Update app.py

app.py

@@ -129,6 +129,8 @@ Primary goals for reconcile rate:
 - Fix MD-3: Check image pages — GLM often emits ![](bbox…) plus <div align="center"> fragments
             (e.g. 2293 $6,721.00 4/29/2025). Collapse detected gallery pages into one Check #|Amount|Date
             table when bbox/image markers + repeated check lines are present; skips chunks with YOUR CHECKS SEQUENCED.
+- Fix MD-4: YOUR CHECKS SEQUENCED wide grid (N×DATE|CHECK#|AMOUNT per row) is unfolded to one DATE|CHECK#|AMOUNT
+            table so narrow viewers show every check, not only the first date column.
 """
 
 # Patch asyncio first (before Gradio imports it) to suppress Python 3.13 cleanup noise
@@ -609,6 +611,59 @@ def _merge_blank_header_text_columns(grid):
     return new_grid
 
 
+def _split_wide_checks_sequenced_grid(grid: List[List[str]]) -> Optional[List[List[str]]]:
+    """
+    Fix MD-4: Banks print YOUR CHECKS SEQUENCED as 2–3 side-by-side DATE|CHECK #|AMOUNT groups per row.
+    OCR HTML keeps that as 6–9 columns; narrow markdown previews only show the first date column.
+    Unfold into a single standard 3-column table (one row per check).
+    """
+    if not grid or len(grid) < 2:
+        return None
+    header = [str(c or "").strip() for c in grid[0]]
+    header_join = " ".join(header).upper()
+    if "SEQUENCED" not in header_join or "CHECK" not in header_join:
+        return None
+    ncols = max(len(r) for r in grid if isinstance(r, list))
+    if ncols < 6 or ncols % 3 != 0:
+        return None
+    amt_cols = sum(1 for c in header if str(c or "").strip().upper() == "AMOUNT")
+    if amt_cols < 2 and ncols < 9:
+        return None
+
+    date_re = re.compile(r"^\d{1,2}/\d{2}(?:/\d{2,4})?$")
+    out: List[List[str]] = [["DATE", "CHECK #", "AMOUNT"]]
+
+    for row in grid[1:]:
+        cells = [str(c or "").strip() for c in row]
+        while len(cells) < ncols:
+            cells.append("")
+        cells = cells[:ncols]
+        c0 = cells[0] if cells else ""
+        joined = " ".join(cells)
+
+        if c0.startswith("=") and ("=" * 3 in c0 or len(c0) >= 8):
+            continue
+        if c0 and "=" in c0 and not c0.replace("=", "").strip():
+            continue
+        if "DATE" in c0.upper() and "CHECK" in c0.upper() and "AMOUNT" in c0.upper():
+            continue
+        if "GAP IN CHECK" in joined.upper() and "SEQUENCE" in joined.upper():
+            continue
+
+        for g in range(0, ncols, 3):
+            if g + 2 >= len(cells):
+                break
+            d, ck, am = cells[g], cells[g + 1], cells[g + 2]
+            if not d and not ck and not am:
+                continue
+            if d and date_re.match(d):
+                out.append([d, ck, am])
+
+    if len(out) < 2:
+        return None
+    return out
+
+
 def _fill_headless_trailing_amount_headers(grid):
     """
     Fix 32a: PDF layouts often omit a header above the rightmost amount column; OCR leaves a
@@ -3142,6 +3197,9 @@ def normalize_html_tables(text: str) -> str:
 
             grid = _drop_truly_empty_columns(grid)
             grid = _merge_blank_header_text_columns(grid)
+            _seq_linear = _split_wide_checks_sequenced_grid(grid)
+            if _seq_linear is not None:
+                grid = _seq_linear
 
             grid = _fill_headless_trailing_amount_headers(grid)
             grid = _normalize_date_desc_money_table_alignment(grid)