Update app.py

app.py

@@ -1166,8 +1166,24 @@ def _patch_ocr_with_textlayer(page_md: str, pdf_path: str, page_num: int) -> str
         return page_md
 
     try:
+        # Extract raw page text once — reused by Fix 4D for JB summary sections
+        try:
+            import pymupdf as _fitz_patch
+            _doc_patch  = _fitz_patch.open(pdf_path)
+            _raw_page_text = _doc_patch[page_num].get_text()
+            _doc_patch.close()
+        except Exception:
+            _raw_page_text = ""
+
         tl_rows = _extract_textlayer_rows(pdf_path, page_num)
         if not tl_rows:
+            # Even if no transaction rows, still run Fix 4D for JB summary sections
+            needs_checks = "checks" not in page_md.lower() or "<th>Number</th>" not in page_md
+            needs_dab    = "daily account balance" not in page_md.lower()
+            if _raw_page_text and (needs_checks or needs_dab):
+                jb_extra = _extract_jb_summary_sections(_raw_page_text, needs_checks, needs_dab)
+                if jb_extra:
+                    return page_md.rstrip() + "\n\n" + jb_extra
             return page_md  # scanned PDF or non-transaction page — safe no-op
 
         def _norm(s):
@@ -1379,17 +1395,8 @@ def _patch_ocr_with_textlayer(page_md: str, pdf_path: str, page_num: int) -> str
         # if missing from OCR but present in PDF text layer.
         needs_checks = "checks" not in result.lower() or "<th>Number</th>" not in result
         needs_dab    = "daily account balance" not in result.lower()
-        if needs_checks or needs_dab:
-            # Use get_text() directly (not clip-based) to preserve multi-column layout order
-            try:
-                import pymupdf as _fitz
-                _doc  = _fitz.open(pdf_path)
-                _page = _doc[page_num]
-                raw_tl = _page.get_text()
-                _doc.close()
-            except Exception:
-                raw_tl = ""
-            jb_extra = _extract_jb_summary_sections(raw_tl, needs_checks, needs_dab)
+        if _raw_page_text and (needs_checks or needs_dab):
+            jb_extra = _extract_jb_summary_sections(_raw_page_text, needs_checks, needs_dab)
             if jb_extra:
                 result = result.rstrip() + "\n\n" + jb_extra