Update app.py

app.py

@@ -1166,24 +1166,8 @@ 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):
@@ -1391,15 +1375,6 @@ def _patch_ocr_with_textlayer(page_md: str, pdf_path: str, page_num: int) -> str
                 log.info("page %d: Case C — new table (%d rows) from text layer",
                          page_num, len(_cnew_rows))
 
-        # Fix 4D: append Johnson Bank Checks + Daily Account Balance
-        # 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 _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
-
         return result
 
     except Exception as e:
@@ -2386,6 +2361,24 @@ def run_ocr(uploaded_file):
                 # Safe no-op for scanned PDFs (no text layer) and non-transaction pages.
                 if is_pdf:
                     stabilized = _patch_ocr_with_textlayer(stabilized, path, page_num)
+
+                # Fix 4D: append Johnson Bank Checks + Daily Account Balance
+                # Run OUTSIDE _patch_ocr_with_textlayer so exceptions there don't block it.
+                if is_pdf:
+                    try:
+                        needs_checks = "checks" not in stabilized.lower() or "<th>Number</th>" not in stabilized
+                        needs_dab    = "daily account balance" not in stabilized.lower()
+                        if needs_checks or needs_dab:
+                            import pymupdf as _fitz4d
+                            _doc4d  = _fitz4d.open(path)
+                            _txt4d  = _doc4d[page_num].get_text()
+                            _doc4d.close()
+                            jb_extra = _extract_jb_summary_sections(_txt4d, needs_checks, needs_dab)
+                            if jb_extra:
+                                stabilized = stabilized.rstrip() + "\n\n" + jb_extra
+                    except Exception:
+                        pass  # safe no-op — never break other PDFs
+
                 parts.append(stabilized)
 
             # Footer extraction: PDF text layer -> band words -> OCR crop