Update app.py

app.py

@@ -1255,6 +1255,64 @@ def _patch_ocr_with_textlayer(page_md: str, pdf_path: str, page_num: int) -> str
                     insert_pos = result.find(new_table_html) + len(new_table_html)
                     result = result[:insert_pos] + extra_html + result[insert_pos:]
 
+        # Case C: NO transaction table exists on this page at all, but text
+        # layer has valid rows. Build a brand-new table and prepend it to
+        # the page content. This handles pages where OCR dropped the entire
+        # transaction section (e.g. East West Bank page 7 with only 2 rows).
+        #
+        # Guard: only fire if no table on this page already has DATE+DESCRIPTION
+        # columns (avoids duplicating non-transaction tables like DAILY BALANCES).
+        table_pattern2 = re.compile(r"<table[^>]*>.*?</table>", re.DOTALL | re.IGNORECASE)
+        has_txn_table = False
+        for tbl in table_pattern2.finditer(result):
+            p2 = TableGridParser()
+            p2.feed(tbl.group(0))
+            g2 = _build_grid(p2.rows)
+            if len(g2) < 2:
+                continue
+            hdr2 = [str(c or "").strip().upper() for c in g2[0]]
+            has_date  = any(re.search(r"DATE", h) for h in hdr2)
+            has_desc  = any(re.search(r"(DESCRIPTION|DETAILS?|NARRATION|PARTICULARS?)", h) for h in hdr2)
+            if has_date and has_desc:
+                has_txn_table = True
+                break
+
+        if not has_txn_table and tl_rows:
+            # Infer date format from text layer to pick column header label
+            _hy = re.compile(r"^\d{1,2}-\d{2}")
+            _sl = re.compile(r"^\d{1,2}/\d{2}")
+            _sample_dates = [_norm(r["date"]) for r in tl_rows[:3]]
+            if any(_hy.match(d) for d in _sample_dates):
+                date_label = "Date"
+            elif any(_sl.match(d) for d in _sample_dates):
+                date_label = "Date"
+            else:
+                date_label = "Trans Date"
+
+            has_post = any(r.get("post_date") for r in tl_rows)
+            if has_post:
+                header_row = [date_label, "Post Date", "Transaction Description", "Amount"]
+            else:
+                header_row = [date_label, "Transaction Description", "Amount"]
+
+            date_ci = 0
+            desc_ci = 2 if has_post else 1
+            amt_ci  = 3 if has_post else 2
+
+            new_rows = []
+            for r in tl_rows:
+                row = [""] * len(header_row)
+                row[date_ci] = r["date"]
+                row[desc_ci] = r["desc"]
+                row[amt_ci]  = r["amount"]
+                if has_post:
+                    row[1] = r.get("post_date", "")
+                new_rows.append(row)
+
+            new_table_html = _grid_to_html([header_row] + new_rows)
+            result = new_table_html + "\n\n" + result
+            log.info("page %d: created new table from text layer (%d rows)", page_num, len(new_rows))
+
         return result
 
     except Exception as e: