feat(HEL-457): 估值字段级质量门、股票主档每日发布和资金流历史回补

- field_gates 按数据集配置关键字段非空率下限/非有限比例/相对上一批次的塌陷保护，
  字段大面积为空的批次拒发并保留上一正式批次，可读失败原因入 batches.error
- 股票主档交易日 20:00/23:10 自动刷新并发布版本化快照（eod_stocks + publications），
  覆盖新上市/简称变化/N前缀摘除；/v1/stocks 携带 batch_id/published_at，无变化跳过
- moneyflow 历史回补（默认 60 交易日，跳过已发布日期）；未发布点查返回
  available_from/available_to 与 history_not_backfilled 标记，缺失不再静默
- eod-refresh 新增 --force --dataset 安全重发（仍走全部质量门，上一批次可回滚）
- 保持 HEL-435 盘后重试机制；新增 22 项测试覆盖字段拒发/正常通过/旧批保留/
  主档新增改名/资金流覆盖/重复执行幂等

Co-authored-by: multica-agent <github@multica.ai>

xiaobai-datahub/README.md

@@ -79,6 +79,38 @@ python -m datahub history-backfill
 
 区间接口在 `meta.coverage` / `meta.incomplete` 标明覆盖是否完整；网站只读接入把不完整区间视为不可用并回旧链路。个股日 K 的 90 天区间查询依赖已核实，本阶段不回补全市场历史。
 
+## 估值字段级质量门
+
+`hub-quality.config.json` 的 `field_gates` 按数据集配置关键字段：非空率下限（支持按字段覆盖，如 `dv_ttm` 合法高空值）、非有限值比例上限、以及相对上一已发布批次的非空率塌陷保护。字段大面积为空的批次会被拒绝发布、保留上一份正常正式数据，失败原因逐字段写入 `batches.error` / `quality_json`。被拒后数据集仍视为缺失，盘后自动重试（HEL-435 机制）会继续尝试直到成功或截止。配置对任意数据集生效，不写死单日或单字段。
+
+## 股票主档每日刷新与发布
+
+交易日 20:00 与 23:10（`stocks_refresh_times` 可配）自动刷新股票主档并发布版本化快照（`eod_stocks` + `publications.dataset='stocks'`），覆盖当日新上市、证券简称变化和上市首日 N/C 前缀摘除；无变化则跳过，重复执行幂等。`/v1/stocks` 从最新已发布快照提供数据并带 `batch_id` / `published_at`；`/v1/datasets/status` 同步展示 stocks 状态。
+
+```bash
+cd xiaobai-datahub
+python -m datahub stocks-refresh            # 手动触发；--force 无变化也重发
+```
+
+## 资金流历史回补
+
+网站会沿真实调用链查最近若干交易日的 moneyflow（个股详情任意日期点查 + 智能选股最近 5 个交易日），默认回补最近 60 个交易日（`moneyflow_history_trading_days` 可配，已发布日期自动跳过）。点查未覆盖的历史日期返回 `DATASET_NOT_PUBLISHED` 并附 `available_from` / `available_to`（低于下界时 `reason=history_not_backfilled`），网站据此明确回退旧链路，不会静默拿到半截数据。
+
+```bash
+cd xiaobai-datahub
+python -m datahub moneyflow-backfill        # --trading-days 60 --end-date --force 可选
+```
+
+## 盘后补跑与强制重发
+
+```bash
+cd xiaobai-datahub
+python -m datahub eod-refresh --trade-date 20260904                  # 只补缺失数据集
+python -m datahub eod-refresh --trade-date 20260904 --force --dataset valuation
+# 强制重取重发：仍走全部质量门，生成新批次，上一批次保留可回滚
+```
+
+
 ## 备份
 
 每日 00:40 任务把 `datahub.db` 备份到 `data/backups/`（保留 14 份）。也可手动：

xiaobai-datahub/config/hub-quality.config.json

@@ -16,5 +16,31 @@
   "index_history_trading_days": 260,
   "eod_retry_start": "15:15",
   "eod_retry_interval_minutes": 30,
-  "eod_retry_cutoff": "23:30"
+  "eod_retry_cutoff": "23:30",
+  "moneyflow_history_trading_days": 60,
+  "stocks_refresh_times": [
+    "20:00",
+    "23:10"
+  ],
+  "field_gates": {
+    "valuation": {
+      "fields": [
+        "turnover_rate",
+        "volume_ratio",
+        "total_mv",
+        "circ_mv",
+        "pe_ttm",
+        "pb",
+        "ps_ttm",
+        "dv_ttm"
+      ],
+      "min_nonnull_rate": 0.9,
+      "min_nonnull_rate_by_field": {
+        "pe_ttm": 0.5,
+        "dv_ttm": 0.3
+      },
+      "max_nonnull_drop_vs_prev": 0.15,
+      "max_nonfinite_rate": 0.01
+    }
+  }
 }

xiaobai-datahub/datahub/admin_api.py

@@ -84,12 +84,14 @@ class AdminAPI:
 
     def jobs(self) -> dict[str, Any]:
         runs = self.db.fetchall("SELECT * FROM job_runs ORDER BY id DESC LIMIT 100")
+        stocks_times = "/".join(self.pipeline.settings.stocks_refresh_times) or "20:00"
         return {
             "jobs": [
                 {"id": "precheck", "at": "08:45", "title": "盘前预检"},
                 {"id": "eod_a", "at": "15:05", "title": "盘后批 A daily/valuation/moneyflow/auction"},
                 {"id": "eod_b", "at": "15:10", "title": "盘后批 B index_daily"},
                 {"id": "eod_retry", "at": "15:15-23:30", "title": "盘后未出数自动重试（每 30 分钟，成功即停）"},
+                {"id": "stocks_refresh", "at": stocks_times, "title": "股票主档刷新与正式发布（新上市/更名，无变化跳过）"},
                 {"id": "history_backfill", "at": "manual", "title": "回补历史日历与指数日 K"},
                 {"id": "cleanup", "at": "00:30", "title": "清理 staging / 日志"},
                 {"id": "backup", "at": "00:40", "title": "SQLite 备份"},

xiaobai-datahub/datahub/cli.py

@@ -7,6 +7,7 @@ import json
 import sys
 
 from datahub.hub import build_hub
+from datahub.pipeline import OFFICIAL_DATASETS
 from datahub.settings import load_settings
 from datahub.timeutil import yyyymmdd
 
@@ -20,6 +21,23 @@ def main(argv: list[str] | None = None) -> int:
     history.add_argument("--force", action="store_true", help="覆盖已发布的指数日期")
     refresh = sub.add_parser("eod-refresh", help="对指定交易日补跑盘后正式数据（跳过已发布数据集，仍走质量门禁）")
     refresh.add_argument("--trade-date", default=None, help="交易日 YYYYMMDD，默认今天")
+    refresh.add_argument(
+        "--force", action="store_true",
+        help="对 --dataset 指定的数据集强制重取重发（生成新批次，保留上一批次可回滚）",
+    )
+    refresh.add_argument(
+        "--dataset", default=None,
+        help="配合 --force 使用：只强制重发该数据集（如 valuation）",
+    )
+    stocks_refresh = sub.add_parser("stocks-refresh", help="刷新股票主档并发布正式快照（幂等：无变化则跳过）")
+    stocks_refresh.add_argument("--trade-date", default=None, help="交易日 YYYYMMDD，默认今天")
+    stocks_refresh.add_argument("--force", action="store_true", help="即使快照无变化也重新发布")
+    moneyflow_backfill = sub.add_parser(
+        "moneyflow-backfill", help="回补资金流历史（默认覆盖网站所需的最近 N 个交易日，跳过已发布日期）",
+    )
+    moneyflow_backfill.add_argument("--end-date", default=None, help="截止交易日 YYYYMMDD，默认今天")
+    moneyflow_backfill.add_argument("--trading-days", type=int, default=None, help="回补交易日数量，默认配置 moneyflow_history_trading_days")
+    moneyflow_backfill.add_argument("--force", action="store_true", help="覆盖已发布的资金流日期")
     args = parser.parse_args(argv)
 
