Why 17px and 38rem
Three numbers carry most of the reading experience on this site: a 17px body size, a 38rem column and a line height of 1.75. They can look like taste. Each has a source, and one of them was wrong once before it was right.
One pixel over a floor
The US Web Design System tells designers: “For most text, including body copy, use at least an effective size of 16px”. Remarque’s --text-body is 1.0625rem, which is 17px at the default root size, and the spec calls it a non-negotiable floor. It lives in the core tier of the tokens, so a site that changes it has forked the system.
The record is plain about what this is. The spec states the 17px as a deliberate step over the USWDS minimum. It does not cite a reading study, and neither does this essay. Matthew Butterick recommends “body text of 15–25 pixels” for websites, so 17 sits well inside the range a working typographer gives, and one pixel above the government floor. The size is authored in rem, which means a reader who raises their browser’s default size gets a proportionally larger page.
The text scale below the floor follows the same logic. Per the spec, 14px and 13px are the USWDS minimums for small text and for fine print, and remarque-audit fails any font-size declaration under 13px.
The measure, counted
The reading column is 38rem, or 608px at the default root size. The target behind it is a character count, and the spec treats the count as the invariant: a median of 66 to 72 characters per line at --text-body, with the 90th percentile under about 78.
Three authorities give ranges for it, and they do not quite agree. Bringhurst, in The Elements of Typographic Style, puts 45 to 75 characters as the generally accepted range for a single-column page set in a serifed text face at a text size, with 66 as the ideal.
Butterick says to “Aim for an average line length of 45–90 characters, including spaces”.
USWDS gives the same 45 to 90 and names 66 as a good target for long texts.
Bringhurst’s 75 is stricter than the other two, and his sentence is about a serifed text face; Inter is not one. Remarque borrows the range rather than applying it. At 17px Inter, 38rem sets a median of 71 characters per line and a 90th percentile of 76, measured on this repo’s demo essays in October 2026. The median sits inside all three ranges. The tail runs one character past Bringhurst’s 75 and well inside the 90 that Butterick and USWDS allow.
The word “measured” matters. The default was 46rem until 0.27.0, documented as about 70 ch at 17px Inter. That figure came from arithmetic with the ch unit, which MDN defines as the advance measure of the glyph “0”. Proportional text averages narrower than a zero, so a line holds more characters than the ch count suggests. When 0.27.0 counted the characters actually rendered, 46rem set a median of 86 and a 90th percentile of 93, past even Butterick’s ceiling.
The method is written down so anyone can repeat it. For each body paragraph, walk its text nodes with a Range per character, group the characters by the top of their first client rect into lines, drop each paragraph’s last and usually short line, and take the median and 90th percentile across the page. Spaces count; code and sup are skipped. A computed-style test now asserts that the demo essays measure a median between 66 and 72.
Because the right width depends on the face, the measure is a palette-tier token. The spec’s Measure Compensation table records Source Serif 4 at 73 and 78 on the reference consumer site, also at 38rem. Change the body font and you re-measure; the character count stays fixed and the rem value follows it.
Leading, and a deviation
Line height is 1.75, or 175% of the type size. Butterick’s guidance for “most text” is “between 120% and 145% of the point size”, and on the same page a sample set at 170% is labelled too loose. Remarque is outside his range, and REMARQUE.md says so in its Butterick lineage table, where the row is marked a deliberate deviation.
The spec makes three arguments for it. First, leading scales with measure, and screen prose at 17px with a column of about 70 characters benefits from more air than a print-derived band allows. USWDS makes the same connection from the other side: text with greater line height can have a longer measure. Second, hierarchy. Body leading at or below 145% would sit under --leading-meta, which is 1.5 because of the USWDS and WCAG floors the system also adopts, and body text should not read tighter than captions. Third, identity: the spacious register is part of what makes a page recognizable.
The third argument is a preference, and the spec treats it as one. It also states the cost: 1.75 spends more vertical space and reads calmer and slower than 1.5 to 1.65. A site that wants Butterick’s tighter rhythm is choosing a different identity, which the spec calls an honorable fork. The design-review panel ratified the deviation 3 to 0 on 2026-07-20, over the alternative of changing the token.
Text Spacing compatibility
WCAG 2.2 success criterion 1.4.12 is often read as a minimum line height. It is not. It requires that “no loss of content or functionality occurs” when a reader sets line height to at least 1.5 times the font size, spacing after paragraphs to at least 2 times, letter spacing to at least 0.12 times and word spacing to at least 0.16 times. The criterion’s first note adds that content is not required to use those values, only to survive a reader applying them.
So 1.75 is not what makes Remarque compatible. What does is how the spacing is authored. Measures and spacing are in relative units, line heights are unitless, overflow-wrap: break-word is set on html, and code blocks and wide tables scroll inside their own box. A Playwright suite, site/tests/reflow.spec.ts, applies the override set from the criterion to selected pages and checks the TOC rail, sidenotes, buttons and form fields for clipped text and the pages for horizontal overflow. That is coverage of chosen elements, not a conformance claim for every page, and the spec assigns conformance claims about a whole site to the site’s owner.
If you change one of these numbers, change the evidence with it. Re-measure the column with the method above, and record the reason for a leading that leaves Butterick’s range somewhere a reviewer will find it.