@opengroundplan/core 0.1.1

package/catalog/sources/frontend-ui.yaml

@@ -0,0 +1,654 @@
+# Groundplan template source catalog — category: frontend-ui
+#
+# Research date: 2026-07-18
+#
+# VERIFICATION NOTES
+#   Stars, last-commit dates, and archived status were verified against primary
+#   GitHub sources on 2026-07-18. The unauthenticated api.github.com budget
+#   (60 req/hr, shared per-IP) was exhausted by concurrent research agents, so
+#   figures were taken from two equally primary, unmetered endpoints:
+#     - stars + license + archived banner: github.com/{owner}/{repo} (HTML)
+#     - last commit timestamp:             github.com/{owner}/{repo}/commits.atom
+#   License class is a GATE, not a note. Every SPDX id below was additionally
+#   confirmed by reading the actual LICENSE/LICENSE.md blob via
+#   raw.githubusercontent.com/{owner}/{repo}/HEAD/LICENSE. No license is guessed.
+#   The archived detector was validated against a known-archived control
+#   (angular/angular.js -> correctly flagged); all entries below are unarchived.
+#   Archived status is further corroborated logically: an archived repo is
+#   read-only, so a 2026 commit timestamp is itself proof of a live repo.
+#
+# LICENSE CLASSES
+#   vendorable     - MIT / Apache-2.0 / BSD / ISC / Unlicense / CC0 / 0BSD.
+#                    Copy and adapt. Apache-2.0 and BSD require the notice be
+#                    carried; MIT requires the copyright line be retained.
+#   reference-only - MPL-2.0 / LGPL / EPL. Read for ideas, copy no files.
+#   excluded       - GPL / AGPL / SSPL / BUSL / source-available / NO LICENSE.
+#
+# All 15 entries below are `vendorable`: 14 MIT + 1 Apache-2.0 (tremor). All
+# carry attribution obligations — MIT requires the copyright line be retained;
+# Apache-2.0 additionally requires the LICENSE (and NOTICE, if present) be
+# carried. See attributionNote per source.
+#
+# ADVERSARIAL RE-VERIFICATION 2026-07-18 (independent second pass)
+#   api.github.com returned HTTP 403 (rate limit, 0/60 remaining, shared per-IP)
+#   for the whole verification window, so the JSON API was NOT used. Every figure
+#   below was independently re-derived from unmetered primary sources:
+#     - stars / archived / license label: github.com/{owner}/{repo} (HTML)
+#     - last commit:                      github.com/{owner}/{repo}/commits.atom
+#     - SPDX id:                          raw.githubusercontent.com LICENSE blob,
+#                                         classified from the license TEXT
+#   The raw LICENSE blob is the stronger source for the gate: the API's
+#   license.spdx_id is a heuristic match, the blob is the instrument itself.
+#   RESULT: all 15 licenses confirmed correct. All 15 unarchived. All star counts
+#   within 1 of recorded. No URL 404s. The AGPL-3.0 trap (origin-space/originui)
+#   was re-confirmed AGPL and correctly excluded; base-ui-components/base-ui was
+#   re-confirmed HTTP 404 and correctly dropped.
+#   Defects found were NOT license defects — they were stale repo-structure and
+#   toolchain claims in shadcn-ui/ui and ixartz/Next-js-Boilerplate; corrected
+#   in place below and marked `verified: 2026-07-18`.
+
+category: frontend-ui
+researchDate: 2026-07-18
+licenseVerification: raw-license-blob-text-classified
+licenseVerificationNote: >-
+  github.com HTML + commits.atom + raw LICENSE blob. api.github.com was rate
+  limited (HTTP 403) and was not used — do not record it as a source here.