     settings = load_settings()
@@ -35,16 +53,56 @@ def main(argv: list[str] | None = None) -> int:
         return 0 if result.get("ok") else 1
     if args.command == "eod-refresh":
         day = yyyymmdd(args.trade_date) if args.trade_date else yyyymmdd()
-        result = hub.pipeline.run_eod_missing(day)
+        if args.force:
+            datasets = tuple(sorted({args.dataset} & OFFICIAL_DATASETS)) if args.dataset else ()
+            if args.dataset and not datasets:
+                parser.error(f"unknown dataset: {args.dataset}")
+            if not datasets:
+                parser.error("--force requires --dataset (e.g. --dataset valuation)")
+            result = {}
+            for dataset in datasets:
+                result[dataset] = hub.pipeline.run_dataset(dataset, day)
+        else:
+            result = hub.pipeline.run_eod_missing(day)
         hub.pipeline.audit("cli", "eod-refresh", f"eod:{day}", json.dumps(
-            {name: item.get("state") for name, item in result.items() if isinstance(item, dict)},
+            {"force": bool(args.force), "dataset": args.dataset,
+             **{name: item.get("state") for name, item in result.items() if isinstance(item, dict)}},
             ensure_ascii=False,
         ))
+        if args.force:
+            payload = {"trade_date": day, "datasets": result}
+            json.dump(payload, sys.stdout, ensure_ascii=False, indent=2, default=str)
+            sys.stdout.write("\n")
+            return 0
         missing = hub.pipeline.missing_official_datasets(day)
         payload = {"trade_date": day, "datasets": result, "missing_after": missing}
         json.dump(payload, sys.stdout, ensure_ascii=False, indent=2, default=str)
         sys.stdout.write("\n")
         return 0 if not missing else 1
+    if args.command == "stocks-refresh":
+        day = yyyymmdd(args.trade_date) if args.trade_date else yyyymmdd()
+        result = hub.pipeline.refresh_stocks(day, force=args.force)
+        hub.pipeline.audit("cli", "stocks-refresh", f"stocks:{day}", json.dumps(
+            {"force": bool(args.force), "state": result.get("state"), "batch_id": result.get("batch_id")},
+            ensure_ascii=False,
+        ))
+        json.dump(result, sys.stdout, ensure_ascii=False, indent=2, default=str)
+        sys.stdout.write("\n")
+        return 0 if result.get("state") != "failed" else 1
+    if args.command == "moneyflow-backfill":
+        result = hub.pipeline.backfill_moneyflow_history(
+            end_date=args.end_date,
+            trading_days=args.trading_days,
+            force=args.force,
+        )
+        hub.pipeline.audit("cli", "moneyflow-backfill", f"moneyflow:{result.get('end')}", json.dumps(
+            {"published": len(result.get("published") or []), "skipped": len(result.get("skipped") or []),
+             "failed": len(result.get("failed") or [])},
+            ensure_ascii=False,
+        ))
+        json.dump(result, sys.stdout, ensure_ascii=False, indent=2, default=str)
+        sys.stdout.write("\n")
+        return 0 if result.get("ok") else 1
     parser.error(f"unknown command: {args.command}")
     return 2
 

xiaobai-datahub/datahub/db.py

@@ -114,6 +114,21 @@ CREATE TABLE IF NOT EXISTS eod_index_bars (
   PRIMARY KEY (ts_code, trade_date, batch_id)
 ) WITHOUT ROWID;
 
+CREATE TABLE IF NOT EXISTS eod_stocks (
+  ts_code TEXT NOT NULL, trade_date TEXT NOT NULL,
+  symbol TEXT, name TEXT, area TEXT, industry TEXT, market TEXT,
+  list_status TEXT, list_date TEXT,
+  batch_id TEXT NOT NULL,
+  PRIMARY KEY (ts_code, trade_date, batch_id)
+) WITHOUT ROWID;
+
+CREATE TABLE IF NOT EXISTS staging_stocks (
+  ts_code TEXT NOT NULL, trade_date TEXT NOT NULL, batch_id TEXT NOT NULL,
+  symbol TEXT, name TEXT, area TEXT, industry TEXT, market TEXT,
+  list_status TEXT, list_date TEXT,
+  PRIMARY KEY (batch_id, ts_code, trade_date)
+);
+
 CREATE TABLE IF NOT EXISTS staging_bars (
   ts_code TEXT NOT NULL, trade_date TEXT NOT NULL, batch_id TEXT NOT NULL,
   open REAL, high REAL, low REAL, close REAL, pct_chg REAL,
@@ -273,6 +288,7 @@ DATASET_TABLES = {
     "moneyflow": ("eod_moneyflow", "staging_moneyflow"),
     "auction": ("eod_auction", "staging_auction"),
     "index_daily": ("eod_index_bars", "staging_index_bars"),
+    "stocks": ("eod_stocks", "staging_stocks"),
 }
 
 

xiaobai-datahub/datahub/pipeline.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import json
+import math
 import time
 from collections.abc import Callable
 from datetime import timedelta
@@ -22,11 +23,21 @@ LOGGER = get_logger()
 HARD_DATASETS = {"daily", "valuation", "index_daily"}
 SOFT_DATASETS = {"moneyflow", "auction"}
 OFFICIAL_DATASETS = HARD_DATASETS | SOFT_DATASETS
+STOCKS_DATASET = "stocks"
+STOCK_SNAPSHOT_FIELDS = ("ts_code", "symbol", "name", "area", "industry", "market", "list_status", "list_date")
 EOD_A_DATASETS = ("daily", "valuation", "moneyflow", "auction")
 EOD_B_DATASETS = ("index_daily",)
 EMPTY_BATCH_ERROR = "empty official batch: 0 valid rows"
 
 STAGING_INSERT = {
+    "stocks": (
+        "INSERT INTO staging_stocks(ts_code,trade_date,batch_id,symbol,name,area,industry,market,list_status,list_date) "
+        "VALUES (?,?,?,?,?,?,?,?,?,?)",
+        lambda r, b: (
+            r["ts_code"], r["trade_date"], b, r.get("symbol"), r.get("name"), r.get("area"),
+            r.get("industry"), r.get("market"), r.get("list_status"), r.get("list_date"),
+        ),
+    ),
     "daily": (
         "INSERT INTO staging_bars(ts_code,trade_date,batch_id,open,high,low,close,pct_chg,volume,amount,adj_factor) "
         "VALUES (?,?,?,?,?,?,?,?,?,?,?)",
@@ -72,6 +83,11 @@ STAGING_INSERT = {
 }
 
 EOD_COPY = {
+    "stocks": (
+        "INSERT OR REPLACE INTO eod_stocks "
+        "SELECT ts_code,trade_date,symbol,name,area,industry,market,list_status,list_date,batch_id "
+        "FROM staging_stocks WHERE batch_id = ?"
+    ),
     "daily": (
         "INSERT OR REPLACE INTO eod_bars "
         "SELECT ts_code,trade_date,open,high,low,close,pct_chg,volume,amount,adj_factor,batch_id "
@@ -101,6 +117,13 @@ EOD_COPY = {
 }
 
 
+def _finite(value: Any) -> bool:
+    try:
+        return math.isfinite(float(value))
+    except (TypeError, ValueError):
+        return False
+
+
 def _staging_row_count(connection: Any, dataset: str, batch_id: str) -> int:
     table = DATASET_TABLES[dataset][1]
     row = connection.execute(
@@ -176,23 +199,7 @@ class Pipeline:
                     """,
                     (row["exchange"], row["cal_date"], row["is_open"], row.get("pretrade_date"), fetched_at),
                 )
-            for row in stocks:
+            self._upsert_stock_master(connection, stocks, fetched_at)
-                connection.execute(
-                    """
-                    INSERT INTO stock_master(ts_code,symbol,name,area,industry,market,list_status,list_date,updated_at)
-                    VALUES (?,?,?,?,?,?,?,?,?)
-                    ON CONFLICT(ts_code) DO UPDATE SET
-                      symbol=excluded.symbol, name=excluded.name, area=excluded.area,
-                      industry=excluded.industry, market=excluded.market,
-                      list_status=excluded.list_status, list_date=excluded.list_date,
-                      updated_at=excluded.updated_at
-                    """,
-                    (
-                        row["ts_code"], row.get("symbol"), row.get("name"), row.get("area"),
-                        row.get("industry"), row.get("market"), row.get("list_status"),
-                        row.get("list_date"), fetched_at,
-                    ),
-                )
         return {
             "calendar": len(calendar),
             "stocks": len(stocks),
@@ -201,6 +208,154 @@ class Pipeline:
             "calendar_to": end,
         }
 
