@dashforge/tw 0.5.0-beta → 0.7.0-beta

package/CHANGELOG.md

@@ -12,6 +12,532 @@ This project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 > duplicated intentionally — no shared "lowest common denominator" headless
 > layer.
 
+## [0.7.0-beta] — 2026-05-19
+
+**Combined Sprint 4.2 + 4.3 release.** Ships **`<DataGrid>`** — a
+virtualized data table for large data sets (500 rows to millions)
+— alongside a **theme identity sweep** across the whole catalog
+(systemic dark-mode contrast fixes on 10 existing components).
+
+Two coherent themes, one release:
+
+1. **DataGrid (Sprint 4.2)** — sibling to `<Table>` with **homemade
+   virtualization** (no `@tanstack/react-virtual`, no
+   `react-window`). Same column model, same cell renderer library,
+   same RBAC integration, same identity-consistent visual design as
+   Table. Differs at the render strategy level: only the window of
+   visible rows mounts in DOM. Adds sticky-left columns, server-side
+   mode (4 independent opt-in flags), `selectAllScope`, and optional
+   internal pagination.
+
+2. **Theme identity sweep (Sprint 4.3)** — applies the identity rule
+   codified in Sprint 4.1 to the WHOLE catalog. 10 existing
+   components (Typography, Box, Dialog, Divider, Pagination,
+   Popover, Skeleton, Tabs, Tooltip, plus Table-side comments)
+   had latent `dark:` variants on the neutral palette → those
+   classes double-inverted via the dashforgePreset CSS-var swap
+   and broke dark mode. Sweep removes the anti-pattern across the
+   catalog and adds a package-level regression test
+   (`_shared/themeIdentity.test.ts`) that scans every source file
+   on every test run.
+
+**Minor bump** for the new DataGrid public API surface. Strictly
+additive — zero breaking changes on existing component APIs.
+Drop-in upgrade from `0.6.0-beta`.
+
+### Added
+
+- **`<DataGrid>`** — virtualized data table for large data sets.
+  Sibling to `<Table>`. Required props: `rows`, `cols`,
+  `getRowId`, `rowHeight` (fixed for v1), `height` (bounded
+  container).
+  - **Homemade virtualization** via the new `useVirtualizer` hook
+    in `_shared/data/`. Scroll-event based with
+    `requestAnimationFrame` debounce (60 fps throttling) +
+    `ResizeObserver` for container resize. Spacer-row approach
+    preserves `<table>` semantics — screen readers see continuous
+    structure with `aria-hidden` spacers above/below the visible
+    window. Variable row height deferred to v1-bis.
+  - **Sticky left column** via `cols[i].sticky: 'left'`. CSS
+    `position: sticky; left: 0` with z-index ladder for the
+    top-left corner intersection with the sticky header
+    (`thead` z-10 < sticky cell z-[1] < sticky header corner z-20).
+    Right-sticky deferred to v1-bis.
+  - **Server-side mode flags** (4 independent opt-ins):
+    - `serverSideSort` — emits `onSortChange` but doesn't sort locally
+    - `serverSideFilter` — emits `onFilterChange` but doesn't filter
+    - `serverSideSearch` — emits `onSearchQueryChange` (debounced)
+    - `serverSidePagination` — uses `totalCount` for the virtual
+      scrollbar math; `rows` is the server-returned page slice
+  - **`selectAllScope`** prop (`'visible' | 'allLoaded'`, default
+    `'allLoaded'`) controls the header "select all" checkbox
+    semantic.
+  - **Optional internal pagination** via the `pagination` prop —
+    renders the Sprint 4 `<Pagination>` component below the
+    virtualized scroll. Composes naturally with virtualization
+    (each page is virtualized internally if large enough).
+  - **Bundle impact**: +8.4 KB gz on top of `0.6.0-beta`
+    (`<Table>` baseline). Zero new runtime deps.
+  - **Variants**: `plain` / `lines` (default) / `striped` /
+    `bordered`. **No `card` variant** — incompatible with
+    `<table>` virtualization spacer-rows.
+  - 24 unit tests + 12 `useVirtualizer` math tests + dash smoke
+    test at `/test-data-grid` with 10 000 generated rows.
+
+- **`useVirtualizer` hook** in `_shared/data/` — exported
+  internally only (not in the public API surface). Other future
+  virtualized components (TreeView, future GanttGrid, etc.) can
+  reuse it.
