feat(data): Tushare 限速退避 + 新浪兜底（财务 getFinanceReport2022 / 日K 前复权），source+adjust 口径标记

- TushareProvider：频率超限按指数退避重试（不再一次 200/min 即中断），最长等待 30s
- SinaProvider 重构（参考 cc-cursor 公开接口实现）：
  · 新增财务通道 CompanyFinanceService.getFinanceReport2022(source=gjzb) → FinancialIndicator
    （report_date / announce_date=publish_date），与 Tushare fina_indicator schema 一致
  · 日 K 保留 jsonp（前复权），统一 UA + 重试
  · 不支持方法仍抛 DataSourceNotSupported（复权因子/交易日历/基础信息）
- FailoverProvider 现在可对 daily 与 financial 兜底（CLI _failover_provider 接 SinaProvider）
- DailyBar + stock_daily 表新增 source/adjust 列：新浪兜底行标记 sina/qfq，
  Tushare 恢复后 --resume 按同键覆盖回不复权 → 两源格式一致且可追溯
- 迁移 91c4e27a03fb 已生成；执行需在全市场同步结束后：uv run alembic upgrade head
- 测试 34+ 项（新浪财务解析/格式一致/限速退避等）通过

backend/app/cli/sync.py

@@ -44,10 +44,17 @@ def _parse_day(text: str) -> date:
 
 
 def _failover_provider(session):
-    """Tushare 单源经 FailoverProvider 包装：每次尝试写 sync_log（AGENT.md §7 审计）。"""
+    """Tushare 首选 + 新浪兜底（FailoverProvider 每次尝试写 sync_log，AGENT.md §7）。
+
+    能力矩阵：新浪仅提供日线（前复权，标记 source=sina/adjust=qfq）；其余接口
+    （adj_factor/financial/basic/calendar）新浪不支持 → 抛错保留单源语义，日志可见。
+    Tushare 恢复后重跑 --resume 会把新浪兜底日按同键覆盖回不复权口径。
+    """
+    from app.infrastructure.data_sources.sina import SinaProvider
+
     audit_repo = SqlAlchemySyncLogRepository(session)
     primary = TushareProvider(token=get_settings().tushare_token)
-    return FailoverProvider(primary, fallback=None, audit=audit_repo.add)
+    return FailoverProvider(primary, fallback=SinaProvider(), audit=audit_repo.add)
 
 
 def _session_ctx():

backend/app/domain/entities/market.py

@@ -42,10 +42,18 @@ class TradingCalendar(BaseModel):
 
 
 class DailyBar(BaseModel):
-    """不复权日线。复权请使用 AdjustFactor 在消费侧显式计算。"""
+    """日线。默认不复权（source=tushare, adjust=none）。
+
+    备用源兜底行会标记 source=sina、adjust=qfq（新浪返回前复权价）。
+    字段统一、可区分、可追溯（AGENT §5.2/§8）：研究侧应优先消费
+    source=tushare 且 adjust=none 的行；新浪行仅在 Tushare 不可用期间作为兜底，
+    Tushare 恢复后重跑 --resume 会按日覆盖回不复权口径。
+    """
 
     symbol: str
     trade_date: date
+    source: str = Field(default="tushare", description="tushare | sina")
+    adjust: str = Field(default="none", description="none 不复权 | qfq 前复权")
     open: Decimal | None = None
     high: Decimal | None = None
     low: Decimal | None = None

backend/app/infrastructure/data_sources/sina.py

@@ -1,39 +1,71 @@
 """新浪财经 Provider —— 备用数据源。
 
-能力边界（AGENT.md §5.2）：
+通道（公开接口方案参考 cc-cursor/finance/data/sources/sina_source.py）：
-- 新浪日 K 接口返回**前复权**数据，口径与 Tushare 不复权不同，
+1. 财务：quotes.sina.cn CompanyFinanceService.getFinanceReport2022（source=gjzb，
-  因此本 Provider 只用于「缺失/不可用时的行情参考与交叉验证」，
+   匿名免费、一次多期），含披露日 publish_date → FinancialIndicator
-  不得把结果直接并入不复权主时序库（禁止静默混口径）。
+   （symbol / report_date=end_date / announce_date=publish_date），schema 与
-- 新浪不提供复权因子 / 财务指标 → 相应方法抛 DataSourceNotSupported。
+   Tushare fina_indicator 一致 —— 用于财务兜底（保留防未来函数所需的公告日）。
+2. 日 K：quotes.sina.cn getKLineData（jsonp，**前复权**）。新浪无「不复权 + 独立复权
+   因子」，因此日线兜底行标记 source=sina、adjust=qfq，与主口径区分；Tushare 恢复
+   后 --resume 会按日覆盖回不复权行。
+
+能力边界（其余接口新浪不支持 → DataSourceNotSupported）：
+get_stock_basic / get_trade_cal / get_adjust_factor。
 """
 
 from __future__ import annotations
 
 import json
 import re
+import time
 import urllib.parse
 import urllib.request
 from datetime import date, datetime
 from decimal import Decimal
 from typing import Any
 