+
+sources:
+  # ---------------------------------------------------------------- simple ---
+  - name: tailwindlabs/headlessui
+    url: https://github.com/tailwindlabs/headlessui
+    stars: 28673
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-04-13"
+    complexity: simple
+    stack: react, vue, typescript, tailwind
+    whatToBorrow: >-
+      The cleanest reference for accessible interaction primitives with zero
+      styling opinions. Borrow the focus-management and keyboard-nav internals
+      under packages/@headlessui-react/src/components/ — Dialog (focus trap +
+      scroll lock + portal), Combobox (typeahead + virtual focus via
+      aria-activedescendant), Menu, and Listbox. The
+      src/hooks/use-outside-click.ts, use-escape.ts, and
+      src/utils/focus-management.ts helpers are self-contained and drop into any
+      design system. Also worth adapting: the `render` prop / `as` polymorphic
+      slot pattern in src/utils/render.ts, which is how a component library
+      stays unstyled without losing type safety.
+    whatToAvoid: >-
+      Deliberately tiny surface — no date picker, no table, no form layer, so it
+      cannot be a whole UI layer on its own. The polymorphic `as` typing produces
+      famously unreadable TypeScript errors when misused. Slowest-moving repo in
+      this list (last commit 2026-04-13, ~3 months); fine, but do not expect
+      rapid support for new React features. Vue and React packages drift in
+      feature parity — verify before promising both.
+    proposedLayerIds:
+      - ui-headless-primitives
+      - a11y-focus-management
+      - styling-tailwind
+
+  - name: recharts/recharts
+    url: https://github.com/recharts/recharts
+    stars: 27383
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-17"
+    complexity: simple
+    stack: react, typescript, d3, svg
+    whatToBorrow: >-
+      The lowest-friction charting layer to scaffold — declarative JSX composition
+      (<LineChart><XAxis/><Tooltip/></LineChart>) means a generated template needs
+      no imperative chart-init code. Borrow the ResponsiveContainer pattern for
+      charts that resize without layout thrash, and the customized Tooltip/Legend
+      content-render props, which are the standard way to make charts match a
+      shadcn/Tailwind theme. Pairs directly with shadcn's chart.tsx wrapper.
+    whatToAvoid: >-
+      SVG-based, so it degrades badly past a few thousand points — use visx or a
+      canvas renderer for dense series. Historically heavy bundle and a long tail
+      of re-render performance issues on frequently-updating data. The
+      composition API leaks d3 concepts inconsistently, and some props are
+      documented only by example. Do not use as the chart layer for realtime
+      dashboards.
+    proposedLayerIds:
+      - charts-recharts
+      - dashboard-analytics-widgets
+
+  # -------------------------------------------------------------- standard ---
+  - name: shadcn-ui/ui
+    url: https://github.com/shadcn-ui/ui
+    stars: 119258
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-17"
+    complexity: standard
+    stack: react, next, typescript, tailwind, radix
+    whatToBorrow: >-
+      The single most important source in this category, and structurally the
+      closest thing to Groundplan's own model: components are copied into the consumer
+      repo, not installed. PATHS VERIFIED 2026-07-18 — the monorepo has moved to
+      apps/v4/; the old apps/www/ tree no longer exists and there is no root
+      registry.json. Borrow the registry architecture wholesale —
+      apps/v4/registry.json plus apps/v4/registry/ (registry:ui, registry:block,
+      registry:hook item types) and the CLI in packages/shadcn/ (src/commands/
+      add.ts, init.ts and src/utils/transformers/ which rewrite import aliases and
+      apply the tsconfig path mapping on install). That transformer chain is
+      exactly the mechanism a scaffolding tool needs. Also take
+      apps/v4/components.json as the config-file schema, the cn()
+      clsx+tailwind-merge helper, the class-variance-authority variant pattern in
+      apps/v4/registry/new-york-v4/ui/button.tsx, and the OKLCH CSS-custom-property
+      theming in apps/v4/app/globals.css that makes dark mode a token swap rather
+      than a class fork. NOTE: theming is OKLCH, not HSL — the Tailwind v4 rewrite
+      changed the color space, and globals.css now contains zero hsl() values.
+      Generating HSL tokens against this theme produces wrong colors.
