Tailwind Scale Accessibility Benchmark 1.0.0
Released: July 22, 2026
Research scope
- Published a frozen 100-seed CPS reference corpus, 20-seed holdout, and 12 synthetic controls.
- Compared HSL, gamma-encoded sRGB, linear-sRGB, and OKLCH scale generation.
- Generated 11 shades from 50 through 950 with the exact seed anchored at 500.
- Measured adjacent Delta E OK variation, monotonic lightness, hue drift, sRGB gamut mapping, and white/black contrast evidence.
- Published mean, median, worst case, per-seed wins, distributions, holdout ordering, and unfavorable samples.
Open evidence
- Complete nested JSON and flat CSV results.
- Seed corpus in JSON and CSV.
- Protocol manifest and SHA-256 checksums.
- Downloadable TypeScript engine and generator.
- Browser-local Web Worker runner and raw-result export.
- BibTeX and plain-text suggested citation.
Known limitations
- The designed CPS corpus is not a random sample of all colors.
- The benchmark does not reproduce Tailwind CSS palette-generation internals.
- The v1.0 OKLCH gamut rule is fixed-lightness, fixed-hue chroma reduction, not a complete CSS local-MINDE implementation.
- Contrast pairs are evidence, not accessibility certification.
Semantic version policy
| Change | Version impact | Example |
|---|---|---|
| Patch | Presentation or documentation correction with no changed raw result | Typo, clearer limitation, corrected broken link |
| Minor | Additive evidence that preserves existing result meaning | New export format or separately labeled control cohort |
| Major | Changed input, algorithm, metric, weights, gamut, or published result | New scale curve, new mapping rule, revised corpus selection |
Corpus, protocol, metric, and runner versions remain independent even when published under one research release.
Correction policy
A material correction does not replace v1.0 files at their versioned URLs. Color Pick will:
- Document the issue, discovery date, affected fields, and practical impact.
- Publish a new versioned dataset or protocol.
- Preserve the older release with a visible superseded notice.
- Update checksums, static findings, structured data, citation files, and regression fixtures together.
- Explain whether aggregate conclusions changed.
Planned research sequence
- Review v1.0 methodology and real-device runner performance.
- Publish a dark-mode palette benchmark with separate surface, text, border, state, and focus evidence.
- Publish an accessible-pastel benchmark with explicit foreground, boundary, and semantic-use contexts.
- Add user-palette cohort percentiles only after cohort sizes, distributions, and interpretation rules are stable.
Future suites will receive their own methods and datasets. They will not retroactively create a universal score for Tailwind Scale Benchmark v1.0.