@dashforge/tw 1.1.0 → 1.2.0

package/PERFORMANCE.md

@@ -1,227 +0,0 @@
-# `@dashforge/tw` performance baseline
-
-> Establishes the performance commitment for `@dashforge/tw`. Every
-> subsequent release should hold these numbers or justify any
-> regression in `CHANGELOG.md`.
-
-## Bundle size
-
-### Full library (`dist/index.esm.js`)
-
-Measured immediately after `nx build @dashforge/tw` on the
-`0.4.0-beta` source tree.
-
-| Metric | Value |
-|---|---|
-| Raw size | **312 KB** (319,924 bytes) |
-| Gzipped | **68.85 KB** (70,499 bytes) |
-| Brotli (estimate) | ~58 KB |
-
-### Historical bundle trajectory
-
-| Version | Raw | Gzipped | Components | Delta vs prev |
-|---|---|---|---|---|
-| `0.1.0-beta` | 255 KB | ~57 KB | 16 (tier-1/2/3) | baseline |
-| `0.2.0-beta` | 272 KB | ~60 KB | 24 (+8 foundation) | +17 KB raw |
-| `0.3.0-beta` | 272 KB | ~60 KB | 24 | 0 (no new components) |
-| `0.4.0-beta` | **312 KB** | **68.85 KB** | 29 (+5 tier-4) | **+40 KB raw / +8.85 KB gz** |
-
-The `0.4.0-beta` delta is the cost of 5 new Tier-4 components +
-their 5 Radix UI primitive dependencies (`@radix-ui/react-dialog`,
-`-tabs`, `-tooltip`, `-popover`, `-accordion`). The Radix runtime
-sub-dependencies (`@radix-ui/react-portal`, `-presence`,
-`-dismissable-layer`, etc.) are shared across primitives so the
-marginal cost of each new primitive after the first is smaller.
-
-### Tree-shaken size — per-component source weight
-
-Tree-shaken bundle size depends on the consumer's bundler (Vite,
-webpack, esbuild, Rollup). The TS source size below is a **proxy**
-for the contribution each component makes — not the precise
-tree-shaken cost, which includes the Radix runtime and shared
-helpers (`cn()`, `tailwind-variants`).
-
-| Component | Source bytes | Tier |
-|---|---:|---|
-| Autocomplete | 49,020 | Tier-3 (heaviest — multi-select + free-solo + async) |
-| LeftNav | 18,002 | F6 (navigation) |
-| RadioGroup | 16,946 | Tier-2 |
-| NumberField | 16,289 | Tier-2 |
-| DateTimePicker | 16,075 | Tier-3 |
-| Checkbox | 15,732 | Tier-1 |
-| Grid | 15,086 | F10 (foundation) |
-| TextField | 14,772 | Tier-1 |
-| Snackbar | 14,770 | F7 |
-| Box | 14,039 | F9 (foundation) |
-| ConfirmDialog | 13,361 | F7 |
-| AppShell | 12,734 | F6 |
-| Button | 12,628 | Tier-1 |
-| OTPField | 12,497 | Tier-2 |
-| Breadcrumbs | 11,694 | F6 |
-| Typography | 11,619 | F9 |
-| Divider | 11,026 | F10 |
-| Textarea | 10,310 | Tier-2 |
-| Stack | 9,131 | F9 |
-| Switch | 8,711 | Tier-1 |
-| **Dialog** | **7,421** | **Tier-4 (new)** |
-| Container | 6,754 | F10 |
-| **Accordion** | **6,150** | **Tier-4 (new)** |
-| **Tabs** | **5,303** | **Tier-4 (new)** |
-| AspectRatio | 4,871 | F10 |
-| TopBar | 4,723 | F6 |
-| **Tooltip** | **4,389** | **Tier-4 (new)** |
-| **Popover** | **3,747** | **Tier-4 (new)** |
-| VisuallyHidden | 3,744 | F10 |
-
-The 5 new Tier-4 components contribute **~27 KB of source** combined
-— roughly aligned with the +40 KB raw bundle delta (the extra
-~13 KB is Radix runtime).
