Update app.py

app.py

@@ -1586,6 +1586,150 @@ def _extract_fused_desc_amount(grid):
     return new_grid
 
 
+def _is_junk_table(grid):
+    """
+    Detect and suppress known junk tables that are OCR artifacts from
+    page headers/footers, not real transaction data.
+
+    Current patterns:
+    1. CUSTOMER INFORMATION fragment tables (First Horizon Bank):
+       Single-column table with header "CUSTOMER INFORMATION" and
+       data rows that are short numeric fragments (e.g. "254", "24")
+       from the partial account number / year printed in the header band.
+    """
+    if not grid or len(grid) < 2:
+        return False
+    ncols = len(grid[0])
+    if ncols != 1:
+        return False
+    header_text = str(grid[0][0] or "").strip().upper()
+    if "CUSTOMER INFORMATION" in header_text:
+        # All data rows should be short (< 10 chars) numeric/alphanumeric fragments
+        data_rows = grid[1:]
+        if all(len(str(r[0] or "").strip()) <= 10 for r in data_rows):
+            return True
+    return False
+
+
+def _split_fused_multicolumn_header(grid):
+    """
+    Fix OCR artifact where multiple column headers are fused into one <th> cell.
+
+    Example (First Horizon Bank alternating pages):
+      Fused:   ['DATE DESCRIPTION CARD #', 'DEPOSIT', 'WITHDRAWAL']
+      Correct: ['DATE', 'DESCRIPTION', 'DEPOSIT', 'WITHDRAWAL', 'CARD #']
+
+    The data rows also have date + description + card# all in col 0:
+      Fused:   ['01/16 PURCHASE - UBER TRIP... 4919', '', '$21.02']
+      Correct: ['01/16', 'PURCHASE - UBER TRIP...', '', '$21.02', '4919']
+
+    Detection: first header cell contains 2+ keyword terms separated by spaces,
+    AND remaining header cells are valid money column keywords.
+    """
+    if not grid or len(grid) < 2:
+        return grid
+
+    ncols = len(grid[0])
+    if ncols < 2:
+        return grid
+
+    first_header = str(grid[0][0] or "").strip()
+    # Split first header cell into parts and check if multiple are keywords
+    parts = first_header.split()
+    kw_hits = [p for p in parts if _HEADER_KW_EXACT_RE.match(p)]
+    if len(kw_hits) < 2:
+        return grid  # not a fused multi-keyword header
+
+    # The remaining header cells must all be keyword columns
+    rest_headers = [str(c or "").strip() for c in grid[0][1:]]
+    if not all(_HEADER_KW_EXACT_RE.match(h) for h in rest_headers if h):
+        return grid
+
+    # Build the new expanded header:
+    # Split the fused first cell into individual keyword columns
+    # Keep non-keyword trailing parts (e.g. "CARD #") as the last new col
+    new_header_cols = []
+    non_kw_parts = []
+    for p in parts:
+        if _HEADER_KW_EXACT_RE.match(p):
+            if non_kw_parts:
+                new_header_cols.append(" ".join(non_kw_parts))
+                non_kw_parts = []
+            new_header_cols.append(p)
+        else:
+            non_kw_parts.append(p)
+    if non_kw_parts:
+        new_header_cols.append(" ".join(non_kw_parts))
+
+    # Full new header = expanded first cell + remaining cells
+    new_header = new_header_cols + rest_headers
+    new_ncols = len(new_header)
+    extra_cols = new_ncols - ncols  # how many new columns were added
+
+    # Identify column roles in the new header
+    _DATE_COL_RE   = re.compile(r"\bDATE\b", re.IGNORECASE)
+    _CARD_COL_RE   = re.compile(r"\bCARD\b", re.IGNORECASE)
+    _DESC_COL_RE   = re.compile(r"\b(DESCRIPTION|DETAILS?|NARRATION|PARTICULARS?)\b", re.IGNORECASE)
+    _MONEY_COL_RE  = re.compile(r"\b(DEPOSIT|WITHDRAWAL|DEBIT|CREDIT|AMOUNT|ADDITION|SUBTRACTION)S?\b", re.IGNORECASE)
+
+    new_date_col = next((i for i,h in enumerate(new_header) if _DATE_COL_RE.search(h)), None)
+    new_desc_col = next((i for i,h in enumerate(new_header) if _DESC_COL_RE.search(h)), None)
+    new_card_col = next((i for i,h in enumerate(new_header) if _CARD_COL_RE.search(h)), None)
+
+    if new_date_col is None or new_desc_col is None:
+        return grid  # can't map columns
+
+    # Re-split each data row: col 0 had "date description... card#" fused together
+    _DATE_PREFIX = re.compile(r"^(\d{1,2}/\d{2}(?:/\d{2,4})?)\s+(.*)", re.DOTALL)
+    _CARD_SUFFIX = re.compile(r"^(.*?)\s+(\d{4})$", re.DOTALL)  # 4-digit card number at end
+
+    def pad(row, n):
+        r = list(row)
+        while len(r) < n: r.append("")
+        return r[:n]
+
+    new_grid = [pad(new_header, new_ncols)]
+    for row in grid[1:]:
+        cells = [str(c or "").strip() for c in row]
+        while len(cells) < ncols: cells.append("")
+
+        fused_cell = cells[0]
+        rest_cells = cells[1:]
+
+        # Extract date from front
+        date_val = ""
+        desc_val = fused_cell
+        card_val = ""
+
+        dm = _DATE_PREFIX.match(fused_cell)
+        if dm:
+            date_val = dm.group(1)
+            desc_val = dm.group(2).strip()
+
+        # Extract card# from end (4-digit number)
+        if new_card_col is not None:
+            cm = _CARD_SUFFIX.match(desc_val)
+            if cm:
+                desc_val = cm.group(1).strip()
+                card_val = cm.group(2)
+
+        # Build new row
+        new_row = [""] * new_ncols
+        new_row[new_date_col] = date_val
+        new_row[new_desc_col] = desc_val
+        if new_card_col is not None:
+            new_row[new_card_col] = card_val
+        # Fill remaining money columns from rest_cells
+        money_col_indices = [i for i,h in enumerate(new_header) if _MONEY_COL_RE.search(h)]
+        for mi, ri in enumerate(range(len(rest_cells))):
+            if mi < len(money_col_indices):
+                new_row[money_col_indices[mi]] = rest_cells[ri]
+
+        new_grid.append(new_row)
+
+    return new_grid
+
+
 def normalize_html_tables(text: str) -> str:
     """
     For every <table>...</table>:
@@ -1612,6 +1756,13 @@ def normalize_html_tables(text: str) -> str:
             p = TableGridParser()
             p.feed(table_html)
             grid = _build_grid(p.rows)
+
+            # Suppress known junk tables (OCR artifacts from page headers)
+            if _is_junk_table(grid):
+                out.append("")
+                last = m.end()
+                continue
+
             grid = _drop_truly_empty_columns(grid)
             grid = _merge_blank_header_text_columns(grid)
 
@@ -1626,6 +1777,9 @@ 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 8: split fused multi-keyword header cell (First Horizon Bank)
+            grid = _split_fused_multicolumn_header(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)