Position

What is measured, what is not, and where practice runs ahead of the evidence.

Known (measured)

  • Line length affects speed and comprehension in opposite directions; both ranges are documented.
  • Type below the fluent range slows reading; larger than the minimum helps comprehension.
  • Common serif and sans faces read equally fast on screen.
  • Dyslexia-specific fonts do not help; wider letter spacing helps dyslexic children.
  • Light text on a dark ground is no better by day and worse at small sizes.
  • Readers scan; layout that supports scanning raises usability.
  • Words belong with the figure they explain; decoration does not help learning.

Known with caveat

  • 45-75 cpl as a safe band: practitioner consensus around one review.
  • Off-white paper, left alignment, no italics: guidance with weak direct evidence.
  • Collapsed content is missed: one company article and design practice, no study.

Not known

  • No study fixes the best line length for mixed text and code.
  • No study tests heading density or paragraph length thresholds.
  • No study tests a logical-line limit for a file or note.
  • Evidence on dark mode comes from older displays.

What this repo adds

  • It states each rule with its grade, so a convention is not mistaken for a result.
  • It measures the rendered line (characters, pixels) instead of trusting a CSS unit.
  • It separates the standard (WCAG, pass or fail) from the research (ranges, trade-offs).

Practice against evidence

Practice runs ahead of the evidence: - Line length ranges in USWDS 2026, GOVUK 2024 and Baymard 2022 come from craft tradition. Baymard's own text credits Ruder and reports no numeric line-length test results. - APCA thresholds are published by their authors with no experimental data on the page. - The 4.5:1 floor has a stated derivation, a loss factor, but the page cites no reading experiment.

Practice ignores the evidence: - 83.9% of home pages fail 4.5:1 (WebAIM 2026): a known rule, widely broken. - Speed and comprehension peak at different lengths (Dyson 2004). None of the three design systems read names the trade-off; each gives one range.

Pending change: - WCAG 3 has not chosen a contrast method (W3C 2026), so 4.5:1 stays the floor until it does. - No rendered-page contrast checker is in this repository's checks; axe-core, pa11y and Lighthouse fill that role in practice.