+    whatToAvoid: >-
+      Not a dependency — you own the copied code forever, including upstream bug
+      fixes, which is a real maintenance tax at scale. Tightly coupled to Tailwind
+      and to Radix primitives; extracting a component without both is substantial
+      work. The New York / Default style split described in most tutorials is
+      OBSOLETE — as of 2026-07-18 the registry ships a single style directory
+      (apps/v4/registry/new-york-v4/) alongside bases/, icons/ and styles/; do not
+      generate a two-style fork. Pin the registry version you vendor from — the
+      registry schema and the directory layout have both changed shape more than
+      once (apps/www/ -> apps/v4/ is the most recent), and blocks are far less
+      stable than the ui primitives.
+    proposedLayerIds:
+      - ui-shadcn
+      - component-registry-cli
+      - theming-css-variables
+      - styling-tailwind
+      - icons-lucide
+
+  - name: radix-ui/primitives
+    url: https://github.com/radix-ui/primitives
+    stars: 19073
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-15"
+    complexity: standard
+    stack: react, typescript
+    whatToBorrow: >-
+      The accessibility substrate under shadcn/ui, and worth understanding
+      directly rather than only through it. Borrow the compound-component
+      contract (Root/Trigger/Portal/Content/Item) which is the de-facto API shape
+      for headless React libraries, plus packages/react/context/ (createContextScope
+      for collision-free nested providers), packages/react/primitive/ (the asChild
+      + Slot merge pattern for prop/ref forwarding), and the collection and
+      dismissable-layer/focus-scope packages that handle layered overlay
+      z-ordering correctly. Its WAI-ARIA implementations are the reference
+      standard for keyboard and screen-reader behavior.
+    whatToAvoid: >-
+      Unstyled by design — no visual layer, so it is a dependency of a UI layer,
+      never the layer itself. Historically ships as many small packages, which
+      bloats package.json and complicates version alignment (the consolidated
+      radix-ui package is the newer answer — prefer it). Portal-rendered content
+      routinely breaks naive E2E selectors and CSS that assumes DOM nesting.
+      Note the LICENSE copyright is WorkOS, not Radix — carry it verbatim.
+    proposedLayerIds:
+      - ui-radix-primitives
+      - a11y-wai-aria
+      - ui-headless-primitives
+
+  - name: react-hook-form/react-hook-form
+    url: https://github.com/react-hook-form/react-hook-form
+    stars: 44800
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-18"
+    complexity: standard
+    stack: react, typescript
+    whatToBorrow: >-
+      The default form layer for a generated React template. Borrow the
+      uncontrolled/ref-registration model (re-renders only the fields that
+      change, which is why it outperforms Formik on large forms), the
+      @hookform/resolvers zod bridge that lets one zod schema drive both client
+      validation and server parsing, and the useFieldArray pattern for
+      repeatable/dynamic sections. The FormProvider + useFormContext pattern is
+      what shadcn's components/ui/form.tsx wraps — vendor the two together so
+      generated forms get labels, descriptions, and aria-invalid error wiring for
+      free.
+    whatToAvoid: >-
+      The uncontrolled model is a genuine mental-model shift; mixing it with
+      controlled third-party inputs requires the Controller wrapper and is the
+      single largest source of user confusion. Deep generic types produce
+      inscrutable errors on nested/conditional schemas. Resolver packages version
+      independently from core — a zod major bump routinely needs a matching
+      @hookform/resolvers bump, so pin both together in any template.
+    proposedLayerIds:
+      - forms-react-hook-form
+      - validation-zod
+      - forms-server-actions
+
+  - name: TanStack/table
+    url: https://github.com/TanStack/table
+    stars: 28198
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-17"
+    complexity: standard
+    stack: react, vue, svelte, solid, typescript
+    whatToBorrow: >-
+      Headless table logic, which is the right split for a scaffolder: the state
+      machine is generic and the markup stays yours. Borrow the columnDef /
+      createColumnHelper typing pattern, and the pluggable row-model composition
+      (getCoreRowModel, getSortedRowModel, getFilteredRowModel,
+      getPaginationRowModel) — importing only the models you need is what keeps
+      the bundle small. The manual* flags (manualPagination, manualSorting,
+      manualFiltering) are the exact seam for swapping a client-side demo table
+      to a server-driven one, so generate both variants from one config. Column
+      visibility, ordering, and row-selection state are all serializable, which
+      makes URL-persisted table state straightforward.
