Update app.py

app.py

@@ -1192,6 +1192,83 @@ def synthesize_running_balance_columns(md: str) -> str:
     return "".join(pieces)
 
 
+def _extract_txn_date_balance_pairs(md: str):
+    pairs = []
+    for tm in re.finditer(r"<table\b[^>]*>.*?</table>", md, flags=re.IGNORECASE | re.DOTALL):
+        t = tm.group(0)
+        rows = _extract_rows_plain_from_table(t)
+        if len(rows) < 2:
+            continue
+        hdr = [c.strip().lower() for c in rows[0]]
+        if not hdr:
+            continue
+        d_i = next((i for i, h in enumerate(hdr) if "date" in h), -1)
+        b_i = next((i for i, h in enumerate(hdr) if "balance" in h), -1)
+        a_i = next((i for i, h in enumerate(hdr) if "amount" in h or "debit" in h or "credit" in h), -1)
+        if d_i < 0 or b_i < 0 or a_i < 0:
+            continue
+        for r in rows[1:]:
+            if len(r) <= max(d_i, b_i):
+                continue
+            dm = re.search(r"\d{1,2}/\d{1,2}(?:/\d{2,4})?", (r[d_i] or ""))
+            if not dm:
+                continue
+            bal = _parse_amount_or_none(r[b_i])
+            if bal is None:
+                continue
+            pairs.append((dm.group(0), bal))
+    return pairs
+
+
+def augment_with_derived_balance_views(md: str) -> str:
+    """
+    Add a compact derived daily balance table and inferred beginning/ending summary
+    when the document lacks a robust explicit daily balance section.
+    """
+    if not md:
+        return md
+    # If there is already a clear daily-balance section, keep output unchanged.
+    if re.search(r"daily\s+(?:ledger\s+)?balances", md, flags=re.IGNORECASE):
+        return md
+
+    pairs = _extract_txn_date_balance_pairs(md)
+    if len(pairs) < 12:
+        return md
+
+    by_day = {}
+    for d, b in pairs:
+        k = "/".join(d.split("/")[:2])
+        by_day[k] = b
+    if len(by_day) < 8:
+        return md
+
+    days = sorted(
+        by_day.keys(),
+        key=lambda x: (int(x.split("/")[0]), int(x.split("/")[1])),
+    )
+    begin = by_day[days[0]]
+    end = by_day[days[-1]]
+
+    has_begin = bool(re.search(r"(?:beginning|starting)\s+balance", md, flags=re.IGNORECASE))
+    has_end = bool(re.search(r"ending\s+balance", md, flags=re.IGNORECASE))
+
+    rows = ["<tr><th>Date</th><th>Balance</th></tr>"]
+    for d in days:
+        rows.append(f"<tr><td>{html.escape(d)}</td><td>{by_day[d]:,.2f}</td></tr>")
+    daily_tbl = "<table>\n" + "\n".join(rows) + "\n</table>"
+
+    extra = ["", "<div align=\"center\">Derived daily balances</div>", daily_tbl]
+    if not has_begin or not has_end:
+        extra.insert(
+            0,
+            (
+                f"Inferred beginning balance: {begin:,.2f}\n"
+                f"Inferred ending balance: {end:,.2f}"
+            ),
+        )
+    return md.rstrip() + "\n\n" + "\n\n".join(extra) + "\n"
+
+
 def render_pdf_pages_to_images(pdf_path: str) -> Tuple[List[str], List[int]]:
     import pymupdf as fitz
     from PIL import Image
@@ -1331,6 +1408,7 @@ def run_ocr(uploaded_file):
             merged = stabilize_table_markup(merged, rounds=4)
             merged = enrich_transaction_tables_with_daily_balances(merged)
             merged = synthesize_running_balance_columns(merged)
+            merged = augment_with_derived_balance_views(merged)
             merged = stabilize_table_markup(merged, rounds=2)
             merged = repair_thead_cell_semantics(merged)
             merged = strip_degenerate_html_tables(merged)