-
-### Representative bundle subsets
-
-If a consumer imports only a subset, expected gzipped cost
-(estimated, depends on bundler):
-
-| Subset | Components | Estimated gzipped |
-|---|---|---|
-| Atomic (Button only) | 1 | ~6 KB |
-| Form bundle | TextField + Checkbox + Switch + RadioGroup + NumberField + Textarea + OTPField | ~28 KB |
-| Layout bundle | Box + Stack + Grid + Container + AppShell + LeftNav + TopBar + Breadcrumbs | ~24 KB |
-| Foundation (F9 + F10) | Typography + Box + Stack + Grid + Container + Divider + AspectRatio + VisuallyHidden | ~18 KB |
-| Tier-4 overlay primitives | Dialog + Tabs + Tooltip + Popover + Accordion | ~12 KB |
-| Full library | 29 | **68.85 KB** |
-
-Estimates are upper bounds — actual tree-shaking can do better if
-the consumer doesn't trigger code paths (e.g. importing
-`<Autocomplete>` without using `loadOptions` won't strip the async
-loader code from the bundle, but other unused components are
-fully eliminable).
-
-## Render performance
-
-### Sprint 2 baseline (in `dash` consumer, React Profiler)
-
-Measured in
-`~/projects/web/learn/dash/src/pages/TestFoundation.tsx` — page
-mounts all 8 Foundation primitives × ~6 instances each (50+
-primitives total) inside `<DashforgeTailwindProvider>` wrapped in
-`<React.Profiler>`:
-
-| Operation | Time | Notes |
-|---|---|---|
-| First mount | **12.1 ms** | Cold cache, no styles compiled yet |
-| Re-render (no state change) | **7.0 ms** | After style cache warm |
-| Re-render (state change) | **8.6 ms** | One field updated, sibling fields untouched |
-
-Within the **60 fps budget** (16.67 ms per frame) for the full page
-mount + every subsequent interaction. The primitives are pure (no
-internal `useState`, no `useEffect`, only className resolution via
-`tv()` + `cn()`), so React's reconciler trivially handles re-renders
-without `React.memo`.
-
-### Tier-4 components — expected render cost
-
-Tier-4 components (Dialog, Tabs, Tooltip, Popover, Accordion) carry
-internal Radix state machines. Render cost is dominated by Radix's
-own reconciliation, which is well-optimized. Expected ranges:
-
-| Component | Expected mount | Notes |
-|---|---|---|
-| Tooltip | <1 ms | Renders nothing until hover; portal mount on open is ~2 ms |
-| Popover | <1 ms | Same as Tooltip |
-| Tabs | ~2 ms | Static — all triggers + active panel mount once |
-| Accordion | ~3 ms | Per-item triggers + lazy panel mount |
-| Dialog | ~3 ms | Portal + focus trap setup on open |
-
-These are estimates based on Radix's published benchmarks. Real
-consumer-app numbers will be captured in Sprint 3 follow-up
-end-to-end validation pass (`/test-tier-4` page in `dash`).
-
-## Regression budget policy
-
-**Every PR that touches a component or adds a new export MUST**:
-
-1. Re-run `nx build @dashforge/tw` and check the new
-   `dist/index.esm.js` size (raw + gzipped).
-2. If the gzipped delta is **> +5%** vs the previous release,
-   include in the PR description:
-   - The bytes added (raw + gzipped)
-   - The reason (new component / new feature / new dep / refactor)
-   - A justification for why the cost is worth it
-3. If the gzipped delta is **> +10%**, the PR also requires a
-   reviewer's sign-off on the bundle impact explicitly (not a
-   silent pass).
-
-The 5% threshold at the current 68.85 KB baseline = 3.4 KB. That's
-enough headroom for a typical component-level change (e.g. adding a
-single Radix primitive's worth of code) without the policy
-triggering, but small enough that a careless dep addition gets
-flagged.
-
-## Render perf test suite (Sprint 6 P3)
-
-CI-enforced render budgets, co-located as `*.perf.test.tsx`. Bounds
-are deliberately generous — they catch **order-of-magnitude**
-regressions (an accidental O(n²) in the pipeline, or a virtualization
-break that mounts the full dataset), not micro-jitter.
