From 1a811d8499f36f376ab4c2b99cac202185d53f6c Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Sun, 27 Sep 2026 21:51:40 -0700
Subject: [PATCH] fix(i18n): the percent sign closes to hanzi, like a unit
 (TASK-180)

---
 .../__tests__/zhNumberUnitSpacing.test.ts     | 27 +++++++++++++++++++
 frontend/src/i18n/locales/zh-CN.json          |  2 +-
 2 files changed, 28 insertions(+), 1 deletion(-)

diff --git a/frontend/src/i18n/__tests__/zhNumberUnitSpacing.test.ts b/frontend/src/i18n/__tests__/zhNumberUnitSpacing.test.ts
index 64114ae88..cebf66624 100644
--- a/frontend/src/i18n/__tests__/zhNumberUnitSpacing.test.ts
+++ b/frontend/src/i18n/__tests__/zhNumberUnitSpacing.test.ts
@@ -187,6 +187,33 @@ describe('zh-CN number and unit spacing', () => {
     expect(zhValues.filter(([, value]) => anyLatin.test(value)).length).toBeGreaterThan(200);
   });
 
+  it('attaches a percent sign to the hanzi after it, the same way as a unit', () => {
+    // ux-lead's render-gate call on this PR (75149), open to lily-shen's overrule: the
+    // `%` belongs to its numeral, so `{{attachRate}}%` closes to hanzi exactly like
+    // `{{days}}天` does. Measured as the class before patching the instance, since a
+    // one-key fix that leaves siblings is how the `256KB` defect happened: this is the
+    // ONLY `%` in the catalog, one value, four events — instance and class coincide.
+    //
+    // Deliberately NOT swept and counted here so they read as a boundary rather than an
+    // omission: `·` separates clauses (19 values / 21 events space it) and `—` likewise
+    // (31 / 31). They are separators, not a numeral's unit, so the `%` reasoning does
+    // not transfer and nothing here asserts either way.
+    const spacedPct = /%[ \u00a0]+[\u4e00-\u9fff]/;
+    const attachedPct = /%[\u4e00-\u9fff]/;
+    expect(zhValues.filter(([, value]) => spacedPct.test(value)).map(([key]) => key)).toEqual([]);
+    // Non-vacuity, both directions and both numeral forms, or the line above would be
+    // satisfied by a pattern that matches nothing at all.
+    expect(spacedPct.test('36% 挂载')).toBe(true);
+    expect(attachedPct.test('36%挂载')).toBe(true);
+    expect(spacedPct.test('{{attachRate}}% 挂载')).toBe(true);
+    expect(attachedPct.test('{{attachRate}}%挂载')).toBe(true);
+    // …and it must NOT fire where a Latin word follows the `%`: that space stays, the
+    // same other half of the rule the Latin test above defends.
+    expect(spacedPct.test('{{rate}}% Growth')).toBe(false);
+    // and the catalog still holds the attached form, so green means "none left".
+    expect(zhValues.filter(([, value]) => attachedPct.test(value)).length).toBeGreaterThan(0);
+  });
+
   it('is not green because the detector matches nothing', () => {
     // Vacuity control. The same detector has to be able to fire — on a hand-written
     // spaced pair and on the attached pairs this catalog does contain — or the test
diff --git a/frontend/src/i18n/locales/zh-CN.json b/frontend/src/i18n/locales/zh-CN.json
index 15de3832b..3ced5a542 100644
--- a/frontend/src/i18n/locales/zh-CN.json
+++ b/frontend/src/i18n/locales/zh-CN.json
@@ -1603,7 +1603,7 @@
     },
     "funnel": {
       "title": "激活漏斗",
-      "summary": "在过去{{days}}天的{{signups}}个注册中：{{attachRate}}% 挂载了智能体 · {{messageRate}}% 发送了消息 · {{d1Rate}}% 在第1天后回访 · {{d7Rate}}% 在第7天后回访。",
+      "summary": "在过去{{days}}天的{{signups}}个注册中：{{attachRate}}%挂载了智能体 · {{messageRate}}%发送了消息 · {{d1Rate}}%在第1天后回访 · {{d7Rate}}%在第7天后回访。",
       "table": {
         "cohort": "群组",
         "signups": "注册数",