看股功能更新

2026-08-16 00:05:26 +08:00
parent 9cce670b74
commit fc86fe0674
28 changed files with 3823 additions and 96 deletions
--- a/backend/scripts/backfill_adj_factor.py
+++ b/backend/scripts/backfill_adj_factor.py
@@ -0,0 +1,122 @@
+"""全量回补历史复权因子（adj_factor 表）。
+
+用法（在 backend 目录下）：
+    uv run python scripts/backfill_adj_factor.py                # 从 candles 最早日期回补到今天
+    uv run python scripts/backfill_adj_factor.py --start 20180101
+    uv run python scripts/backfill_adj_factor.py --force        # 已有日期也重拉
+
+- 按交易日逐日拉取全市场因子（pro.adj_factor(trade_date=...)），幂等可断点续跑；
+- 交易日取自本地 trade_calendar（缓存不到的区间自动刷新一次日历）；
+- Tushare 每分钟限频由 _call_retry 自动等待 62s 重试。
+"""
+from __future__ import annotations
+
+import argparse
+import asyncio
+import sys
+import time
+from datetime import datetime
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from sqlalchemy import delete, func, insert, select
+
+from app.db import async_session
+from app.models import AdjFactor, Candle, TradeCalendar
+from app.screener.market_sync import _call_retry, _get_pro, _norm_date, _parse_d
+
+_INTERVAL_MSG = 20  # 每完成 N 个交易日打印一次进度
+
+
+async def _calendar_dates(start: str, end: str) -> list[str]:
+    """[start, end] 交易日（升序）。本地日历覆盖不足时直接拉宽范围日历并回写缓存。"""
+    async with async_session() as session:
+        all_cached = set((await session.execute(select(TradeCalendar.trade_date))).scalars().all())
+    cached = sorted(d for d in all_cached if start <= d <= end)
+    if cached and min(cached) <= start:
+        return cached
+
+    # 覆盖不到起点：按需拉宽范围日历（trade_cal 低积分限频 1 次/小时，失败沿用缓存）
+    pro = _get_pro()
+    try:
+        cal = await asyncio.to_thread(
+            _call_retry, pro.trade_cal, exchange="SSE", start_date=start, end_date=end, is_open="1"
+        )
+        dates = sorted(cal["cal_date"].tolist())
+    except Exception as e:  # noqa: BLE001
+        if not cached:
+            raise
+        print(f"交易日历拉取受限（{str(e)[:100]}），沿用本地缓存")
+        return cached
+    fresh = [d for d in dates if d not in all_cached]
+    if fresh:
+        async with async_session() as session:
+            await session.execute(insert(TradeCalendar), [{"trade_date": d} for d in fresh])
+            await session.commit()
+    return dates
+
+
+async def _existing_dates() -> set[str]:
+    async with async_session() as session:
+        res = await session.execute(select(func.distinct(AdjFactor.trade_date)))
+        return {_norm_date(r[0]) for r in res}
+
+
+async def main(start: str, end: str, force: bool) -> None:
+    # 默认起点：candles 最早日线（因子只需覆盖有 K 线的区间）
+    if start is None:
+        async with async_session() as session:
+            first = await session.scalar(select(func.min(Candle.ts)).where(Candle.timeframe == "1d"))
+        start = first.strftime("%Y%m%d") if first else "20050101"
+    if end is None:
+        end = datetime.now().strftime("%Y%m%d")
+
+    dates = await _calendar_dates(start, end)
+    have = set() if force else await _existing_dates()
+    todo = [d for d in dates if d not in have]
+    print(f"区间 {start}~{end} 共 {len(dates)} 个交易日，待回补 {len(todo)} 个（已有 {len(dates) - len(todo)}）")
+    if not todo:
+        return
+
+    pro = _get_pro()
+    done = 0
+    for d in todo:
+        time.sleep(0.15)  # 轻微控频；分钟级限频由 _call_retry 自动等待重试
+        df = None
+        for attempt in range(5):  # 网络抖动（超时/断连）也重试，_call_retry 只兜限频
+            try:
+                df = _call_retry(pro.adj_factor, trade_date=d)  # noqa: 线性脚本直接同步调用
+                break
+            except Exception as e:  # noqa: BLE001
+                wait = min(30 * (attempt + 1), 120)
+                print(f"  {d} 拉取异常（{str(e)[:80]}），{wait}s 后重试 {attempt + 1}/5")
+                time.sleep(wait)
+        if df is None:
+            print(f"  {d} 连续 5 次失败，跳过（断点续跑可补）")
+            continue
+        if df is None or df.empty:
+            print(f"  {d} 无数据（非交易日或未生成），跳过")
