Update app.py

app.py

@@ -3114,6 +3114,9 @@ def _dedupe_subset_transaction_tables_html(text: str) -> str:
       - Earlier table is a strict subset of a later one → drop earlier (UCB-style
         "Deposits (continued)" preview before the same rows appear in the full list).
       - Identical row sets → drop the later table.
+
+    Row matching uses normalized keys so OCR variants (e.g. "0 27" vs "027", backticks)
+    do not prevent duplicate tables from being detected.
     """
     if not text or "<table" not in text.lower():
         return text
@@ -3135,22 +3138,39 @@ def _dedupe_subset_transaction_tables_html(text: str) -> str:
     for m in matches:
         grids.append((m, _parse_grid(m.group(0))))
 
-    hdr_ok = ["date", "description", "amount"]
-
     def _header_matches(grid):
         if not grid or len(grid) < 2:
             return False
         h = [re.sub(r"\s+", " ", str(c or "").strip().lower()) for c in grid[0]]
-        return len(h) == 3 and h == hdr_ok
+        if len(h) != 3:
+            return False
+        d_ok = h[0] == "date"
+        desc_ok = "description" in h[1]
+        amt_ok = h[2] == "amount"
+        return d_ok and desc_ok and amt_ok
+
+    def _norm_row_key(cells):
+        """Stable key for same logical row despite minor OCR/HTML differences."""
+        parts = [str(c or "").strip() for c in cells[:3]]
+        while len(parts) < 3:
+            parts.append("")
+        date_s, desc_s, amt_s = parts[0], parts[1], parts[2]
+        date_s = re.sub(r"\s+", "", date_s)
+        amt_s = re.sub(r"[\s$,]", "", amt_s).lower().replace("−", "-")
+        desc_s = html.unescape(desc_s)
+        desc_s = desc_s.replace("`", "'").replace("’", "'")
+        desc_s = re.sub(r"\s+", "", desc_s.lower())
+        return (date_s, desc_s, amt_s)
 
     def _row_set(g):
         s = set()
         for r in g[1:]:
             if not any(str(c or "").strip() for c in r):
                 continue
-            t = tuple(str(c or "").strip() for c in r)
-            if any(t):
-                s.add(t)
+            cells = [str(c or "").strip() for c in r]
+            if not any(cells):
+                continue
+            s.add(_norm_row_key(cells))
         return s
 
     drop_idx = set()