+
+- **`sticky?: 'left' | 'right'`** added to `TableColumn<T>` for
+  DataGrid integration. Currently ignored by Table (non-virtualized);
+  honored by DataGrid v1 for `'left'` only.
+
+- **Doc MDX `/tw/docs/components/data-grid`** — emphasizes the
+  Table vs DataGrid decision matrix (24-row comparison table +
+  "Use Table when..." / "Use DataGrid when..." + "Switch later"
+  path showing the 3-line diff to migrate).
+
+### Changed
+
+- **Internal refactor — shared data helpers moved**:
+  `Table/_internal/` → `_shared/data/` (13 files). The helpers
+  (`getNestedValue`, `useTableSearch`, `useTableSort`,
+  `useTableSelection`, `useTableFilter`, `useColumnAutoDetect`,
+  `useDebouncedValue`, `useControllableState`) are now shared
+  between Table and DataGrid. **Pure path move** — function names
+  unchanged. Public API surface unchanged (only `getNestedValue`
+  was ever re-exported; the re-export path updated transparently).
+
+- **Theme identity sweep** — 10 existing components had latent
+  `dark:` Tailwind variants on the neutral palette, which the
+  dashforgePreset CSS-var swap double-inverted and broke in dark
+  mode. Removed across the catalog:
+
+  | Component | Sites fixed | Pattern category |
+  |---|---|---|
+  | Accordion (`accordion.variants.ts`) | 3 | A (drop `dark:` — auto-invert) |
+  | Box (`box.variants.ts`) | 4 | A + B (corrected dark targets) + D (solid neutral now auto-inverts as accent) |
+  | Dialog (`dialog.variants.ts`) | 6 | A × 5 + B × 1 |
+  | Divider (`divider.variants.ts` + `Divider.tsx`) | 2 | A × 2 |
+  | Pagination (`pagination.variants.ts` + `Pagination.tsx`) | 10 | A × 7 + B × 3 (`dark:bg-neutral-900` → `bg-neutral-50` auto-invert) |
+  | Popover (`popover.variants.ts`) | 4 | A × 3 + B × 1 |
+  | Skeleton (`skeleton.variants.ts`) | 1 | A |
+  | Tabs (`tabs.variants.ts`) | 8 | A × 7 + B × 1 (`dark:bg-neutral-900` → `dark:bg-neutral-200` for proper active-pill elevation) |
+  | Tooltip (`tooltip.variants.ts`) | 2 | **Bug fix** — was previously invisible in dark mode (double-inverted to dark surface on dark page). Now auto-inverts: dark tooltip on light page, light tooltip on dark page (high contrast preserved). |
+  | Typography (`typography.variants.ts`) | 2 | A × 2 |
+
+  **Box solid neutral** now uses `bg-neutral-900 text-neutral-50`
+  (both auto-invert) — the accent surface flips with the page in
+  dark mode (light surface + dark text) for identity consistency.
+  Previously was `bg-neutral-900 text-white dark:bg-neutral-100
+  dark:text-neutral-900` — the `text-white` (no auto-invert) had
+  forced the double-invert pattern to stay readable.
+
+  Affected test files updated to assert the new canonical
+  patterns (Box × 4 + Typography × 1 + Divider × 1 = 6 tests
+  retargeted).
+
+### Internal
+
+- **`THEME-AUDIT.md`** at package root — categorizes every site
+  in the catalog as **A** (drop `dark:` — auto-invert), **B**
+  (keep `dark:` because `bg-white`/`fill-white` is hardcoded),
+  **C** (color palette — intentional design choice, not affected
+  by the rule), or **D** (component-specific exception). 42
+  violations catalogued across 10 components.
+
+- **Package-level theme identity regression test** —
+  `_shared/themeIdentity.test.ts` replaces the Table-only one
+  from Sprint 4.1. Scans every `.ts`/`.tsx` source file under
+  `src/components/**` (excluding tests) on every test run.
+  Allows the Category B legitimate pattern
+  (`bg-white dark:bg-neutral-N` — `bg-white` doesn't auto-invert,
+  the `dark:` is required). Catches accidental re-introductions
+  of the anti-pattern at PR time.
+
+- **+76 new unit tests** for Sprint 4.2 + 4.3:
+  - `useVirtualizer` math (12) — window calc, overscan, padding
+    invariant, edge cases (totalCount=0, very large dataset)
+  - `<DataGrid>` (24) — rendering, virtualized window, sort,
+    search, selection, selectAllScope, sticky left column,
+    server-side mode flags, RBAC, internal pagination, smart
+    defaults, i18n
+  - `_shared/themeIdentity.test.ts` package-level scanner (40
+    file-scan tests; 0 violations reported)
+
+  Full TW suite at **961/961 passing** across **48 files**.
