fix(i18n): check the script of short labels, not only of sentences

untranslated() is the one criterion that knows a catalog's script, and it
never fired. It returned False below LATIN_FLOOR = 6 words, while the
catalogs are mostly one- to four-word UI labels, so the guard sat above
its own corpus.

Measured across the 62 non-Latin catalogs: the floor condemns 0 entries.
Without it, 536, concentrated in the languages the model cannot write:

  av 99   as 92   ba 87   be 84   ab 84   am 56   bh 13

Every entry in the low-count tail was read before the floor was removed,
because that tail is where a false positive would live. There is none:

  ko 'Applications'            -> a refusal sentence from the model
  zh 'Updates & Package Mgmt'  -> the string plus a version changelog
  fa 'Cloud'                   -> 'Cloud' and a markdown rule
  ab 'Communication'           -> 'Comunicacion'

A word count was the wrong instrument. reason() consults this only after
an exact echo is ruled out, so for a non-Latin target, Latin text that
DIFFERS from the source cannot be a term the language keeps in English.
That distinction is what an earlier attempt at floor=2 lacked, which is
why it condemned 4053 entries including legitimate ones.

Scope, stated because the numbers invite a larger reading: this guards
the 62 non-Latin catalogs only. For the ~122 Latin-script languages a
wrong-language answer still passes, and catching those needs a detector
this repository does not have.

LATIN_FLOOR and its re-export are deleted rather than left at 1; nothing
else consumed them.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

tests/unit/python/utils/i18n/test_damage.py

@@ -1,4 +1,5 @@
-"""Unit tests for the criterion that condemns an answer carrying its source.
+"""Unit tests for the criteria that condemn an answer carrying its source or
+coming back in the wrong script.
 
 The shape is one the gateway produced across twenty languages: the English
 source, a markdown rule, then a rendering. Every other criterion passed it,
@@ -42,5 +43,30 @@ class TestEchoAndContinue(unittest.TestCase):
         self.assertEqual(reason(source, "Mail-Konto des Benutzers", "de"), "")
 
 
+class TestWrongScript(unittest.TestCase):
+    """A language written in another script came back in Latin letters.
+
+    The catalogs are mostly one- to four-word labels, and a six-word floor
+    once stood in front of this criterion, so it never fired on them: a whole
+    run filled Belarusian with Polish and Abkhaz with Spanish while every
+    criterion passed.
+    """
+
+    def test_a_one_word_answer_in_the_wrong_script_is_condemned(self) -> None:
+        self.assertEqual(reason("Applications", "Aplikacje", "be"), "untranslated")
+
+    def test_a_label_answered_in_another_language_is_condemned(self) -> None:
+        self.assertEqual(reason("Communication", "Comunicación", "ab"), "untranslated")
+
+    def test_an_answer_in_the_language_s_own_script_is_kept(self) -> None:
+        self.assertEqual(reason("Applications", "Праграмы", "be"), "")
+
+    def test_a_latin_script_target_is_not_judged_by_script(self) -> None:
+        self.assertEqual(reason("Applications", "Anwendungen", "de"), "")
+
+    def test_a_translation_without_a_word_is_kept(self) -> None:
+        self.assertEqual(reason("Status", "100 %", "be"), "")
+
+
 if __name__ == "__main__":
     unittest.main()

utils/i18n/damage.py

@@ -29,7 +29,6 @@ MAX_SHARING_SOURCES = 2
 ECHO_FLOOR = 30
 STUTTER = re.compile(r"\b(\w+)(?:\s+\1\b){2,}", re.IGNORECASE | re.UNICODE)
 WORDS = re.compile(r"[^\W\d_]{2,}", re.UNICODE)
-LATIN_FLOOR = 6
 LATIN_SHARE = 0.9
 NON_LATIN_SCRIPTS = non_latin_languages(PROJECT_ROOT)
 
@@ -89,6 +88,12 @@ def untranslated(translation: str, language: str) -> bool:
     English. Comparing scripts catches it wherever the target does not use the
     Latin one; for a Latin-script target it would need a language detector.
 
+    There is no length floor. :func:`verdict` reaches this only once an exact
+    echo is ruled out, so Latin text that differs from the source cannot be a
+    term the language keeps in English. The catalogs are mostly one- to
+    four-word labels, which is where the wrong language hid for as long as a
+    six-word floor stood here.
+
     Args:
         translation: what came back for the entry.
         language: ISO 639-1 code of the catalog being filled.
@@ -96,7 +101,7 @@ def untranslated(translation: str, language: str) -> bool:
     if language not in NON_LATIN_SCRIPTS:
         return False
     words = WORDS.findall(prose(translation))
-    if len(words) < LATIN_FLOOR:
+    if not words:
         return False
     latin = sum(1 for word in words if (latin_share(word) or 0) > 0.5)
     return latin / len(words) >= LATIN_SHARE

utils/i18n/placeholders.py

@@ -14,7 +14,6 @@ import html
 from utils.i18n.damage import COLLAPSE_FLOOR as COLLAPSE_FLOOR
 from utils.i18n.damage import COLLAPSE_RATIO as COLLAPSE_RATIO
 from utils.i18n.damage import ECHO_FLOOR as ECHO_FLOOR
-from utils.i18n.damage import LATIN_FLOOR as LATIN_FLOOR
 from utils.i18n.damage import LATIN_SHARE as LATIN_SHARE
 from utils.i18n.damage import MARKUP as MARKUP
 from utils.i18n.damage import MAX_SHARING_SOURCES as MAX_SHARING_SOURCES