+            continue
+        rows = [
+            {"trade_date": _parse_d(d), "ts_code": r["ts_code"], "adj_factor": float(r["adj_factor"])}
+            for _, r in df.iterrows()
+        ]
+        async with async_session() as session:
+            dt = _parse_d(d)
+            await session.execute(delete(AdjFactor).where(AdjFactor.trade_date == dt))
+            await session.execute(insert(AdjFactor), rows)
+            await session.commit()
+        done += 1
+        if done % _INTERVAL_MSG == 0 or done == len(todo):
+            print(f"  进度 {done}/{len(todo)}（{d}，+{len(rows)} 行）")
+    print(f"回补完成：{done} 个交易日")
+
+
+if __name__ == "__main__":
+    ap = argparse.ArgumentParser(description="全量回补历史复权因子")
+    ap.add_argument("--start", default=None, help="YYYYMMDD，默认 candles 最早日期")
+    ap.add_argument("--end", default=None, help="YYYYMMDD，默认今天")
+    ap.add_argument("--force", action="store_true", help="已有日期也重拉")
+    a = ap.parse_args()
+    asyncio.run(main(a.start, a.end, a.force))
--- a/backend/scripts/backfill_turnover.py
+++ b/backend/scripts/backfill_turnover.py
@@ -0,0 +1,162 @@
+"""全量回补换手率（candles.turnover，单位 %）。
+
+用法（在 backend 目录下）：
+    uv run python scripts/backfill_turnover.py               # 从 2000-01-01（daily_basic 起点）回补到今天
+    uv run python scripts/backfill_turnover.py --start 20200101
+    uv run python scripts/backfill_turnover.py --force       # 已回补的交易日也重拉
+
+- 数据源：Tushare daily_basic(trade_date=..., fields='ts_code,turnover_rate')，按日全市场；
+- 幂等可断点续跑：某交易日 candles 已有非空 turnover 即跳过（--force 强制重做）；
+- 交易日取自本地 trade_calendar（缓存覆盖不到起点时自动拉一次宽范围日历）；
+- 每日一条 UPDATE ... FROM unnest(...) 批量写回，仅更新 turnover 列；
+- Tushare 每分钟限频由 _call_retry 自动等待 62s 重试。
+
+注意：与 import_tdx_day.py（回填 amount 会整行 upsert）串行运行，避免同表行锁竞争。
+"""
+from __future__ import annotations
+
+import argparse
+import asyncio
+import sys
+import time
+from datetime import datetime
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from app.screener.market_sync import _call_retry, _get_pro
+
+import asyncpg
+
+
+def load_db_url() -> str:
+    """与 import_tdx_day.py 相同的 .env -> libpq URL 解析（本地复制避免跨脚本导入）。"""
+    env = Path(__file__).resolve().parent.parent / ".env"
+    if env.exists():
+        for line in env.read_text(encoding="utf-8").splitlines():
+            line = line.strip()
+            if line.startswith("DATABASE_URL=postgresql+asyncpg://"):
+                return "postgresql://" + line.split("://", 1)[1]
+    return "postgresql://postgres:postgres@localhost:5432/stock"
+
+_DAILY_BASIC_FLOOR = "20000101"  # daily_basic 最早覆盖 2000-01-04，更早的交易日无换手数据
+_INTERVAL_MSG = 20
+
+
+async def _calendar_dates(conn: asyncpg.Connection, start: str, end: str) -> list[str]:
+    """[start, end] 交易日（升序）。本地缓存覆盖不到起点时拉一次宽范围日历并回写。"""
+    cached = [r[0] for r in await conn.fetch(
+        "SELECT trade_date FROM trade_calendar WHERE trade_date >= $1 AND trade_date <= $2 "
+        "ORDER BY trade_date", start, end)]
+    if cached and cached[0] <= start:
+        return cached
+
+    pro = _get_pro()
+    try:
+        cal = await asyncio.to_thread(
+            _call_retry, pro.trade_cal, exchange="SSE", start_date=start, end_date=end, is_open="1"
+        )
+        dates = sorted(cal["cal_date"].tolist())
+    except Exception as e:  # noqa: BLE001
+        if not cached:
+            raise
+        print(f"交易日历拉取受限（{str(e)[:100]}），沿用本地缓存")
+        return cached
+    have = set(cached)
+    fresh = [d for d in dates if d not in have]
+    if fresh:
+        await conn.executemany(
+            "INSERT INTO trade_calendar (trade_date) VALUES ($1) ON CONFLICT DO NOTHING", [(d,) for d in fresh]
+        )
+    return dates
+
+
+async def _day_status(conn: asyncpg.Connection, d: str) -> tuple[int, int]:
+    """(已有换手的行数, 当日总行数)。无行情的日子 total=0 直接跳过。"""
+    row = await conn.fetchrow(
+        "SELECT count(*) FILTER (WHERE turnover IS NOT NULL) AS done, count(*) AS total "
