Update app.py

app.py

@@ -2422,6 +2422,148 @@ def _parse_nfcu_page(pdf_path: str, page_num: int) -> str:
         log.warning("_parse_nfcu_page failed (page %d): %s", page_num, e)
         return ""
 
+def _extract_nfcu_summary_table(pdf_path: str, page_num: int) -> str:
+    """
+    Navy-only additive helper:
+    Rebuild "Summary of your deposit accounts" from PDF text-layer words.
+    Returns a clean HTML table or "" (safe no-op on any miss/error).
+    """
+    try:
+        import pymupdf as fitz
+
+        doc = fitz.open(pdf_path)
+        if page_num >= len(doc):
+            doc.close()
+            return ""
+        page = doc[page_num]
+        words = page.get_text("words") or []
+        doc.close()
+        if not words:
+            return ""
+
+        # Group tokens into reading lines by y-position.
+        buckets = []
+        for w in words:
+            if len(w) < 5:
+                continue
+            x0 = float(w[0])
+            y0 = float(w[1])
+            tok = str(w[4] or "").strip()
+            if not tok:
+                continue
+            bi = None
+            for i, (yb, _) in enumerate(buckets):
+                if abs(yb - y0) <= 2.0:
+                    bi = i
+                    break
+            if bi is None:
+                buckets.append((y0, [(x0, tok)]))
+            else:
+                buckets[bi][1].append((x0, tok))
+
+        lines = []
+        for y, arr in sorted(buckets, key=lambda x: x[0]):
+            arr2 = sorted(arr, key=lambda x: x[0])
+            toks = [t for _, t in arr2]
+            text = " ".join(toks).strip()
+            lines.append((y, toks, text))
+
+        anchor_idx = None
+        for i, (_, _, text) in enumerate(lines):
+            if "summary of your deposit accounts" in text.lower():
+                anchor_idx = i
+                break
+        if anchor_idx is None:
+            return ""
+
+        money_re = re.compile(r"^[\d,]+\.\d{2}-?$")
+        date_re = re.compile(r"^\d{2}-\d{2}$")
+        stop_re = re.compile(
+            r"(business checking\s*-|mbr business savings\s*-|date\s+transaction\s+detail|items paid|checking|savings)",
+            re.IGNORECASE,
+        )
+
+        rows = []
+        for _, toks, text in lines[anchor_idx + 1:]:
+            if not text:
+                continue
+            if stop_re.search(text):
+                break
+            if not toks:
+                continue
+            if not date_re.match(toks[0]):
+                continue
+
+            date = toks[0]
+            rest = toks[1:]
+            mvals = [t for t in rest if money_re.match(t)]
+            desc_tokens = [t for t in rest if not money_re.match(t)]
+            desc = " ".join(desc_tokens).strip()
+            if not desc:
+                continue
+
+            amount = ""
+            balance = ""
+            if len(mvals) >= 2:
+                amount = mvals[0]
+                balance = mvals[-1]
+            elif len(mvals) == 1:
+                # Beginning/ending balance rows usually carry the number in Balance.
+                if re.search(r"\b(beginning|ending)\s+balance\b", desc, re.IGNORECASE):
+                    balance = mvals[0]
+                else:
+                    amount = mvals[0]
+
+            rows.append([date, desc, "", amount, balance])
+
+        if not rows:
+            return ""
+
+        hdr = ["Date", "Transaction", "Detail", "Amount($)", "Balance($)"]
+        return _grid_to_html([hdr] + rows)
+    except Exception:
+        return ""
+
+def _replace_nfcu_summary_table(page_md: str, summary_table_html: str) -> str:
+    """
+    Replace the first table after 'Summary of your deposit accounts' with
+    a rebuilt NFCU summary table. Safe no-op if anchor/table is missing.
+    """
+    if not page_md or not summary_table_html:
+        return page_md
+    table_pattern = re.compile(r"<table[^>]*>.*?</table>", re.DOTALL | re.IGNORECASE)
+    lower = page_md.lower()
+    anchor = lower.find("summary of your deposit accounts")
+
+    # Primary path: replace first table after the summary heading.
+    if anchor >= 0:
+        for m in table_pattern.finditer(page_md):
+            if m.start() > anchor:
+                old_table = m.group(0)
+                return page_md.replace(old_table, summary_table_html, 1)
+
+    # Fallback path: heading may be malformed/absent in OCR text.
+    # For NFCU, summary table is the 5-column table with
+    # Date + Transaction + Detail + Amount + Balance headers.
+    for m in table_pattern.finditer(page_md):
+        tbl = m.group(0)
+        p = TableGridParser()
+        p.feed(tbl)
+        g = _build_grid(p.rows)
+        if not g:
+            continue
+        h = " ".join(str(c or "").strip().lower() for c in g[0])
+        if (
+            "date" in h
+            and "transaction" in h
+            and "detail" in h
+            and "amount" in h
+            and "balance" in h
+        ):
+            return page_md.replace(tbl, summary_table_html, 1)
+
+    return page_md
+
 def normalize_html_tables(text: str) -> str:
     """
     For every <table>...</table>:
@@ -2638,6 +2780,12 @@ def run_ocr(uploaded_file):
                 # Safe no-op for scanned PDFs (no text layer) and non-transaction pages.
                 if is_pdf and not is_nfcu_doc:
                     stabilized = _patch_ocr_with_textlayer(stabilized, path, page_num)
+                elif is_pdf and is_nfcu_doc:
+                    # Navy-only additive fix: rebuild malformed summary table from
+                    # text layer, but keep GLM body/header/footer unchanged.
+                    nfcu_summary = _extract_nfcu_summary_table(path, page_num)
+                    if nfcu_summary:
+                        stabilized = _replace_nfcu_summary_table(stabilized, nfcu_summary)
 
                 # Fix 4D: append Johnson Bank Checks + Daily Account Balance
                 # Run OUTSIDE _patch_ocr_with_textlayer so exceptions there don't block it.