-from app.domain.entities.market import DailyBar
+from app.domain.entities.market import DailyBar, FinancialIndicator
 from app.infrastructure.data_sources.errors import (
     DataSourceError,
     DataSourceNotSupported,
 )
 
+_UA = (
+    "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
+    "(KHTML, like Gecko) Chrome/138.0.0.0 Safari/537.36"
+)
 _KLINE_JSONP = (
     "https://quotes.sina.cn/cn/api/jsonp_v2.php/var%20data=/CN_MarketDataService"
     ".getKLineData?symbol={sina_symbol}&scale=240&ma=no&datalen={datalen}"
 )
+_FIN_BASE = "https://quotes.sina.cn/cn/api/openapi.php/CompanyFinanceService.getFinanceReport2022"
+
+# 新浪「关键指标」中文项名 → 本项目 FinancialIndicator 字段（None 表示已具备/忽略）
+_FIN_FIELD_MAP = {
+    "基本每股收益": "eps",
+    "净资产收益率(ROE)": "roe",
+    "加权净资产收益率": "roe",
+    "销售毛利率": "gross_margin",
+    "毛利率": "gross_margin",
+    "营业总收入": "total_revenue",
+    "净利润": "net_profit",
+}
 
 
 def _to_sina_symbol(symbol: str) -> str:
-    """600519.SH -> sh600519；000001.SZ -> sz000001。"""
+    """600519.SH -> sh600519；000001.SZ -> sz000001；无后缀时按规则猜测。"""
-    code, _, suffix = symbol.partition(".")
+    code = symbol.strip().upper()
-    prefix = {"SH": "sh", "SZ": "sz", "BJ": "bj"}.get(suffix.upper(), "sh")
+    if code.endswith(".SH"):
-    return f"{prefix}{code}"
+        return "sh" + code[:-3]
+    if code.endswith(".SZ"):
+        return "sz" + code[:-3]
+    if code.endswith(".BJ"):
+        return "bj" + code[:-3]
+    if code.startswith(("6", "9")):
+        return "sh" + code
+    if code.startswith(("4", "8")):
+        return "bj" + code
+    return "sz" + code
 
 
 def _extract_jsonp(payload: str) -> list[dict[str, Any]]:
@@ -43,23 +75,110 @@ def _extract_jsonp(payload: str) -> list[dict[str, Any]]:
     return json.loads(match.group(1))
 
 
+def _d(value) -> Decimal | None:
+    if value is None:
+        return None
+    try:
+        return Decimal(str(value))
+    except (ValueError, TypeError):
+        return None
+
+
+def _to_date(value: str) -> date:
+    """兼容 20240831 / 2024-08-31 等格式。"""
+    digits = re.sub(r"\D", "", str(value))[:8]
+    return datetime.strptime(digits, "%Y%m%d").date()
+
+
 class SinaProvider:
-    """新浪财经备用数据源（仅日线参考 / 交叉验证）。"""
+    """新浪财经备用数据源：财务（与 Tushare schema 一致）+ 日线（前复权）。"""
 
     name = "sina"
 
-    def __init__(self, *, timeout: float = 10.0, urlopen=urllib.request.urlopen) -> None:
+    def __init__(
+        self,
+        *,
+        timeout: float = 10.0,
+        retries: int = 2,
+        urlopen=urllib.request.urlopen,
+    ) -> None:
         self._timeout = timeout
+        self._retries = retries
         self._urlopen = urlopen
 
