This is genuinely good work — it's a real Traditional translation, not a Simplified copy with a few swaps (程式碼庫, 相容, 專案, 儲存 are the right Taiwan terms), TestCatalogsComplete and TestCatalogsAgreeOnPlaceholders both pass, and the frontend thoughtfully maps zh-hk/zh-mo to zh-TW too. Thank you for that.

One blocker before it can merge: your own TestNormalize fails locally:

normalize("zh_TW.UTF-8") = "zh", want "zh-TW"
normalize("  ZH_TW  ")   = "zh", want "zh-TW"

normalize() lowercases the input but the zh-TW branch only matches the hyphen form zh-tw (plus zh_hant/zh-hant) — it's missing the underscore form zh_tw, which is exactly what POSIX env locales use (LANG=zh_TW.UTF-8). So those fall through to the generic zh branch.

Simplest fix — normalize separators once at the top so both forms (and zh_hant) are covered:

func normalize(s string) string {
	s = strings.ToLower(strings.TrimSpace(s))
	s = strings.ReplaceAll(s, "_", "-")   // zh_TW.UTF-8 → zh-tw.utf-8
	if s == "" {
		return ""
	}
	if strings.HasPrefix(s, "zh-tw") || strings.HasPrefix(s, "zh-hant") || strings.Contains(s, "chinese traditional") || strings.Contains(s, "繁體") {
		return "zh-TW"
	}
	...

(The frontend is fine as-is — navigator.language is always the hyphen form, never the POSIX underscore one.)

Push that and CI should go green — happy to merge once it does.

Fixed. Added strings.ReplaceAll(s, "_", "-") at the top of normalize() so POSIX underscore forms (zh_TW.UTF-8) are normalized before prefix matching. Also removed the now-redundant zh_hant underscore check since the separator normalization covers it.

func normalize(s string) string {
	s = strings.ToLower(strings.TrimSpace(s))
	s = strings.ReplaceAll(s, "_", "-") // zh_TW.UTF-8 → zh-tw.utf-8
	...

All tests pass:

• TestNormalize — zh_TW.UTF-8, ZH_TW, zh-TW, zh-Hant, Chinese Traditional, 繁體 all resolve to zh-TW
• TestCatalogsComplete — zh-TW catalog validated
• TestCatalogsAgreeOnPlaceholders — placeholder counts match

Thank you for this — the Taiwan terminology is genuinely well done. I rebased it onto current main-v2 as #4199 and merged it from there. main-v2 had added ~560 i18n keys since you opened this (bots, hooks, memory, palette, status bar, music, custom font, /rename, /diff-fold, self-update, …), and the completeness checks on both sides require the full key set, so I backfilled them: OpenCC s2twp from the Simplified catalogues, aligned to your existing terms (權限 / 存取 / 參數 / 介面, kept your 當前) with a residual sweep, and dropped 13 keys en.ts has since renamed away. Your normalize fix is included. Credit kept to you.