+    @staticmethod
+    def _upsert_stock_master(connection: Any, stocks: list[dict[str, Any]], fetched_at: str) -> None:
+        for row in stocks:
+            connection.execute(
+                """
+                INSERT INTO stock_master(ts_code,symbol,name,area,industry,market,list_status,list_date,updated_at)
+                VALUES (?,?,?,?,?,?,?,?,?)
+                ON CONFLICT(ts_code) DO UPDATE SET
+                  symbol=excluded.symbol, name=excluded.name, area=excluded.area,
+                  industry=excluded.industry, market=excluded.market,
+                  list_status=excluded.list_status, list_date=excluded.list_date,
+                  updated_at=excluded.updated_at
+                """,
+                (
+                    row["ts_code"], row.get("symbol"), row.get("name"), row.get("area"),
+                    row.get("industry"), row.get("market"), row.get("list_status"),
+                    row.get("list_date"), fetched_at,
+                ),
+            )
+
+    def latest_stocks_publication(self, on_or_before: str | None = None) -> dict[str, Any] | None:
+        if on_or_before:
+            row = self.db.fetchone(
+                "SELECT * FROM publications WHERE dataset = ? AND trade_date <= ? ORDER BY trade_date DESC LIMIT 1",
+                (STOCKS_DATASET, yyyymmdd(on_or_before)),
+            )
+        else:
+            row = self.db.fetchone(
+                "SELECT * FROM publications WHERE dataset = ? ORDER BY trade_date DESC LIMIT 1",
+                (STOCKS_DATASET,),
+            )
+        return row
+
+    def published_stock_snapshot(self, on_or_before: str | None = None) -> tuple[str | None, list[dict[str, Any]]]:
+        pub = self.latest_stocks_publication(on_or_before)
+        if not pub:
+            return None, []
+        rows = self.db.fetchall(
+            f"SELECT {','.join(STOCK_SNAPSHOT_FIELDS)} FROM eod_stocks WHERE batch_id = ? ORDER BY ts_code",
+            (pub["active_batch"],),
+        )
+        return str(pub["active_batch"]), rows
+
+    def refresh_stocks(self, trade_date: str | None = None, force: bool = False) -> dict[str, Any]:
+        """Refresh stock master from upstream and publish a versioned snapshot.
+
+        Runs on trading days (scheduler) and via CLI. Idempotent: when the
+        latest published snapshot already matches the upstream list, nothing
+        is published; ``force`` re-publishes unconditionally. New listings,
+        renames (incl. N/C prefix removal) and status changes all flow into
+        the snapshot, which carries batch_id/published_at metadata.
+        """
+        day = yyyymmdd(trade_date or self.clock())
+        rows = self._fetch_dataset(STOCKS_DATASET, day)
+        with self.db.write() as connection:
+            self._upsert_stock_master(connection, rows, isoformat(self.clock()))
+        if not force:
+            active, snapshot = self.published_stock_snapshot(day)
+            if active is not None:
+                upstream = sorted(
+                    tuple(str(row.get(field)) for field in STOCK_SNAPSHOT_FIELDS) for row in rows
+                )
+                published = sorted(tuple(str(row.get(field)) for field in STOCK_SNAPSHOT_FIELDS) for row in snapshot)
+                if upstream == published:
+                    return {
+                        "dataset": STOCKS_DATASET,
+                        "trade_date": day,
+                        "state": "skipped",
+                        "reason": "unchanged",
+                        "batch_id": active,
+                        "rows": len(snapshot),
+                    }
+        result = self.run_dataset(STOCKS_DATASET, day, prepared_rows=rows)
+        self.audit(
+            "pipeline", "stocks-refresh", f"{STOCKS_DATASET}:{day}",
+            json.dumps({"batch_id": result["batch_id"], "rows": result["rows"]}, ensure_ascii=False),
+        )
+        return result
+
+    def backfill_moneyflow_history(
+        self,
+        end_date: str | None = None,
+        trading_days: int | None = None,
+        force: bool = False,
+    ) -> dict[str, Any]:
+        """Incrementally publish official moneyflow history for the website window.
+
+        The website queries moneyflow for any navigable trade date (stock
+        detail point queries) and for the screener's last-5-days window, so
+        the hub must cover a trailing window of trading days instead of only
+        days published since go-live. Already published dates are skipped
+        unless ``force``; per-date failures are recorded without aborting.
+        """
+        end = yyyymmdd(end_date or self.clock())
+        limit = int(trading_days or self.settings.moneyflow_history_trading_days)
+        open_dates = self.open_trade_dates(end, limit)
+        if not open_dates:
+            return {
+                "start": None,
+                "end": end,
+                "requested_days": 0,
+                "published": [],
+                "skipped": [],
+                "failed": [{"error": "calendar has no open dates on or before end"}],
+                "ok": False,
+            }
+        start = open_dates[0]
+        published_dates: set[str] = set()
+        if not force:
+            pubs = self.db.fetchall(
+                "SELECT trade_date FROM publications WHERE dataset = 'moneyflow' AND trade_date >= ? AND trade_date <= ?",
+                (start, end),
+            )
+            published_dates = {str(row["trade_date"]) for row in pubs}
+        targets = [day for day in open_dates if day not in published_dates]
+        skipped = [day for day in open_dates if day in published_dates]
+        published: list[dict[str, Any]] = []
+        failed: list[dict[str, Any]] = []
+        for day in targets:
+            try:
+                raw = retry_call(
+                    lambda day=day: self._guarded_fetch("moneyflow", {"trade_date": day}),
+                    attempts=self.settings.max_publish_attempts,
+                    base_delay=0.05,
+                    sleeper=lambda _d: time.sleep(_d),
+                )
+                rows = self.adapter.normalize("moneyflow", raw)
+                result = self.run_dataset("moneyflow", day, prepared_rows=rows)
+                published.append(
+                    {
+                        "trade_date": day,
+                        "batch_id": result["batch_id"],
+                        "rows": result["rows"],
+                        "state": result["state"],
+                    }
+                )
+            except Exception as exc:
+                failed.append({"trade_date": day, "error": str(exc)})
+        return {
+            "start": start,
+            "end": end,
+            "requested_days": len(open_dates),
+            "published": published,
+            "skipped": skipped,
+            "failed": failed,
+            "ok": not failed,
+        }
+
     def open_trade_dates(self, end: str, limit: int) -> list[str]:
         end = yyyymmdd(end)
         rows = self.db.fetchall(
@@ -464,13 +619,14 @@ class Pipeline:
             if null_rate >= float(quality.get("null_rate_max") or 0.01):
                 errors.append(f"null rate {null_rate:.4f}")
         empty = row_n == 0
-        if empty and dataset in OFFICIAL_DATASETS:
+        if empty and (dataset in OFFICIAL_DATASETS or dataset == STOCKS_DATASET):
             errors.append(EMPTY_BATCH_ERROR)
+        field_report = self._field_gate(dataset, trade_date, rows, errors)
         if dataset in SOFT_DATASETS:
             hard_fail = bool(dup or bad_date or empty)
         else:
-            hard_fail = bool(errors) and dataset in HARD_DATASETS
+            hard_fail = bool(errors) and (dataset in HARD_DATASETS or dataset == STOCKS_DATASET)
-        return {
+        report = {
             "rows": row_n,
             "listed": listed_n,
             "ratio": round(ratio, 4),
@@ -480,6 +636,103 @@ class Pipeline:
             "soft_fail": bool(warnings) and not hard_fail,
             "batch_id": batch_id,
         }
+        if field_report is not None:
+            report["fields"] = field_report
+        return report
+
+    def _field_gate(
+        self,
+        dataset: str,
+        trade_date: str,
+        rows: list[dict[str, Any]],
+        errors: list[str],
+    ) -> dict[str, Any] | None:
+        """Config-driven per-field completeness gate.
+
+        Catches field-level half-products (rows complete, key columns empty)
+        that row-count gates miss: non-null rate floors per field, non-finite
+        share, and a collapse guard against the previous published batch so
+        legitimately sparse fields (e.g. dv_ttm) are not false-flagged.
+        Generic for any dataset configured under quality["field_gates"].
+        """
+        gate = dict((self.settings.quality.get("field_gates") or {}).get(dataset) or {})
+        if not gate or not rows:
+            return None
+        fields = [str(item) for item in (gate.get("fields") or []) if str(item)]
+        if not fields:
+            return None
+        min_rate = float(gate.get("min_nonnull_rate") or 0.9)
+        by_field = {str(k): float(v) for k, v in dict(gate.get("min_nonnull_rate_by_field") or {}).items()}
+        max_drop = float(gate.get("max_nonnull_drop_vs_prev") or 0.15)
+        max_nonfinite = float(gate.get("max_nonfinite_rate") or 0.01)
+        row_n = len(rows)
+        prev_batch, prev_stats = self._prev_field_stats(dataset, trade_date, fields)
+        report: dict[str, Any] = {}
+        for field in fields:
+            values = [row.get(field) for row in rows]
+            nulls = sum(1 for value in values if value is None)
+            nonfinite = sum(1 for value in values if value is not None and not _finite(value))
+            rate = (row_n - nulls) / row_n
+            stats = {
+                "nonnull": row_n - nulls,
+                "null": nulls,
+                "nonnull_rate": round(rate, 4),
+                "nonfinite": nonfinite,
+            }
+            floor = by_field.get(field, min_rate)
+            if rate < floor:
+                errors.append(
+                    f"field gate: {dataset}.{field} non-null rate {rate:.4f} < {floor}"
+                )
+            if nonfinite / row_n > max_nonfinite:
+                errors.append(
+                    f"field gate: {dataset}.{field} non-finite rate {nonfinite / row_n:.4f} > {max_nonfinite}"
+                )
+            prev_rate = prev_stats.get(field) if prev_stats else None
+            if prev_rate is not None:
+                stats["prev_nonnull_rate"] = round(prev_rate, 4)
+                if prev_rate - rate > max_drop:
+                    errors.append(
+                        f"field gate: {dataset}.{field} non-null rate {rate:.4f} dropped > {max_drop:.2f} "
+                        f"vs prev batch {prev_batch} ({prev_rate:.4f})"
+                    )
+            report[field] = stats
+        return report
+
+    def _prev_field_stats(
+        self,
+        dataset: str,
+        trade_date: str,
+        fields: list[str],
+    ) -> tuple[str | None, dict[str, float]]:
+        """Non-null rates per field from the latest earlier published batch."""
+        if dataset not in DATASET_TABLES:
+            return None, {}
+        table = DATASET_TABLES[dataset][0]
+        columns = {
+            str(row["name"])
+            for row in self.db.fetchall(f"PRAGMA table_info({table})")
+        }
+        usable = [field for field in fields if field in columns]
+        if not usable:
+            return None, {}
+        pub = self.db.fetchone(
+            """
+            SELECT active_batch FROM publications
+            WHERE dataset = ? AND trade_date < ? ORDER BY trade_date DESC LIMIT 1
+            """,
+            (dataset, trade_date),
+        )
+        if not pub:
+            return None, {}
+        batch_id = str(pub["active_batch"])
+        selects = ",".join(f"AVG({field} IS NOT NULL) AS {field}" for field in usable)
+        row = self.db.fetchone(
+            f"SELECT {selects} FROM {table} WHERE batch_id = ?",
+            (batch_id,),
+        )
+        stats = {field: float(row[field]) for field in usable if row.get(field) is not None}
+        return batch_id, stats
 