+    whatToAvoid: >-
+      Headless means you write every <tr> and <td> yourself — significant
+      boilerplate, so ship it as a pre-built data-table block rather than raw.
+      Virtualization is a separate package (TanStack Virtual); combining it with
+      sticky headers and resizing is where most implementations break. The v7 to
+      v8 rewrite invalidated nearly all older tutorials, so LLM-recalled and
+      blog-sourced snippets are frequently for the wrong major version — verify
+      against current docs.
+    proposedLayerIds:
+      - table-tanstack
+      - data-table-server-side
+      - table-virtualization
+
+  - name: satnaing/shadcn-admin
+    url: https://github.com/satnaing/shadcn-admin
+    stars: 12662
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-06-11"
+    complexity: standard
+    stack: react, vite, typescript, tailwind, shadcn, tanstack-router
+    whatToBorrow: >-
+      The best-proportioned shadcn admin reference — genuinely complete without
+      being a framework. Borrow the whole application shell: src/components/layout/
+      (collapsible sidebar with persisted state, breadcrumb header, nav-user
+      menu), the src/features/ vertical-slice folder convention (tasks, users,
+      apps, chats each owning their components/data/context), and the
+      src/components/data-table/ implementation which is a working TanStack Table
+      + shadcn pairing with faceted filters, column toggling, and a pagination
+      bar. Also take the light/dark/system theme provider, the RTL support, and
+      the 401/403/404/500 error-page set — all of which templates usually omit
+      and users always need.
+    whatToAvoid: >-
+      Vite + TanStack Router, not Next.js — porting the routing and layout tree to
+      an App Router template is real work, so keep it as the SPA-flavored variant
+      rather than forcing a conversion. Auth screens are UI-only mockups with no
+      backend wiring; do not present them as a working auth layer. Data is local
+      mock fixtures, so pagination and filtering are all client-side and must be
+      re-pointed at a real API. Single-maintainer project — lower bus factor than
+      the library entries here.
+    proposedLayerIds:
+      - dashboard-admin-shell
+      - layout-sidebar-nav
+      - ui-shadcn
+      - table-tanstack
+      - theming-dark-mode
+      - app-vite-spa
+
+  - name: Kiranism/next-shadcn-dashboard-starter
+    url: https://github.com/Kiranism/next-shadcn-dashboard-starter
+    stars: 6689
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-18"
+    complexity: standard
+    stack: next, react, typescript, tailwind, shadcn, clerk, zustand
+    whatToBorrow: >-
+      The Next.js App Router counterpart to shadcn-admin, and the most actively
+      maintained template in this list (committed the day of research). Borrow the
+      app/ route-group layout split (auth pages vs dashboard shell), the
+      parallel-routes + Suspense streaming setup on the overview page, which is a
+      clean demonstration of per-widget loading states, and the nuqs-based
+      URL-as-state search-param pattern for table filters — that is the piece most
+      admin templates get wrong. The Clerk integration and the Sentry + auth
+      middleware wiring are both isolated enough to lift or delete cleanly.
+    whatToAvoid: >-
+      Opinionated toward Clerk for auth; swapping to Auth.js or Better Auth means
+      touching middleware, layout, and several server components, so make auth an
+      explicit swappable layer rather than inheriting the default. Kitchen-sink
+      by nature (feature flags, Sentry, Zustand, nuqs, KBar) — strip aggressively
+      or generated projects inherit dependencies they never use. Follows shadcn
+      canary/registry changes quickly, which is good for freshness but means
+      pinning a commit is safer than tracking main.
