Typography Systems

A type system is the set of rules that turns raw fonts into a coherent, scalable voice across every screen. Get it right and hierarchy, rhythm, and readability come for free; get it wrong and no amount of color or layout rescues it. This is the bible: anatomy → scale → spacing → selection → variable/OpenType → rhythm → hierarchy → tokens → accessibility → mistakes.

1. Anatomy (the vocabulary you need to make decisions)

• Baseline — the invisible line letters sit on. The anchor for vertical alignment.
• x-height — height of a lowercase x. The perceived size of a font; two fonts at 16px look different sizes because of it. High x-height (Inter, Source Sans) reads larger and better at small sizes; low x-height (Garamond) feels elegant but needs more size.
• Cap height — top of a capital letter. Usually below ascender height. Aligning UI elements to cap height (not the font's em-box) is what makes icons + text look centered.
• Ascender / descender — strokes above x-height (h, b, k) / below baseline (g, p, y). They eat into line-height; this is why a 16px font occupies ~22px of vertical space.
• Counter — enclosed/partially-enclosed space inside o, e, a. Open counters survive small sizes and low resolution.
• Aperture — the opening of a counter (the gap in c, e, a). Wide apertures (Frutiger, Inter) = legible at distance/small size; closed apertures hurt accessibility.
• Terminal — the end of a stroke not finished with a serif. Shape signals the typeface's character/era.
• Em / em-box — the design grid the font is drawn on. 1em = the font-size. The reference for all relative metrics.

> Why care: you can't pick a UI font, set a scale, or align to a grid without knowing x-height, cap height, and how ascenders/descenders consume line-height.

2. The modular scale

Pick a base size and a ratio; multiply repeatedly to generate steps. This produces sizes that relate harmonically instead of arbitrary 14/15/17/19px noise.

Base size: 16px on web — it's the browser default, the minimum for comfortable body text, and the anchor for rem. Never set html { font-size: 14px } (it shrinks everyone's accessibility baseline).

Common ratios (smaller ratio = subtle, dense UI; larger = dramatic editorial):

| Ratio | Name | Feel | Use for | |------:|------|------|---------| | 1.125 | Major second | Tight, calm | Dense dashboards, data UI | | 1.200 | Minor third | Gentle | Product UI, forms | | 1.250 | Major third | Balanced (most popular) | General web/app UI | | 1.333 | Perfect fourth | Confident | Marketing + app hybrid | | 1.414 | Augmented fourth (√2) | Strong | Editorial | | 1.500 | Perfect fifth | Bold | Landing pages | | 1.618 | Golden ratio | Dramatic | Hero-driven sites |

Generating a scale (base 16, ratio 1.25): step_n = 16 × 1.25ⁿ → 10.24, 12.8, 16, 20, 25, 31.25, 39, 48.8, 61… Round to sane px, then store in rem.

css
:root {
  --step--2: 0.64rem;   /* 10px caption-min   */
  --step--1: 0.8rem;    /* 13px caption       */
  --step-0:  1rem;       /* 16px body          */
  --step-1:  1.25rem;    /* 20px h5            */
  --step-2:  1.5625rem;  /* 25px h4            */
  --step-3:  1.953rem;   /* 31px h3            */
  --step-4:  2.441rem;   /* 39px h2            */
  --step-5:  3.052rem;   /* 49px h1            */
}

Pro move: use two ratios — a tight one (1.2) for body/UI steps, a wider one (1.333+) for display/headings. Headlines need more drama than paragraph text. A single ratio across the whole range either makes body cramped or headlines timid.

Fluid type with clamp()

Fixed steps break between a phone and a 27" monitor. clamp(MIN, PREFERRED, MAX) scales type with the viewport, no media queries.

css
/* fluid from 16px @360px viewport → 20px @1240px viewport */
--step-0: clamp(1rem, 0.93rem + 0.31vw, 1.25rem);
--step-5: clamp(2.25rem, 1.1rem + 5.1vw, 4rem); /* hero */

The math: preferred = yIntercept + slope·vw, where slope = (maxPx − minPx) / (maxVw − minVw) and slope·100 → the vw value; the rem term is the intercept so it lands exactly at your min at the min viewport. Use a generator (utopia.fyi) rather than hand-tuning. Always keep rem in the preferred term so the value still responds to the user's browser zoom/root-size — a pure vw clamp defeats zoom and fails WCAG 1.4.4.