     def publish(self, dataset: str, trade_date: str, batch_id: str, state: str = "published") -> None:
         copy_sql = EOD_COPY[dataset]
@@ -636,6 +889,11 @@ class Pipeline:
         )
 
     def _fetch_dataset(self, dataset: str, trade_date: str) -> list[dict[str, Any]]:
+        if dataset == STOCKS_DATASET:
+            rows = self.adapter.normalize("stocks", self._guarded_fetch("stocks", {"list_status": "L"}))
+            for row in rows:
+                row["trade_date"] = trade_date
+            return rows
         if dataset == "daily":
             raw = self._guarded_fetch("daily", {"trade_date": trade_date})
             factors = {

xiaobai-datahub/datahub/scheduler.py

@@ -49,6 +49,7 @@ class Scheduler:
             "eod_a": self._eod_a,
             "eod_b": self._eod_b,
             "eod_retry": self._eod_retry,
+            "stocks_refresh": self._stocks_refresh,
             "cleanup": self._cleanup,
             "backup": self._backup,
             "history_backfill": self._history_backfill,
@@ -89,6 +90,8 @@ class Scheduler:
             ("cleanup", time(0, 30)),
             ("backup", time(0, 40)),
         ]
+        for refresh_at in self.pipeline.settings.stocks_refresh_times:
+            plan.append(("stocks_refresh", _hhmm(refresh_at)))
         open_day = is_open_day(self.db, day)
         for job_id, at in plan:
             if current < at:
@@ -96,7 +99,7 @@ class Scheduler:
             key = (job_id, day, at.strftime("%H%M"))
             if key in self._fired:
                 continue
-            if job_id in {"eod_a", "eod_b"} and not open_day:
+            if job_id in {"eod_a", "eod_b", "stocks_refresh"} and not open_day:
                 self._fired.add(key)
                 continue
             self._fired.add(key)
@@ -107,7 +110,7 @@ class Scheduler:
             try:
                 self.run_job(job_id, day)
             except Exception:
-                if job_id not in {"eod_a", "eod_b"}:
+                if job_id not in {"eod_a", "eod_b", "stocks_refresh"}:
                     raise
                 # Keep the tick alive; evening retries take over.
                 LOGGER.exception("scheduled job %s failed for %s", job_id, day)
@@ -317,6 +320,9 @@ class Scheduler:
     def _eod_retry(self, trade_date: str) -> dict[str, Any]:
         return self.pipeline.run_eod_missing(trade_date)
 
+    def _stocks_refresh(self, trade_date: str) -> dict[str, Any]:
+        return self.pipeline.refresh_stocks(trade_date)
+
     def _history_backfill(self, trade_date: str) -> dict[str, Any]:
         return self.pipeline.backfill_history(trade_date)
 

xiaobai-datahub/datahub/serving.py

@@ -135,6 +135,29 @@ class V1API:
 
     def stocks(self, updated_since: str, q: dict[str, str]) -> dict[str, Any]:
         limit, offset = self._page(q)
+        today = yyyymmdd(now_shanghai())
+        batch_id, snapshot = self.pipeline.published_stock_snapshot(today)
+        if batch_id:
+            # Formal view: the latest published stock snapshot, with batch
+            # metadata. Filters are applied in-memory on the snapshot.
+            pub = self.pipeline.latest_stocks_publication(today) or {}
+            rows = snapshot
+            if updated_since:
+                rows = []
+            rows = rows[offset: offset + limit]
+            return envelope(
+                rows,
+                {
+                    "tier": "official",
+                    "trade_date": pub.get("trade_date"),
+                    "published_at": pub.get("published_at"),
+                    "source": "tushare:stock_basic",
+                    "batch_id": batch_id,
+                    "stale": False,
+                    "staleness_seconds": 0,
+                    "state": pub.get("state"),
+                },
+            )
         if updated_since:
             rows = self.db.fetchall(
                 "SELECT * FROM stock_master WHERE updated_at >= ? ORDER BY ts_code LIMIT ? OFFSET ?",
@@ -176,7 +199,7 @@ class V1API:
 
     def dataset_status(self, date: str) -> dict[str, Any]:
         trade_date = yyyymmdd(date or now_shanghai())
-        datasets = ("daily", "valuation", "moneyflow", "auction", "index_daily")
+        datasets = ("daily", "valuation", "moneyflow", "auction", "index_daily", "stocks")
         items = []
         for dataset in datasets:
             pub = self.db.fetchone(
@@ -251,7 +274,7 @@ class V1API:
                 raise ApiError(
                     "DATASET_NOT_PUBLISHED",
                     f"{dataset} {start} 尚未发布",
-                    extra={"expected_at": "15:05+08:00"},
+                    extra=self._unpublished_extra(dataset, start),
                 )
             limit, offset = self._page(q)
             sql = f"SELECT * FROM {table} WHERE trade_date = ? AND batch_id = ?"
@@ -281,7 +304,11 @@ class V1API:
             (dataset, start, end),
         )
         if not pubs:
-            raise ApiError("DATASET_NOT_PUBLISHED", f"{dataset} {start}-{end} 尚未发布")
+            raise ApiError(
+                "DATASET_NOT_PUBLISHED",
+                f"{dataset} {start}-{end} 尚未发布",
+                extra=self._unpublished_extra(dataset, end),
+            )
         rows: list[dict[str, Any]] = []
         limit, offset = self._page(q)
         for pub in pubs:
@@ -350,6 +377,20 @@ class V1API:
         offset = max(0, offset)
         return limit, offset
 
