Summary

The Korean UI mode shows large amounts of untranslated English copy — header counters, search filter buttons, tag empty states, settings labels, and timeline filters all leak through. The existing i18n guard test (tests/test_i18n.py) misses them because it only inspects showToast(...) and showConfirm(...) call sites in 6 specific JS modules, not textContent / innerHTML assignments or HTML literals.

Found by Playwright UX review of v0.1.34 prod (2026-05-02). See docs/reports/mm-web-prod-v0.1.34-playwright-review.md (P1 — Korean UI contains many untranslated English strings).

Evidence — strings bypassing the dictionary

Hardcoded textContent templates:

• packages/memtomem/src/memtomem/web/static/app.js:931-932 — header counter

  qs('stat-chunks').textContent  = `${data.total_chunks} chunks`;
  qs('stat-sources').textContent = `${data.total_sources} sources`;

  Translation keys (header.stat_chunks, header.stat_sources) exist in the locales but are never invoked.

• packages/memtomem/src/memtomem/web/static/app.js:2561 — ${n} chunks · ${m} files template (no plural key in either locale).

• packages/memtomem/src/memtomem/web/static/settings-config.js:187 — Max chunk: ${n} tokens.

Hardcoded innerHTML empty states:

• packages/memtomem/src/memtomem/web/static/app.js:3793 — emptyState('🏷', 'No tags yet', 'Run Auto-Tag to generate tags') — both strings absent from en.json / ko.json.

HTML literal button text without data-i18n:

• "Advanced", "Delete Selected" buttons in packages/memtomem/src/memtomem/web/static/index.html — no data-i18n attribute.

Evidence — guard scope

packages/memtomem/tests/test_i18n.py scans only:

• showToast(\...`)` template literals
• showToast('English ...', ...) literals
• showConfirm({ title: 'English...' }) titles

…in 6 explicit files (app.js, settings-*.js, context-gateway.js). Direct DOM updates (textContent =, innerHTML =, attribute assignment) and HTML literals are unscanned.

Suggested fix

1. Extend test_i18n.py to AST-scan / regex-scan for:
   • el.textContent = "<English string>" or `${expr} <English suffix>` patterns.
   • el.innerHTML = "..." containing English sentences.
   • HTML files with element text not wrapped in data-i18n.
2. Cover the static HTML templates as well as JS — the audit shows HTML-side hardcoding is just as common.
3. Patch the offenders listed above. The header counters likely need plural-form keys ({n, plural, ...} or two keys for one/other) since "1 chunks" reads poorly.

This is the canonical-fixture approach (per repo's feedback_docs_parity_canonical_fixture.md): the guard should fail loudly with the file:line so contributors don't have to grep the whole tree.

References

• Review: docs/reports/mm-web-prod-v0.1.34-playwright-review.md
• Tracking umbrella: TBD
• Earlier i18n PR (feat(web): complete i18n coverage for showConfirm/showToast strings #281): introduced the original guard — this issue extends that scope.