-
-### `<Table>` — non-virtualized (`Table.perf.test.tsx`)
-
-| Scenario | Budget | Asserts |
-|---|---|---|
-| Mount 500 rows (Table's row-count ceiling) | < 800 ms | render path stays linear |
-| 500 rows in DOM | exactly 500 `<tbody tr>` | non-virtualized contract |
-| Default sort over 500 rows | < 800 ms | sort pipeline is not O(n²) |
-
-### `<DataGrid>` — virtualized (`DataGrid.perf.test.tsx`)
-
-| Scenario | Budget | Asserts |
-|---|---|---|
-| Mount 10 000 rows | < 500 ms | render is O(window) |
-| 10 000-row DOM node count | < 100 `<tbody tr>` | only the window is mounted |
-| Mount 100 000 rows | < 800 ms | cost does NOT scale with the dataset |
-
-The 100 000-row test is the load-bearing one: if a regression makes
-`renderedRows` slice the full array, the DOM-node-count assertion
-fails immediately (100k `<tr>` ≫ 100).
-
-## Strict gate assessment (Sprint 6 P3)
-
-A probe enabled `exactOptionalPropertyTypes` + `noUncheckedIndexedAccess`
-+ `noUnusedParameters` on `@dashforge/tw`. Result: **42 errors, zero
-real bugs.** Every hit was either pedantic (`exactOptionalPropertyTypes`
-flagging `fn | undefined` passed to an optional prop) or invariant-safe
-but unprovable (`noUncheckedIndexedAccess` flagging array accesses
-already guarded by a `.length` check — AppShell focus trap, RadioGroup
-option access, Autocomplete highlight index).
-
-**Decision**: enabling those flags is a workspace-wide change (the
-`@dashforge/source` path mapping pulls the bridge layer's source into
-the same program) for near-zero bug-catching value. Deferred to the
-`1.0.0-rc` hardening sprint. The codebase is already solid under the
-current `strict: true` baseline + `noImplicitOverride` /
-`noImplicitReturns` / `noUnusedLocals` / `noFallthroughCasesInSwitch`.
-ESLint is on `typescript-eslint recommended` + `react-hooks
-recommended` + Nx boundary/dependency checks; promotion to
-`strict-type-checked` (type-aware `no-unsafe-*` / `no-floating-promises`)
-is the same workspace-wide 1.0-rc item.
-
-## Console hygiene (Sprint 6 P3)
-
-The full test run is audited for stray `console.error` / `console.warn`.
-Sprint 6 P3 fixed the one finding: Radix logged 8× *"Missing
-`Description` or `aria-describedby` for DialogContent"* — `<Dialog>`
-now passes `aria-describedby={undefined}` on the content when no
-`description` is supplied (Radix's documented opt-out). Console is
-clean apart from a single jsdom-only *"Not implemented: navigation"*
-(an environment limitation, not a component defect).
-
-## Reference
-
-- **Build command**: `nx build @dashforge/tw`
-- **Measurement script**: `gzip -c dist/index.esm.js | wc -c`
-- **Render perf setup**: see Sprint 2 P1
-  (`/test-foundation` route in `dash` consumer)
-- **Render perf suite**: `*.perf.test.tsx` (Sprint 6 P3 — Table + DataGrid)
-- **CHANGELOG entries**: every release records the bundle delta
-  in the `Compatibility` section

package/REFINEMENT-NOTES.md

@@ -1,50 +0,0 @@
-# REFINEMENT-NOTES — `@dashforge/tw` Table + DataGrid
-
-> Sprint 6 P2 deliverable (2026-05-20). Blind technical review of
-> `<Table>` + `<DataGrid>` — what was fixed, and what was explicitly
-> deferred.