+    def _unpublished_extra(self, dataset: str, trade_date: str) -> dict[str, Any]:
+        """Identifiable coverage info: is this a history gap or today-not-yet?"""
+        extra: dict[str, Any] = {"expected_at": "15:05+08:00"}
+        row = self.db.fetchone(
+            "SELECT MIN(trade_date) AS a, MAX(trade_date) AS b FROM publications WHERE dataset = ?",
+            (dataset,),
+        )
+        if row and row.get("a"):
+            extra["available_from"] = row["a"]
+            extra["available_to"] = row["b"]
+            if str(trade_date) < str(row["a"]):
+                extra["reason"] = "history_not_backfilled"
+        return extra
+
     def _official_meta(self, dataset: str, trade_date: str, source: str) -> dict[str, Any]:
         pub = self.db.fetchone(
             "SELECT * FROM publications WHERE dataset = ? AND trade_date = ?",

xiaobai-datahub/datahub/settings.py

@@ -56,6 +56,17 @@ class Settings:
     def index_history_trading_days(self) -> int:
         return int(self.quality.get("index_history_trading_days") or 260)
 
+    @property
+    def moneyflow_history_trading_days(self) -> int:
+        return int(self.quality.get("moneyflow_history_trading_days") or 60)
+
+    @property
+    def stocks_refresh_times(self) -> tuple[str, ...]:
+        raw = self.quality.get("stocks_refresh_times") or ["20:00", "23:10"]
+        if isinstance(raw, str):
+            raw = [raw]
+        return tuple(str(item) for item in raw)
+
     @property
     def eod_retry_start(self) -> str:
         return str(self.quality.get("eod_retry_start") or "15:15")

xiaobai-datahub/tests/fixtures.py

@@ -25,7 +25,7 @@ RAW = {
     ],
     "daily_basic": [
         {"ts_code": "600000.SH", "trade_date": "20240902", "turnover_rate": 1.2, "volume_ratio": 0.8, "total_mv": 1000.0, "circ_mv": 800.0, "pe_ttm": 5.1, "pb": 0.6, "ps_ttm": 1.1, "dv_ttm": 4.0},
-        {"ts_code": "000001.SZ", "trade_date": "20240902", "turnover_rate": 2.2, "volume_ratio": 1.1, "total_mv": 2000.0, "circ_mv": 1500.0, "pe_ttm": 6.2, "pb": 0.7, "ps_ttm": 1.2, "dv_ttm": 3.0},
+        {"ts_code": "000001.SZ", "trade_date": "20240902", "turnover_rate": 2.2, "volume_ratio": 1.1, "total_mv": 2000.0, "circ_mv": 1500.0, "pe_ttm": 6.2, "pb": 0.7, "ps_ttm": 1.2, "dv_ttm": None},
     ],
     "adj_factor": [
         {"ts_code": "600000.SH", "trade_date": "20240902", "adj_factor": 1.1},

xiaobai-datahub/tests/test_eod_retry.py

@@ -60,7 +60,11 @@ class EodRetryTests(unittest.TestCase):
         return db.fetchall("SELECT * FROM job_runs WHERE job_id = ? ORDER BY id", (job_id,))
 
     def _batches(self, db: HubDB, day: str) -> list[dict]:
-        return db.fetchall("SELECT * FROM batches WHERE trade_date = ?", (day,))
+        placeholders = ",".join("?" for _ in OFFICIAL)
+        return db.fetchall(
+            f"SELECT * FROM batches WHERE trade_date = ? AND dataset IN ({placeholders})",
+            (day, *sorted(OFFICIAL)),
+        )
 
     @staticmethod
     def _batch_ids(db: HubDB, day: str) -> set[str]:
@@ -71,7 +75,11 @@ class EodRetryTests(unittest.TestCase):
         return [name for name in transport.calls if name in DelayedTransport.DATE_APIS]
 
     def _published(self, db: HubDB, day: str) -> set[str]:
-        rows = db.fetchall("SELECT dataset FROM publications WHERE trade_date = ?", (day,))
+        placeholders = ",".join("?" for _ in OFFICIAL)
+        rows = db.fetchall(
+            f"SELECT dataset FROM publications WHERE trade_date = ? AND dataset IN ({placeholders})",
+            (day, *sorted(OFFICIAL)),
+        )
         return {str(row["dataset"]) for row in rows}
 
     def test_first_empty_then_retry_succeeds(self) -> None:
@@ -105,12 +113,12 @@ class EodRetryTests(unittest.TestCase):
 
         # success stops all further same-day requests
         batches_before = len(self._batches(db, day))
-        calls_before = len(transport.calls)
+        eod_calls_before = len(self._eod_calls(transport))
         sched.tick(clock_at(day, 17, 0))
         sched.tick(clock_at(day, 23, 0))
         self.assertEqual(len(self._job_runs(db, "eod_retry")), 2)
         self.assertEqual(len(self._batches(db, day)), batches_before)
-        self.assertEqual(len(transport.calls), calls_before)
+        self.assertEqual(len(self._eod_calls(transport)), eod_calls_before)
 
     def test_never_ready_marks_cutoff_failed_and_stops(self) -> None:
         day = "20240902"
@@ -144,9 +152,20 @@ class EodRetryTests(unittest.TestCase):
         sched.tick(clock_at(day, 15, 5))   # eod_a publishes 4 datasets
         sched.tick(clock_at(day, 15, 10))  # eod_b publishes index
         self.assertEqual(self._published(db, day), OFFICIAL)
+
+        def official_batches() -> list[str]:
+            placeholders = ",".join("?" for _ in OFFICIAL)
+            return [
+                str(row["batch_id"])
+                for row in db.fetchall(
+                    f"SELECT batch_id FROM batches WHERE trade_date = ? AND dataset IN ({placeholders})",
+                    (day, *sorted(OFFICIAL)),
+                )
+            ]
+
         active = db.fetchall("SELECT dataset, active_batch FROM publications WHERE trade_date = ?", (day,))
-        active_map = {row["dataset"]: row["active_batch"] for row in active}
+        active_map = {row["dataset"]: row["active_batch"] for row in active if row["dataset"] in OFFICIAL}
-        batches_before = self._batch_ids(db, day)
+        batches_before = set(official_batches())
         calls_before = self._eod_calls(transport)
 
         # container restart: fresh scheduler, missed-time catch-up fires eod_a/eod_b
@@ -157,8 +176,11 @@ class EodRetryTests(unittest.TestCase):
         self.assertNotIn("eod_retry", ran)
         self.assertEqual(self._published(db, day), OFFICIAL)
         after = db.fetchall("SELECT dataset, active_batch FROM publications WHERE trade_date = ?", (day,))
-        self.assertEqual({row["dataset"]: row["active_batch"] for row in after}, active_map)
+        self.assertEqual(
-        self.assertEqual(self._batch_ids(db, day), batches_before)  # no duplicate batches
+            {row["dataset"]: row["active_batch"] for row in after if row["dataset"] in OFFICIAL},
+            active_map,
+        )
+        self.assertEqual(set(official_batches()), batches_before)  # no duplicate batches
         self.assertEqual(self._eod_calls(transport), calls_before)  # no duplicate upstream EOD calls
         self.assertEqual(sched2.eod_status(day, clock=clock_at(day, 21, 0))["state"], "done")
 