+    # ---- HTTP（统一 UA / 重试） ----
+
+    def _open(self, url: str) -> bytes:
+        req = urllib.request.Request(url, headers={"User-Agent": _UA})
+        last_error: Exception | None = None
+        for attempt in range(self._retries):
+            try:
+                with self._urlopen(req, timeout=self._timeout) as resp:
+                    return resp.read()
+            except Exception as exc:  # noqa: BLE001
+                last_error = exc
+                if attempt < self._retries - 1:
+                    time.sleep(0.5 * (attempt + 1))
+        raise DataSourceError(f"sina 请求失败: {last_error}") from last_error
+
+    # ---- 财务（兜底 Tushare fina_indicator） ----
+
+    def get_financial(self, symbol: str) -> list[FinancialIndicator]:
+        """新浪关键指标（source=gjzb），含披露日 publish_date → announce_date。"""
+        params = {
+            "paperCode": _to_sina_symbol(symbol),
+            "source": "gjzb",
+            "type": "0",
+            "page": "1",
+            "num": "100",
+        }
+        url = f"{_FIN_BASE}?{urllib.parse.urlencode(params)}"
+        payload = json.loads(self._open(url).decode("utf-8", errors="replace"))
+        try:
+            data = payload["result"]["data"]
+            report_dates = [item["date_value"] for item in data["report_date"]]
+        except (KeyError, TypeError) as exc:
+            raise DataSourceError(f"新浪财务返回结构异常（{symbol}）: {exc}") from exc
+
+        rows: list[FinancialIndicator] = []
+        for rd in report_dates:
+            entry = data["report_list"].get(rd)
+            if not entry:
+                continue
+            announce = entry.get("publish_date")
+            if not announce:
+                continue  # 无披露日不可用于研究（防未来函数）
+            fields: dict[str, Decimal | None] = {
+                "eps": None,
+                "roe": None,
+                "total_revenue": None,
+                "net_profit": None,
+                "gross_margin": None,
+            }
+            for item in entry.get("data", []):
+                std = _FIN_FIELD_MAP.get(item.get("item_title", ""))
+                if std and fields.get(std) is None:
+                    fields[std] = _d(item.get("item_value"))
+            rows.append(
+                FinancialIndicator(
+                    symbol=symbol,
+                    report_date=_to_date(str(rd)),
+                    announce_date=_to_date(str(announce)),
+                    eps=fields["eps"],
+                    roe=fields["roe"],
+                    total_revenue=fields["total_revenue"],
+                    net_profit=fields["net_profit"],
+                    gross_margin=fields["gross_margin"],
+                )
+            )
+        return rows
+
+    # ---- 日 K（前复权兜底，标记 adjust=qfq） ----
+
     def get_daily(self, symbol: str, start: date, end: date, datalen: int = 320) -> list[DailyBar]:
         """拉取前复权日 K（新浪仅支持最近 datalen 个自然日窗口）。"""
         url = _KLINE_JSONP.format(sina_symbol=_to_sina_symbol(symbol), datalen=datalen)
-        try:
+        payload = self._open(url).decode("utf-8", errors="replace")
-            with self._urlopen(url, timeout=self._timeout) as resp:
-                payload = resp.read().decode("utf-8", errors="replace")
-        except Exception as exc:  # noqa: BLE001
-            raise DataSourceError(f"sina 请求失败: {exc}") from exc
 
         bars: list[DailyBar] = []
         for rec in _extract_jsonp(payload):
@@ -70,6 +189,8 @@ class SinaProvider:
                 DailyBar(
                     symbol=symbol,
                     trade_date=day,
+                    source="sina",
+                    adjust="qfq",
                     open=_d(rec.get("open")),
                     high=_d(rec.get("high")),
                     low=_d(rec.get("low")),
@@ -79,6 +200,8 @@ class SinaProvider:
             )
         return bars
 
+    # ---- 不支持 ----
+
     def get_stock_basic(self):
         raise DataSourceNotSupported("新浪不提供股票基础信息列表")
 
@@ -86,16 +209,4 @@ class SinaProvider:
         raise DataSourceNotSupported("新浪不提供交易日历")
 
     def get_adjust_factor(self, symbol, start, end):
-        raise DataSourceNotSupported("新浪不提供复权因子（返回数据为前复权口径）")
+        raise DataSourceNotSupported("新浪不提供复权因子（日线接口为前复权口径）")
-
-    def get_financial(self, symbol):
-        raise DataSourceNotSupported("新浪不提供财务指标")
-
-
-def _d(value) -> Decimal | None:
-    if value is None:
-        return None
-    try:
-        return Decimal(str(value))
-    except (ValueError, TypeError):
-        return None

backend/app/infrastructure/data_sources/tushare.py

@@ -8,6 +8,8 @@ tushare 库（pyproject optional：uv sync --extra datasource-tushare）。
 from __future__ import annotations
 
 import importlib
+import logging
+import time
 from datetime import date, datetime
 from decimal import Decimal
 from typing import Any
@@ -24,6 +26,7 @@ from app.infrastructure.data_sources.errors import (
     DataSourceError,
 )
 