+    proposedLayerIds:
+      - app-next-app-router
+      - dashboard-admin-shell
+      - auth-clerk
+      - ui-shadcn
+      - state-zustand
+      - url-state-nuqs
+      - observability-sentry
+
+  # ---------------------------------------------------------- professional ---
+  - name: mui/base-ui
+    url: https://github.com/mui/base-ui
+    stars: 10373
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-17"
+    complexity: professional
+    stack: react, typescript
+    whatToBorrow: >-
+      The current-generation headless primitive library, built by the MUI team
+      together with people who worked on Radix and Floating UI — the most credible
+      forward-looking alternative to Radix. Borrow its render-prop composition API
+      (a cleaner successor to asChild), its Floating UI-based positioning for
+      popovers/tooltips/menus, and its explicit CSS-first styling hooks — state is
+      exposed as data-* attributes rather than requiring a styling engine, which
+      makes it unusually easy to theme from plain CSS or Tailwind.
+    whatToAvoid: >-
+      The youngest library here and the highest churn risk — API shape has moved
+      between releases, so pin exact versions and expect migration work. Smallest
+      component inventory of the primitive libraries; verify the specific
+      components a template needs actually exist before committing. Ecosystem is
+      thin compared to Radix — far fewer worked examples, and most shadcn blocks
+      assume Radix, so a base-ui variant means rewriting rather than swapping.
+    proposedLayerIds:
+      - ui-base-ui
+      - ui-headless-primitives
+      - a11y-wai-aria
+
+  - name: airbnb/visx
+    url: https://github.com/airbnb/visx
+    stars: 20961
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-06-22"
+    complexity: professional
+    stack: react, typescript, d3, svg
+    whatToBorrow: >-
+      The right layer when Recharts' abstraction is too rigid — unopinionated
+      React wrappers over d3 with no chart components imposed. Borrow the
+      granular package split (@visx/scale, @visx/axis, @visx/shape, @visx/tooltip,
+      @visx/gradient), which lets a generated bundle include only the primitives
+      used, and the @visx/responsive + useTooltip/useParentSize hook patterns.
+      Best source for bespoke or brand-specific visualizations where the chart is
+      a designed artifact rather than a stock widget.
+    whatToAvoid: >-
+      Not a charting library — you assemble scales, axes, and marks yourself, so
+      a simple bar chart costs meaningfully more code than in Recharts. Requires
+      real d3 fluency (scales, domains, ranges), which raises the floor for
+      whoever inherits the template. Development pace is measured and Airbnb-driven
+      rather than community-driven. Still SVG, so it shares Recharts' density
+      ceiling — visx does not solve large-dataset rendering for you.
+    proposedLayerIds:
+      - charts-visx
+      - charts-custom-d3
+      - dashboard-analytics-widgets
+
+  - name: tremorlabs/tremor
+    url: https://github.com/tremorlabs/tremor
+    stars: 3525
+    spdx: Apache-2.0
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2025-10-10"
+    complexity: professional
+    stack: react, typescript, tailwind, radix, recharts
+    whatToBorrow: >-
+      Dashboard-specific composites that the general component libraries do not
+      cover: KPI/stat cards, delta badges, spark charts, category bars, progress
+      circles, and date-range pickers, all Tailwind-native and copy-pasteable in
+      the same spirit as shadcn. Borrow the chart-theming approach — a shared
+      color-palette map applied consistently across chart types — which is the
+      thing most hand-rolled dashboards get inconsistent.
+    whatToAvoid: >-
+      STALENESS RISK — last commit 2025-10-10, roughly 9 months before this
+      research. Under the 18-month stale threshold, so recorded as maintained,
+      but it is by a wide margin the slowest-moving entry here and the project has
+      shifted toward a paid Tremor Blocks offering, so treat the OSS component set
+      as feature-frozen and verify before adopting. Apache-2.0, not MIT: the only
+      such entry — copying files obliges you to retain the license and NOTICE
+      attribution, which the MIT entries do not require in the same form. Also
+      layers on Recharts, so it inherits Recharts' dense-data ceiling.