-
-## Fixed in Sprint 6 P2
-
-| ID | Area | Fix |
-|----|------|-----|
-| D1 | DataGrid a11y | `aria-rowcount` on the `<table>` (true dataset size, not the virtualized window) + `aria-rowindex` on the header row (1) and every data row (`absoluteIndex + 2`). A screen reader now announces the real position. |
-| D3 | DataGrid a11y | Column resize is now keyboard-operable: the resize handle is `tabIndex=0`, `ArrowLeft`/`ArrowRight` nudge the width 16px (64px with Shift), `aria-valuemin/max/now` announce the bounds. New `nudge()` on `useColumnResize`. |
-| D6 | DataGrid bug | The resize handle on a reorderable column could also start a column-reorder drag (the `<th>` is `draggable`). Handle now sets `draggable={false}` — grabbing it always resizes. |
-| D2 | DataGrid touch | Column reorder rides native HTML5 drag-and-drop, whose events don't fire on touch. The grid now detects a coarse pointer (`matchMedia('(pointer: coarse)')`) and disables the `draggable` affordance there, so it isn't a dead interaction. (Full touch reorder — see Deferred below.) |
-| D4 | DataGrid UX | `Enter` in any filter Popover input now applies the filter (was: must click "Apply"). |
-| D7 | Table + DataGrid UX | Empty state distinguishes a genuinely empty dataset (`noData`) from one emptied by an active search / filter (`noResults`). New `TableLabels.noResults`. |
-| D8 | Table | Verified non-issue — Table is auto-height by design; skeleton rows inherit cell density padding. Added a clarifying code comment. |
-| M4 | Cell library | Cleaned up a self-contradictory comment in `RenderChip` (the `solid neutral` variant). |
-| T1 | Table | `getRowId` stays optional (back-compat) but now emits a dev-only `console.warn` when omitted while sort / search / selection / expandable is active — the positional-index fallback breaks row identity on reorder. Expanded the prop JSDoc. |
-
-## Deferred — tracked follow-ups
-
-### T3 — Per-column filter UI on `<Table>`
-`<DataGrid>` got the in-header filter Popover in 0.8.0-beta (`filterable: true`).
-`<Table>` still only accepts a controlled `filterModel` with no UI to
-produce it. Porting `ColumnFilterTrigger` + the filter pipeline to
-Table is a feature, not a refinement — it belongs in a dedicated
-feature sprint (or post-1.0). Until then, Table's `filterModel` is
-consumer-driven only.
-
-### C2 — Cell renderer library additions
-The library ships `RenderText`, `RenderTwoLine`, `RenderChip`,
-`RenderButton`, `RowActionsMenu`. Common table-cell patterns still
-missing:
-- `RenderDate` — locale-aware date formatting (`Intl.DateTimeFormat`,
-  zero-dep)
-- `RenderAvatar` — avatar image / initials + label, the canonical
-  "user" cell
-- `RenderLink` — a styled anchor honoring the theme
-
-Purely additive — slated for a feature sprint, not refinement.
-
-### D2-full — Pointer-events column reorder (touch support)
-The D2 fix above is *graceful degradation* — reorder is cleanly
-unavailable on touch rather than silently broken. A proper touch
-implementation means replacing `useColumnReorder`'s HTML5
-drag-and-drop with pointer events (à la `useColumnResize`): track
-`pointerdown` → `pointermove` with hit-testing (`elementFromPoint`)
-for the drop target → `pointerup` to commit. Meaningful rewrite —
-deferred to a follow-up so P2 stays a refinement pass.

package/THEME-AUDIT.md

@@ -1,174 +0,0 @@
-# THEME-AUDIT — @dashforge/tw catalog identity sweep
-
-> Audit of every `@dashforge/tw` component against the **default
-> `dashforgePreset()` identity rule**.
->
-> **Rule recap.** The `dashforgePreset()` CSS-variable swap
-> auto-inverts the neutral palette:
-> `bg-neutral-50` / `text-neutral-900` etc. resolve to the
-> appropriate physical color via `--df-tw-color-neutral-N`. Adding a
-> `dark:` variant on a neutral class **double-inverts** the value
-> and breaks dark mode. The Table fix established the
-> canonical patterns; Sprint 4.3 applies them to the whole catalog.
-
-## Scoring legend
-
-- **A** — Pure double inversion. REMOVE `dark:` variant entirely.
-- **B** — `bg-white` / `text-white` paired with a wrong `dark:`
-  target. KEEP the `dark:` variant (white doesn't auto-invert) but
-  FIX the target — `dark:bg-neutral-900` in dark mode resolves to
-  near-white (broken); use `dark:bg-neutral-100` for proper dark
-  elevated surface.
-- **C** — Intentional `dark:` on color palette (`primary`,
-  `success`, `warning`, `danger`, `info`, `secondary`). KEEP — color
-  palettes don't auto-invert; explicit `dark:` is a design choice.
