Update app.py

app.py

@@ -25,7 +25,8 @@ Primary goals for reconcile rate:
             Beginning/Ending balance rows (UCB-style wide summary tables)
 - Fix 14: UCB 3-col tables where Beginning Balance amount is fused into description and Amount is $0.00
 - Fix 15: UCB (and similar) 3-col rows where Date+Description are fused in col1 and Amount sits in col2 with col3 empty
-- Post-pass: drop later duplicate subset Date|Description|Amount tables (repeated deposit blocks)
+- Post-pass: dedupe 3-col Date|Description|Amount tables when one is a duplicate fragment
+            (later subset of earlier, earlier subset of later, or identical row sets)
 """
 
 # Patch asyncio first (before Gradio imports it) to suppress Python 3.13 cleanup noise
@@ -3108,9 +3109,11 @@ def normalize_html_tables(text: str) -> str:
 
 def _dedupe_subset_transaction_tables_html(text: str) -> str:
     """
-    Remove duplicate 3-column transaction tables where a later table's rows are a
-    strict subset of an earlier table's rows (same Date|Description|Amount header).
-    Used when OCR repeats the same deposit block under a section heading.
+    Remove duplicate 3-column Date|Description|Amount tables:
+      - Later table is a strict subset of an earlier one → drop later (original).
+      - 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.
     """
     if not text or "<table" not in text.lower():
         return text
@@ -3132,42 +3135,55 @@ 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
+
+    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)
+        return s
+
     drop_idx = set()
     for i in range(1, len(grids)):
+        if i in drop_idx:
+            continue
         mi, gi = grids[i]
-        if not gi or len(gi) < 2:
+        if not _header_matches(gi):
             continue
-        h = [re.sub(r"\s+", " ", str(c or "").strip().lower()) for c in gi[0]]
-        if len(h) != 3 or h != ["date", "description", "amount"]:
+        si = _row_set(gi)
+        if not si:
             continue
 
         for j in range(i):
             if j in drop_idx:
                 continue
             mj, gj = grids[j]
-            if not gj or len(gj) < 2:
+            if not _header_matches(gj):
                 continue
-            hj = [re.sub(r"\s+", " ", str(c or "").strip().lower()) for c in gj[0]]
-            if len(hj) != 3 or hj != ["date", "description", "amount"]:
+            sj = _row_set(gj)
+            if not sj:
                 continue
 
-            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)
-                return s
-
-            si, sj = _row_set(gi), _row_set(gj)
-            if not si or not sj:
+            # Earlier j fully contained in later i (preview + full list) → drop earlier.
+            if sj <= si and len(sj) < len(si):
+                drop_idx.add(j)
                 continue
-            # Drop later table if it is a strict subset of an earlier one (repeated block).
+
+            # Later i strict subset of earlier j → drop later.
             if si <= sj and len(si) < len(sj):
                 drop_idx.add(i)
                 break
+
             if si == sj:
                 drop_idx.add(i)
                 break