+    attributionNote: >-
+      Apache-2.0 — retain the LICENSE file and state changes on modified files
+      when vendoring. Verified 2026-07-18: the repo ships NO NOTICE file (NOTICE,
+      NOTICE.txt, NOTICE.md all 404), so the Apache-2.0 section 4(d) NOTICE
+      propagation duty does not attach — carrying LICENSE is sufficient. Note the
+      root package.json omits a "license" field; the LICENSE blob (Apache License
+      2.0, verbatim) is the authoritative source and is what the SPDX id records.
+    proposedLayerIds:
+      - dashboard-analytics-widgets
+      - charts-tremor
+      - ui-kpi-cards
+      - styling-tailwind
+
+  - name: storybookjs/storybook
+    url: https://github.com/storybookjs/storybook
+    stars: 90616
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-17"
+    complexity: professional
+    stack: react, vue, svelte, angular, vite, typescript
+    whatToBorrow: >-
+      The workbench layer that turns a pile of components into a maintained design
+      system — this is a config pattern to generate, not code to copy. Borrow the
+      .storybook/main.ts + preview.ts setup (framework preset, autodocs, a global
+      theme decorator wired to the same CSS variables the app uses so light/dark
+      is togglable in the toolbar), the CSF3 story format with Meta/StoryObj
+      typing, and the play() function pattern which makes stories double as
+      interaction tests. The a11y addon plus test-runner turns the component
+      catalog into an automated accessibility gate — high value for any template
+      claiming a11y.
+    whatToAvoid: >-
+      Heavy: significant install size, a second build pipeline to keep in sync
+      with the app's bundler, and a recurring source of CI slowness. Major-version
+      upgrades are historically disruptive and addon compatibility lags each
+      release — do not scaffold third-party addons casually. Real risk of
+      abandonment inside generated projects: an unmaintained Storybook with stale
+      stories is worse than none, so only include it in templates where a
+      component catalog is genuinely the point.
+    proposedLayerIds:
+      - storybook-workbench
+      - design-system-docs
+      - a11y-testing
+      - visual-regression-testing
+
+  - name: ixartz/Next-js-Boilerplate
+    url: https://github.com/ixartz/Next-js-Boilerplate
+    stars: 13023
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-08"
+    complexity: professional
+    stack: next, react, typescript, tailwind, drizzle, playwright, vitest
+    whatToBorrow: >-
+      The strongest reference for the non-UI scaffolding that surrounds a UI
+      template — the tooling spine most component-focused repos skip. Borrow the
+      full quality gate: oxlint config (oxlint.config.ts), Vitest unit setup,
+      Playwright E2E config, Lefthook pre-commit hooks, commitlint/semantic-release,
+      and the
+      typed environment-variable validation (T3-style env schema) that keeps
+      config out of business logic. The next-intl i18n routing setup and the
+      Drizzle + database layer are cleanly separated and make good optional
+      layers. Its Sentry/monitoring and bundle-analyzer wiring are also isolated
+      enough to include or omit independently.
+    whatToAvoid: >-
+      Genuinely kitchen-sink — a new project inherits a large dependency surface
+      and several vendor SDKs it may never use, so treat nearly every concern here
+      as an opt-in layer rather than a default. Opinionated vendor choices (Clerk
+      auth, Sentry) mean lock-in if copied wholesale — note PostHog and Crowdin
+      are named in older descriptions of this repo but are NOT in current
+      dependencies (verified 2026-07-18). Moves fast
+      and tracks Next.js canary features closely, so pin a commit. Its UI layer is
+      deliberately minimal — take the tooling from here and the components from
+      shadcn, not the reverse. LINTER CHANGED: this repo has fully migrated off
+      ESLint to oxlint (oxlint + oxlint-tsgolint in devDependencies; no eslint
+      dependency and no eslint.config.* file remain as of 2026-07-18). Any
+      generated template copying an "ESLint flat config" from here will not match
+      the source.
