logos-web/docs/code-quality-followups.md

Code-Quality Follow-ups

This document tracks items that were identified during the code-quality pass but need design or infrastructure decisions before code can land. Each section spells out the gap, the proposed fix, and the artifacts the implementer needs.

---

i18n: single-locale today, multi-locale infra in place

Status

Configured for English only.

• apps/web/i18n/routing.ts declares locales: ['en'].
• Only apps/web/messages/en.json exists.
• packages/content/src/locales/registry.ts defaults to ['en'] until the app calls setActiveLocales with more.
• apps/web/scripts/strip-default-locale-prefix.sh runs after next build to strip the /en prefix from the static-export output (single-locale optimization).

When adding fr / ko / etc

1. Add the locale code to apps/web/i18n/routing.ts locales array.
2. Create apps/web/messages/<locale>.json mirroring the en.json keys.
3. Boot the registry at startup with the new list (the next-intl provider already does this for the routing.locales array).
4. For each PageCopy-driven route, create <locale>.json files alongside the existing English copy under content/pages/<route>/<locale>.json.
5. Re-run the build; strip-default-locale-prefix.sh continues to strip the default locale (en) prefix only.

Why not yet

No translated content shipped. Locking single-locale wiring keeps the codebase honest until that arrives.

---

knip dead-export audit

Status

Open. 102 export keywords appear under apps/web/components/. Likely some are unused after the previous refactors (sub-component extractions, re-exports from barrel files).

Run path

pnpm dlx knip --workspace apps/web

A typical first-run baseline:

• Unused exports flagged → review one batch per feature area; some are intentional public API for sibling files in the same barrel and should be marked with // @public or moved into a private file.
• Unused dependencies flagged → confirm in package.json before removing.

Why not yet

Running knip casually creates noise; the right pass is once the structural refactors settle, then bake knip --no-progress into CI as a warn-only step.

---

Bundle analyzer

Status

Open. Heavy dependencies (motion, leaflet, react-leaflet-cluster) ship to clients but actual cost is unmeasured.

Wiring (when ready)

pnpm add -D @next/bundle-analyzer --filter web

Then in next.config.mjs:

import withBundleAnalyzer from '@next/bundle-analyzer'
const analyzer = withBundleAnalyzer({ enabled: process.env.ANALYZE === 'true' })
export default withNextIntl(analyzer(nextConfig))

pnpm --filter web build with ANALYZE=true emits HTML reports under .next/analyze/.

Why not yet

devDep install needs explicit user approval; once installed, also worth adding a CI bundle-size budget (e.g. bundlewatch).

---

Typography token migration

Status

Open. Tokens exist in @repo/tokens; raw font-* + text-[Npx] patterns appear ~171 times across the codebase. Migrating without a concurrent Figma spec audit risks visible regressions on typography-heavy frames (press, design-systems, blog).

What "done" looks like

1. Inventory the top-five token equivalences from raw classes:

   Raw pattern	Proposed token
   font-mono text-[10px] font-semibold leading-[1.35] uppercase	text-eyebrow
   font-mono text-[10px] leading-[1.3]	text-mono-s
   font-display text-[36px] leading-none tracking-[-0.03em]	text-h3-serif
   font-sans text-[18px] leading-[1.15] tracking-[-0.01em]	text-subhead-sans
   font-sans text-[14px] leading-[1.2]	text-body-sans

2. Cross-check each pattern against the corresponding Figma frame; if a frame uses a one-off variant, keep the raw class and add an inline comment so the next reader doesn't "fix" it.

3. Replace the raw classes with the token name file by file. Keep one PR per feature area (press, builders-hub, circles) for safer Figma diffing.

4. After migration, add an ESLint rule (no-restricted-syntax) that flags font-mono/font-sans/font-display outside the tokens file.

Why not yet

The work is mechanical only if the existing raw classes already match the token spec. A spot check found at least three near-matches that diverge from the token by 1px or 0.05em — those would silently change visuals when the token replaces them.

---

alt="" audit

Status

Open. 45 instances of alt="" across apps/web.

Required content-schema change

In packages/content/src/schemas/press.ts and circles.ts, the image field currently allows alt: ''. Make alt required + non-empty for content images so articlesToCards can drop its || article.title fallback.

