Fix translation corruptions: LLM sentinel markers and capitalized Liquid tags (#60527)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: heiskr

src/languages/lib/correct-translation-content.ts

@@ -428,6 +428,16 @@ export function correctTranslatedContentStrings(
   }
 
   // --- Generic fixes (all languages) ---
+
+  // Strip leaked LLM sentinel markers (e.g. `<|endoftext|>`) that
+  // occasionally survive the translation pipeline. Replace the marker
+  // and any surrounding whitespace with a single space so adjacent
+  // words don't concatenate.
+  content = content.replace(/\s*<\|endoftext\|>\s*/g, ' ')
+
+  // Capitalized Liquid keyword: `{% Data ` → `{% data `
+  content = content.replaceAll('{% Data ', '{% data ')
+
   // These run after per-language fixes so that e.g. `{{% данных variables`
   // first becomes `{{% data variables` and then gets caught here.
 

src/languages/tests/correct-translation-content.ts

@@ -591,6 +591,20 @@ describe('correctTranslatedContentStrings', () => {
   // ─── GENERIC FIXES ────────────────────────────────────────────────
 
   describe('Generic fixes (all languages)', () => {
+    test('strips LLM sentinel markers and preserves word boundaries', () => {
+      expect(fix('Hello<|endoftext|>World', 'es')).toBe('Hello World')
+      expect(fix('Hello <|endoftext|> World', 'es')).toBe('Hello World')
+      expect(fix('end of sentence.<|endoftext|>Start', 'es')).toBe('end of sentence. Start')
+    })
+
+    test('fixes capitalized Data Liquid keyword', () => {
+      expect(fix('{% Data variables.product.github %}', 'es')).toBe(
+        '{% data variables.product.github %}',
+      )
+      expect(fix('{% Data reusables.foo %}', 'es')).toBe('{% data reusables.foo %}')
+      expect(fix('{% Data ifversion ghec %}', 'es')).toBe('{% data ifversion ghec %}')
+    })
+
     test('fixes AUTOTITLE corruption patterns', () => {
       expect(fix('["AUTOTITLE](/path)', 'es')).toBe('"[AUTOTITLE](/path)')
       expect(fix('[ AUTOTITLE](/path)', 'es')).toBe('[AUTOTITLE](/path)')