A legible face is not automatically readable
Aliases: display versus text · discriminable but unreadable · construct split
What it is
A wayfinding face makes every letter unlike every other, so the name can still be called at fifty metres — a win for legibility. Set the same face on a three-thousand-word manual and every fixation is full of corners and oversized counters, word-shapes shout, two pages tire you. Passing discrimination does not pass efficiency. The two constructs differ in materials, time scale, and success criterion. A face can be specialised for one and throw the other away.
The reverse also holds: some text faces are unremarkable on isolated letter tests and then, set as paragraphs, win on speed and comprehension because the rhythm is steady and the texture even. Choosing a long-text face from a legibility specimen, or a one-time-code face from “it reads comfortably,” is using the wrong ruler.
Why it happens
Legibility rewards distinctive features: exaggerated apertures, oversized counters, a deliberately asymmetric l and I, extra decorative strokes on Han. In a short identification those features are signal; in a chain of fixations they are noise — every letter shouting “look at this hook,” the word as a clump torn apart, the visual span filling with souvenirs instead of words.
Readability rewards regularity that can be ignored: a moderate x-height, a stable width rhythm, a weight that does not perform, so periphery can walk on word-shape rather than on the alphabet. The best faces for long copy are often the ones you cannot describe, and you do not remember fighting, after you finish. Display cuts and geometric sans whose “every letter is a logo” walk the opposite road.
All-caps, overly wide tracking, and an oversized x-height can manufacture the same split: letters become easier to name (legibility up), word outlines become flatter and more scattered (readability down). That is why the same variables can move the two constructs in opposite directions. They are not two ways of saying “clean.”
Studying it
To show they come apart, run two tasks on the same face: a confusion matrix or timed short-string identification, and paragraph speed plus comprehension. The independent is still the face; the dependents split into two columns. If face A wins on short strings and loses on paragraphs, and face B the reverse, the split holds. Tinker already reported the two tasks separately; later work that reports only one column cannot give a single verdict that “this type is better.”
Do not mix materials. Inferring code-legibility from paragraph scores, or novel-readability from alphabet scores, is a construct error.
Where it stops holding
Interfaces of three to five characters — buttons, nav, badges — live almost only on the discrimination layer; a face whose every letter is unique is reasonable, and there is no long text whose readability to worry about. Novels, terms, and help centres live almost only on the efficiency layer; a wayfinding face’s discrimination advantage is unused. Mid-length settings copy needs both: serials and units still have to split, then three paragraphs still have to be finishable, which often means two specs rather than one “good typeface” that wins everywhere. Change the reader’s job and the advantage flips: the same geometric sans may be right for airport signs and wrong for an ebook.
Applying it
- Split type uses into two lists — “short, must split” and “long, must finish” — and allow two faces. Do not hunt for one that ranks first on both specimens.
- Short uses get confusion pairs; long uses get paragraph speed and completion. Do not let one scoreboard eliminate candidates from the other list.
- Wayfinding, serials, and codes may lean legible; body, help, and terms lean readable. Titles may take a display cut; do not pour that cut into the paragraph style.
- If a face is perfect on short strings and widely abandoned across three pages of explanation, write down that the construct was reversed, not that “the type is bad.” Keep it for short uses; give long copy a quieter rhythm.