+logger = logging.getLogger(__name__)
 _TS_DATE = "%Y%m%d"
 
 
@@ -56,9 +59,11 @@ class TushareProvider:
         *,
         pro: object | None = None,
         max_retries: int = 3,
+        rate_limit_wait: float = 30.0,
     ) -> None:
         self._pro = pro if pro is not None else _build_pro(token)
         self._max_retries = max_retries
+        self._rate_limit_wait = rate_limit_wait
 
     # ---- 归一化（纯函数，输入 list[dict]，可单测） ----
 
@@ -101,6 +106,8 @@ class TushareProvider:
                 DailyBar(
                     symbol=str(rec.get("ts_code") or ""),
                     trade_date=_to_date(rec.get("trade_date")) or date.min,
+                    source="tushare",
+                    adjust="none",
                     open=_to_decimal(rec.get("open")),
                     high=_to_decimal(rec.get("high")),
                     low=_to_decimal(rec.get("low")),
@@ -182,9 +189,12 @@ class TushareProvider:
 
     # ---- 内部 ----
 
+    _RATE_LIMIT_MARKERS = ("频率超限", "每分钟", "frequenc", "too many")
+
     def _call(self, api: str, **kwargs) -> list[dict[str, Any]]:
+        """带限速退避的调用：频率超限按指数退避（最长 _rate_limit_wait）等待后重试。"""
         last_error: Exception | None = None
-        for _ in range(self._max_retries):
+        for attempt in range(self._max_retries):
             try:
                 fn = getattr(self._pro, api)
                 result = fn(**kwargs)
@@ -200,6 +210,10 @@ class TushareProvider:
                 msg = str(exc)
                 if "权限" in msg or "积分" in msg or "token" in msg.lower():
                     raise DataSourceAuthenticationError(msg) from exc
+                if any(marker in msg for marker in self._RATE_LIMIT_MARKERS):
+                    wait = min(self._rate_limit_wait, 2 ** (attempt + 1))
+                    logger.warning("tushare.%s 频率超限，退避 %.1fs 后重试", api, wait)
+                    time.sleep(wait)
         raise DataSourceError(
             f"tushare.{api} 重试 {self._max_retries} 次仍失败: {last_error}"
         ) from last_error

backend/app/infrastructure/persistence/migrations/versions/91c4e27a03fb_stock_daily_source_and_adjust.py

@@ -0,0 +1,37 @@
+"""stock_daily source/adjust 来源与口径标记
+
+Revision ID: 91c4e27a03fb
+Revises: 53113c80257f
+Create Date: 2026-09-06
+
+新浪兜底行带 source=sina / adjust=qfq 标记；现有 648 万行回填默认
+tushare / none（SQLite ADD COLUMN 带常量默认值，不重写现有数据）。
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "91c4e27a03fb"
+down_revision: str | None = "53113c80257f"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "stock_daily",
+        sa.Column("source", sa.String(length=16), nullable=False, server_default="tushare"),
+    )
+    op.add_column(
+        "stock_daily",
+        sa.Column("adjust", sa.String(length=8), nullable=False, server_default="none"),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("stock_daily", "adjust")
+    op.drop_column("stock_daily", "source")

backend/app/infrastructure/persistence/sqlalchemy/models/market.py

@@ -61,6 +61,8 @@ class StockDailyModel(Base):
     id: Mapped[int] = mapped_column(PK_INT, primary_key=True, autoincrement=True)
     symbol: Mapped[str] = mapped_column(String(SYMBOL_LEN), index=True)
     trade_date: Mapped[date] = mapped_column(Date, index=True)
+    source: Mapped[str] = mapped_column(String(16), default="tushare", server_default="tushare")
+    adjust: Mapped[str] = mapped_column(String(8), default="none", server_default="none")
     open: Mapped[Decimal | None] = mapped_column(Numeric(12, 4), nullable=True)
     high: Mapped[Decimal | None] = mapped_column(Numeric(12, 4), nullable=True)
     low: Mapped[Decimal | None] = mapped_column(Numeric(12, 4), nullable=True)

backend/tests/test_sina_provider.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import json
 from datetime import date
 
