chore: author import-ready Tokens Studio JSON for the missing beta colour tokens

[pomfrida]

Part of #5221, follow-up from the Button migration test case (#5208).

What

Import-ready Tokens Studio proposal sets in packages/eds-tokens/token-proposals/5221-beta-colour-gaps/, generated deterministically from the current studio tokens pull by build-proposal.mjs (re-run it after a new pull to rebase the proposal). The three JSON files are complete replacement sets for their counterparts in src/tokens/raw/, so the diff against the raw sets is exactly the proposal. README.md in the folder is the design-review document: change list with rationale, computed-value table, open questions, and import instructions.

Content

• Alpha step twins {accent,neutral,danger}-alpha.{3,7} in the scheme sets as explicit hex8 values (🎨 Implement Alpha‑Channel approach for disabled states #4741 — the platform only evaluates colour formulas in the CSS export, so modify formulas are not an option). Each twin composites to its opaque step exactly over white (light) / black (dark); deviation over the actual canvas is ≤ ±12/255 and reported per twin. Colour math matches scripts/generate-ts-tokens.mjs.
• Semantic fill group background.fill.{tone}.{default,hover,pressed,disabled} → steps 9/10/11 + alpha-3 (the values the migrated Button consumes directly today).
• Neutral/danger equivalents of the accent-only background.surface group, mirrored step for step.
• Tone text/border: text.{tone} → step 12, border.{tone} → step 8. text.accent repointed from {info.8} (link blue, still available as text-link) to {accent.12}.
• Name-collision fixes: the semantic layer becomes the single owner of border-focus (→ {info.7}, removing the self-reference), border-disabled and text-disabled (→ {neutral-alpha.7}); the scheme sets drop their duplicates. text-disabled had the same double-value collision as border-disabled and is fixed the same way.

The generated artifacts are added to .prettierignore so they keep the studio-pull JSON style and diff cleanly against src/tokens/raw/.

Verification

build-proposal.mjs resolves all 376 references across the proposal sets, checks that no flattened --eds-* name is declared by both the semantic and a scheme set, and asserts each alpha twin composites back to its opaque step exactly.

Out of scope

• The :root-scoping export problem (Emit the semantic token layer under the [data-color-scheme] scopes #5226) — this PR is token content only.
• The Studio UI import itself: after design review, the sets are imported as replacements on a Studio work branch (removals must apply), released to beta, and verified after studio tokens pull.

[pomfrida]

@claude review

[claude]

Claude finished @pomfrida's task in 2m 49s —— View job

---

Review: import-ready Tokens Studio JSON for beta colour gaps

• Read build-proposal.mjs (generator + verification logic)
• Read README.md (design-review document)
• Spot-check the generated JSON sets (semantic + scheme)
• Check .prettierignore changes
• Verify the generator's claims by re-running it

I ran the generator and checked its claims against the raw sources. This is a high-quality, self-verifying PR — the approach is sound and the diff is trustworthy.

Verified ✅

• Idempotent generation. Re-running node build-proposal.mjs produces a zero-line diff against the committed JSON, so the artifacts genuinely match the generator.
• Verification passes. 376 references verified, no CSS name collisions. — every {ref} resolves against the combined scheme+semantic sets, and the flattened-CSS-name collision check is real (it maps paths to --eds-* names exactly the way resolveToChannels does, segments.join('-')).
• Alpha-twin table is reproducible. The console table matches README.md row-for-row (values, alpha %, canvas deviation). The composite-exactness assertion in alphaTwin (composite.every((c,i) => c === target[i])) is a genuine guarantee, and it picks the minimal alpha deterministically.
• Collision/double-value claims are accurate. Confirmed against src/tokens/raw/: semantic border.focus = {border-focus} (the self-reference), semantic border.disabled = {neutral.2} vs scheme {neutral.7}, semantic text.disabled = {neutral.5} vs scheme {neutral.7}. The fix (semantic becomes sole owner, scheme sets drop the duplicates) is correctly implemented.
• retone string-replace is safe. background.surface.accent references only {accent.*} steps, so JSON.stringify(...).replaceAll('{accent.', '{neutral.') mirrors cleanly with no accidental cross-tone matches ({accent-alpha. is not matched by {accent.).

Observations / minor suggestions (non-blocking)

1. Duplicated colour math (maintenance risk). oklchToChannels / gamut-mapping is copied verbatim from scripts/generate-ts-tokens.mjs (the header says so). If that source ever changes its conversion, this copy diverges silently — the exactness assertions here are relative to this copy, not the shipped tokens. Low risk since the proposal is regenerated on demand, but worth a note, or importing the shared helper if it's exportable.

2. Strict oklch() regex. ^oklch\(([\d.]+) ([\d.]+) ([\d.]+)\)$ (line 92) won't match percentage lightness, none hue, or / alpha forms. It fails loudly via fail() rather than producing wrong output, which is the right default — but given the "re-run after each studio tokens pull" workflow, a future export-format tweak would break the rebase. Fine as-is; just flagging the coupling.

3. icon.disabled is a value change, not just a collision fix. Line 322 repoints icon.disabled ({neutral.5} → {neutral-alpha.7}) alongside the true collision fixes. It's not a duplicated CSS name like the other three, so it's a disabled-strength change bundled in. It is covered by open question fake-bot added new tokens #2, so this is intentional — just make sure design reviewers see it as a behaviour change, not a no-op.

4. .prettierignore glob scope. packages/eds-tokens/token-proposals/**/*.json ignores all JSON under any future proposal folder, not just this one. Presumably intended (all proposals are studio-pull-style), but confirm that's the desired blanket scope. Also note the file has pre-existing duplication (packages/eds-icons/src/data.ts appears twice) and no trailing newline — not introduced here, but the new block lands in a slightly messy file if you want to tidy it in passing.

Design sign-off still needed

The README's open questions are the right ones to gate on — derived alpha percentages becoming canonical (#1), standardising disabled strength on step 7 (#2), and the text.accent → {accent.12} repoint being covered by text-link (#3). None are code issues; they're design decisions this PR correctly defers.

Nice work — the generator-as-proof-of-diff pattern makes this genuinely reviewable.
• chore/tokens-studio-beta-colour-proposal