TendUI supplies semantic HTML, native controls, visible keyboard focus and appearance tokens. An accessible application also depends on its content, headings, labels, navigation, validation flow and any custom CSS or optional HTMX behavior. Passing the checks below is not a WCAG conformance claim.
The Go tests read the embedded stylesheet's literal colors and calculate sRGB relative luminance. The screen matrix covers light/dark page, card and subtle surfaces, including tinted cards and striped rows:
- Body text, muted text and the four feedback tones: at least 4.5:1.
- All four presets' links, including hover colors: at least 4.5:1.
- Control-border color against all three surfaces: at least 3:1.
- Filled-button labels against normal/hover backgrounds: at least 4.5:1.
- System dark and explicit dark declarations must match, including token aliases.
The chosen thresholds follow W3C's explanations of text contrast and non-text contrast. The tests use unrounded values. Inactive controls are excluded from these thresholds; the current disabled opacity remains unchanged. Focus rings use the primary accent, whose contrast against these surfaces is covered by the link checks. Their visibility, clipping, area and placement still need browser review.
For the current default tokens, the lowest tested screen text ratio is 5.42:1 (light muted text on subtle background); the lowest tested border ratio is 3.21:1 (light border on subtle background). Dark minima are 6.36:1 for text and 3.40:1 for borders. These are source-color calculations, not pixel measurements. The checks do not model arbitrary consumer overrides, inherited opacity, background images, transparency, antialiasing or browser forced-color replacement.
Run the palette checks with:
go test ./components -run 'Test(ScreenPalette|SystemDark|PrintPalette|PresetText)' -vThey also run through go run ./cmd/tendui-dev check with the full race suite.
Print uses a white canvas, black body text, dark neutral labels and visible gray
borders regardless of the screen theme. Feedback retains its text label and
border structure while using a neutral print color. The root print palette uses
!important deliberately: the explicit and system-dark selectors have higher
specificity than a plain :root, and consumer appearance exports load later.
A normal-priority print root would leave parts of a dark palette active.
Buttons and button-like links print with dark text, a transparent background and
a border. The current-step marker also retains dark text and a stronger outline
without a fill. Switches use their native checkbox appearance so checked state does
not rely on a colored CSS thumb. This avoids requiring background graphics just
to read a label or identify a checked control. Browsers and users may adjust
printed colors; see MDN's print-color-adjust reference.
TendUI does not force print-color-adjust: exact or override print-dialog choices.
The print rules are consolidated at the end of the stylesheet. Animation and hover displacement stop, shadows disappear, the application shell becomes a normal block without a viewport-height minimum, and tables leave their scroll containers. Sidebars, skip links, theme toggles, native overlays and pagination remain omitted as before. Main content, ordinary navigation links, table captions and form controls remain available. Printing does not fetch hidden pages, expand all disclosures or include closed overlays; applications should offer a separate complete document when that is needed.
Custom print styling can override TendUI's important print palette with matching important declarations in a later print stylesheet. Review the result with both background-graphics settings. The tests guard source rules and palette ratios; review rendered PDFs for pagination, readable controls and background omission. Automated source checks do not establish every page/theme combination.
Use the same built demo, record its -version, browser/version and operating
system, and retain only the screenshots/PDFs needed to explain a finding.
| Review | Suggested demo and expected observation |
|---|---|
| Keyboard | /forms and /patterns?section=composition: visible focus, source-order navigation, labels and native submit/reset controls |
| Error recovery | /wizard and /uploads: linked summaries lead to the relevant field; validation text remains available |
| Zoom and narrow views | /catalog, /dashboard, /patterns?section=composition: content remains usable at narrow widths and 200%/400% zoom; long rows stay reachable |
| RTL | A consumer page with dir="rtl" on its content wrapper: start/end placement follows direction without reversing DOM order |
| Forced colors | /playground: native checkbox/switch state, control boundaries, focus and inline icons remain distinguishable |
| Reduced motion | /motion: content remains present; entrances, lift and skeleton animation stop under the OS preference and explicit Off |
/playground, /dashboard, /workspace, /motion: repeat light/dark/system with background graphics on/off; inspect labels, switches, surfaces, clipped tables and extra pages |
|
| Assistive technology | Check names, landmarks, status/error announcements, table captions/headers and native overlay focus with the target screen reader |
Modern native dialog, popover and grouped-details support varies by browser; retain the documented ordinary-page/inline fallbacks. A successful HTTP response or render test proves neither keyboard behavior nor screen-reader output. The local sandbox cannot launch browsers or bind a TCP socket, so the release suite runs real Chromium, Firefox and WebKit engines in GitHub Actions. It passed 53 checks, including optional HTMX execution; one unsupported WebKit forced-color emulation check is skipped. Automated axe rules cover all 37 component pages in both themes. Keyboard, native overlays, narrow layouts, reduced motion and print have focused coverage. See release verification. The broader consumer matrix above, especially screen-reader output, zoom and physical devices, still needs review for each assembled application.
Use the manual session checklist and result template to record real screen-reader and physical-device observations for the public site. It starts with untested results and must not be treated as evidence of a pass.