+
+- **`TestDataGrid.tsx`** added to the `dash` consumer
+  (`/test-data-grid`) — 10 000 generated rows + sticky col +
+  selectAllScope toggle + variant/density/size switcher + RBAC
+  per-column demo + custom empty state + React.Profiler logging.
+
+- **Sidebar entries** for the new `DataGrid` doc page added to
+  `dashforge-docs-lab/src/tw-docs/sidebar.model.ts` (new "Data"
+  subgroup under "UI Components" now contains Table + DataGrid).
+
+### Compatibility
+
+| Axis | Pre-`0.7.0` | Post-`0.7.0` |
+|---|---|---|
+| Public API surface | 32 components | **+1 `<DataGrid>` + types (`DataGridProps`, `DataGridSelectAllScope`, `DataGridServerSideFlags`, `DataGridPaginationConfig`, `DataGridSlotProps`) + `dataGridVariants` recipe** + `sticky` axis on `TableColumn<T>` (additive, optional) |
+| Peer deps | `react ^18 \|\| ^19`, `tw-theme workspace`, `tw-tokens workspace` | unchanged |
+| Bridge deps | `forms` / `rbac` / `ui-core` `workspace:*` | unchanged |
+| **New runtime deps** | — | **none** (homemade virtualization — constraint honored) |
+| Breaking changes | — | **Zero** on public APIs. Internal refactor: `Table/_internal/` helpers moved to `_shared/data/`. Only externally-visible re-export was `getNestedValue` (path updated internally; export name preserved). |
+| Bundle size | 402 KB raw / 91 KB gz | **445 KB raw / 98.4 KB gz** (+43 KB raw / +7.4 KB gz / **+8.1% gz**) — just under the 10% reviewer-sign-off threshold |
+| Tests passing | 828/828 (46 files) | **961/961 (48 files)** — +133 from Sprint 4.2 + 4.3 |
+
+> **Bundle sign-off** : the +8.1% gz delta is JUST under the 10%
+> threshold from PERFORMANCE.md. Justification: DataGrid is the
+> highest-leverage addition before 1.0 — it unlocks the entire
+> "1 000+ row admin views" use case at €0 vs MUI X DataGrid Pro's
+> €499/year. The theme identity sweep is net-neutral on bundle
+> (removed redundant classes, replaced with smaller ones); the
+> delta is entirely DataGrid + virtualizer code (+8 KB gz).
+
+### Migration
+
+No code changes required:
+
+```bash
+pnpm up @dashforge/tw@^0.7.0-beta
+```
+
+To adopt the new DataGrid:
+
+```tsx
+import {
+  DataGrid,
+  RenderTwoLine,
+  RenderChip,
+  RowActionsMenu,
+} from '@dashforge/tw';
+
+<DataGrid
+  rows={users}                  // can be 10k, 100k, millions
+  cols={columns}                // SAME shape as Table's TableColumn<T>
+  getRowId={(r) => r.id}
+  rowHeight={48}                // required for virtualization
+  height="600px"                // required for bounded container
+  enableSearch
+  rowSelection="multiple"
+  selectAllScope="allLoaded"
+  bulkActions={(rows) => (
+    <Button color="danger">Delete {rows.length}</Button>
+  )}
+  rowActions={(row) => (
+    <RowActionsMenu
+      row={row}
+      actions={[
+        { label: 'Edit',   onClick: edit },
+        { label: 'Delete', onClick: del, color: 'danger' },
+      ]}
+    />
+  )}
+/>
+```
+
+To switch an existing Table to DataGrid when you outgrow the row
+count (~500+ rows), the migration is mechanical (3 lines):
+
+```diff
+- <Table
++ <DataGrid
+    rows={users}
+    cols={columns}
+    getRowId={(row) => row.id}
++   rowHeight={48}
++   height="600px"
+  />
+```
+
+Everything else (sort / search / selection / row actions / cell
+renderers / RBAC / i18n labels / sx / slotProps) keeps working
+identically. **Expandable rows** are the one feature you lose
+on Table→DataGrid swap — they ship in DataGrid v1-bis.
+
+### Heads-up — companion releases coming
+
+- **Sprint 4.2-bis → `0.8.0-beta`** (estimated): DataGrid v1-bis —
+  per-column filter UI chips (text/number/boolean/date), column
+  resize via drag, column reorder via drag, column visibility
+  dialog, right-sticky columns.