xiaobai-datahub/tests/test_moneyflow_backfill.py

@@ -0,0 +1,128 @@
+from __future__ import annotations
+
+import unittest
+from datetime import date, timedelta
+from pathlib import Path
+import tempfile
+
+from datahub.adapters.tushare import TushareAdapter
+from datahub.crypto import SecretVault
+from datahub.db import HubDB
+from datahub.pipeline import Pipeline
+from datahub.serving import ApiError, V1API
+from datahub.settings import Settings
+from tests.fixtures import fake_transport
+
+OPEN_DATES = ["20240826", "20240827", "20240828", "20240829", "20240830", "20240902", "20240903"]
+EMPTY_UPSTREAM = {"20240828"}  # one date the upstream cannot serve
+
+
+def build_calendar(open_dates: list[str], span_days: int = 16) -> list[dict]:
+    start = date(int(open_dates[0][:4]), int(open_dates[0][4:6]), int(open_dates[0][6:8]))
+    rows = []
+    open_set = set(open_dates)
+    for offset in range(span_days):
+        cursor = start + timedelta(days=offset)
+        compact = cursor.strftime("%Y%m%d")
+        rows.append(
+            {"exchange": "SSE", "cal_date": compact, "is_open": 1 if compact in open_set else 0, "pretrade_date": compact}
+        )
+    return rows
+
+
+def moneyflow_rows(day: str) -> list[dict]:
+    return [
+        {
+            "ts_code": "600000.SH", "trade_date": day,
+            "buy_sm_amount": 10 + int(day[-2:]), "sell_sm_amount": 8, "buy_md_amount": 20, "sell_md_amount": 15,
+            "buy_lg_amount": 30, "sell_lg_amount": 25, "buy_elg_amount": 40, "sell_elg_amount": 35, "net_mf_amount": 17,
+        },
+        {
+            "ts_code": "000001.SZ", "trade_date": day,
+            "buy_sm_amount": 11, "sell_sm_amount": 9, "buy_md_amount": 21, "sell_md_amount": 16,
+            "buy_lg_amount": 31, "sell_lg_amount": 26, "buy_elg_amount": 41, "sell_elg_amount": 36, "net_mf_amount": 18,
+        },
+    ]
+
+
+class MoneyflowHistoryTransport:
+    def __init__(self) -> None:
+        self.calendar = build_calendar(OPEN_DATES)
+        self.moneyflow_fetches: list[str] = []
+
+    def __call__(self, api_name: str, params: dict, fields: str):
+        if api_name == "trade_cal":
+            start = str(params.get("start_date") or "")
+            end = str(params.get("end_date") or "99999999")
+            return [row for row in self.calendar if start <= row["cal_date"] <= end]
+        if api_name == "moneyflow":
+            day = str(params.get("trade_date") or "")
+            self.moneyflow_fetches.append(day)
+            if day in EMPTY_UPSTREAM:
+                return []
+            return moneyflow_rows(day)
+        return fake_transport(api_name, params, fields)
+
+
+class MoneyflowBackfillTests(unittest.TestCase):
+    def setUp(self) -> None:
+        self.transport = MoneyflowHistoryTransport()
+        tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        self.db = HubDB(Path(tmp.name) / "hub.db")
+        adapter = TushareAdapter("x", transport=self.transport)
+        settings = Settings(
+            encryption_key=SecretVault.generate_key(),
+            db_path=self.db.path,
+            backup_dir=Path(tmp.name) / "backups",
+            quality={"max_publish_attempts": 2, "publication_generations": 3},
+        )
+        self.pipe = Pipeline(self.db, adapter, settings)
+        self.pipe.ingest_reference("20240903")
+        self.api = V1API(self.db, self.pipe, settings)
+
+    def test_backfill_publishes_window_and_reports_failures(self) -> None:
+        result = self.pipe.backfill_moneyflow_history(end_date="20240903", trading_days=5)
+        published = [item["trade_date"] for item in result["published"]]
+        self.assertEqual(published, ["20240829", "20240830", "20240902", "20240903"])
+        self.assertEqual(result["failed"][0]["trade_date"], "20240828")
+        self.assertFalse(result["ok"])
+        rows = self.db.fetchall("SELECT * FROM eod_moneyflow WHERE trade_date='20240830'")
+        self.assertEqual(len(rows), 2)
+
+    def test_backfill_is_idempotent(self) -> None:
+        self.pipe.backfill_moneyflow_history(end_date="20240903", trading_days=5)
+        fetches_after_first = list(self.transport.moneyflow_fetches)
+        second = self.pipe.backfill_moneyflow_history(end_date="20240903", trading_days=5)
+        # only the still-missing date is re-fetched; published dates are skipped
+        self.assertEqual(self.transport.moneyflow_fetches[len(fetches_after_first):], ["20240828"])
+        self.assertEqual(len(second["skipped"]), 4)
+
+    def test_point_query_on_backfilled_date_serves_data(self) -> None:
+        self.pipe.backfill_moneyflow_history(end_date="20240903", trading_days=5)
+        payload = self.api.handle("/v1/moneyflow", {"date": ["20240830"]})
+        self.assertEqual(len(payload["data"]), 2)
+        self.assertEqual(payload["data"][0]["net_mf_amount"], 180000.0)
+
+    def test_unpublished_point_below_window_is_identifiable(self) -> None:
+        self.pipe.backfill_moneyflow_history(end_date="20240903", trading_days=5)
+        with self.assertRaises(ApiError) as ctx:
+            self.api.handle("/v1/moneyflow", {"date": ["20240801"]})
+        extra = ctx.exception.extra
+        self.assertEqual(extra["available_from"], "20240829")  # window starts at the first published date
+        self.assertEqual(extra["available_to"], "20240903")
+        self.assertEqual(extra["reason"], "history_not_backfilled")
+        self.assertEqual(extra["expected_at"], "15:05+08:00")
+
+    def test_range_query_flags_missing_dates(self) -> None:
+        self.pipe.backfill_moneyflow_history(end_date="20240903", trading_days=5)
+        payload = self.api.handle("/v1/moneyflow", {"from": ["20240828"], "to": ["20240903"]})
+        coverage = payload["meta"]["coverage"]
+        self.assertFalse(coverage["complete"])
+        self.assertEqual(coverage["missing_count"], 1)
+        self.assertEqual(coverage["missing_sample"], ["20240828"])
+        self.assertTrue(payload["meta"]["incomplete"])
+
+
+if __name__ == "__main__":
+    unittest.main()

xiaobai-datahub/tests/test_quality_gates.py

@@ -0,0 +1,253 @@
+from __future__ import annotations
+
+import unittest
+from pathlib import Path
+import tempfile
+
+from datahub.adapters.tushare import TushareAdapter
+from datahub.crypto import SecretVault
+from datahub.db import HubDB
+from datahub.pipeline import Pipeline, QualityError
+from datahub.scheduler import Scheduler
+from datahub.settings import Settings
+from tests.fixtures import TRADE_DATE, fake_transport
+from tests.test_eod_retry import clock_at
+
+FIELD_GATES = {
+    "valuation": {
+        "fields": [
+            "turnover_rate", "volume_ratio", "total_mv", "circ_mv",
+            "pe_ttm", "pb", "ps_ttm", "dv_ttm",
+        ],
+        "min_nonnull_rate": 0.9,
+        "min_nonnull_rate_by_field": {"pe_ttm": 0.5, "dv_ttm": 0.3},
+        "max_nonnull_drop_vs_prev": 0.15,
+        "max_nonfinite_rate": 0.01,
+    },
+}
+
+
+class ValuationTransport:
+    """fake_transport with switchable daily_basic degradation modes."""
+
+    def __init__(self) -> None:
+        self.mode = "ok"
+
+    def __call__(self, api_name: str, params: dict, fields: str):
+        rows = fake_transport(api_name, params, fields)
+        if api_name != "daily_basic":
+            return rows
+        trade_date = str(params.get("trade_date") or "")
+        if trade_date:
+            rows = [{**row, "trade_date": trade_date} for row in rows]
+        if self.mode == "ok":
+            return rows
+        patched = []
+        for row in rows:
+            item = dict(row)
+            if self.mode == "fields_all_null":
+                item["volume_ratio"] = None
+                item["dv_ttm"] = None
+            elif self.mode == "vr_all_null":
+                item["volume_ratio"] = None
+            elif self.mode == "dv_all_null":
+                item["dv_ttm"] = None
+            elif self.mode == "nonfinite":
+                item["volume_ratio"] = float("inf")
+            patched.append(item)
+        return patched
+
+
+def make_pipe(transport, quality_extra=None, clock=None):
+    tmp = tempfile.TemporaryDirectory()
+    db = HubDB(Path(tmp.name) / "hub.db")
+    adapter = TushareAdapter("test-token", transport=transport)
+    quality = {
+        "daily_row_ratio": 0.98,
+        "null_rate_max": 0.01,
+        "max_publish_attempts": 3,
+        "publication_generations": 3,
+        "job_run_retain_days": 90,
+        "staging_retain_days": 14,
+        "field_gates": FIELD_GATES,
+    }
+    if quality_extra:
+        quality.update(quality_extra)
+    settings = Settings(
+        encryption_key=SecretVault.generate_key(),
+        api_token="t" * 32,
+        admin_password="admin-pass",
+        tushare_token="test-token",
+        db_path=db.path,
+        quality=quality,
+        scheduler_enabled=False,
+    )
+    pipe = Pipeline(db, adapter, settings, clock=clock)
+    pipe._tmp = tmp
+    return pipe, db
+
+
+class ValuationFieldGateTests(unittest.TestCase):
+    def setUp(self) -> None:
+        self.transport = ValuationTransport()
+        self.pipe, self.db = make_pipe(self.transport)
+        self.pipe.ingest_reference(TRADE_DATE)
+
+    def _active(self) -> str | None:
+        row = self.db.fetchone(
+            "SELECT active_batch FROM publications WHERE dataset='valuation' AND trade_date=?",
+            (TRADE_DATE,),
+        )
+        return str(row["active_batch"]) if row else None
+
+    def test_normal_batch_with_legit_dv_nulls_passes(self) -> None:
+        result = self.pipe.run_dataset("valuation", TRADE_DATE)
+        self.assertEqual(result["state"], "published")
+        fields = result["quality"]["fields"]
+        # fixture: 1 of 2 stocks has null dv_ttm → 0.5 non-null ≥ 0.3 floor
+        self.assertEqual(fields["dv_ttm"]["nonnull_rate"], 0.5)
+        self.assertEqual(fields["volume_ratio"]["nonnull_rate"], 1.0)
+        self.assertFalse(result["quality"]["errors"])
+
+    def test_all_null_fields_rejected_and_prev_batch_kept(self) -> None:
+        first = self.pipe.run_dataset("valuation", TRADE_DATE)
+        self.transport.mode = "fields_all_null"
+        with self.assertRaises(QualityError) as ctx:
+            self.pipe.run_dataset("valuation", TRADE_DATE)
+        errors = "; ".join(ctx.exception.report["errors"])
+        self.assertIn("field gate: valuation.volume_ratio non-null rate 0.0000 < 0.9", errors)
+        self.assertIn("field gate: valuation.dv_ttm non-null rate 0.0000 < 0.3", errors)
+        # previous good publication stays active
+        self.assertEqual(self._active(), first["batch_id"])
+        # rejected batch left staged with readable error + field stats
+        rejected = self.db.fetchone(
+            "SELECT * FROM batches WHERE state='staged' AND dataset='valuation' ORDER BY started_at DESC",
+        )
+        self.assertIsNotNone(rejected)
+        self.assertIn("field gate: valuation.volume_ratio", str(rejected["error"]))
+        import json
+
+        quality = json.loads(rejected["quality_json"])
+        self.assertEqual(quality["fields"]["volume_ratio"]["nonnull"], 0)
+        self.assertEqual(quality["fields"]["dv_ttm"]["nonnull"], 0)
+
+    def test_volume_ratio_all_null_alone_rejected(self) -> None:
+        self.pipe.run_dataset("valuation", TRADE_DATE)
+        self.transport.mode = "vr_all_null"
+        with self.assertRaises(QualityError):
+            self.pipe.run_dataset("valuation", TRADE_DATE)
+        self.assertEqual(
+            self.db.fetchone(
+                "SELECT active_batch FROM publications WHERE dataset='valuation' AND trade_date=?",
+                (TRADE_DATE,),
+            )["active_batch"],
+            "20240902-valuation-001",
+        )
+
+    def test_dv_ttm_all_null_rejected_by_floor_and_collapse(self) -> None:
+        prev_day = "20240830"
+        prev = self.pipe.run_dataset("valuation", prev_day)  # prev dv nonnull 0.5
+        self.transport.mode = "dv_all_null"
+        with self.assertRaises(QualityError) as ctx:
+            self.pipe.run_dataset("valuation", TRADE_DATE)
+        errors = "; ".join(ctx.exception.report["errors"])
+        self.assertIn("field gate: valuation.dv_ttm non-null rate 0.0000 < 0.3", errors)
+        self.assertIn(f"dropped > 0.15 vs prev batch {prev['batch_id']}", errors)
+
+    def test_nonfinite_values_rejected(self) -> None:
+        self.pipe.run_dataset("valuation", TRADE_DATE)
+        rows = self.pipe.adapter.normalize(
+            "valuation", self.pipe._guarded_fetch("valuation", {"trade_date": TRADE_DATE})
+        )
+        for row in rows:
+            row["volume_ratio"] = float("inf")
+        with self.assertRaises(QualityError) as ctx:
+            self.pipe.run_dataset("valuation", TRADE_DATE, prepared_rows=rows)
+        errors = "; ".join(ctx.exception.report["errors"])
+        self.assertIn("field gate: valuation.volume_ratio non-finite rate 1.0000 > 0.01", errors)
+
+    def test_gate_off_when_not_configured(self) -> None:
+        pipe, _db = make_pipe(ValuationTransport(), quality_extra={"field_gates": {}})
+        pipe.ingest_reference(TRADE_DATE)
+        pipe.adapter._transport.mode = "fields_all_null"
+        result = pipe.run_dataset("valuation", TRADE_DATE)
+        self.assertEqual(result["state"], "published")  # legacy behavior when unconfigured
+
+    def test_gate_applies_to_any_configured_dataset(self) -> None:
+        gates = {"daily": {"fields": ["volume"], "min_nonnull_rate": 0.9, "max_nonfinite_rate": 0.01}}
+        pipe, _db = make_pipe(ValuationTransport(), quality_extra={"field_gates": gates})
+        pipe.ingest_reference(TRADE_DATE)
+
+        def null_volume(api_name, params, fields):
+            if api_name != "daily":
+                return fake_transport(api_name, params, fields)
+            rows = fake_transport(api_name, params, fields)
+            for row in rows:
+                row["vol"] = None
+            return rows
+
+        pipe.adapter._transport = null_volume
+        with self.assertRaises(QualityError) as ctx:
+            pipe.run_dataset("daily", TRADE_DATE)
+        errors = "; ".join(ctx.exception.report["errors"])
+        self.assertIn("field gate: daily.volume non-null rate 0.0000 < 0.9", errors)
+
+
+class GateRetryInterplayTests(unittest.TestCase):
+    def test_rejected_valuation_stays_missing_and_retry_publishes_later(self) -> None:
+        transport = ValuationTransport()
+        transport.mode = "fields_all_null"
+        tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        db = HubDB(Path(tmp.name) / "hub.db")
+        adapter = TushareAdapter("x", transport=transport)
+        settings = Settings(
+            encryption_key=SecretVault.generate_key(),
+            db_path=db.path,
+            backup_dir=Path(tmp.name) / "backups",
+            quality={"field_gates": FIELD_GATES, "max_publish_attempts": 2},
+        )
+        pipe = Pipeline(db, adapter, settings)
+        pipe.ingest_reference("20240902")
+        sched = Scheduler(db, pipe)
+
+        sched.tick(clock_at("20240902", 15, 5))   # valuation rejected by field gate
+        sched.tick(clock_at("20240902", 15, 10))
+        self.assertIn("valuation", pipe.missing_official_datasets("20240902"))
+        self.assertEqual(
+            pipe.active_batch("valuation", "20240902"),
+            None,
+        )
+
+        transport.mode = "ok"
+        sched.tick(clock_at("20240902", 15, 45))  # retry passes the gate
+        self.assertNotIn("valuation", pipe.missing_official_datasets("20240902"))
+        rows = db.fetchall("SELECT * FROM eod_valuation WHERE trade_date='20240902'")
+        self.assertTrue(rows)
+        self.assertTrue(all(row["volume_ratio"] is not None for row in rows))
+
+
+class ForceRepublishTests(unittest.TestCase):
+    def test_run_dataset_over_published_keeps_prev_for_rollback(self) -> None:
+        transport = ValuationTransport()
+        pipe, db = make_pipe(transport)
+        pipe.ingest_reference(TRADE_DATE)
+        first = pipe.run_dataset("valuation", TRADE_DATE)
+        transport.mode = "vr_all_null"
+        with self.assertRaises(QualityError):
+            pipe.run_dataset("valuation", TRADE_DATE)  # gate holds: bad re-publish refused
+        transport.mode = "ok"
+        second = pipe.run_dataset("valuation", TRADE_DATE)  # CLI --force path
+        self.assertNotEqual(first["batch_id"], second["batch_id"])
+        pub = db.fetchone(
+            "SELECT * FROM publications WHERE dataset='valuation' AND trade_date=?",
+            (TRADE_DATE,),
+        )
+        self.assertEqual(pub["active_batch"], second["batch_id"])
+        self.assertEqual(pub["prev_batch"], first["batch_id"])
+        rolled = pipe.rollback("valuation", TRADE_DATE, actor="cli")
+        self.assertEqual(rolled["active_batch"], first["batch_id"])
+
+
+if __name__ == "__main__":
+    unittest.main()