-- **D** — Component-specific exception (e.g. Tooltip wants
-  high-contrast inversion against the page). Document why kept.
-
-## Per-file findings
-
-### Accordion (`accordion.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 6 | `border-b border-neutral-200 dark:border-neutral-800` | A | `border-b border-neutral-200` |
-| 10 | `text-sm font-medium text-neutral-900 dark:text-neutral-50` | A | `text-sm font-medium text-neutral-900` |
-| 18 | `overflow-hidden text-sm text-neutral-700 dark:text-neutral-300` | A | drop `dark:text-neutral-300` |
-
-### Box (`box.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 58 | `bg-white dark:bg-neutral-900` (elevated) | B | `bg-white dark:bg-neutral-100` (dark-mode elevation tier above page surface) |
-| 168 | `border-neutral-200 bg-white dark:border-neutral-700 dark:bg-neutral-900` (outlined) | A + B | `border-neutral-200 bg-white dark:bg-neutral-100` (border auto-inverts; bg keeps `bg-white` + corrected dark target) |
-| 184 | `bg-neutral-100 text-neutral-900 dark:bg-neutral-800 dark:text-neutral-100` (soft) | A | `bg-neutral-100 text-neutral-900` |
-| 200 | `bg-neutral-900 text-white dark:bg-neutral-100 dark:text-neutral-900` (solid neutral) | D | `bg-neutral-900 text-neutral-50` (auto-invert — solid neutral becomes inverted accent: dark surface in light mode, light surface in dark mode; consistent with identity rule) |
-
-### Dialog (`dialog.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 29 | `w-full bg-white dark:bg-neutral-900 rounded-lg shadow-xl` | B | `dark:bg-neutral-100` |
-| 30 | `border border-neutral-200 dark:border-neutral-800` | A | drop `dark:` |
-| 37 | `text-lg ... text-neutral-900 dark:text-neutral-50` | A | drop `dark:` |
-| 38 | `text-sm text-neutral-600 dark:text-neutral-400` | A | drop `dark:` |
-| 42 | `dark:text-neutral-400 dark:hover:text-neutral-50` (close button) | A | drop both `dark:` |
-| 43 | `hover:bg-neutral-100 dark:hover:bg-neutral-800` | A | drop `dark:` |
-
-### Divider (`divider.variants.ts` + `Divider.tsx`)
-
-| File / Line | Class | Category | Fix |
-|---|---|---|---|
-| `divider.variants.ts:88` | `border-neutral-200 dark:border-neutral-800` (neutral color variant) | A | drop `dark:` |
-| `Divider.tsx:131` | `text-sm text-neutral-500 dark:text-neutral-400` (label) | A | drop `dark:` |
-
-### Pagination (`pagination.variants.ts` + `Pagination.tsx`)
-
-10 violations — full sweep of borders / text / bg / hover / active-button.
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| `variants:26` summary | `text-neutral-600 dark:text-neutral-400` | A | drop |
-| `variants:30, 44, 58` borders | `border border-neutral-300 dark:border-neutral-700` | A | drop |
-| `variants:31, 45, 59` button/selector/input bg | `bg-white dark:bg-neutral-900` | B → drop entirely | use `bg-neutral-50` (auto-inverts; border provides separation against page) |
-| `variants:32, 46, 55` button text | `text-neutral-700 dark:text-neutral-300` | A | drop |
-| `variants:33, 47` button hover bg | `hover:bg-neutral-50 dark:hover:bg-neutral-800` | A | `hover:bg-neutral-100` (1 elevation tier; auto-inverts) |
-| `variants:60` jumpInput text | `text-neutral-900 dark:text-neutral-50` | A | drop |
-| `Pagination.tsx:241` select bg+border | `border ... dark:border-neutral-700 bg-white dark:bg-neutral-900` | B → drop | `border border-neutral-300 bg-neutral-50` |
-
-### Popover (`popover.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 6 | `border border-neutral-200 dark:border-neutral-800` | A | drop |
-| 7 | `bg-white dark:bg-neutral-900 p-4 shadow-lg` | B | `dark:bg-neutral-100` |
-| 8 | `text-sm text-neutral-900 dark:text-neutral-50` | A | drop |
