Retrofit: tighten yod-bet pattern, add cover-block fallback

The "על כן" pattern for block-yod-bet was too greedy and matched mid-discussion
transitional sentences (e.g. "על כן, במקום בו..."), which caused forward-scan
to skip block-yod-alef ("סוף דבר") via the pointer advance.

Tightened to require an operative subject (אנו / הערר / הוועדה / ועדת הערר)
so terminal "על כן, אנו מחליטים" still matches but mid-block transitions don't.

Added structural_fallback for cover blocks (alef/bet/gimel/dalet) — these are
template metadata not present in user-edited DOCX bodies. Inject zero-content
anchors so apply_user_edit can still target them later. The frontend toast
distinguishes real content gaps from fallback anchors.

Also expanded heading patterns based on training corpus inspection:
- block-vav: על המקרקעין חלות / במצב התכנוני / התכניות החלות
- block-zayin: טענות העוררת
- block-chet: עיקר תגובת המשיב
- block-tet: הדיון בוועדת הערר

For case 1130-25, this raises detection from 6/12 to 11/12 blocks — only
block-yod-bet remains missing (Daphna's edit ends at "סוף דבר" + numbered
ruling, no terminal "ההחלטה" or "על כן אנו מחליטים" paragraph).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

mcp-server/src/legal_mcp/services/docx_retrofit.py

@@ -85,13 +85,40 @@ _BLOCK_HEADING_PATTERNS: list[tuple[str, list[str]]] = [
     ("block-gimel", [r"^נגד\s*$", r"^—\s*נגד\s*—"]),
     ("block-dalet", [r"^החלטה\s*$"]),
     ("block-heh",  [r"^רקע\s*$", r"^רקע\s+עובדתי", r"^פתח\s+דבר"]),
-    ("block-vav",  [r"^תכניות\s+חלות", r"^ההליכים?\s+שבפנינו", r"^ההליכים?\s+בפני\s+הוועדה\s+המקומית"]),
-    ("block-zayin", [r"^תמצית\s+טענות", r"^טענות\s+הצדדים", r"^טענות\s+העוררי"]),
-    ("block-chet", [r"^תגובת\s+המשיב", r"^עמדת\s+הוועדה\s+המקומית", r"^תשובת"]),
-    ("block-tet",  [r"^ההליכים?\s+בפני\s+ועדת\s+הערר", r"^הדיון\s+בפנינו"]),
+    ("block-vav",  [
+        r"^תכניות\s+חלות",
+        r"^ההליכים?\s+שבפנינו",
+        r"^ההליכים?\s+בפני\s+הוועדה\s+המקומית",
+        r"^על\s+המקרקעין\s+חלות",
+        r"^התכניות?\s+החלות",
+        r"^במצב\s+התכנוני",
+    ]),
+    ("block-zayin", [
+        r"^תמצית\s+טענות",
+        r"^טענות\s+הצדדים",
+        r"^טענות\s+העוררי",
+        r"^טענות\s+העוררת",
+    ]),
+    ("block-chet", [
+        r"^תגובת\s+המשיב",
+        r"^עמדת\s+הוועדה\s+המקומית",
+        r"^תשובת",
+        r"^עיקר\s+תגובת\s+המשיב",
+    ]),
+    ("block-tet",  [
+        r"^ההליכים?\s+בפני\s+ועדת\s+הערר",
+        r"^הדיון\s+בפנינו",
+        r"^הדיון\s+בוועדת\s+הערר",
+    ]),
     ("block-yod",  [r"^דיון\s+והכרעה", r"^דיון\s*$", r"^ההכרעה"]),
     ("block-yod-alef", [r"^סוף\s+דבר", r"^סיכום\s*$"]),
-    ("block-yod-bet", [r"^ההחלטה\s*$", r"^על\s+כן[,\.]?"]),
+    # block-yod-bet "על כן" must be operative — paired with אנו/הערר/הוועדה.
+    # Loose `^על כן` alone matches mid-discussion transitions ("על כן, במקום בו...")
+    # and steals the bookmark from block-yod-alef via forward-scan.
+    ("block-yod-bet", [
+        r"^ההחלטה\s*$",
+        r"^על\s+כן[,\.\s]+(?:אנו|הערר|הוועדה|ועדת\s+הערר)\b",
+    ]),
 ]
 
 _COMPILED_HEADING_PATTERNS: list[tuple[str, list[re.Pattern[str]]]] = [
@@ -252,6 +279,20 @@ def retrofit_bookmarks(
 
     block_starts = _detect_block_starts(paragraphs)
 
+    # Cover-block fallback: alef/bet/gimel/dalet are template metadata
+    # (judges, case number, parties, "החלטה" title) that don't appear in
+    # the body of user-edited DOCX files — they live in headers/template.
+    # Inject zero-content anchors at paragraph 0 so apply_user_edit can
+    # still target them later.
+    structural_fallback: list[str] = []
+    cover_blocks = ["block-alef", "block-bet", "block-gimel", "block-dalet"]
+    first_detected_idx = min(block_starts.values()) if block_starts else 0
+    for i, name in enumerate(cover_blocks):
+        if name not in block_starts:
+            idx = min(i, max(0, first_detected_idx - 1))
+            block_starts[name] = idx
+            structural_fallback.append(name)
+
     # Calculate end_idx for each block = paragraph before the next block's start,
     # or last paragraph if this is the last block found.
     ordered_found = sorted(block_starts.items(), key=lambda kv: kv[1])
@@ -280,11 +321,16 @@ def retrofit_bookmarks(
 
     _save_docx_xml(members, doc_tree, settings_tree, output_path)
 
-    missing = [n for n, _ in BLOCK_ORDER if n not in block_starts and n not in existing_names]
-    logger.info("retrofit %s: added=%s missing=%s",
-                docx_path.name, added, missing)
+    missing = [
+        n for n, _ in BLOCK_ORDER
+        if n not in block_starts
+        and n not in existing_names
+    ]
+    logger.info("retrofit %s: added=%s missing=%s structural=%s",
+                docx_path.name, added, missing, structural_fallback)
     return {
         "bookmarks_added": added,
         "missing_blocks": missing,
+        "structural_fallback": structural_fallback,
         "existing_bookmarks": existing_names,
     }

mcp-server/src/legal_mcp/tools/drafting.py

@@ -576,6 +576,7 @@ async def apply_user_edit(case_number: str, edit_filename: str) -> str:
             "active_draft_path": str(edit_path),
             "bookmarks_added": retrofit_result.get("bookmarks_added", []),
             "missing_blocks": retrofit_result.get("missing_blocks", []),
+            "structural_fallback": retrofit_result.get("structural_fallback", []),
             "existing_bookmarks": retrofit_result.get("existing_bookmarks", []),
         }, ensure_ascii=False, indent=2)
     except Exception as e: