import argparse import csv import hashlib import re from collections import Counter from datetime import datetime, timedelta, timezone from pathlib import Path PAGE_RE = re.compile(r"\[\[PAGE\s+(\d+)\]\]") VALUE_RE = re.compile( r"[-+]?\d+(?:\.\d+)?\s*(?:%|\u4e07\u5428|\u5428|\u4ebf\u5143|\u5143/\u5428|\u7f8e\u5143/\u5428|GWh)" ) TABLE_HINTS = [ "\u8868", "\u56fe", "\u8d44\u6599\u6765\u6e90", "\u5355\u4f4d\uff1a", "\u5e93\u5b58", "\u4ef7\u683c", "\u4ea7\u91cf", "\u4f9b\u7ed9", ] def read_csv(path): with path.open("r", encoding="utf-8-sig", newline="") as handle: return list(csv.DictReader(handle)) def write_csv(path, fieldnames, rows): path.parent.mkdir(parents=True, exist_ok=True) with path.open("w", encoding="utf-8-sig", newline="") as handle: writer = csv.DictWriter(handle, fieldnames=fieldnames) writer.writeheader() writer.writerows(rows) def sha256_file(path): return hashlib.sha256(path.read_bytes()).hexdigest() def normalize(text): return re.sub(r"\s+", "", text or "") def read_lines(path): try: return path.read_text(encoding="utf-8", errors="replace").splitlines() except FileNotFoundError: return [] def page_marks(lines): marks = [] for line_no, line in enumerate(lines, start=1): match = PAGE_RE.search(line) if match: marks.append((line_no, match.group(1))) return marks def page_for_line(marks, line_no): current = "" for mark_line, page_no in marks: if mark_line <= line_no: current = page_no else: break return current def table_score(lines): score = 0 joined = "\n".join(lines) if VALUE_RE.search(joined): score += 1 if sum(1 for line in lines if len(re.findall(r"\s{2,}", line)) >= 2) >= 2: score += 1 if any(hint in joined for hint in TABLE_HINTS): score += 1 return score def find_line(lines, evidence, declared_page): needle = normalize(evidence) if needle: for line_no, line in enumerate(lines, start=1): if needle in normalize(line): return line_no, "EXACT_NORMALIZED_LINE" prefix = needle[:80] if prefix: for idx in range(len(lines)): window = "".join(normalize(x) for x in lines[max(0, idx - 2) : min(len(lines), idx + 3)]) if prefix in window: return idx + 1, "WINDOW_PREFIX_MATCH" if declared_page and declared_page.isdigit(): for line_no, page_no in page_marks(lines): if page_no == declared_page: return line_no, "PAGE_MARK_FALLBACK" return None, "NO_MATCH" def main(): parser = argparse.ArgumentParser() parser.add_argument("--project-root", default=".") parser.add_argument("--input", required=True) parser.add_argument("--output", required=True) parser.add_argument("--manifest", required=True) parser.add_argument("--summary", required=True) parser.add_argument("--run-id", default="RUN-ANA-YS-FACT-LOCATION-003") args = parser.parse_args() project_root = Path(args.project_root).resolve() cards = read_csv(project_root / args.input) created_at = datetime.now(timezone(timedelta(hours=8))).isoformat(timespec="seconds") rows = [] for card in cards: converted_rel = card.get("converted_text_path", "") lines = read_lines(project_root / converted_rel) marks = page_marks(lines) declared_page = card.get("matched_page_no") or card.get("declared_page_no") line_no, match_method = find_line(lines, card.get("evidence_text", ""), declared_page) if line_no: context = lines[max(0, line_no - 4) : min(len(lines), line_no + 5)] location_status = "LOCATED_IN_CONVERTED_TEXT" location = f"line:{line_no}" precheck_page_no = page_for_line(marks, line_no) or declared_page location_type = "TABLE_CANDIDATE" if table_score(context) >= 2 else "PARAGRAPH_CANDIDATE" else: context = [] location_status = "NEEDS_MANUAL_SOURCE_PAGE_REVIEW" location = "" precheck_page_no = declared_page location_type = "UNLOCATED" rows.append( { "location_precheck_id": f"YS-FACT-LOC-003-{len(rows) + 1:04d}", "case_id": card.get("case_id", ""), "batch_id": card.get("batch_id", ""), "run_id": args.run_id, "evidence_card_id": card.get("evidence_card_id", ""), "fact_id": card.get("fact_id", ""), "doc_id": card.get("doc_id", ""), "converted_text_path": converted_rel, "declared_page_no": card.get("declared_page_no", ""), "matched_page_no_pass002": card.get("matched_page_no", ""), "precheck_page_no": precheck_page_no, "precheck_location": location, "match_method": match_method, "location_type_candidate": location_type, "location_status": location_status, "evidence_text": card.get("evidence_text", ""), "location_context": "\\n".join(context)[:1200], "external_cross_check_status": card.get("external_cross_check_status", ""), "review_status": "DRAFT_FOR_REVIEW", "created_at": created_at, } ) output_path = project_root / args.output fieldnames = list(rows[0].keys()) if rows else [ "location_precheck_id", "case_id", "batch_id", "run_id", "review_status", ] write_csv(output_path, fieldnames, rows) output_sha = sha256_file(output_path) manifest_path = project_root / args.manifest manifest_rows = [ { "case_id": "ANA-YS-INDUSTRY-001", "batch_id": "BATCH-001+BATCH-003", "run_id": args.run_id, "artifact_type": "key_fact_location_precheck", "artifact_path": output_path.relative_to(project_root).as_posix(), "row_count": str(len(rows)), "sha256": output_sha, "review_status": "DRAFT_FOR_REVIEW", "created_at": created_at, } ] write_csv( manifest_path, ["case_id", "batch_id", "run_id", "artifact_type", "artifact_path", "row_count", "sha256", "review_status", "created_at"], manifest_rows, ) status_counts = Counter(row["location_status"] for row in rows) type_counts = Counter(row["location_type_candidate"] for row in rows) summary_path = project_root / args.summary summary_lines = [ "# \u5173\u952e\u4e8b\u5b9e\u5b9a\u4f4d\u590d\u6838\u5305 PASS-003 \u6458\u8981", "", "\u72b6\u6001\uff1aDRAFT_FOR_REVIEW", f"\u751f\u6210\u65f6\u95f4\uff1a{created_at}", "", "## \u8f93\u51fa", "", f"- \u5b9a\u4f4d\u590d\u6838\u8868\uff1a`{output_path.relative_to(project_root).as_posix()}`", f"- manifest\uff1a`{manifest_path.relative_to(project_root).as_posix()}`", f"- \u8bb0\u5f55\u6570\uff1a{len(rows)}", f"- sha256\uff1a`{output_sha}`", "", "## \u5b9a\u4f4d\u72b6\u6001", "", "| \u72b6\u6001 | \u6570\u91cf |", "|---|---:|", ] for key, count in status_counts.most_common(): summary_lines.append(f"| {key} | {count} |") summary_lines.extend(["", "## \u5b9a\u4f4d\u7c7b\u578b\u5019\u9009", "", "| \u7c7b\u578b | \u6570\u91cf |", "|---|---:|"]) for key, count in type_counts.most_common(): summary_lines.append(f"| {key} | {count} |") summary_lines.extend( [ "", "## \u8fb9\u754c", "", "\u672c\u8f6e\u53ea\u5b8c\u6210 converted text \u5c42\u7684\u9875\u7801\u3001\u884c\u53f7\u548c\u4e0a\u4e0b\u6587\u5b9a\u4f4d\u9884\u590d\u6838\u3002\u8868\u683c\u5019\u9009\u4e0d\u7b49\u4e8e PDF \u539f\u9875\u8868\u683c\u5df2\u590d\u6838\uff1b\u6b63\u5f0f\u8bc1\u636e\u5347\u7ea7\u524d\u4ecd\u9700\u8981\u4eba\u5de5\u590d\u6838\u548c\u5916\u90e8\u4ea4\u53c9\u9a8c\u8bc1\u3002", ] ) summary_path.parent.mkdir(parents=True, exist_ok=True) summary_path.write_text("\n".join(summary_lines) + "\n", encoding="utf-8") print(f"rows={len(rows)}") print(f"located={status_counts.get('LOCATED_IN_CONVERTED_TEXT', 0)}") print(f"unlocated={status_counts.get('NEEDS_MANUAL_SOURCE_PAGE_REVIEW', 0)}") print(f"output={output_path.relative_to(project_root).as_posix()}") print(f"manifest={manifest_path.relative_to(project_root).as_posix()}") print(f"summary={summary_path.relative_to(project_root).as_posix()}") if __name__ == "__main__": main()