-| 14 | `fill-white dark:fill-neutral-900` (arrow) | B | `fill-white dark:fill-neutral-100` (match content bg) |
-
-### Skeleton (`skeleton.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 38 | `bg-neutral-200 dark:bg-neutral-800` | A | drop `dark:` (auto-inverts — skeleton placeholder slightly more visible than page surface in both modes) |
-
-### Tabs (`tabs.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 22 | `border-b border-neutral-200 dark:border-neutral-800` (underline list) | A | drop |
-| 24 | `... text-neutral-600 dark:text-neutral-400` (underline trigger) | A | drop |
-| 25 | `hover:text-neutral-900 dark:hover:text-neutral-50` | A | drop |
-| 31 | `bg-neutral-100 dark:bg-neutral-800 p-1` (pill list) | A | drop |
-| 33 | `... text-neutral-600 dark:text-neutral-400` (pill trigger) | A | drop |
-| 34 | `hover:text-neutral-900 dark:hover:text-neutral-50` | A | drop |
-| 35 | `data-[state=active]:bg-white dark:data-[state=active]:bg-neutral-900` (pill active) | B | `dark:bg-neutral-100` |
-| 36 | `data-[state=active]:text-neutral-900 dark:data-[state=active]:text-neutral-50` | A | drop |
-
-### Tooltip (`tooltip.variants.ts`)
-
-Tooltip historically used a "fixed dark surface in both modes" pattern.
-Per the identity rule, **auto-invert** is the right behavior: dark
-tooltip on light page (light mode), light tooltip on dark page (dark
-mode) — the inversion **is** the high-contrast effect.
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 8 | `bg-neutral-900 text-neutral-50` + `dark:bg-neutral-50 dark:text-neutral-900` | A | drop the `dark:` line entirely (the base classes already auto-invert and give the high-contrast effect against the page) |
-| 14 | `fill-neutral-900 dark:fill-neutral-50` (arrow) | A | drop `dark:` (arrow auto-inverts with content bg) |
-
-### Typography (`typography.variants.ts`)
-
-| Line | Class | Category | Fix |
-|---|---|---|---|
-| 65 | `neutral: 'text-neutral-900 dark:text-neutral-100'` | A | `'text-neutral-900'` |
-| 66 | `muted: 'text-neutral-600 dark:text-neutral-400'` | A | `'text-neutral-600'` |
-
-### Table-side (Sprint 4.1 — already fixed)
-
-Per-file already corrected in Sprint 4.1. Sprint 4.3 just verifies
-the regression guard (themeIdentity test) is promoted to global so
-the rest of the catalog is covered too.
-
-## Summary
-
-| Component | Violations | Fix categories |
-|---|---|---|
-| Accordion | 3 | A × 3 |
-| Box | 4 | A + B + D |
-| Dialog | 6 | A × 5 + B × 1 |
-| Divider | 2 | A × 2 |
-| Pagination | 10 | A × 7 + B × 3 |
-| Popover | 4 | A × 3 + B × 1 |
-| Skeleton | 1 | A × 1 |
-| Tabs | 8 | A × 7 + B × 1 |
-| Tooltip | 2 | A × 2 (was "double-inverted to preserve fixed look" — auto-invert is now the canonical choice) |
-| Typography | 2 | A × 2 |
-| **Total** | **42** | **31 A + 5 B + 1 D + 5 fixed already** |
-
-## Color palettes — NOT touched
-
-`dark:` variants on `primary`, `secondary`, `success`, `warning`,
-`danger`, `info` palettes are **intentional** — those palettes do
-not auto-invert (same physical color in both modes), so the
-`dark:` shift is a design choice for tone refinement. They stay.
-
-Affected sites (kept as-is):
-
-- `Box` compound variants (`outlined/soft/solid` × color) for the
-  6 color palettes (lines 188-198 in `box.variants.ts`)
-- `RenderChip` color rows (Sprint 4.1 decision)
-- `RowActionsMenu` `text-danger-700 dark:text-danger-300` and same
-  for warning
-
-## Sprint 4.3 deliverables
-
-1. ✅ THEME-AUDIT.md (this file)
-2. ⏳ Apply fixes per-file (P3)
-3. ⏳ Promote `themeIdentity.test.ts` from `Table/_internal/` to
-   package-level scanner walking `libs/dashforge/tw/src/components/**`
-   (P2). Excludes `*.test.*` files. The test fails any future
-   `dark:*-neutral-N` reintroduction.