+        "FROM candles WHERE timeframe = '1d' AND ts = $1::timestamp", datetime.strptime(d, "%Y%m%d")
+    )
+    return row["done"], row["total"]
+
+
+async def main(start: str, end: str, force: bool) -> None:
+    conn = await asyncpg.connect(load_db_url())
+    try:
+        # 默认起点：daily_basic 覆盖范围与 candles 最早日线的较大者（更早的日期拉了也是空）
+        if start is None:
+            first = await conn.fetchval(
+                "SELECT min(ts) FROM candles WHERE timeframe = '1d' AND symbol <> 'DEMO'")
+            start = max(first.strftime("%Y%m%d"), _DAILY_BASIC_FLOOR) if first else _DAILY_BASIC_FLOOR
+        if end is None:
+            end = datetime.now().strftime("%Y%m%d")
+
+        dates = await _calendar_dates(conn, start, end)
+        todo: list[str] = []
+        for d in dates:
+            if force:
+                done, total = await _day_status(conn, d)
+                if total:
+                    todo.append(d)
+                continue
+            done, total = await _day_status(conn, d)
+            if total and done < total // 2:  # 过半缺换手才重做（容忍个别股票无快照）
+                todo.append(d)
+        print(f"区间 {start}~{end} 共 {len(dates)} 个交易日，待回补 {len(todo)} 个")
+
+        pro = _get_pro()
+        done = 0
+        t0 = time.time()
+        for d in todo:
+            time.sleep(0.15)  # 轻微控频；分钟级限频由 _call_retry 自动等待重试
+            df = None
+            for attempt in range(5):  # 网络抖动（超时/断连）也重试，_call_retry 只兜限频
+                try:
+                    df = _call_retry(
+                        pro.daily_basic, trade_date=d, fields="ts_code,trade_date,turnover_rate"
+                    )
+                    break
+                except Exception as e:  # noqa: BLE001
+                    wait = min(30 * (attempt + 1), 120)
+                    print(f"  {d} 拉取异常（{str(e)[:80]}），{wait}s 后重试 {attempt + 1}/5")
+                    time.sleep(wait)
+            if df is None:
+                print(f"  {d} 连续 5 次失败，跳过（断点续跑可补）")
+                continue
+            if df.empty:
+                continue
+
+            syms: list[str] = []
+            vals: list[float] = []
+            for _, r in df.iterrows():
+                tr = r["turnover_rate"]
+                if tr is None or tr != tr:  # None / NaN
+                    continue
+                syms.append(str(r["ts_code"]).split(".")[0])
+                vals.append(float(tr))
+            if not syms:
+                continue
+            n = await conn.execute(
+                "UPDATE candles AS c SET turnover = v.t "
+                "FROM unnest($1::text[], $2::float8[]) AS v(sym, t) "
+                "WHERE c.symbol = v.sym AND c.timeframe = '1d' AND c.ts = $3::timestamp",
+                syms, vals, datetime.strptime(d, "%Y%m%d"),
+            )
+            done += 1
+            if done % _INTERVAL_MSG == 0 or done == len(todo):
+                elapsed = time.time() - t0
+                eta = elapsed / done * (len(todo) - done) if done else 0
+                print(f"  进度 {done}/{len(todo)}（{d}，{len(syms)} 只，{n}），"
+                      f"{elapsed:.0f}s 已用，预计还需 {eta/60:.0f}m")
+        print(f"回补完成：{done} 个交易日")
+    finally:
+        await conn.close()
+
+
+if __name__ == "__main__":
+    ap = argparse.ArgumentParser(description="全量回补换手率 candles.turnover")
+    ap.add_argument("--start", default=None, help="YYYYMMDD，默认 max(candles 最早, 20000101)")
+    ap.add_argument("--end", default=None, help="YYYYMMDD，默认今天")
+    ap.add_argument("--force", action="store_true", help="已有换手的交易日也重拉")
+    a = ap.parse_args()
+    asyncio.run(main(a.start, a.end, a.force))
--- a/backend/scripts/import_tdx_day.py
+++ b/backend/scripts/import_tdx_day.py
@@ -0,0 +1,160 @@
+"""通达信「沪深京日线数据完整包」全量导入 candles 表。
+
+用法（在 backend 目录下）：
+    uv run python scripts/import_tdx_day.py C:/Users/cirry/Downloads/hsjday [symbol ...]