+    proposedLayerIds:
+      - app-next-app-router
+      - test-vitest
+      - e2e-playwright
+      - lint-oxlint
+      - config-typed-env
+      - i18n-next-intl
+      - db-postgres-drizzle
+      - ci-github-actions
+
+  # ------------------------------------------------------------ enterprise ---
+  - name: refinedev/refine
+    url: https://github.com/refinedev/refine
+    stars: 35333
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-06-05"
+    complexity: enterprise
+    stack: react, typescript, next, remix, vite, antd, mui, chakra, mantine
+    whatToBorrow: >-
+      The best MIT-licensed answer for large internal-tool and admin estates,
+      because it is headless at the architecture level rather than the component
+      level. Borrow the dataProvider contract (getList/getOne/create/update/
+      deleteOne/getMany with filters, sorters, pagination) — a single well-designed
+      interface that decouples every CRUD screen from the backend and is the most
+      reusable idea in this repo. Also take the authProvider and accessControlProvider
+      abstractions (RBAC/ACL via Casbin or Permify hook in cleanly), the
+      auditLogProvider for regulated contexts, and the useTable/useForm/useSelect
+      hooks that bind list and edit screens to the provider layer. Its UI-kit
+      agnosticism is the model to copy: the same screens render against Ant
+      Design, MUI, Chakra, or Mantine.
+    whatToAvoid: >-
+      Substantial framework buy-in — the provider abstractions are pervasive and
+      migrating away later is expensive, so only reach for it when CRUD volume
+      actually justifies it. Over-engineered for small apps: for a handful of
+      screens the indirection costs more than it saves. Large surface area and
+      docs that lag the code in places. Open-core: the OSS project is MIT, but
+      enterprise-oriented add-ons and support sit behind a commercial offering —
+      confirm any specific package is in the MIT tree before vendoring.
+    proposedLayerIds:
+      - crud-admin-refine
+      - data-provider-rest
+      - rbac-access-control
+      - audit-log
+      - auth-provider-abstraction
+      - dashboard-admin-shell
+
+  - name: ant-design/ant-design-pro
+    url: https://github.com/ant-design/ant-design-pro
+    stars: 38561
+    spdx: MIT
+    licenseClass: vendorable
+    maintained: true
+    lastCommit: "2026-07-15"
+    complexity: enterprise
+    stack: react, typescript, umi, antd, less
+    whatToBorrow: >-
+      The reference for large-org admin conventions, battle-tested inside Alipay
+      and widely deployed across enterprise China. Borrow the ProLayout shell
+      (config-driven menu tree, multi-tab navigation, breadcrumb generation from
+      route config) and especially the ProComponents patterns — ProTable and
+      ProForm generate a full CRUD screen from a column/field schema, which is the
+      highest-leverage idea here for a scaffolder producing many similar pages.
+      Its config/ route-level access and permission model, i18n across many
+      locales, and the mock-service conventions are all mature and worth adapting
+      even outside the Ant ecosystem.
+    whatToAvoid: >-
+      Heavy end-to-end lock-in: Umi as the framework, Ant Design as the component
+      layer, Less for styling — this is an all-or-nothing stack, and extracting
+      pieces into a Tailwind/Next template is largely a rewrite. Very large bundle
+      by default and real work to trim. A meaningful share of docs, issues, and
+      code comments are Chinese-first, which is a practical barrier for some
+      teams. The strong visual identity is unmistakably Ant Design and resists
+      rebranding. Choose this for enterprise back-office, never for a public
+      product surface.
+    proposedLayerIds:
+      - dashboard-admin-shell
+      - crud-schema-driven
+      - ui-ant-design
+      - rbac-access-control
+      - i18n-multi-locale
+      - layout-sidebar-nav
+
+droppedCandidates:
+  - name: origin-space/originui
+    reason: >-
+      EXCLUDED BY LICENSE. Verified AGPL-3.0 (GitHub reports "AGPL-3.0 license";
+      raw LICENSE blob is the GNU Affero General Public License v3 verbatim).