-4. ⏳ Visual regression catch in `dash` (manual check in both
-   light + dark mode)
-5. ⏳ Release `@dashforge/tw@0.7.0-beta` — expected bundle to
-   **shrink** by 1-3 KB gz (removing redundant classes).

package/eslint.config.mjs

@@ -1,62 +0,0 @@
-import baseConfig from '../../../eslint.config.mjs';
-import reactHooks from 'eslint-plugin-react-hooks';
-
-export default [
-  ...baseConfig,
-  // React Hooks plugin — every component in this package authors
-  // hooks, so the rule set is required for `// eslint-disable-next-line
-  // react-hooks/exhaustive-deps` directives to be valid AND to catch
-  // real rules-of-hooks violations (hook called after early-return).
-  // Scoped here (not at workspace root) to avoid surfacing pre-existing
-  // unrelated bugs in sibling packages that fall outside this sprint's
-  // scope — those should be cleaned up in a dedicated workspace pass.
-  {
-    files: ['**/*.{ts,tsx,js,jsx,cts,mts,cjs,mjs}'],
-    plugins: { 'react-hooks': reactHooks },
-    rules: {
-      ...reactHooks.configs.recommended.rules,
-    },
-  },
-  {
-    files: ['**/*.json'],
-    rules: {
-      '@nx/dependency-checks': [
-        'error',
-        {
-          // Non-shipped files: their imports must NOT count toward the
-          // published `dependencies` set. Spec/test files import the
-          // test runner; the build/test config files import build
-          // tooling — all of which belong in `devDependencies`.
-          ignoredFiles: [
-            '{projectRoot}/eslint.config.{js,cjs,mjs,ts,cts,mts}',
-            '{projectRoot}/rollup.config.{js,ts,mjs,mts,cjs,cts}',
-            '{projectRoot}/vite.config.{js,ts,mjs,mts}',
-            '{projectRoot}/vitest.config.{js,ts,mjs,mts}',
-            '{projectRoot}/**/*.spec.{ts,tsx}',
-            '{projectRoot}/**/*.test.{ts,tsx}',
-          ],
-          // tw-tokens (CSS vars + Tailwind preset) and tw-theme
-          // (Provider that injects the runtime vars) are CONSUMER-FACING
-          // peer dependencies: the runtime requires them, but @dashforge/tw
-          // itself never imports them. Declaring them as peers documents
-          // the requirement without triggering the "not used" check.
-          // `tailwindcss` is the same case — a build-time peer (the
-          // consumer's Tailwind compiles the class strings + reads the
-          // `darkMode: 'selector'` config from `dashforgePreset()`); it
-          // is never an `import` site in this package's code.
-          ignoredDependencies: [
-            '@dashforge/tw-tokens',
-            '@dashforge/tw-theme',
-            'tailwindcss',
-          ],
-        },
-      ],
-    },
-    languageOptions: {
-      parser: await import('jsonc-eslint-parser'),
-    },
-  },
-  {
-    ignores: ['**/out-tsc'],
-  },
-];

package/project.json

@@ -1,15 +0,0 @@
-{
-  "$schema": "../../../node_modules/nx/schemas/project-schema.json",
-  "name": "@dashforge/tw",
-  "targets": {
-    "typecheck": {
-      "options": {
-        "command": "tsc --build tsconfig.lib.json --emitDeclarationOnly"
-      }
-    },
-    "build": {
-      "_comment": "typecheck excluded from dependsOn: TS bundler resolver fails when consuming `@dashforge/forms/dist/index.d.ts` (a 1-line `export * from \"./src/index\"` re-export) through nx's composite project chain. `nx run typecheck` separately works; `nx run-many -t test` works (it bundles types via vitest). Re-attach `typecheck` once the publishables' dist d.ts wrappers are switched to inline `export ...` (out of scope F3).",
-      "dependsOn": ["^build"]
-    }
-  }
-}

package/rollup.config.cjs

@@ -1,16 +0,0 @@
-const { withNx } = require('@nx/rollup/with-nx');
-
-module.exports = withNx(
-  {
-    main: './src/index.ts',
-    outputPath: './dist',
-    tsConfig: './tsconfig.lib.json',
-    compiler: 'swc',
-    format: ['esm'],
-  },
-  {
-    // Provide additional rollup configuration here. See: https://rollupjs.org/configuration-options
-    // e.g.
