Update app.py

app.py

@@ -1417,6 +1417,111 @@ def _reconstruct_separator_rows(grid):
 # Normalizer entry point
 # --------------------------
 
+def _merge_split_rows(grid):
+    """
+    Fix OCR artifact where a transaction row is split across two <tr> rows
+    because the description wrapped to a second line in the source PDF.
+
+    Two patterns detected (Hardin County Bank and similar monospace statements):
+
+    Pattern A — description + empty continuation:
+      Row N:   [desc, '', '', '', '']          ← description, no date/money
+      Row N+1: ['', debit, credit, date, bal]  ← money/date, no description
+      → Merge into: [desc, debit, credit, date, bal]
+
+    Pattern B — description + text continuation + money:
+      Row N:   [desc, '', '', '', '']               ← first line of description
+      Row N+1: [desc_cont, debit, credit, date, bal] ← continuation + money
+      → Merge into: [desc + ' ' + desc_cont, debit, credit, date, bal]
+
+    Guard conditions (no-op unless both rows match the pattern):
+      - Row N must have non-empty col 0 (description).
+      - Row N must have empty date column (col 3 or wherever DATE is).
+      - Row N must have empty debit AND credit columns.
+      - Row N+1 must have non-empty date column.
+      - Row N+1 must have non-empty debit OR credit column.
+      - Row N balance (if present) must be a short fragment without a decimal
+        (1-6 chars, no '.') — confirms it's a split-off ref number, not a balance.
+    """
+    if not grid or len(grid) < 3:
+        return grid
+
+    ncols = len(grid[0])
+    if ncols < 3:
+        return grid
+
+    # Identify column indices from header
+    header = [str(c or "").strip().upper() for c in grid[0]]
+    date_col = desc_col = bal_col = None
+    debit_cols = []
+    credit_cols = []
+
+    for ci, h in enumerate(header):
+        if re.search(r"\bDATE\b", h) and date_col is None:
+            date_col = ci
+        if re.search(r"\b(DESCRIPTION|DETAILS?|NARRATION|PARTICULARS?|TRANSACTION)\b", h) and desc_col is None:
+            desc_col = ci
+        if re.search(r"\bBALANCE\b", h) and bal_col is None:
+            bal_col = ci
+        if re.search(r"\b(DEBIT|DEBITS|DR|WITHDRAWAL|SUBTRACTIONS?)\b", h):
+            debit_cols.append(ci)
+        if re.search(r"\b(CREDIT|CREDITS|CR|DEPOSIT|ADDITIONS?)\b", h):
+            credit_cols.append(ci)
+
+    # Need at least date + description + one money column
+    if date_col is None or desc_col is None:
+        return grid
+    if not debit_cols and not credit_cols:
+        return grid
+
+    money_cols = debit_cols + credit_cols
+    _MONEY_RE2 = re.compile(r"^-?\$?\d{1,3}(?:,\d{3})*\.\d{1,4}$")
+    _DATE_RE2  = re.compile(r"^\d{1,2}/\d{2}(?:/\d{2,4})?$|^\d{1,2}-\d{2}(?:-\d{2,4})?$")
+    # Short fragment guard: 1-6 non-space chars, no decimal point → split-off ref tail
+    _FRAGMENT_RE = re.compile(r"^[^\s\.]{1,6}$")
+
+    new_grid = [grid[0]]
+    i = 1
+    while i < len(grid):
+        row = [str(c or "").strip() for c in grid[i]]
+        while len(row) < ncols:
+            row.append("")
+
+        # Check if this row is a "description-only" split row candidate
+        has_desc    = bool(row[desc_col])
+        no_date     = not row[date_col]
+        no_money    = all(not row[c] for c in money_cols)
+        bal_is_frag = (bal_col is not None and
+                       (_FRAGMENT_RE.match(row[bal_col]) or not row[bal_col]))
+
+        if has_desc and no_date and no_money and bal_is_frag and (i + 1) < len(grid):
+            next_row = [str(c or "").strip() for c in grid[i + 1]]
+            while len(next_row) < ncols:
+                next_row.append("")
+
+            next_has_date  = bool(next_row[date_col]) and _DATE_RE2.match(next_row[date_col])
+            next_has_money = any(bool(next_row[c]) for c in money_cols)
+
+            if next_has_date and next_has_money:
+                # Pattern A or B — merge
+                merged = list(next_row)
+                # Append description (and possible continuation from next row)
+                if next_row[desc_col]:
+                    # Pattern B: next row has desc continuation
+                    merged[desc_col] = row[desc_col] + " " + next_row[desc_col]
+                else:
+                    # Pattern A: next row has no description
+                    merged[desc_col] = row[desc_col]
+                new_grid.append(merged)
+                i += 2  # skip both rows, emit merged
+                continue
+
+        new_grid.append(row)
+        i += 1
+
+    return new_grid
+
+
 def normalize_html_tables(text: str) -> str:
     """
     For every <table>...</table>:
@@ -1457,6 +1562,10 @@ def normalize_html_tables(text: str) -> str:
             # Fix 2: misplaced header row promotion (keyword-score guard — safe on clean PDFs)
             grid = _promote_misplaced_header_row(grid)
 
+            # Fix 6: merge split rows where description wraps to next line
+            # (Hardin County Bank and similar monospace statement formats)
+            grid = _merge_split_rows(grid)
+
             # Fix 3: fused key-value rows in summary sections (no-space guard — safe on all PDFs)
             grid = _fix_fused_keyvalue_rows(grid)