3. Spacing: leading, measure, tracking, kerning

Line-height (leading) — the readability lever

| Context | line-height (unitless) | |---|---| | Body text | 1.4–1.6 (1.5 safe default) | | Long-form reading | 1.5–1.7 | | UI labels / dense | 1.3–1.4 | | Headings (large) | 1.1–1.25 | | Display / hero | 0.95–1.1 (can go sub-1) |

Rules: set line-height unitless (1.5, not 24px) so it scales with each element's font-size. Bigger type needs less relative leading (the eye travels less to find the next line); small text needs more. Wider columns need more leading; narrow columns can use less. Tighten line-height as font-size grows — a single global 1.5 makes h1s look airy and broken.

Measure (line length) — the silent readability killer

• 45–75 characters per line; ~66 is the optimum. Below ~45 the eye jumps too often; above ~75 it loses the return point.
• Set with max-width in ch (the 0 glyph width): max-width: 66ch;. Re-check per font — ch varies by typeface.
• Multi-column? 40–50ch each. Captions can run shorter.

Tracking (letter-spacing)

• Large/display: tighten slightly (-0.01em to -0.03em) — default spacing looks loose at big sizes.
• Small caps & ALL-CAPS: open up (+0.05em to +0.1em) — caps have no ascender/descender cues, so they need air to be legible.
• Body text: leave at 0. Designers over-track body; it reduces readability by breaking word shapes.
• Use em, never px, so tracking scales with size.

Kerning & word-spacing

• Kerning = space between specific pairs (AV, To, We). font-kerning: normal (default) uses the font's built-in pairs. text-rendering: optimizeLegibility enables optical kerning + ligatures but can hurt perf on huge text blocks — reserve for headings.
• Metric kerning uses the designer's pair table (trust it for good fonts); optical lets the renderer space by glyph shape (rescues fonts with poor metrics). Headlines benefit most.
• Word-spacing: rarely touch it. Justified text is where it goes wrong (see mistakes).

4. Font selection & pairing

Classification (so you can reason about pairing): Serif (Old-style/Garamond, Transitional/Times, Slab/Roboto Slab), Sans (Grotesque/Helvetica, Neo-grotesque/Inter, Humanist/Frutiger, Geometric/Futura), Mono (JetBrains Mono, SF Mono), Display/Script (headlines only).

Pairing principles:

• Max 2–3 families. One workhorse for body/UI, one for display, optionally one mono for code/data.
• Contrast by role, harmonize by feel. A clean geometric sans heading over a humanist serif body works because roles differ but proportions agree. Pairing two similar sans-serifs reads like a mistake, not a choice.
• Superfamily (Source Sans + Source Serif + Source Code; IBM Plex family) gives guaranteed harmony — same designer, matched x-heights and metrics. Lowest-risk pairing.
• Match x-heights across paired fonts or the size relationship feels off.
• Never pair two display faces, two geometrics, or fonts from the same era doing the same job.

System font stack (zero network cost, instant render, native feel):

css
font-family: system-ui, -apple-system, "Segoe UI", Roboto,
  Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji";
/* mono */
font-family: ui-monospace, "SF Mono", "Cascadia Code", "JetBrains Mono",
  Menlo, Consolas, monospace;

Loading performance

• woff2 only for web fonts — ~30% smaller than woff, universal support. Don't serve ttf/otf to browsers.
• font-display:
• swap — render fallback immediately, swap in webfont when ready (FOUT). Default choice; text always visible.
• optional — use webfont only if it's near-instant (cached), else stick with fallback. Best for body text / Core Web Vitals (no layout shift).
• block (FOIT) — invisible text up to ~3s. Avoid; it hides content.
• preload the 1–2 critical fonts (body regular, maybe body bold): <link rel="preload" href="/inter.woff2" as="font" type="font/woff2" crossorigin>. Don't preload everything — it starves other resources.
• Subset to the glyphs/languages you actually use (glyphhanger, fonttools) and use unicode-range to split by script — Latin-only subsets cut a font from ~200KB to ~15KB.
• FOUT vs FOIT: FOUT (flash of unstyled text) is preferable — content is readable instantly. Minimize the swap jolt by choosing a fallback whose metrics match (use size-adjust, ascent-override, descent-override on an @font-face fallback to eliminate layout shift).

css
@font-face {
  font-family: "Inter";
  src: url("/fonts/inter-var.woff2") format("woff2-variations");
  font-weight: 100 900;       /* variable range */
  font-display: swap;
}

5. Variable fonts & OpenType

Variable fonts ship every weight/width in one file along a continuous axis, controlled at runtime. One ~70KB file replaces 6+ static weights.

Registered axes: wght (weight 1–1000), wdth (width %), opsz (optical size), ital (0/1), slnt (slant degrees). Plus font-specific custom axes (grade, etc.).

css
h1 { font-variation-settings: "wght" 720, "wdth" 105; }
body { font-variation-settings: "wght" 400; }

Prefer the high-level properties (font-weight: 720) when they map to an axis — they're inherited and animatable; reach for font-variation-settings only for custom axes.

