Phase 13: Grammar SV/gender whitelist, preposition/nasb regex, pronoun-h guard

- Expanded grammar pattern whitelist for SV suffix additions (ذهب→ذهبوا, يذهب→يذهبون)
- Added gender agreement suffix (جميل→جميلة) to whitelist
- Added preposition case (ون→ين after في/من) and nasb (ون→وا after لن/لم) regex fallbacks
- Smart pronoun-ه guard: uses IV check instead of hardcoded consonant list

src/app.py

@@ -872,13 +872,19 @@ def _is_small_spelling_change(orig_word, corr_word, vocab_manager=None):
             #    E.g., فتأملته (fataamaltahu) → فتأملتة is WRONG.
             if (orig_word.endswith('ه') and corr_word.endswith('ة')
                     and orig_word[:-1] == corr_word[:-1]):
-                # Guard: if word ends in ته, the ه is likely a pronoun suffix
-                if len(orig_word) >= 3 and orig_word[-2] == 'ت':
-                    logger.info(
-                        f"[SPELLING] Blocked ه→ة at pronoun suffix: "
-                        f"'{orig_word}'→'{corr_word}' (ته pattern = pronoun 'him/it')"
-                    )
-                    return 0.0
+                # Guard: if ه is a pronoun suffix, block conversion to ة.
+                # Heuristic: if removing ه gives a valid standalone word,
+                # the ه is likely pronoun 'his/it' (عمله = عمل+ه).
+                # If removing ه gives an invalid word, it's ta-marbuta (مدرسه→مدرسة).
+                if len(orig_word) >= 3 and vocab_manager:
+                    stem_without_h = orig_word[:-1]
+                    if vocab_manager.is_iv(stem_without_h):
+                        logger.info(
+                            f"[SPELLING] Blocked ه→ة (pronoun suffix): "
+                            f"'{orig_word}'→'{corr_word}' "
+                            f"('{stem_without_h}' is valid IV → ه = 'his/it')"
+                        )
+                        return 0.0
                 return 0.9
             # 2. ة→ه at word end (less common but valid)
             if (orig_word.endswith('ة') and corr_word.endswith('ه')
@@ -1671,6 +1677,28 @@ def analyze_text():
                         # Demonstrative: هذان→هاتان, هاتان→هذان
                         elif ({orig_text, corr_text} <= {'هذان', 'هاتان'}):
                             _is_grammar_pattern = True
+                        # ── NEW: SV agreement suffix additions ──
+                        # Past tense masc plural: verb→verb+وا (ذهب→ذهبوا, حضر→حضروا)
+                        elif (corr_text.endswith('وا') and corr_text[:-2] == orig_text
+                                and len(orig_text) >= 3):
+                            _is_grammar_pattern = True
+                        # Past tense fem plural: verb→verb+ن (ذهب→ذهبن, حضر→حضرن)
+                        elif (corr_text.endswith('ن') and corr_text[:-1] == orig_text
+                                and len(orig_text) >= 3):
+                            _is_grammar_pattern = True
+                        # Present tense masc plural: يفعل→يفعلون (adding ون)
+                        elif (corr_text.endswith('ون') and corr_text[:-2] == orig_text
+                                and len(orig_text) >= 3):
+                            _is_grammar_pattern = True
+                        # Gender: adjective→adjective+ة (جميل→جميلة, كبير→كبيرة)
+                        elif (corr_text.endswith('ة') and corr_text[:-1] == orig_text
+                                and len(orig_text) >= 3):
+                            _is_grammar_pattern = True
+                        # Gender with ي: ذكي→ذكية
+                        elif (corr_text.endswith('ية') and corr_text[:-1] == orig_text[:-1] + 'ي'
+                                and orig_text.endswith('ي') and len(orig_text) >= 3):
+                            _is_grammar_pattern = True
+
 
                     if not _is_grammar_pattern:
                         if len(orig_text.split()) == 1 and len(corr_text.split()) == 1:

src/nlp/grammar/grammar_rules.py

@@ -350,8 +350,18 @@ class ArabicGrammarGuard:
         # حروف الجر المتصلة بدون مسافة (بأخوك، لأبوك -> بأخيك، لأبيك)
         text = re.sub(r'\b([وف]?[بل])(أبوك|أباك|أخوك|أخاك|ذو|ذا)\b',
                       lambda m: f"{m.group(1)}{m.group(2).replace('و', 'ي').replace('ا', 'ي')}", text)
+
+        # ── NEW: Preposition + ون→ين case (في المهندسون→المهندسين) ──
+        text = re.sub(r'\b([وف]?(?:في|من|إلى|على|عن|حتى))\s+([أ-ي]{4,})(ون|ان)\b', r'\1 \2ين', text)
+        text = re.sub(r'\b([وف]?[بلكف])ال([أ-ي]{4,})(ون|ان)\b', r'\1ال\2ين', text)
+        text = re.sub(r'\b([وف]?ل)([أ-ي]{4,})(ون|ان)\b', r'\1\2ين', text)
+
+        # ── NEW: Nasb/Jazm ون→وا (لن يذهبون→يذهبوا) ──
+        text = re.sub(r'\b(لن|لم|كي|لكي|حتى|أن)\s+([أ-ي]+)(ون)\b', r'\1 \2وا', text)
+
         return text
 
+
     def process(self, original_text, generated_text):
         """Apply all grammar rules to model output."""
         text = self.preserve_numbers(original_text, generated_text)