+      10,273 stars and actively developed (last commit 2026-07-15), and it is
+      widely mistaken for a permissive shadcn-style copy-paste collection — which
+      makes it precisely the trap to document. Copying components would impose
+      AGPL network-copyleft terms on every repo Groundplan generates. Do not vendor;
+      do not read-and-reimplement from it either.
+  - name: base-ui-components/base-ui
+    reason: >-
+      NO SUCH REPOSITORY — returns HTTP 404. The real project is mui/base-ui,
+      which is included above. Recorded here so the dead path is not retried.
+  - name: mantinedev/mantine
+    reason: >-
+      Verified good (31,451 stars, MIT, last commit 2026-07-13) but trimmed for
+      overlap — a full styled component library whose role is already covered by
+      the Ant Design and shadcn entries. Strong fallback if a non-Tailwind,
+      non-Ant option is needed; it has the best built-in hooks library of the group.
+  - name: heroui-inc/heroui
+    reason: >-
+      Verified good (29,946 stars, Apache-2.0 confirmed from raw LICENSE, last
+      commit 2026-07-18) but trimmed for overlap with the styled-library slot.
+      Note for the record: formerly NextUI, and now Apache-2.0 rather than MIT —
+      attribution obligations differ from what most people assume.
+  - name: mui/material-ui
+    reason: >-
+      Verified good (98,614 stars, MIT, last commit 2026-07-17) but trimmed —
+      enterprise slot is better served by refine and ant-design-pro, which supply
+      application architecture rather than only components. Also note MUI X
+      (data grid, date pickers) is dual-licensed with commercial tiers, so the
+      MIT status of the core repo does not extend to every MUI package.
+  - name: ant-design/ant-design
+    reason: >-
+      Verified MIT from the raw LICENSE blob (98,763 stars, last commit
+      2026-07-17) after the HTML license parse initially returned nothing.
+      Trimmed because ant-design-pro already carries the Ant patterns in
+      application form.
+  - name: marmelab/react-admin
+    reason: >-
+      Verified good (26,838 stars, MIT, last commit 2026-07-16) but trimmed for
+      direct overlap with refine in the enterprise CRUD slot. Note it is open-core
+      — react-admin core is MIT while several enterprise modules are commercial.
+  - name: colinhacks/zod
+    reason: >-
+      Verified good (43,286 stars, MIT, last commit 2026-06-10). Trimmed from
+      this category as a validation library rather than a UI one — it belongs to
+      a validation/backend category, and is referenced from the react-hook-form
+      entry via the @hookform/resolvers pairing.
+  - name: tailwindlabs/tailwindcss
+    reason: >-
+      Verified good (95,955 stars, MIT, last commit 2026-07-16). Trimmed as
+      infrastructure rather than template source material — it is a dependency of
+      several entries above, captured as the styling-tailwind layer id.
+  - name: plouc/nivo
+    reason: >-
+      Verified MIT (14,066 stars, last commit 2026-07-11); LICENSE.md is verbatim
+      MIT text titled only with the copyright line, which is why an automated
+      title match initially failed. Trimmed — charts slot already covered by
+      recharts and visx.
+  - name: themesberg/flowbite-react
+    reason: >-
+      Verified MIT and active (2,146 stars, last commit 2026-06-27) but the
+      weakest adoption signal of the set; dropped in favor of higher-confidence
+      Tailwind component sources.
+  - name: TanStack/form
+    reason: >-
+      Verified MIT and active (6,608 stars, last commit 2026-07-13). Trimmed
+      because react-hook-form remains the higher-adoption default for the forms
+      layer; revisit if the TanStack-native stack becomes a template variant.
+  - name: magicuidesign/magicui
+    reason: >-
+      Verified MIT (21,593 stars, last commit 2026-07-02). Trimmed as
+      marketing/landing-page animation components rather than application UI, and
+      it operates a paid pro tier alongside the OSS set — scope the MIT boundary
+      carefully if revisited.