// proposed
image: z.object({
  src: z.string().min(1),
  alt: z.string().min(1, 'image.alt must describe the image; use a decorative container if the image is purely visual'),
}),

Component-side rule

Image purpose	alt value
Article thumbnail / podcast cover	{title} (or schema-supplied caption)
Decorative blur / pattern background	alt="" (intentional, leave a comment)
Hero foreground portrait	descriptive copy from translations

Action items

1. Tighten the schemas above.
2. Re-run getPageCopy integration to surface any data files that violate.
3. Switch articlesToCards to read article.image.alt directly (no fallback).
4. Audit all 45 sites; add a comment next to each intentional alt="".
5. Enable eslint-plugin-jsx-a11y/alt-text at error level.

---

E2E happy-path coverage

Status

Open. No test runner is configured in apps/web yet.

Bootstrap path

pnpm add -D playwright @playwright/test --filter web
pnpm exec playwright install --with-deps

Add to apps/web/package.json:

"scripts": {
  "test:e2e": "playwright test",
  "test:e2e:ui": "playwright test --ui"
}

First flows to cover (in order of incident risk)

1. / — home renders, hero loads, navigation overlay opens/closes.
2. /press — articles list renders ≥ one article from the API; cards link externally.
3. /circles — settings + circles + events all resolve; map renders.
4. /active-circles — Hasura fetch path works; stat cards show non-zero data.
5. /builders-hub/ideas/[slug] and /builders-hub/rfps/[slug] — both happy path and notFound() path (use a known-bad slug).

Why not yet

Playwright bootstrap and CI integration is its own multi-hour task; bundling it into the quality pass would obscure the diff.

---

default → named export cleanup

Status

Mostly safe. Constraint: Next.js requires export default on page.tsx, layout.tsx, loading.tsx, error.tsx, not-found.tsx, template.tsx, route.ts, and middleware.ts.

Conversion targets (safe)

• apps/web/components/locale/locale-switcher-select.tsx — default LocaleSwitcherSelect → named export.
• apps/web/components/site-header/site-header-client.tsx — SiteHeaderClient.
• All components/sections/**/*.tsx files using export default — switch to named export, update barrel re-exports accordingly.

Conversion targets (must stay default)

• apps/web/app/**/page.tsx, layout.tsx, loading.tsx, error.tsx, not-found.tsx.

Why not yet

Each conversion is mechanical but cascades through the per-section barrels. Doing it incrementally (one feature area per PR) avoids a 50-file diff.

---

Vitest infra

Already done

• apps/web/vitest.config.ts and the first test file under apps/web/lib/__tests__/ are in place.

Still required

• Add the test script to apps/web/package.json (vitest run).
• Wire CI to run it (turbo run test).
• Author tests for lib/reward.ts, lib/cn.ts, lib/page-sections.ts.
• Configure coverage thresholds (suggested: 80% for lib/).

---

cn() adoption sweep

Status

Open. lib/cn.ts exists; ~25 components still use template-literal className composition.

Codemod

Many call sites match the pattern:

className={`base-classes ${dynamicClass} ${className ?? ''}`}

A safe rewriter is jscodeshift with a small transform that detects this exact shape and replaces it with cn('base-classes', dynamicClass, className). Hand-written variants will need manual review.

Why not yet

A naive sed across 25 files breaks template literals that interpolate non-className expressions; a real codemod is the right tool but requires setup.

---

resolveLocale + LocaleParams adoption

Status

Open. Helpers in lib/route-params.ts; 13 page.tsx files still inline the isActiveLocale(locale) guard, 36 still inline params: Promise<{ locale }>.

Codemod

The transformation is uniform:

- export default async function FooPage({
-   params,
- }: {
-   params: Promise<{ locale: string }>
- }) {
-   const { locale } = await params
-   if (!isActiveLocale(locale)) {
-     throw new Error(`FooPage received non-active locale "${locale}"`)
-   }
+ export default async function FooPage({ params }: LocaleParams) {
+   const locale = await resolveLocale(params, 'FooPage')

A jscodeshift transform that recognises the function-name string in the error message keeps the page-name argument correct.