The Aristeia Scale
Six steps with deliberate holes in the middle, set in themselves, plus a measure comparison that shows where reading breaks.
Hector: Solid, honourable, high-craft work.
Exhibit case
Why it matters
Most type scales fail by being complete. A ratio applied eight times produces steps that are all available, which means the middle of the scale fills with sizes barely distinguishable from each other, and hierarchy collapses into a gradient. The reader can no longer tell what is important because everything is slightly different from everything else.
This scale is built around its gaps. Monument sits far above Statement; Body sits far below Display; nothing occupies the mushy region where a heading and a paragraph look nearly the same. Six steps, each with an assigned job, and no size available for a decision you have not made.
Key techniques
- Six named steps with explicit uses, not a ratio —
Monument,Statement,Display,Lede,Body,Label. - Deliberate discontinuity: the jump from
Display(1.75rem) toLede(1.1875rem) is large enough that the two can never be confused at a glance. - Optical tracking compensation:
-0.035emat Monument,0at Body,+0.16emat Label. Large serif type needs tightening; small mono caps need opening. A single tracking value across a scale is always wrong at both ends. clamp()on the two largest steps only. Body text should not be fluid — it should be the right size and stay there.line-height: 1on display steps,1.68on documentation. Leading is a function of measure and size, not a global setting.- The specimen is set in the actual step, at the actual tracking, so the row is the evidence.
What to steal
The transferable part.Name the steps by job, not by size. Lede tells you when to use it; text-xl does not. Naming by role is what stops a scale from being used arbitrarily.
Track per step. Three values — negative at display, zero at body, strongly positive at label caps — covers ninety percent of the need and is the single fastest upgrade to any existing type system.
Delete steps. If two steps in your scale are never chosen deliberately over each other, one of them is decoration. Removing it makes every remaining choice legible.
The measure comparison. Setting the same paragraph at 46, 66, and 84 characters side by side settles the argument permanently. 84ch fails visibly, and once seen it cannot be unseen.
What not to copy
Context-specific or costly.Do not use line-height: 1 below the Display step. It works at large sizes where descenders are the only risk; at body size it makes multi-line text unreadable and the fix is not obvious to whoever inherits it.
Do not fluid-scale body text with clamp(). Users choose their own body size through browser and OS settings, and a viewport-derived body size overrides that preference for no benefit.
Inclusion rationale
This is the typographic constitution the rest of the canon is set in — every page here uses these six steps and nothing else, so the entry is simultaneously the documentation and the proof. It earns its place on the strength of one transferable idea: that a scale is defined by what it refuses to offer.
Technical register
Performance
Three families, four total weights, all loaded with display=swap from a single stylesheet link. clamp() is resolved by the compositor with no JavaScript involvement, so there is no resize listener and no layout thrash on viewport change. The specimen switch toggles which subtree renders — no font is loaded conditionally, so there is no flash on mode change.
Mobile
Monument is clamp(3rem, 10.5vw, 9rem), which lands near 40px on a 375px viewport — large enough to read as monumental, small enough that a two-word title does not break mid-word. The measure comparison stacks to one column below lg, where each column then gets full width and the 84ch example wraps naturally, which weakens the demonstration slightly but keeps it honest rather than horizontally scrolling.
Accessibility
Every step is expressed in rem, so browser font-size preferences scale the whole system proportionally. Body sits at 1.0625rem — above the 16px floor rather than at it. Label at 0.6875rem is the smallest step and is used only for structural labels, never for content, and its 0.16em tracking keeps small caps legible at that size. The specimen rows use real headings and real paragraphs rather than dummy text nodes, so the document outline is intact for anyone navigating by structure.
Study next