+    # symbol 为可选的 6 位代码过滤（如 000001 002671），只重导这些标的
+    uv run python scripts/import_tdx_day.py <目录> --no-clear
+    # --no-clear：不清空任何行，纯 upsert（用于给已导入的底座回补 amount 成交额）
+
+- 解析 vipdoc 的 .day 二进制文件（每条 32 字节）：
+    日期(YYYYMMDD) 开 高 低 收（×100） 成交额(元, float32) 成交量(股) 保留
+- 只导入 stock_basic 里登记的股票（自动排除指数/基金/可转债/回购）；
+  sh000001(上证指数) 与 sz000001(平安银行) 这类代码冲突也由此化解。
+- 价格为**不复权**：全量模式导入前清空已有的非 DEMO 行情；指定 symbol 过滤时
+  只清空这些标的（用于修复被复权口径污染的个别股票），其余不动。
+- amount 为 TDX 原生 float32（元），精度 ~6 位有效数字，展示用途足够；
+  ON CONFLICT 时仅更新 amount 列，不动 OHLCV/turnover（避免与换手率回补互相干扰）。
+- 写入用 asyncpg execute_many + ON CONFLICT DO UPDATE，可重复执行（幂等）。
+"""
+from __future__ import annotations
+
+import argparse
+import asyncio
+import struct
+import sys
+import time
+from pathlib import Path
+
+import asyncpg
+
+# .env 里的 DATABASE_URL 是 SQLAlchemy 格式，asyncpg 需要 libpq 格式
+DEFAULT_URL = "postgresql://postgres:postgres@localhost:5432/stock"
+BATCH = 20_000  # 每批 upsert 行数
+
+
+def load_db_url() -> str:
+    env = Path(__file__).resolve().parent.parent / ".env"
+    if env.exists():
+        for line in env.read_text(encoding="utf-8").splitlines():
+            line = line.strip()
+            if line.startswith("DATABASE_URL=postgresql+asyncpg://"):
+                return "postgresql://" + line.split("://", 1)[1]
+    return DEFAULT_URL
+
+
+def parse_day_file(path: Path) -> list[tuple[int, float, float, float, float, float, float]]:
+    """解析单个 .day 文件 -> [(date, open, high, low, close, volume(股), amount(元)), ...]"""
+    raw = path.read_bytes()
+    unpack = struct.Struct("<IIIIIfII").unpack_from
+    out = []
+    for i in range(len(raw) // 32):
+        date, o, h, l, c, amount, vol, _reserved = unpack(raw, i * 32)
+        out.append((date, o / 100.0, h / 100.0, l / 100.0, c / 100.0, float(vol), float(amount)))
+    return out
+
+
+async def main(root: Path, symbols: list[str] | None = None, no_clear: bool = False) -> None:
+    if not root.exists():
+        sys.exit(f"目录不存在: {root}")
+
+    conn = await asyncpg.connect(load_db_url())
+    try:
+        # 股票清单：ts_code 形如 000001.SZ，用于过滤指数/基金/转债
+        rows = await conn.fetch("SELECT ts_code, symbol FROM stock_basic WHERE list_status = 'L'")
+        by_exchange: dict[str, set[str]] = {"sh": set(), "sz": set(), "bj": set()}
+        for r in rows:
+            suffix = r["ts_code"].split(".")[-1].lower()  # SH/SZ/BJ -> sh/sz/bj
+            if suffix in by_exchange:
+                by_exchange[suffix].add(r["symbol"])
+        print(f"stock_basic 在市股票: " + ", ".join(f"{k}={len(v)}" for k, v in by_exchange.items()))
+
+        files = sorted(root.glob("*/lday/*.day"))
+        print(f"发现 .day 文件: {len(files)} 个")
+
+        if no_clear:
+            print("--no-clear：不清空任何行，纯 upsert 回补 amount")
+        elif symbols:
+            # 清空旧行情（保留 DEMO 合成数据），避免 qfq/不复权混用；
+            # 带 symbol 过滤时只清空目标标的（修复个别被污染的股票，不动其余底座）
+            deleted = await conn.execute(
+                "DELETE FROM candles WHERE symbol = ANY($1)", symbols
