fix: make format* take s/l/t on the 0–1 scale

[tenphi]

Closes #93.

Takes suggestion 1 from the issue — the stated first preference: make the writers accept 0–1 so the library is self-consistent — plus suggestion 4's doc fixes.

What changed

formatOkhsl / formatOkhst / formatRgb / formatHsl / formatOklch now read s / l / t on the 0–1 factor scale every producer in the library returns (resolve(), variantToOkhsl, srgbToOkhsl, oklabToOkhsl, okhslToSrgb). The percentages are an output detail: the writers scale by 100 themselves where the CSS syntax asks for one. So the composition variantToOkhsl's doc points at is now correct with nothing in between:

const v = glaze.color('#7A4DBF').resolve().light;

formatOkhst(v.h, v.s, v.t);                    // 'okhst(298.52 70.41% 45.02%)'
formatOkhsl(...Object.values(variantToOkhsl(v))); // 'okhsl(298.52 70.41% 45.27%)'

and it round-trips: glaze.color(formatOkhst(v.h, v.s, v.t)) comes back with the same s / t.

The old call shape no longer fails silently either. A value above 1 cannot be a factor, so it can only be pre-2.0 percentage-scale input — the writers console.warn once per writer (deduped for the process) and emit an obviously-broken 7040.68% rather than a plausible near-black.

Docs, per suggestion 4: okhslToSrgb now states its scale (the one converter that stated no units), and variantToOkhsl's comment spells out that s / l come back on the same 0–1 scale the writers take, so it no longer points at a bug.

Compatibility

Breaking for direct callers of the writers, so the changeset is major. The fix is to drop the * 100:

- formatOkhst(v.h, v.s * 100, v.t * 100);
+ formatOkhst(v.h, v.s, v.t);

Nothing else moves — h was always 0–360, alpha always 0–1 — and no export method changes its output: css() / tokens() / tasty() / json() / dtcg() / tailwind() / glaze.format() were compensating internally, and now just stop.

I diffed the full export surface (4 theme seeds × every format × css/json/tasty/tailwind/dtcg/splitHue, before vs. after) to confirm. One string differs anywhere: a fully-desaturated shadow emits hsl(0 0% 18.07% / 0.3554) where it used to emit hsl(340 0% 18.07% / 0.3554). Dropping the redundant ×100 ÷100 round-trip moves the sRGB channels by ~1.7e-16, which flips which channel formatHsl picks as max when deriving hue — and at 0% saturation the hue term names no color either way. Same color, one fewer float round-trip.

If you'd rather ship this as a minor (these are the advanced-use re-exports, not the main surface), the only edit needed is the bump in .changeset/plain-scales-align.md.

Tests

New src/okhsl-color-math.test.ts (12 tests) covers the scale contract, agreement with the converters each writer wraps, the pastel recompute on the new scale, the percentage-scale guard (warns on 0–100 input, once per writer, silent at the 1 endpoint), and the resolve() → writer → parser round-trip from the issue.

pnpm typecheck, pnpm lint, pnpm format:check, pnpm build, and pnpm test (386 tests) all pass; the pre-existing 374 are untouched.

---

Generated by Claude Code

[github-actions]

📦 Snapshot release

Published 0.0.0-snapshot.bc6c362.

pnpm add @tenphi/glaze@0.0.0-snapshot.bc6c362