Skip to content

CHANGELOG.md

Latest commit

 

History

History
59 lines (49 loc) · 3.06 KB

CHANGELOG.md

File metadata and controls

59 lines (49 loc) · 3.06 KB
Raw

Changelog

All notable changes to Railing. The format follows Keep a Changelog; versions follow SemVer once packages are published.

Scores are not changelog entries. A library's number changing is a result, recorded in results/history/; this file records changes to the instrument.

Unreleased

Changed

  • Project renamed from Handrail to Railing (5 Aug 2026); site domain is now railing.dev and packages are scoped @railing-dev/*. No packages had been published under the old scope, so nothing external breaks.

Added

  • @railing-dev/announce: screen reader spot checks driving real VoiceOver via Guidepup, capturing what was actually spoken. Observations, never score inputs; see docs/SCREENREADERS.md and decision 021.
  • Specifications for dialog, combobox, menu, tabs and accordion: 64 assertions, each citing a W3C APG clause or WCAG success criterion.
  • Adapters for React Spectrum (calibration control), Radix UI, MUI, Headless UI, Chakra UI, Ant Design and shadcn/ui, plus a deliberately broken fixture with a catalogued defect list for measuring the runner's own error rate.
  • The runner: Chromium-only by design, reading the accessibility tree over the Chrome DevTools Protocol. --repeat discards unstable runs, --expect calibrates against the broken fixture, --only reproduces a single assertion without writing anything.
  • A replayable Playwright trace per published run, covering exactly the run that was scored and none of the stability repeats.
  • Result history: every run is archived by date alongside the latest file, so a regression can be traced to the version that introduced it.
  • The maintainer notification pipeline: a private report per library with findings grouped by authored cause, the full adapter source, a reproduction command per finding, and fourteen days before anything is published.
  • The publication gate, enforced in the pages, the API emission and the deploy, tested to fail closed on unparseable dates.
  • The public site at railing.dev: method, scoring, decision log, disclosure policy and contribution guide, generated from the documents in docs/ so the site and the repository cannot disagree.
  • An accessibility audit that holds the site and the maintainer reports to the standards the project measures in others, in both colour schemes, judging text at the worst point of any gradient it sits on.

Fixed

  • The publication gate failed open on a notifiedOn date that would not parse. Found by its own test; now fails closed. (Decision 018.)
  • Six results that would have accused libraries falsely, each caught before publication and recorded in docs/DECISIONS.md (decisions 007, 012, 016, 017): a 19% that was a 78%, a 27% and a 42% that were our selectors against the wrong major version, a focus trap reported broken because focus was read too early, and a containment check that judged shadcn's own close button to be outside its dialog.