicety, it is the difference between one chunk and sixteen. Same for phone numbers, sort codes, licence keys, timestamps. Whoever removes those spaces to make the field "cleaner" has made the product measurably harder to use, and the field usually looks tidier in the mockup.
Don't make people carry anything between screens. Every value someone has to remember from step two to use in step four is capacity you have spent on your own navigation. Show it to them instead. Recognition costs nothing; recall costs everything.
Confirmation codes are the honest test. Six digits in two groups of three, on screen at the same time as the field, is a solved problem. Six digits in an email you have to switch apps to read, ungrouped, with a form that clears on blur, is the same six digits and a different product.
I have argued for a design on the grounds that people can only hold seven things. I was quoting a number I had not read the paper behind, to win a meeting, and it worked, which is the part I think about.