-    // output: { sourcemap: true },
-  }
-);

package/src/components/Accordion/Accordion.test.tsx

@@ -1,95 +0,0 @@
-// @vitest-environment jsdom
-import * as React from 'react';
-import { describe, it, expect, afterEach } from 'vitest';
-import { render, screen, fireEvent, cleanup } from '@testing-library/react';
-import { Accordion } from './Accordion.js';
-
-void React;
-afterEach(() => cleanup());
-
-const items = [
-  { value: 'one',   header: 'Section one',   content: <p>One body</p> },
-  { value: 'two',   header: 'Section two',   content: <p>Two body</p> },
-  { value: 'three', header: 'Section three', content: <p>Three body</p> },
-];
-
-function regionIsOpen(bodyText: string): boolean {
-  const node = screen.queryByText(bodyText);
-  if (!node) return false;
-  const region = node.closest('[role="region"]');
-  if (!region) return false;
-  // Radix sets `hidden` attr on closed regions in jsdom.
-  return region.getAttribute('hidden') === null && !(region as HTMLElement).hidden;
-}
-
-describe('Accordion — single mode (default)', () => {
-  it('renders all triggers', () => {
-    render(<Accordion items={items} />);
-    expect(screen.queryByText('Section one')).not.toBeNull();
-    expect(screen.queryByText('Section two')).not.toBeNull();
-    expect(screen.queryByText('Section three')).not.toBeNull();
-  });
-
-  it('opens a panel on trigger click', () => {
-    render(<Accordion items={items} />);
-    expect(regionIsOpen('Two body')).toBe(false);
-    fireEvent.click(screen.getByText('Section two'));
-    expect(regionIsOpen('Two body')).toBe(true);
-  });
-
-  it('closes the open panel when clicked again (collapsible default)', () => {
-    render(<Accordion items={items} />);
-    const trigger = screen.getByText('Section two');
-    fireEvent.click(trigger);
-    expect(regionIsOpen('Two body')).toBe(true);
-    fireEvent.click(trigger);
-    expect(regionIsOpen('Two body')).toBe(false);
-  });
-
-  it('respects defaultValue', () => {
-    render(<Accordion items={items} defaultValue="three" />);
-    expect(regionIsOpen('Three body')).toBe(true);
-  });
-
-  it('emits onValueChange', () => {
-    let last: string | null = null;
-    render(<Accordion items={items} onValueChange={(v) => (last = v)} />);
-    fireEvent.click(screen.getByText('Section one'));
-    expect(last).toBe('one');
-  });
-});
-
-describe('Accordion — multiple mode', () => {
-  it('allows multiple panels open simultaneously', () => {
-    render(<Accordion type="multiple" items={items} defaultValue={['one', 'three']} />);
-    expect(regionIsOpen('One body')).toBe(true);
-    expect(regionIsOpen('Three body')).toBe(true);
-  });
-
-  it('emits onValueChange with the updated array', () => {
-    let last: string[] = [];
-    render(
-      <Accordion
-        type="multiple"
-        items={items}
-        defaultValue={['one']}
-        onValueChange={(v) => (last = v)}
-      />
-    );
-    fireEvent.click(screen.getByText('Section two'));
-    expect(last).toEqual(expect.arrayContaining(['one', 'two']));
-  });
-});
-
-describe('Accordion — disabled items', () => {
-  it('does not open a disabled section', () => {
-    const disabledItems = [
-      ...items.slice(0, 2),
-      { ...items[2]!, disabled: true },
-    ];
-    render(<Accordion items={disabledItems} />);
-    const trigger = screen.getByText('Section three');
-    fireEvent.click(trigger);
-    expect(regionIsOpen('Three body')).toBe(false);
-  });
-});