 import pytest
@@ -72,6 +73,84 @@ class TestCapabilities:
         with pytest.raises(DataSourceNotSupported):
             provider.get_stock_basic()
         with pytest.raises(DataSourceNotSupported):
-            provider.get_adjust_factor("600519.SH", date(2024, 1, 1), date(2024, 1, 31))
+            provider.get_trade_cal(date(2024, 1, 1), date(2024, 1, 31))
         with pytest.raises(DataSourceNotSupported):
+            provider.get_adjust_factor("600519.SH", date(2024, 1, 1), date(2024, 1, 31))
+
+
+_FIN_PAYLOAD = json.dumps(
+    {
+        "result": {
+            "data": {
+                "report_date": [{"date_value": "2024-06-30"}, {"date_value": "2024-03-31"}],
+                "report_list": {
+                    "2024-06-30": {
+                        "publish_date": "2024-08-31",
+                        "data": [
+                            {"item_title": "基本每股收益", "item_value": "1.23"},
+                            {"item_title": "净资产收益率(ROE)", "item_value": "15.5"},
+                            {"item_title": "净利润", "item_value": "1230000000"},
+                        ],
+                    },
+                    "2024-03-31": {"publish_date": "", "data": []},
+                },
+            }
+        }
+    },
+    ensure_ascii=False,
+)
+
+
+class _FinResp:
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *exc):
+        return False
+
+    def read(self):
+        return _FIN_PAYLOAD.encode("utf-8")
+
+
+class TestFinancial:
+    def test_parse_schema_parity_with_tushare(self) -> None:
+        """财务 schema 与 Tushare 对齐：report_date / announce_date / 指标。"""
+        provider = SinaProvider(urlopen=lambda _req, **kw: _FinResp())
+        rows = provider.get_financial("600519.SH")
+        assert len(rows) == 1  # 无 publish_date 的行被跳过（防未来函数）
+        row = rows[0]
+        assert row.symbol == "600519.SH"
+        assert row.report_date == date(2024, 6, 30)
+        assert row.announce_date == date(2024, 8, 31)
+        assert float(row.eps) == 1.23
+        assert float(row.roe) == 15.5
+        assert float(row.net_profit) == 1230000000
+        # FinancialIndicator 校验能通过 → 格式与 Tushare 一致
+        assert row.model_dump()["eps"] is not None
+
+    def test_bad_structure_raises(self) -> None:
+        class _BadResp(_FinResp):
+            def read(self):
+                return b'{"result": {}}'
+
+        provider = SinaProvider(urlopen=lambda _req, **kw: _BadResp())
+        try:
             provider.get_financial("600519.SH")
+        except DataSourceError:
+            return
+        raise AssertionError("应抛 DataSourceError")
+
+
+class TestFormatParity:
+    """两源格式一致：统一 DailyBar schema，新浪兜底行带 source/adjust 标记。"""
+
+    def test_sina_bar_marks_source_and_adjust(self) -> None:
+        provider = SinaProvider(urlopen=lambda _url, **kw: _FakeResp())
+        bars = provider.get_daily("600519.SH", date(2024, 8, 30), date(2024, 8, 30))
+        assert len(bars) == 1
+        bar = bars[0]
+        assert bar.source == "sina"
+        assert bar.adjust == "qfq"  # 新浪为前复权，与 Tushare 不复权主口径区分
+        # 与 Tushare 同 schema：必备字段齐全
+        assert bar.symbol == "600519.SH"
+        assert bar.close is not None

backend/tests/test_tushare_provider.py

@@ -148,3 +148,24 @@ class TestCall:
     def test_missing_token_rejected(self) -> None:
         with pytest.raises(DataSourceAuthenticationError, match="TUSHARE_TOKEN"):
             TushareProvider(token="")
+
+
+class TestRateLimitBackoff:
+    def test_rate_limit_retries_with_sleep(self) -> None:
+        """频率超限：按退避等待后重试，最终抛错带原始信息（不当作鉴权错误）。"""
+        import app.infrastructure.data_sources.tushare as ts_mod
+
+        orig_sleep = ts_mod.time.sleep
+        sleeps: list[float] = []
+        ts_mod.time.sleep = lambda w: sleeps.append(w)  # noqa: SLF001 —— 测试桩
+        try:
+            provider = _pro(
+                error=RuntimeError("抱歉，您访问接口(adj_factor)频率超限(200次/分钟)"), retries=3
+            )
+            provider._rate_limit_wait = 0.1  # noqa: SLF001
+            with pytest.raises(DataSourceError, match="频率超限"):
+                provider.get_stock_basic()
+        finally:
+            ts_mod.time.sleep = orig_sleep
+        assert len(provider._pro.calls) == 3  # noqa: SLF001 —— 完整重试 3 次
+        assert len(sleeps) >= 2