A type scale without the math

I have generated a lot of modular scales. Pick a base, pick a ratio, multiply until you have twelve sizes, put them in a token file, feel briefly like a typographer. Then ship an interface and find that eleven of the twelve are wrong and the one you actually need is between two of them.

The problem is not the mathematics. It is that a ratio does not know what your page is for.

Scales are for documents

A perfect fourth works beautifully in a book, where there are three levels of heading and one size of body text. Interfaces are not shaped like that. An interface has a lot of small text doing different jobs — labels, metadata, helper copy, table cells — and then two or three big sizes it uses once each.

The sizes cluster at the bottom. A geometric ratio spreads them evenly. Those two facts are in direct conflict, and the ratio always wins, because it is in the token file and you are in a hurry.

What I do instead

I write down the jobs first, in words, before any numbers exist:

• Body — the thing people read for more than a sentence
• Meta — dates, counts, captions; anything that is a fact rather than a sentence
• Subhead — a break in a page of body
• Title — the name of the thing you are looking at
• Display — used once, at the top, and never again on the same screen

Five jobs. Then I pick a size for each by looking at it, and I stop.

The two rules that survived

Only two pieces of scale theory have earned their place for me.

Steps have to be obvious. If two sizes are close enough that you have to check, they are the same size and one of them should go. A reader should never have to compare.

Line height moves the other way. Big type wants tight leading, small type wants generous leading. Most scale generators hand you one ratio for both and it is wrong at one end.

A scale you can hold in your head beats a scale you have to look up.

Five sizes fit in your head. Twelve do not, which is how you end up with three of them unused and one of them used for everything.