Update app.py

app.py

@@ -99,6 +99,71 @@ def _build_table_from_entries(entries: List[dict]) -> str:
     return "\n".join(table)
 
 
+def _is_transaction_start(line: str) -> bool:
+    return bool(re.match(r"\d{2}/\d{2}", line.strip()))
+
+
+def _ends_with_amount(text: str) -> bool:
+    return bool(re.search(r"\d{1,3}(?:,\d{3})*\.\d{2}$", text.strip()))
+
+
+def _is_noise_line(line: str) -> bool:
+    s = line.strip().lower()
+    if not s:
+        return True
+    if s == "---page-separator---":
+        return True
+    if "call 1-800-937-2000" in s:
+        return True
+    if s in ("<table border=\"1\"><tr><td></td></tr></table>", "<table><tr><td></td></tr></table>"):
+        return True
+    return False
+
+
+def _merge_transaction_lines(rows: List[str]) -> List[str]:
+    """
+    Step 1/2: Merge multiline OCR rows into single transaction rows.
+    A new row begins when a line starts with MM/DD.
+    """
+    merged: List[str] = []
+    current = ""
+
+    for raw in rows:
+        line = " ".join(raw.strip().split())
+        if not line or _is_noise_line(line):
+            continue
+
+        if _is_transaction_start(line):
+            if current:
+                merged.append(current.strip())
+            current = line
+        else:
+            if current:
+                current += " " + line
+            else:
+                current = line
+
+    if current:
+        merged.append(current.strip())
+
+    return merged
+
+
+def _repair_orphan_amounts(rows: List[str]) -> List[str]:
+    """
+    Step 3: Attach lines without ending amount to prior transaction line when possible.
+    """
+    fixed: List[str] = []
+    for line in rows:
+        if _ends_with_amount(line):
+            fixed.append(line)
+        elif fixed:
+            fixed[-1] = f"{fixed[-1]} {line}".strip()
+        else:
+            fixed.append(line)
+    return fixed
+
+
 def _normalize_transactions(text: str) -> str:
     """
     Convert loose transaction lines into structured markdown tables using:
@@ -119,8 +184,9 @@ def _normalize_transactions(text: str) -> str:
         if in_section and (line.strip() == "" or line.strip().startswith("Subtotal")):
             if table_buffer:
                 clean_rows = []
-                for raw in table_buffer:
-                    normalized = " ".join(raw.split())
+                merged_rows = _merge_transaction_lines(table_buffer)
+                repaired_rows = _repair_orphan_amounts(merged_rows)
+                for normalized in repaired_rows:
                     match = pattern.search(normalized)
                     if match:
                         date, desc, amount = match.groups()
@@ -143,8 +209,9 @@ def _normalize_transactions(text: str) -> str:
 
     if table_buffer:
         clean_rows = []
-        for raw in table_buffer:
-            normalized = " ".join(raw.split())
+        merged_rows = _merge_transaction_lines(table_buffer)
+        repaired_rows = _repair_orphan_amounts(merged_rows)
+        for normalized in repaired_rows:
             match = pattern.search(normalized)
             if match:
                 date, desc, amount = match.groups()