+- **Sprint 5 → `0.9.0-beta`** + starter kits v1: separate repos
+  `dashforge-starter-mui` + `dashforge-starter-tw` with Auth + RBAC
+  + form CRUD + dashboard with DataGrid admin views.
+- **Sprint 6 → `1.0.0-rc.1 → 1.0.0`**: final A11Y audit, bundle
+  lockdown, beta freeze, cut `1.0.0`.
+
+## [0.6.0-beta] — 2026-05-19
+
+**Sprint 4.1 release.** Ships the central piece of the data layer
+— a market-grounded `<Table>` built from scratch with NO new
+runtime deps. The design references Stripe (visual style),
+Atlassian / Jira (column UX), and Pencil & Paper UX research
+(density tiers, hover-revealed row actions, selection patterns).
+The existing MUI Table inherited from `hub-ws/admin` informed the
+public API shape but not the implementation (we discarded the
+`@mui/x-data-grid` backing — Table v1 is a hand-rolled native
+`<table>`).
+
+**Companion to Sprint 4.2** (DataGrid, virtualization, 10k+ rows)
+— Table v1 covers the 80% admin-list use case up to ~500 rows.
+
+**Minor bump** for the new public API surface — 15+ new exports.
+Strictly additive — zero breaking changes. Drop-in upgrade from
+`0.5.0-beta`.
+
+### Added
+
+- **`<Table>`** — declarative-first data table component.
+  - **Smart defaults**: column types auto-detected from the first
+    non-null value across visible rows. Number columns get
+    `text-right` + `tabular-nums` (digit grid alignment), date
+    columns get `tabular-nums` too, booleans get `text-center`,
+    strings get `text-left`. **The library never changes the font
+    family** — `tabular-nums` is a font-feature setting that
+    preserves the consumer's theme `font-sans`. Explicit
+    `align` / `tabularNums` / `monospace` per column wins.
+  - **Sort**: per-column `sortable: true` (default comparator
+    handles string / number / Date / boolean / bigint with
+    null-last invariant) or custom `(a, b) => number`. Click
+    cycles asc → desc → none. **Shift-click** adds the column to
+    a multi-sort model. Controlled via `sortModel` /
+    `onSortChange` or uncontrolled with internal state.
+  - **Search**: `enableSearch` renders a debounced input above
+    the table (default 200 ms). Matches any column flagged
+    `searchable: true`. **Nested keys supported** at runtime via
+    `getNestedValue(row, 'address.city')` — the type-level
+    autocomplete from `NestedKeyOf<T>` matches the runtime
+    behavior. Stringification covers string / number / boolean /
+    Date / array / object.
+  - **Selection**: `rowSelection: 'none' | 'single' | 'multiple'`,
+    optional select-all checkbox in the header (multiple mode),
+    `selectedRowIds` controlled state, **sticky bulk-action
+    footer** rendered only when at least one row is selected
+    (Pencil & Paper UX pattern: "appear only when selected").
+  - **Expandable rows**: `expandable={{ render, expandedRowIds?,
+    onExpandChange? }}`. Chevron toggle in a dedicated column;
+    `aria-expanded` updates; expanded detail renders as a
+    full-width sibling `<tr>` below the row.
+  - **Row actions revealed on hover**: `rowActions={(row) => …}`
+    slot — by default hidden via opacity, revealed on
+    `tr:hover` / `tr:focus-within` (Stripe pattern — reduces
+    visual density while keeping actions discoverable).
+  - **RBAC at 3 levels**: table-level `access` (hides / disables
+    the whole table), per-column `cols[i].access` (`hide` removes
+    the column entirely from header AND every row cell), and
+    per-action `actions[i].access` on `<RowActionsMenu>`.
+  - **A11Y**: semantic `<table>` + `<th scope="col">` + `aria-sort`
+    on sortable headers + `aria-selected` on selected rows +
+    `aria-expanded` on expand toggles + `<caption>` (optional,
+    `sr-only` by default) + keyboard nav via Tab + Enter/Space.
+  - **Loading state**: `loading={true}` renders N `<Skeleton>` rows
+    (the Sprint 4 component) with `aria-busy="true"`. Count
+    configurable via `loadingRowCount` (default 5).