+            )
+            print(f"清空目标标的 {symbols}: {deleted}")
+            keep = set(symbols)
+            files = [p for p in files if p.name[2:8] in keep]
+            print(f"过滤后待导入 .day 文件: {len(files)} 个")
+        else:
+            deleted = await conn.execute("DELETE FROM candles WHERE symbol <> 'DEMO'")
+            print(f"清空旧行情: {deleted}")
+
+        if no_clear:
+            # 回填模式：只写 amount，不动 OHLCV/turnover（底座已就位，避免全表重写）
+            upsert_sql = """
+                INSERT INTO candles (symbol, timeframe, ts, open, high, low, close, volume, amount)
+                VALUES ($1, '1d', to_timestamp($2::text, 'YYYYMMDD')::timestamp, $3, $4, $5, $6, $7, $8)
+                ON CONFLICT (symbol, timeframe, ts) DO UPDATE
+                SET amount = EXCLUDED.amount
+            """
+        else:
+            upsert_sql = """
+                INSERT INTO candles (symbol, timeframe, ts, open, high, low, close, volume, amount)
+                VALUES ($1, '1d', to_timestamp($2::text, 'YYYYMMDD')::timestamp, $3, $4, $5, $6, $7, $8)
+                ON CONFLICT (symbol, timeframe, ts) DO UPDATE
+                SET open = EXCLUDED.open, high = EXCLUDED.high, low = EXCLUDED.low,
+                    close = EXCLUDED.close, volume = EXCLUDED.volume, amount = EXCLUDED.amount
+            """
+
+        t0 = time.time()
+        total_stocks = 0
+        skipped = 0
+        batch: list[tuple] = []
+
+        rows_done = 0
+
+        async def flush() -> None:
+            nonlocal batch, rows_done
+            if batch:
+                await conn.executemany(upsert_sql, batch)
+                rows_done += len(batch)
+                batch = []
+
+        for n, path in enumerate(files, 1):
+            market = path.name[:2].lower()  # sh / sz / bj
+            code = path.name[2:8]
+            if code not in by_exchange.get(market, set()):
+                skipped += 1
+                continue
+            for date, o, h, l, c, v, amount in parse_day_file(path):
+                batch.append((code, str(date), o, h, l, c, v, amount))
+            total_stocks += 1
+            if len(batch) >= BATCH:
+                await flush()
+            if n % 500 == 0:
+                elapsed = time.time() - t0
+                print(f"  进度 {n}/{len(files)} 文件, 已入库 {total_stocks} 只股票, "
+                      f"{rows_done + len(batch):,} 行, {elapsed:.0f}s")
+        await flush()
+
+        cnt = await conn.fetchval("SELECT count(*) FROM candles WHERE symbol <> 'DEMO'")
+        span = await conn.fetchrow(
+            "SELECT min(ts) AS lo, max(ts) AS hi FROM candles WHERE symbol <> 'DEMO'"
+        )
+        with_amt = await conn.fetchval(
+            "SELECT count(*) FROM candles WHERE symbol <> 'DEMO' AND amount IS NOT NULL"
+        )
+        print(f"\n完成: {total_stocks} 只股票, {cnt:,} 行日线, "
+              f"范围 {span['lo']:%Y-%m-%d} ~ {span['hi']:%Y-%m-%d}, "
+              f"含成交额 {with_amt:,} 行, "
+              f"跳过非股票文件 {skipped} 个, 耗时 {time.time() - t0:.0f}s")
