@adia-ai/web-components 0.8.14 → 0.8.16

package/CHANGELOG.md

@@ -1,5 +1,28 @@
 # Changelog — @adia-ai/web-components
 
+## [0.8.16] — 2026-07-26
+
+### Fixed
+- **Text on the default palette's filled interaction states now meets WCAG AA** (`styles/colors/material-static.css`; gh#427) — scope of the claim: the 48 verified pairs are `on-<family>` over `<family>`, `-hover`, and `-active` for all 8 families in both schemes. Container, disabled, and per-`[theme]` fills are *not* covered by that run. — `on-<family>` text is the near-white `<family>-050` stop, and the resting fill was `<family>-500`, which measured **1.97:1 for `warning`** and 2.90–4.28:1 for the other seven: all 8 families shipped sub-AA text at rest. Worse, dark-mode `-active` resolved to the *light* `250` stop under that same near-white text, measuring **1.32–1.76:1** across all 8 — effectively invisible text while a button is pressed. The fill roles move to one monotonically darkening ladder, identical in both schemes: rest `600` / hover `650` / active `700`, with `secondary` at `650`/`700`/`750` and `warning` at `700`/`750`/`800` (its 1.97 start needed three steps). Measured after the change: every family clears 4.5:1 in all three states, worst cell 4.73:1 (`primary` at rest).
+
+  Two deliberate consequences. Fills are visibly darker, `warning` most of all. And dark mode loses its lighter-on-press behaviour — that `250` stop *was* the 1.32:1 defect, so press feedback now darkens in both schemes, which also removes the light/dark stop divergence in the fill ladder entirely.
+
+- **`styles/themes.css`'s contrast claim corrected** — its header stated the hue-rotation "preserves the verified WCAG AA 4.5:1 thresholds (`npm run verify:contrast`, 42/42 pairs, **every theme**)". Two things were wrong: that 42-pair roster never checked a fill at all (gh#427), and the gate resolves only the default ramps — it does not iterate `[theme]` blocks, so no theme was ever measured. The header now states the real scope: 90 pairs proving the default palette, with per-theme ratios resting on the rotation argument (per-stop lightness untouched, and WCAG luminance is lightness-dominated) rather than on a measured run. Mechanizing per-theme verification is not done here.
+
+### Maintenance
+- **`dist/` bundles rebuilt** in this cut's window (5 file(s)) — regenerated from the source changes described above, not independent edits.
+
+## [0.8.15] — 2026-07-26
+
+### Changed
+- **Hover fills darken toward the ladder in both schemes** (`styles/colors/material-static.css`) — all 8 families' `--md-sys-color-<family>-hover` move from `light-dark(<family>-650, <family>-350)` to `light-dark(<family>-600, <family>-550)` (base fill is `<family>-500` in both modes; ladder stops are 50 points, higher = darker). Light-mode hover lands two stops darker than base (`500 → 600`, softened from the old three-stop `-650`); dark-mode hover lands one stop darker (`500 → 550`), where it previously *lightened* three stops to `-350`. Measured with the gate's own culori converter: light-mode white hover text is 6/8 families ≥ AA 4.5:1 (from 8/8 at the old stronger jump); dark-mode — which failed AA for **all 8 families** at `-350` (ratios ≈ 2.0–2.5) — now passes for 5/8. Flows through the `--a-*-hover` alias layer and all 12 `[theme]` recolors with no component changes. Filled buttons bypassed this role via `--a-<family>-strong` — corrected in the Fixed entry below, same release.
+
+### Fixed
+- **Filled buttons hover on the hover role, not the active role** (`components/button/button.css`) — `button-ui[variant="primary"]` and the `danger`/`success`/`info` filled colors set their hover fill from `--a-<family>-strong`, which aliases `--md-sys-color-<family>-active` (`light-dark(750, 250)`). In dark mode that meant hover *lightened five stops* to the pale `-250` tint — the washed-out hover the role change above couldn't reach, because these rules bypassed the hover role entirely. They now ride `--a-<family>-hover` (`light-dark(600, 550)`). Verified live in dark mode: primary idle computes exactly `primary-500`, hover exactly `primary-550`. `warning` keeps its documented `-bg` tint exception (dark-label contrast, dogfood-audited); ghost/outline **text** hovers keep `-strong` deliberately — lighter text on a dark canvas is more contrast, not less; `-strong` remains the pressed/active fill.
+
+### Maintenance
+- **`dist/` bundles rebuilt** from the token change above.
+
 ## [0.8.14] — 2026-07-25
 
 ### Changed

package/components/button/button.css

@@ -19,10 +19,16 @@ button-ui:active { transform: scale(0.97); }
    untouched (primary/danger/etc. keep their light label). */
 button-ui:not([disabled]):hover { --button-bg: var(--button-bg-hover); }
 
-button-ui[variant="primary"]:not([disabled]):hover { --button-bg: var(--a-primary-strong); }
-button-ui[color="danger"]:not([disabled]):hover     { --button-bg: var(--a-danger-strong); }
-button-ui[color="success"]:not([disabled]):hover    { --button-bg: var(--a-success-strong); }
-button-ui[color="info"]:not([disabled]):hover       { --button-bg: var(--a-info-strong); }
+/* Filled hovers ride the HOVER role (light-dark 600/550 — one rule: darker
+   than the 500 base in both schemes). They previously rode `-strong`, which
+   aliases the ACTIVE role (750/250): in dark mode that hover LIGHTENED five
+   stops to the pale -250 tint — the washed-out hover the 600/550 role change
+   couldn't reach because these rules bypassed it. `-strong` remains the
+   pressed/active fill. */
+button-ui[variant="primary"]:not([disabled]):hover { --button-bg: var(--a-primary-hover); }
+button-ui[color="danger"]:not([disabled]):hover     { --button-bg: var(--a-danger-hover); }
+button-ui[color="success"]:not([disabled]):hover    { --button-bg: var(--a-success-hover); }
+button-ui[color="info"]:not([disabled]):hover       { --button-bg: var(--a-info-hover); }
 /* warning fill → -bg (NOT -strong like the rest): warning's -strong mid-tone
    reads muddy under the dark --a-warning-fg, so the hover surface uses the
    -20 tint (--a-warning-bg) for clean contrast. (dogfood-audit: warning-strong-vs-bg) */