+  - **i18n**: column `header` accepts plain strings (pass
+    `t('...')`); all internal default strings configurable via
+    `labels` prop with English defaults (search placeholder, a11y
+    announcements for sort / select / expand, selection counter
+    with `{count}` placeholder, density / filter labels). Same
+    pattern as `<Pagination>`.
+  - **5 variants** (`plain` · `lines` default — Stripe-style ·
+    `striped` · `bordered` · `card`) × **3 sizes** (`sm` · `md` ·
+    `lg`) × **3 densities** (`compact` 40px · `comfortable`
+    default 48px · `spacious` 56px — Pencil & Paper UX research).
+  - **Sticky header** by default (overridable).
+  - **`sx` + 15 slot props** for the standard customization
+    escape hatches.
+
+- **Cell renderer library** — pre-built renderers for common
+  patterns, exported from `@dashforge/tw`:
+  - `RenderText` — one-line, optionally truncated / muted
+  - `RenderTwoLine` — bold primary + muted secondary
+    (name + email pattern)
+  - `RenderChip` — internal status badge, 7 intent colors × 3
+    variants (soft / solid / outline) × 2 sizes (sm / md)
+  - `RenderButton` — inline button wrapper (defaults `ghost` +
+    `sm`)
+  - `RowActionsMenu` — 3-dot Popover-backed menu with per-action
+    RBAC
+
+- **Helpers exported for power use**:
+  - `getNestedValue(row, path)` — dotted-path lookup powering
+    cell rendering. Useful in consumer custom renderers.
+
+- **Theme identity regression guard** —
+  `_internal/themeIdentity.test.ts` scans every Table source file
+  and fails if `dark:*-neutral-N` classes are introduced. The
+  dashforgePreset auto-inverts the neutral palette via CSS var
+  swap; adding `dark:` variants on neutral creates double
+  inversion and breaks dark mode. The Sprint 4.1 fix removes the
+  anti-pattern from Table; **Sprint 4.3 will sweep the rest of
+  the catalog** (Typography, Box, etc. carry the same latent bug).
+
+- **Smoke-test page** in `dash` consumer at `/test-table` —
+  exercises every Table feature with 30 realistic users + nested
+  meta + chip-rendered status + RBAC per-column + i18n labels in
+  Italian + 5 variants × 3 sizes × 3 densities switcher.
+
+- **Doc MDX** `/tw/docs/components/table` — full feature catalog,
+  i18n example with `react-i18next`, override matrix
+  (align / tabularNums / monospace), 15+ slot props reference,
+  cell renderer library, A11Y guarantees, roadmap.
+
+### Internal
+
+- **Theme identity rule** (now codified): the dashforgePreset
+  default IS the Dashforge visual identity. Consumer apps consume
+  it as-is and never override. Adding `dark:` variants on the
+  neutral palette = double inversion = breaks dark mode. Use
+  canonical patterns (LeftNav `itemActive` for selected,
+  auto-invert for `bg-neutral-*` and `text-neutral-*`).
+- **Font family rule**: the library never picks a font family —
+  `tabular-nums` is safe to auto-apply (font-feature setting),
+  `font-mono` is consumer opt-in only via `col.monospace: true`.
+  The dashforgePreset does not own the `fontFamily` axis; the
+  consumer configures their mono stack in their own
+  `tailwind.config.ts` `theme.extend.fontFamily.mono`.
+- **+147 new unit tests** for Table:
+  - `getNestedValue` (11) — nested keys, null-safety, zero / empty
+    string / false preservation
+  - `useTableSearch` (16) — stringification across primitive types
+    + nested keys
+  - `useTableSort` (15) — null-last invariant (direction-independent),
+    multi-column tie-breaking, custom comparator
+  - `useTableSelection` (13) — single / multiple / none modes
+  - `useColumnAutoDetect` (14) — type inference + align /
+    tabularNums / monospace resolution
+  - `themeIdentity` (18 file scanner) — regression guard for
+    `dark:` on neutral palette
+  - `Table.test.tsx` (57) — rendering, smart defaults, sort,
+    search, selection, expandable, row actions, RBAC, variants,
+    densities, sizes, sx + slotProps, i18n
+- Full TW suite at **828/828 passing** (46 files; +147 from
+  Table, +newer guard).