+    finally:
+        await conn.close()
+
+
+if __name__ == "__main__":
+    ap = argparse.ArgumentParser(description="TDX 沪深京日线全量导入 candles")
+    ap.add_argument("root", help="hsjday 目录（其下 */lday/*.day）")
+    ap.add_argument("symbols", nargs="*", help="可选的 6 位代码过滤")
+    ap.add_argument("--no-clear", action="store_true",
+                    help="不清空任何行，纯 upsert（amount 回补模式）")
+    a = ap.parse_args()
+    asyncio.run(main(Path(a.root), a.symbols or None, a.no_clear))
--- a/backend/scripts/test_trades_parser.py
+++ b/backend/scripts/test_trades_parser.py
@@ -0,0 +1,113 @@
+"""交割单解析器离线自测：不碰数据库，直接调 app.trades.parse_statement。
+
+覆盖四类真实导出格式 + 边界行（转账/配号/利息跳过、费用合计列去重、日期多格式）。
+运行：uv run python scripts/test_trades_parser.py
+"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from app.trades import parse_statement  # noqa: E402
+
+FAIL: list[str] = []
+
+
+def check(name: str, cond: bool, detail: str = "") -> None:
+    mark = "ok " if cond else "FAIL"
+    print(f"[{mark}] {name}{(' — ' + detail) if detail and not cond else ''}")
+    if not cond:
+        FAIL.append(name)
+
+
+# ---------- 1) 通达信式：GBK + 制表符 + 标题行在前 ----------
+tdx = (
+    "交割单\n"
+    "股东账号: A123456789    起始日期: 20240102    终止日期: 20240105    币种: 人民币\n"
+    "\t交割日期\t业务名称\t证券代码\t证券名称\t成交价格\t成交数量\t成交金额\t手续费\t印花税\t过户费\t发生金额\t资金余额\t合同号\n"
+    "\t20240102\t证券买入\t600519\t贵州茅台\t1680.00\t100\t168000.00\t5.00\t0.00\t1.68\t-168006.68\t200000.00\t1000001\n"
+    "\t20240102\t银行转存\t\t\t\t\t\t\t\t\t50000.00\t250000.00\t\n"
+    "\t20240103\t证券卖出\t600519\t贵州茅台\t1700.50\t100\t170050.00\t5.00\t170.05\t1.70\t169873.25\t419873.25\t1000002\n"
+    "\t20240105\t利息归本\t\t\t\t\t\t\t\t\t1.25\t419874.50\t\n"
+)
+r = parse_statement(tdx.encode("gbk"), "交割单.txt")
+check("tdx: 2 笔成交", len(r.trades) == 2, f"got {len(r.trades)}")
+check("tdx: 跳过 2 行非交易", r.skipped_other == 2, f"got {r.skipped_other}")
+t0, t1 = r.trades[0], r.trades[1]
+check("tdx: 日期/代码/后缀", (t0.trade_date.isoformat(), t0.ts_code) == ("2024-01-02", "600519.SH"), f"{t0.trade_date} {t0.ts_code}")
+check("tdx: 买入方向+费用合计", t0.direction == "buy" and abs(t0.fee - 6.68) < 1e-9, f"{t0.direction} fee={t0.fee}")
+check("tdx: 卖出费用含印花税", t1.direction == "sell" and abs(t1.fee - 176.75) < 1e-9, f"fee={t1.fee}")
+check("tdx: 金额取绝对值", t0.amount == 168000.0, f"amount={t0.amount}")
+
+# ---------- 2) 恒生柜台式：UTF-8 CSV，交收日期/交易类别/费用合计 ----------
+hs = (
+    "序号,交收日期,证券代码,证券名称,交易类别,成交价格,成交数量,证券余额,成交金额,资金发生数,资金余额,流水序号,业务标志,业务名称,发生金额,后资金额,货币类别,费用合计,净佣金,规费,印花税,过户费,合同号\n"
+    "1,2024-06-07,000858,五粮液,证券买入,132.50,200,200,26500.00,-26505.80,73494.20,1,0101,证券买入,-26505.80,73494.20,人民币,5.80,4.20,1.60,0.00,0.00,66778001\n"
+    "2,2024-06-07,,,\t,,,,5120.00,78614.20,2,2041,银行转存,5120.00,78614.20,人民币,0,0,0,0,0,\n"
+    "3,2024-06-10,000858,五粮液,证券卖出,135.00,200,0,27000.00,26975.30,105589.50,3,0102,证券卖出,26975.30,105589.50,人民币,24.70,4.20,1.60,18.90,0.00,66779001\n"
+)
+r2 = parse_statement(hs.encode("utf-8"), "hsi.csv")
+check("hs: 2 笔成交", len(r2.trades) == 2, f"got {len(r2.trades)}")