Optical sizing (opsz) adjusts letterforms for size: small text gets thicker strokes, more open spacing; large text gets refined, higher contrast. font-optical-sizing: auto (default) ties it to font-size automatically — keep it on. This is the single biggest quality jump variable fonts offer.

OpenType features (font-feature-settings, or semantic font-variant-*):

| Feature | Tag | Use | |---|---|---| | Ligatures | liga/dlig | On by default; discretionary for display only | | Tabular numerals | tnum | Tables, prices, anything that must align in columns | | Proportional numerals | pnum | Running prose | | Oldstyle numerals | onum | Numbers that flow with lowercase in body text | | Small caps | smcp | True small caps (never CSS font-variant: small-caps fakery if real ones exist) | | Fractions | frac | 1/2 → proper fraction | | Slashed zero | zero | Code, IDs, anywhere 0/O confusion matters | | Stylistic sets | ss01… | Alternate glyph designs (e.g. Inter's single-story a) |

css
.data-table td { font-variant-numeric: tabular-nums; } /* prefer semantic */
code { font-feature-settings: "zero" 1, "ss01" 1; }

> Tabular vs proportional numerals is the classic system-design tell: numbers in tables, dashboards, and timers must be tnum so digits share one width and columns don't shimmer when values change. Body prose uses proportional for even color.

6. Vertical rhythm & baseline grid

Consistent vertical spacing makes a page feel composed. Tie everything to a base unit — typically 4pt or 8pt. Margins, line-heights, and component spacing become multiples of it.

• Set body line-height to a grid multiple: 16px body × 1.5 = 24px line → snaps to a 4/8 grid perfectly. This is why 16/24 is a near-universal default.
• Space between blocks (paragraphs, headings) in grid units (8, 16, 24, 32, 40…).
• Baseline grid: align text baselines to evenly spaced horizontal lines. Strict baseline alignment is purist and hard on the web (line-height rounding, mixed sizes); aim for rhythm (consistent multiples) rather than pixel-perfect baseline lock unless doing editorial/print.
• Cap-height alignment vs baseline: in UI, vertically center elements by cap height, not the em-box, so a label looks centered next to an icon or inside a button. CSS leading-trim/text-box-trim (shipping in modern browsers) finally lets you trim the half-leading above cap height and below baseline so components space by glyph, not by invisible line-box padding — adopt it where supported, with fallbacks.

7. Hierarchy & the type ramp

Hierarchy is built from five tools, in order of strength: size → weight → color/contrast → spacing → case. Use the fewest that work. Reaching for ALL-CAPS + bold + big + colored at once is shouting.

• Size: primary signal; from the modular scale.
• Weight: 400 body, 600/700 emphasis, 300 only at large sizes (thin text fails small).
• Color/contrast: demote secondary text with a lighter ink — but never below WCAG contrast.
• Spacing: whitespace groups and separates more cleanly than rules/boxes.
• Case: small caps / uppercase for overlines, eyebrows — with added tracking.

Reference type ramp (tokens)

| Token | size (rem/px) | line-height | weight | tracking | use | |---|---|---|---|---|---| | display | clamp → 48–72 | 1.0 | 700 | -0.02em | hero | | h1 | 39 / 2.441rem | 1.1 | 700 | -0.01em | page title | | h2 | 31 / 1.953rem | 1.15 | 700 | -0.01em | section | | h3 | 25 / 1.563rem | 1.2 | 600 | 0 | subsection | | h4 | 20 / 1.25rem | 1.3 | 600 | 0 | card title | | body-lg | 18 / 1.125rem | 1.6 | 400 | 0 | lead paragraph | | body | 16 / 1rem | 1.5 | 400 | 0 | default | | body-sm | 14 / 0.875rem | 1.45 | 400 | 0 | secondary | | caption | 13 / 0.8rem | 1.4 | 400 | 0 | meta, timestamps | | overline | 12 / 0.75rem | 1.3 | 600 | +0.08em | UPPERCASE eyebrow | | code | 14 / 0.875rem | 1.5 | 400 | 0 | mono, tnum+zero |

Responsive ramps: don't scale every token equally. Headings should compress more than body on small screens (a 49px hero at 360px is unreadable; 16px body stays 16px). Fluid clamp() per token handles this; or define a mobile ramp that shrinks display/h1/h2 while body holds.

