Audit fixes: narrow FIX-35 suffixes, raise FIX-37 threshold, remove duplicate sort

- FIX-35: Narrowed suffix set from {ن,ت,ا,ي,ة,و,ه} to {ن,ت} only.
Only verb conjugation suffixes are blocked. ة/ا/ي/و/ه are legitimate
typo corrections that should not be blocked.
- FIX-37: Raised word threshold from 2 to 4. Prevents noisy period
suggestions while user is still typing short phrases.
- FIX-32b: Removed duplicate .sort() in applyAllSuggestions — already
sorted at line 679, second sort at 695 was redundant.
- Tests updated to match narrowed suffix expectations.

src/app.py

@@ -1062,11 +1062,13 @@ def _is_small_spelling_change(orig_word, corr_word, vocab_manager=None):
                 _orig_oov = not vocab_manager.is_iv(orig_word)
                 _corr_iv = vocab_manager.is_iv(corr_word)
                 if _orig_oov and _corr_iv:
-                    # FIX-35: Don't strip valid conjugation suffixes.
-                    # If the "extra char" is at the END and is a common
-                    # Arabic verb/noun suffix, this is likely a grammar
-                    # conjugation, not a typo. Skip it.
-                    _CONJUGATION_SUFFIXES = {'ن', 'ت', 'ا', 'ي', 'ة', 'و', 'ه'}
+                    # FIX-35: Don't strip verb conjugation suffixes.
+                    # Only block ن (feminine plural: ذهبن→ذهب) and
+                    # ت (feminine past: كتبت→كتب) — these are the
+                    # suffixes grammar commonly adds that spelling
+                    # would try to strip. Other endings (ة,ا,ي,و,ه)
+                    # are more likely genuine typos than grammar fixes.
+                    _CONJUGATION_SUFFIXES = {'ن', 'ت'}
                     _removed_char = None
                     for _di2 in range(len(orig_word)):
                         if orig_word[:_di2] + orig_word[_di2 + 1:] == corr_word:
@@ -2314,12 +2316,14 @@ def analyze_text():
             # The punctuation model often fails to add a period at the end
             # of longer sentences. If no terminal punctuation exists after
             # model processing, inject a period suggestion for the last word.
+            # Threshold=4 words to avoid noisy suggestions while user is
+            # still typing short phrases.
             import re as _re_punc
             _TERMINAL_PUNCT = set('.،؛؟!?!')
             _current_stripped = ctx.current_text.rstrip()
             _has_terminal = _current_stripped and _current_stripped[-1] in _TERMINAL_PUNCT
             _word_count_fb = len(_re_punc.findall(r'[\u0600-\u06FFa-zA-Z]+', ctx.current_text))
-            if not _has_terminal and _word_count_fb >= 2:
+            if not _has_terminal and _word_count_fb >= 4:
                 # Find the last word's position in current_text
                 _last_word_match = _re_punc.search(r'([\u0600-\u06FF]+)\s*$', _current_stripped)
                 if _last_word_match:

src/js/editor.js

@@ -689,10 +689,6 @@ function applyAllSuggestions() {
     // Reverse-order text replacement is safe — each patch's offsets are
     // valid because we haven't modified anything before them yet.
     let text = getEditorText();
-    // FIX-32b: MUST sort by descending offset — replacing from end-to-start
-    // ensures earlier offsets remain valid. Without this, applying a fix at
-    // offset 5 shifts everything after it, corrupting offset 20's fix.
-    suggestions.sort((a, b) => b.start - a.start);
     suggestions.forEach((s) => {
       if (s.start >= 0 && s.end <= text.length && s.start <= s.end) {
         text = text.substring(0, s.start) + s.correction + text.substring(s.end);

tests/test_recent_fixes.py

@@ -86,7 +86,7 @@ for input_text, expected in plurals:
 # ══════════════════════════════════════════════════════════════
 print("\n═══ FIX-35: Spelling — conjugation suffix protection ═══")
 
-_CONJUGATION_SUFFIXES = {'ن', 'ت', 'ا', 'ي', 'ة', 'و', 'ه'}
+_CONJUGATION_SUFFIXES = {'ن', 'ت'}
 
 def simulate_insertion_fix_check(orig_word, corr_word):
     """Simulate the FIX-35 suffix strip check logic from app.py."""
@@ -105,7 +105,7 @@ def simulate_insertion_fix_check(orig_word, corr_word):
             return "allowed"
     return "not_applicable"
 
-# BLOCKED cases (suffix stripping)
+# BLOCKED cases (verb conjugation suffix stripping)
 test("'ذهبن'→'ذهب' blocked (ن suffix)", 
      simulate_insertion_fix_check("ذهبن", "ذهب") == "blocked",
      f"got {simulate_insertion_fix_check('ذهبن', 'ذهب')}")
@@ -114,15 +114,16 @@ test("'كتبت'→'كتب' blocked (ت suffix)",
      simulate_insertion_fix_check("كتبت", "كتب") == "blocked",
      f"got {simulate_insertion_fix_check('كتبت', 'كتب')}")
 
-test("'درسة'→'درس' blocked (ة suffix)",
-     simulate_insertion_fix_check("درسة", "درس") == "blocked",
+# NOW ALLOWED cases (narrowed set — ة, و no longer blocked)
+test("'درسة'→'درس' allowed (ة not in narrowed set)",
+     simulate_insertion_fix_check("درسة", "درس") == "allowed",
      f"got {simulate_insertion_fix_check('درسة', 'درس')}")
 
-test("'جلسوا'→'جلسو' not applicable (len diff ok but وا→و)",
-     simulate_insertion_fix_check("جلسوا", "جلسو") == "blocked",
+test("'جلسوا'→'جلسو' allowed (ا not in narrowed set)",
+     simulate_insertion_fix_check("جلسوا", "جلسو") == "allowed",
      f"got {simulate_insertion_fix_check('جلسوا', 'جلسو')}")
 
-# ALLOWED cases (mid-word insertion fix)
+# ALLOWED cases (mid-word insertion fix — always allowed)
 test("'الكتتاب'→'الكتاب' allowed (mid-word extra ت)",
      simulate_insertion_fix_check("الكتتاب", "الكتاب") == "allowed",
      f"got {simulate_insertion_fix_check('الكتتاب', 'الكتاب')}")