+check("hs: 费用合计不重复累加", abs(r2.trades[1].fee - 24.70) < 1e-9, f"fee={r2.trades[1].fee}")
+check("hs: 深市后缀", r2.trades[0].ts_code == "000858.SZ", r2.trades[0].ts_code)
+check("hs: 日期 YYYY-MM-DD", r2.trades[0].trade_date.isoformat() == "2024-06-07")
+
+# ---------- 3) HTML 伪 .xls（同花顺导出常见真身） ----------
+html = """<html><head><meta charset="gbk"></head><body>
+<table>
+<tr><td>客户姓名</td><td>测试</td></tr>
+<tr><td>成交日期</td><td>业务名称</td><td>证券代码</td><td>证券名称</td><td>成交价格</td><td>成交数量</td><td>成交金额</td><td>手续费</td></tr>
+<tr><td>2024/03/15</td><td>证券买入</td><td>300750</td><td>宁德时代</td><td>182.30</td><td>300</td><td>54,690.00</td><td>16.41</td></tr>
+<tr><td>2024/03/18</td><td>证券卖出</td><td>300750</td><td>宁德时代</td><td>185.00</td><td>300</td><td>55,500.00</td><td>5.55</td></tr>
+</table></body></html>"""
+r3 = parse_statement(html.encode("gbk"), "jiaogedan.xls")
+check("html: 2 笔成交", len(r3.trades) == 2, f"got {len(r3.trades)}")
+check("html: 千分位金额", r3.trades[0].amount == 54690.0, f"{r3.trades[0].amount}")
+check("html: 创业板后缀", r3.trades[0].ts_code == "300750.SZ", r3.trades[0].ts_code)
+check("html: 斜杠日期", r3.trades[1].trade_date.isoformat() == "2024-03-18")
+
+# ---------- 4) 无业务名称列：发生金额正负判方向（招商式） ----------
+zh = (
+    "证券名称,成交日期,成交价格,成交数量,发生金额,资金余额,合同编号\n"
+    "贵州茅台,20240102,1680.00,100,-168005.00,200000.00,SZ1000001\n"
+    "贵州茅台,20240103,1700.50,100,170049.50,370049.50,SZ1000002\n"
+)
+r4 = parse_statement(zh.encode("utf-8"), "zszs.csv")
+check("sign: 2 笔成交", len(r4.trades) == 2, f"got {len(r4.trades)}")
+check("sign: 负金额=买入", (r4.trades[0].direction, r4.trades[1].direction) == ("buy", "sell"),
+      f"{r4.trades[0].direction}/{r4.trades[1].direction}")
+
+# ---------- 5) xlsx（openpyxl 内存构造） ----------
+import io  # noqa: E402
+from openpyxl import Workbook  # noqa: E402
+
+wb = Workbook()
+ws = wb.active
+ws.append(["对账单", None, None])
+ws.append(["成交日期", "业务名称", "证券代码", "证券名称", "成交均价", "成交股数", "成交金额", "佣金", "过户费"])
+from datetime import datetime as dt  # noqa: E402
+ws.append([dt(2024, 2, 28, 14, 35, 0), "证券买入", "688981", "中芯国际", 52.80, 200, 10560.00, 2.50, 1.06])
+ws.append([dt(2024, 3, 1, 9, 31, 0), "证券卖出", "688981", "中芯国际", 54.10, 200, 10820.00, 2.50, 1.06])
+buf = io.BytesIO()
+wb.save(buf)
+r5 = parse_statement(buf.getvalue(), "sm.xlsx")
+check("xlsx: 2 笔成交", len(r5.trades) == 2, f"got {len(r5.trades)}")
+check("xlsx: datetime 日期", r5.trades[0].trade_date.isoformat() == "2024-02-28")
+check("xlsx: 科创板后缀", r5.trades[0].ts_code == "688981.SH", r5.trades[0].ts_code)
+check("xlsx: 佣金+过户费", abs(r5.trades[0].fee - 3.56) < 1e-9, f"fee={r5.trades[0].fee}")
+
+# ---------- 6) 错误分支 ----------
+from fastapi import HTTPException  # noqa: E402
+try:
+    parse_statement("随便一串不是交割单的文字,1,2,3".encode("utf-8"), "x.csv")
+    check("garbage: 应 422", False)
+except HTTPException as e:
+    check("garbage: 422", e.status_code == 422)
+
+print()
+if FAIL:
+    print(f"FAIL {len(FAIL)}: {FAIL}")
+    sys.exit(1)
+print("PASS: 交割单解析器全部用例通过")