Per-platform ramps (native systems)

• iOS — Dynamic Type / SF: semantic text styles (largeTitle 34, title1 28, title2 22, title3 20, headline 17-semibold, body 17, callout 16, subhead 15, footnote 13, caption1 12, caption2 11). Never hardcode point sizes — use the named styles so they respond to the user's accessibility text-size slider (up to XXXL). SF Pro switches optical variants (Display ≥20pt, Text <20pt) automatically.
• Android — Material 3 / Roboto: scale roles Display/Headline/Title/Body/Label, each in Large/Medium/Small (e.g. Body Large 16/24, Title Medium 16/24 medium-weight, Label Small 11). Map your tokens to these roles so Material components inherit correctly; respects the OS font-scale setting.
• Windows / WinUI: type ramp Caption 12 → Body 14 → Subtitle 20 → Title 28 → Display 68, on Segoe UI Variable (which uses opsz for size-appropriate forms).

> Map your single semantic token set (display, h1…caption, code) onto each platform's native ramp rather than inventing per-platform names. One mental model, native behavior everywhere.

8. Accessibility (non-negotiable)

• Minimum sizes: body ≥16px on web (14px floor for genuinely secondary text; never below 12px for anything a user must read). iOS/Android: respect the system text-size setting — don't cap it.
• Use rem, not px, for font-size. px font-size ignores the user's browser default; rem honors it. This is the #1 accessibility-via-typography rule.
• Support 200% zoom (WCAG 1.4.4) and 400% reflow (1.4.10) without loss of content or horizontal scroll. rem sizing + fluid layouts + clamp() with a rem term get you there; fixed vw-only type breaks zoom.
• Contrast (WCAG 1.4.3): normal text ≥4.5:1, large text (≥24px, or ≥18.66px bold) ≥3:1. AAA wants 7:1 / 4.5:1. Don't demote secondary text into low-contrast gray-on-white.
• Don't rely on weight/color alone to convey meaning (1.4.1) — pair with size, label, or icon. Bold-only emphasis disappears for low-vision users.
• Letter-spacing override (1.4.12): content must survive user-applied spacing (letter ≥0.12em, word ≥0.16em, line ≥1.5×). Don't lock layouts that break when users adjust spacing.
• Dyslexia-friendly: generous line-height (1.5+), left-aligned (ragged-right) not justified, measure ≤70ch, high x-height humanist fonts with open apertures and distinct letterforms (Inter, Atkinson Hyperlegible, Lexend), avoid all-caps for long runs, avoid pure-black-on-pure-white (slight off-white reduces glare), don't italicize whole paragraphs. "Dyslexia fonts" help some users but good general typography helps more.

9. Common mistakes (do / don't)

| Don't | Do | Why | |---|---|---| | Justify text on the web (text-align: justify) | Left-align, ragged right | Browsers lack hyphenation/spacing control → ugly "rivers" of whitespace | | Use 4+ font families | 2–3 max (body, display, mono) | Each family is a download + a cognitive cost; more reads as chaos | | Set sizes in px | rem (+ clamp() for fluid) | px ignores user zoom/root-size → accessibility failure | | Single global line-height | Tighter for display, looser for body | Leading must scale inversely with size | | Run lines 100+ chars wide | max-width: 66ch | Eye loses the return line beyond ~75ch | | Faux bold/italic (font-weight:bold on a font with no bold file) | Load the real weight/italic, or use a variable font | Browser synthesizes by skewing/smearing → muddy, off-brand glyphs | | Track body text | Track only caps/display | Spacing breaks word shapes, hurting readability | | Hardcode iOS point sizes | Dynamic Type styles | Breaks the user's accessibility text scaling | | Tabular data with proportional nums | font-variant-numeric: tabular-nums | Columns shimmer/misalign as digits change width | | Thin weight (200/300) on small text | 400+ for body | Thin strokes vanish at small sizes / low DPI | | font-display: block on body | swap or optional | FOIT hides content for seconds |

Synthesis check before shipping a system: one base + one (or two) ratios → rem scale; tokens with size+line-height+weight+tracking each; ≤3 families in woff2 with font-display + preload + subsetting; tabular nums wired for data; fluid clamp() with rem terms; contrast + 200% zoom verified; mapped to native ramps on the platforms you target.

Cross-device quick reference

• Web: rem + clamp() + ch measure + variable woff2. Logical properties (margin-block, padding-inline) for i18n/RTL.
• iOS: Dynamic Type semantic styles, SF Pro, respect text-size + bold-text accessibility settings.
• Android: Material 3 type roles, Roboto/sp units (which scale with user font setting — use sp for text, dp for layout).
• Desktop (Electron/native): treat like web but lean on system-ui stacks for native feel and zero load cost; honor OS scaling.