Update app.py

app.py

@@ -32,6 +32,10 @@ Primary goals for reconcile rate:
             Date|Description|Amount table (common when Deposits (continued) is absent from OCR), split after a
             leading ITM DEPOSIT run when the next row matches generic electronic-debit patterns (POS purchase,
             ATM withdrawal, etc.). Safe no-op when there is no ITM prefix or no such debit row after it.
+- Fix 17c: When Electronic Debits appears *before* the table but every data row is credit-side (no strong
+            debit pattern) and at least one ITM DEPOSIT row exists, OCR stacked headings wrong — emit
+            Electronic Credits, then the table, then Electronic Debits (section continues below). Safe no-op
+            if the Debits heading was not before the table or any row looks like a card/ATM debit.
 - 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)
 - Fix 18: adjacent DA3 tables — if the last K data rows of table N match the first K rows of table N+1
@@ -2926,6 +2930,28 @@ def _is_strong_ucb_electronic_debit_row(grid, r: int) -> bool:
     return False
 
 
+def _ucb_has_itm_deposit_data_row(grid) -> bool:
+    for r in range(1, len(grid)):
+        if _is_itm_deposit_row_ucb(grid, r):
+            return True
+    return False
+
+
+def _ucb_table_credits_only_no_strong_debits(grid) -> bool:
+    """True if DA3 table has real data rows but none match strong electronic-debit patterns."""
+    if not grid or len(grid) < 2 or not _is_da3_header(grid):
+        return False
+    has_data = False
+    for r in range(1, len(grid)):
+        row_txt = _row_text_full_ucb(grid, r).strip()
+        if not row_txt:
+            continue
+        has_data = True
+        if _is_strong_ucb_electronic_debit_row(grid, r):
+            return False
+    return bool(has_data and _ucb_has_itm_deposit_data_row(grid))
+
+
 def _ucb_split_grid_electronic_credits_debits_after_itm_prefix(grid):
     """
     Fix 17b: one fused DA3 table under Electronic Credits (+ optional empty EC/ED headings) where OCR
@@ -2952,7 +2978,9 @@ def _ucb_split_grid_electronic_credits_debits_after_itm_prefix(grid):
     return g_ec, g_ed
 
 
-def _try_split_ucb_ec_ed_from_fused_table_html(lead_before_ec: str, table_html: str):
+def _try_split_ucb_ec_ed_from_fused_table_html(
+    lead_before_ec: str, table_html: str, had_ed_heading_immediately_before_table: bool
+):
     try:
         p = TableGridParser()
         p.feed(table_html)
@@ -2960,22 +2988,42 @@ def _try_split_ucb_ec_ed_from_fused_table_html(lead_before_ec: str, table_html:
     except Exception:
         return None
     pair = _ucb_split_grid_electronic_credits_debits_after_itm_prefix(grid)
-    if not pair:
-        return None
-    g_ec, g_ed = pair
-    sect = '<div align="center">\n\n{title}\n\n</div>\n\n'
-    parts = []
-    if lead_before_ec and lead_before_ec.strip():
-        parts.append(lead_before_ec.rstrip())
-    parts.extend(
-        [
-            sect.format(title="Electronic Credits"),
-            _grid_to_html(g_ec),
-            sect.format(title="Electronic Debits"),
-            _grid_to_html(g_ed),
-        ]
-    )
-    return "\n\n".join(parts)
+    if pair:
+        g_ec, g_ed = pair
+        sect = '<div align="center">\n\n{title}\n\n</div>\n\n'
+        parts = []
+        if lead_before_ec and lead_before_ec.strip():
+            parts.append(lead_before_ec.rstrip())
+        parts.extend(
+            [
+                sect.format(title="Electronic Credits"),
+                _grid_to_html(g_ec),
+                sect.format(title="Electronic Debits"),
+                _grid_to_html(g_ed),
+            ]
+        )
+        return "\n\n".join(parts)
+
+    # Fix 17c: headings stacked as EC / ED / table but rows are all credits (e.g. ITM-only) —
+    # put the table under Electronic Credits, then leave Electronic Debits for continuation.
+    if (
+        had_ed_heading_immediately_before_table
+        and _ucb_table_credits_only_no_strong_debits(grid)
+    ):
+        sect = '<div align="center">\n\n{title}\n\n</div>\n\n'
+        parts = []
+        if lead_before_ec and lead_before_ec.strip():
+            parts.append(lead_before_ec.rstrip())
+        parts.extend(
+            [
+                sect.format(title="Electronic Credits"),
+                _grid_to_html(grid),
+                sect.format(title="Electronic Debits"),
+            ]
+        )
+        return "\n\n".join(parts)
+
+    return None
 
 
 # Optional empty centered banner only; EC/ED title divs are re-emitted after a successful split
@@ -2983,7 +3031,7 @@ def _try_split_ucb_ec_ed_from_fused_table_html(lead_before_ec: str, table_html:
 _UCB_EC_ED_SINGLE_FUSED_TABLE = re.compile(
     r'(?P<lead>(?:<div\s+align=["\']center["\']\s*>\s*</div>\s*)?)'
     r'<div\s+align=["\']center["\']\s*>\s*Electronic\s+Credits\s*</div>\s*'
-    r'(?:<div\s+align=["\']center["\']\s*>\s*Electronic\s+Debits\s*</div>\s*)?'
+    r'(?P<ed_before>(?:<div\s+align=["\']center["\']\s*>\s*Electronic\s+Debits\s*</div>\s*)?)'
     r'(?P<table><table[^>]*>.*?</table>)',
     re.DOTALL | re.IGNORECASE,
 )
@@ -2997,8 +3045,10 @@ def _split_ucb_fused_ec_ed_headings_single_table_html(text: str) -> str:
 
     def _repl(m):
         lead = m.group("lead")
+        ed_before = m.group("ed_before") or ""
         table_html = m.group("table")
-        new = _try_split_ucb_ec_ed_from_fused_table_html(lead, table_html)
+        had_ed = bool(ed_before.strip())
+        new = _try_split_ucb_ec_ed_from_fused_table_html(lead, table_html, had_ed)
         return new if new is not None else m.group(0)
 
     return _UCB_EC_ED_SINGLE_FUSED_TABLE.sub(_repl, text)