+
+### Compatibility
+
+| Axis | Pre-`0.6.0` | Post-`0.6.0` |
+|---|---|---|
+| Public API surface | 31 components | **+ 1 (`Table`)** + 5 cell renderers + `getNestedValue` helper + `Table*` types (`TableProps`, `TableColumn`, `TableSortModel`, `TableFilterModel`, `TableLabels`, `TableCellContext`, `NestedKeyOf`, `TableRowAction`, …) |
+| Peer deps | `react ^18 \|\| ^19`, `tw-theme workspace`, `tw-tokens workspace` | unchanged |
+| Bridge deps | `forms` / `rbac` / `ui-core` `workspace:*` | unchanged |
+| New runtime deps | — | **none** (no `@tanstack/*`, no DnD libs — constraint honored) |
+| Breaking changes | — | Zero |
+| Bundle size | 336 KB raw / 73.9 KB gzipped | **402 KB raw / 91 KB gzipped** (+66 KB raw / +17 KB gz / **+23% gz**) |
+| Migration | — | Drop-in. Zero code changes required on existing usages. |
+
+> **Bundle regression note**: the +23% gz delta is **above the
+> 10% reviewer-sign-off threshold** documented in
+> `PERFORMANCE.md`. Justification:
+> Table is the lib's central data-display primitive — it includes
+> sort + search + selection + expandable + RBAC at 3 levels + 5
+> cell renderers + RowActionsMenu (using Popover) + 5 variants ×
+> 3 sizes × 3 densities. The bundle weight is proportional to
+> the feature surface. Sign-off rationale: this is the
+> "MUI X DataGrid Community alternative" component — the single
+> highest-value addition before 1.0. Sprint 4.3 (theme identity
+> sweep) will recover -1 to -3 KB gz by removing redundant
+> `dark:` variants across the catalog.
+
+### Migration
+
+No code changes required:
+
+```bash
+pnpm up @dashforge/tw@^0.6.0-beta
+```
+
+To adopt the new Table:
+
+```tsx
+import { Table, RenderTwoLine, RenderChip, RowActionsMenu } from '@dashforge/tw';
+import { useTranslation } from 'react-i18next';
+
+function UsersTable({ users }) {
+  const { t } = useTranslation();
+  const [selected, setSelected] = useState<string[]>([]);
+
+  return (
+    <Table
+      rows={users}
+      cols={[
+        {
+          field: 'name',
+          header: t('users.fields.name'),
+          sortable: true,
+          searchable: true,
+          cellRenderer: ({ row }) => (
+            <RenderTwoLine primary={row.name} secondary={row.email} />
+          ),
+        },
+        {
+          field: 'salary',
+          header: t('users.fields.salary'),
+          sortable: true,
+          // Right-aligned + tabular-nums automatically. Font family
+          // stays whatever the consumer's theme provides — opt into
+          // `monospace: true` if you want font-mono explicitly.
+          access: {
+            resource: 'employee.salary',
+            action: 'read',
+            onUnauthorized: 'hide',
+          },
+        },
+        {
+          field: 'status',
+          header: t('users.fields.status'),
+          cellRenderer: ({ value }) => (
+            <RenderChip
+              color={value === 'active' ? 'success' : 'warning'}
+            >
+              {String(value)}
+            </RenderChip>
+          ),
+        },
+      ]}
+      getRowId={(r) => r.id}
+      enableSearch
+      rowSelection="multiple"
+      selectedRowIds={selected}
+      onSelectionChange={setSelected}
+      bulkActions={(rows) => (
+        <Button color="danger">
+          {t('actions.delete', { count: rows.length })}
+        </Button>
+      )}
+      rowActions={(row) => (
+        <RowActionsMenu
+          row={row}
+          actions={[
+            { label: t('actions.edit'), onClick: edit },
+            { label: t('actions.delete'), onClick: del, color: 'danger' },
+          ]}
+        />
+      )}
+      labels={{
+        searchPlaceholder: t('table.search'),
+        selectedCount: t('table.selected', { defaultValue: '{count} selected' }),
+      }}
+    />
+  );
+}
+```
+
+### Heads-up — companion releases coming
+
+- **Sprint 4.3 → `0.7.0-beta`**: theme identity audit across the
+  whole catalog (Typography, Box, etc. carry the same latent
+  `dark:` anti-pattern Table just fixed). Expected bundle delta:
+  **−1 to −3 KB gz** (removing redundant classes).
+- **Sprint 4.2 → `0.8.0-beta`**: DataGrid with homemade
+  virtualization (`IntersectionObserver`-based, no new deps) for
+  10k+ row data sets, advanced filter model, sticky columns,
+  per-column RBAC.
+
 ## [0.5.0-beta] — 2026-05-19
 
 **Sprint 4 release.** Two TW-only utility primitives — `<Skeleton>`