"""Reproducible comparison with the supplied legacy parser and ingestion benchmark. Usage: .venv/bin/python scripts/qualify_tool.py --legacy ../reference_20261117/component_trade_promax --rows 100000 All generated data is synthetic. Timings do not estimate production throughput. """ import argparse import csv import importlib.util import json import sys import time from pathlib import Path from decimal import Decimal ROOT=Path(__file__).resolve().parents[1] sys.path.insert(0,str(ROOT/'tools-service')) from toolsvc.customs.engine.rules import Rules from toolsvc.customs.engine.storage import ingest, Store def main(): parser=argparse.ArgumentParser() parser.add_argument('--legacy',type=Path,required=True) parser.add_argument('--rows',type=int,default=100000) parser.add_argument('--before',type=Path) args=parser.parse_args() out=ROOT/'.run/qualification'; out.mkdir(parents=True,exist_ok=True) folder=ROOT/'tools-service/toolsvc/customs/rules' mapping=folder/'数据映射规则.xlsx'; fields=folder/'进出口数据字段需求20260819.xlsx' sys.path.insert(0,str(args.legacy.resolve())) from trade0909.rules import Rules as LegacyRules legacy=LegacyRules(mapping, fields); current=Rules(mapping, fields) # Independent baseline: agent implementation before this change. original=args.before before=None if original and original.exists(): spec=importlib.util.spec_from_file_location('before_rules',original) mod=importlib.util.module_from_spec(spec); spec.loader.exec_module(mod) before=mod.Rules(mapping, fields) texts=[ '4|0|多层|陶瓷|MURATA|TEST-1|||', '品牌:MURATA;型号:TEST-1', '品牌名称:村田;型号规格:TEST-1', '村田(MURATA)|TEST-1', 'MURATA/TEST-1/计算机用', 'MURATA_TEST-1', '0 0 多层片式 瓷介 MURATA TEST-1', '品牌为MURATA,型号为TEST-1', '(品牌)MURATA(型号)TEST-1', '品牌|MURATA|型号|TEST-1', '品牌:MURTA;型号:TEST-1', '品牌:MURATA,SAMSUNG;型号:TEST-1', '品牌:ABACUS;型号:A5E30295495R2W', '品牌:无品牌;型号:无型号', '品牌:MURATA;型号:1/10W', '品牌:MURATA;用途:5G基站', ] records=[] keys=['brand','model','brand_country','c1','c2','c3','c4','app','mapped'] for text in texts: old=legacy.parse('8532241000',text); new=current.parse('8532241000',text) prev=before.parse('8532241000',text) if before else None records.append(dict(raw=text,legacy={k:old[k] for k in keys},current={k:new[k] for k in keys}, legacy_equal=all(old[k]==new[k] for k in keys), before_pair=[prev['brand'],prev['model']] if prev else None)) source=out/'synthetic.csv' with source.open('w',encoding='utf-8-sig',newline='') as f: w=csv.writer(f);w.writerow(['I_E_FLAG','HS_CODE','YEAR','WEEK','MONORSUM','COUNT','G_MODEL']) for i in range(args.rows): w.writerow(['I','8532241000','2025',str(i%52+1),'0.10','1',texts[i%len(texts)]]) bench=out/f'bench-{time.time_ns()}';bench.mkdir() start=time.perf_counter() db,stats=ingest(dict(input_csv=str(source),directions=['I'],start_year=2025,end_year=2026,cutoff_date='2026-12-31'),current,bench) elapsed=time.perf_counter()-start totals=Store(db).total({});db.close() assert stats['included']==args.rows assert totals['amount']==Decimal('0.1')*args.rows report=dict(comparison=records, equal_to_legacy=sum(r['legacy_equal'] for r in records), cases=len(records),before_recognized_pairs=sum(bool(r['before_pair'] and all(r['before_pair'])) for r in records), current_recognized_pairs=sum(bool(r['current']['brand'] and r['current']['model']) for r in records), synthetic_benchmark=dict(rows=args.rows,distinct_declarations=len(texts),seconds=round(elapsed,3), rows_per_second=round(args.rows/elapsed),amount=str(totals['amount']),included=stats['included'], scope='stream reading, parsing, audited SQLite ingestion, monthly aggregation and indexes; excludes full reports and uploads')) (out/'report.json').write_text(json.dumps(report,ensure_ascii=False,indent=2)) print(json.dumps({k:v for k,v in report.items() if k!='comparison'},ensure_ascii=False,indent=2)) assert report['equal_to_legacy']==len(records) if __name__=='__main__':main()