v1.15.0 — categorical (non-severity) tone set

[aicodebar]

Adds a categorical colour vocabulary alongside the existing severity ramp, so a component that means "this is a kind of thing" stops borrowing success/warning/danger, which mean "this is good / needs attention / broken".

Why

A survey of the consuming app found ~70 call sites across 11 semantic axes borrowing the severity ramp with no severity intended — on/off state (9 definitions, ~25 render sites, uniformly enabled → success), lifecycle stage, entity kind, field traits, duplicate/replay, provenance. The existing escape hatch made it worse rather than better: info in this design system is monochrome ink, so the variant="purple" → info workaround rendered an essentially grey badge. The workaround was a visual no-op.

The set: indigo | purple | magenta — and why only three

The set was not chosen by eye; candidates were swept across the full Tailwind ramp and gated on four measurements simultaneously:

• AA as text (≥4.5:1) on white, on --color-bg, and on its own -soft background
• OKLCH chroma ≥0.10 (i.e. actually reads as a colour, not a grey)
• OKLab ΔE ≥15 from both other categories and every severity token
• ΔE ≥8 under simulated protanopia and deuteranopia (Machado–Oliveira–Fernandes 2009, severity 1.0)

Most of the wheel fails:

rejected	why
red / rose / pink	collapse into danger — ΔE 6.8 / 8.5 / 11.4
orange / amber / yellow	collapse into warning — ΔE 9.6 / 8.5 / 9.1
lime / green / emerald	collapse into success — ΔE 10.0 / 8.4 / 6.6
teal / cyan	not colours at text weight — dark enough for AA they measure chroma 0.059 / 0.066, under the floor; they render grey
sky	collapses into --color-muted — ΔE 12.5

What survives is a single ~60° blue→magenta arc holding exactly three separable steps; only 5 valid 3-sets exist in the entire space. A fourth category drops the worst pair to ΔE ≈ 3.0 under simulated red-green colour blindness — which is why the set is three and the ceiling is documented in the tokens, the catalog and the CHANGELOG rather than left for someone to discover by adding a fourth.

The first hand-picked guess, for the record, would have shipped three bugs: azure↔violet at ΔE 0.31, teal-700 reading as grey, and lime-700 colliding with success.

Measured contrast (white / --color-bg / own -soft):

• cat-indigo #4f46e5 — 6.29 / 5.87 / 5.62
• cat-purple #581c87 — 10.88 / 10.16 / 10.14
• cat-magenta #a21caf — 6.32 / 5.91 / 5.89

Measured separation: indigo↔purple 17.3 normal / 16.6 CVD · indigo↔magenta 18.2 / 9.5 · purple↔magenta 16.3 / 9.6 · nearest severity pair 21.3 / 10.8. Verified in the browser that the computed colours resolve to exactly these hexes.

Breaking changes: none

Tone is untouched at five values, and palette: Record<Tone, string> was not widened — categoryPalette: Record<Category, string> is a separate map, so no consumer's own Record<Tone, X> breaks.

One deliberate subtlety: the variant prop defaults were removed from Badge and StatusBadge. That is load-bearing, not cosmetic — with a default in place, an unset prop is indistinguishable from an explicit variant="neutral", which makes the tone/category mutual-exclusivity check impossible to write. resolveTone(undefined, 'neutral') already applied the identical fallback one level down, so rendered output is byte-identical for every existing call site.

Tokens are namespaced --color-cat-* rather than --color-indigo, so they cannot shadow Tailwind's own scales in a consuming app.

Known limit, recorded not fixed

The consuming app's 15-kind graph palette is not salvageable as colour alone — several of its hues (amber-500, orange-500, emerald-500, rose-500) sit directly on warning/success/danger, which is the reported bug itself. Finer kind distinction needs a non-colour channel.

Separately: the existing warning and danger bases are themselves close — ΔE 6.7 normal, 2.9 under CVD. Both always carry a text label, so meaning is never colour-alone, but separating them would change existing rendered output. Recorded in the CHANGELOG as a decision for a major rather than acted on here.

Gates

typecheck ✓ · lint ✓ (0 errors, 5 warnings — exactly the v1.14.0 baseline; the 4 vue/require-default-prop warnings this change introduces are suppressed inline with the reason, since the omission is deliberate) · build ✓ (incl. verify:externals and verify:dev-warnings) · build-storybook ✓ · test:ci ✓ 226 passed, including 7 new stories.

No dark: variants and no import.meta.env in the diff; both new warnings use warnOnce. package.json bumped to 1.15.0 — the release workflow hard-fails on a tag/manifest mismatch — and the workflow's awk extraction was confirmed to pull the ## v1.15.0 body.

🤖 Generated with Claude Code