xiaobai-datahub/tests/test_stocks_refresh.py

@@ -0,0 +1,168 @@
+from __future__ import annotations
+
+import unittest
+from pathlib import Path
+import tempfile
+
+from datahub.adapters.tushare import TushareAdapter
+from datahub.crypto import SecretVault
+from datahub.db import HubDB
+from datahub.pipeline import Pipeline
+from datahub.scheduler import Scheduler
+from datahub.serving import V1API
+from datahub.settings import Settings
+from tests.fixtures import TRADE_DATE, fake_transport
+from tests.test_eod_retry import clock_at
+
+
+class StockMasterTransport:
+    """fake_transport with a mutable stock_basic list (new listings / renames)."""
+
+    def __init__(self) -> None:
+        self.stocks = [
+            {"ts_code": "600000.SH", "symbol": "600000", "name": "浦发银行", "area": "上海", "industry": "银行", "market": "主板", "list_status": "L", "list_date": "19991110"},
+            {"ts_code": "920071.BJ", "symbol": "920071", "name": "N金钛", "area": "辽宁", "industry": "小金属", "market": "北交所", "list_status": "L", "list_date": "20240901"},
+        ]
+
+    def __call__(self, api_name: str, params: dict, fields: str):
+        if api_name == "stock_basic":
+            return [dict(row) for row in self.stocks]
+        return fake_transport(api_name, params, fields)
+
+    def rename_and_add(self) -> None:
+        for row in self.stocks:
+            if row["ts_code"] == "920071.BJ":
+                row["name"] = "金钛股份"  # N-prefix removed the day after listing
+        self.stocks.append(
+            {"ts_code": "920289.BJ", "symbol": "920289", "name": "N华汇", "area": "广东", "industry": "专用机械", "market": "北交所", "list_status": "L", "list_date": "20240902"}
+        )
+
+
+def make_pipe(transport):
+    tmp = tempfile.TemporaryDirectory()
+    db = HubDB(Path(tmp.name) / "hub.db")
+    adapter = TushareAdapter("test-token", transport=transport)
+    settings = Settings(
+        encryption_key=SecretVault.generate_key(),
+        api_token="t" * 32,
+        admin_password="admin-pass",
+        tushare_token="test-token",
+        db_path=db.path,
+        quality={"max_publish_attempts": 3, "publication_generations": 3},
+        scheduler_enabled=False,
+    )
+    pipe = Pipeline(db, adapter, settings)
+    pipe._tmp = tmp
+    return pipe, db
+
+
+class StocksRefreshTests(unittest.TestCase):
+    def setUp(self) -> None:
+        self.transport = StockMasterTransport()
+        self.pipe, self.db = make_pipe(self.transport)
+        self.pipe.ingest_reference(TRADE_DATE)
+
+    def _stocks_api(self) -> dict:
+        return V1API(self.db, self.pipe, self.pipe.settings).handle("/v1/stocks", {})
+
+    def test_first_refresh_publishes_snapshot_with_meta(self) -> None:
+        result = self.pipe.refresh_stocks(TRADE_DATE)
+        self.assertEqual(result["state"], "published")
+        self.assertEqual(result["rows"], 2)
+        self.assertTrue(result["batch_id"].startswith("20240902-stocks-"))
+        payload = self._stocks_api()
+        self.assertEqual(payload["meta"]["batch_id"], result["batch_id"])
+        self.assertIsNotNone(payload["meta"]["published_at"])
+        self.assertEqual(len(payload["data"]), 2)
+        names = {row["ts_code"]: row["name"] for row in payload["data"]}
+        self.assertEqual(names["920071.BJ"], "N金钛")
+        self.assertNotIn("batch_id", payload["data"][0])
+
+    def test_new_listing_and_rename_publish_new_batch(self) -> None:
+        first = self.pipe.refresh_stocks(TRADE_DATE)
+        self.transport.rename_and_add()
+        second = self.pipe.refresh_stocks(TRADE_DATE)
+        self.assertEqual(second["state"], "published")
+        self.assertNotEqual(second["batch_id"], first["batch_id"])
+        payload = self._stocks_api()
+        names = {row["ts_code"]: row["name"] for row in payload["data"]}
+        self.assertEqual(names["920071.BJ"], "金钛股份")
+        self.assertIn("920289.BJ", names)
+        self.assertEqual(names["920289.BJ"], "N华汇")
+        # stock_master is refreshed too (code resolution stays current)
+        master = self.db.fetchone("SELECT name FROM stock_master WHERE ts_code='920289.BJ'")
+        self.assertEqual(master["name"], "N华汇")
+
+    def test_unchanged_refresh_is_idempotent(self) -> None:
+        first = self.pipe.refresh_stocks(TRADE_DATE)
+        again = self.pipe.refresh_stocks(TRADE_DATE)
+        self.assertEqual(again["state"], "skipped")
+        self.assertEqual(again["reason"], "unchanged")
+        self.assertEqual(again["batch_id"], first["batch_id"])
+        count = self.db.fetchone(
+            "SELECT COUNT(*) AS n FROM batches WHERE dataset='stocks' AND trade_date=?",
+            (TRADE_DATE,),
+        )["n"]
+        self.assertEqual(count, 1)
+
+    def test_force_republishes_even_unchanged(self) -> None:
+        first = self.pipe.refresh_stocks(TRADE_DATE)
+        forced = self.pipe.refresh_stocks(TRADE_DATE, force=True)
+        self.assertEqual(forced["state"], "published")
+        self.assertNotEqual(forced["batch_id"], first["batch_id"])
+
+    def test_snapshot_pinned_until_next_publish(self) -> None:
+        first = self.pipe.refresh_stocks(TRADE_DATE)
+        self.transport.rename_and_add()
+        # upstream changed but no refresh ran: published snapshot is untouched
+        _, snapshot = self.pipe.published_stock_snapshot(TRADE_DATE)
+        names = {row["ts_code"]: row["name"] for row in snapshot}
+        self.assertEqual(names["920071.BJ"], "N金钛")
+        self.assertNotIn("920289.BJ", names)
+        self.assertEqual(len(snapshot), 2)
+
+    def test_dataset_status_includes_stocks(self) -> None:
+        result = self.pipe.refresh_stocks(TRADE_DATE)
+        payload = V1API(self.db, self.pipe, self.pipe.settings).handle(
+            "/v1/datasets/status", {"date": [TRADE_DATE]}
+        )
+        by_name = {item["dataset"]: item for item in payload["data"]}
+        self.assertIn("stocks", by_name)
+        self.assertEqual(by_name["stocks"]["batch_id"], result["batch_id"])
+        self.assertIsNotNone(by_name["stocks"]["published_at"])
+
+
+class StocksRefreshSchedulingTests(unittest.TestCase):
+    def _make(self):
+        tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        db = HubDB(Path(tmp.name) / "hub.db")
+        transport = StockMasterTransport()
+        adapter = TushareAdapter("x", transport=transport)
+        settings = Settings(
+            encryption_key=SecretVault.generate_key(),
+            db_path=db.path,
+            backup_dir=Path(tmp.name) / "backups",
+            quality={"stocks_refresh_times": ["20:00", "23:10"]},
+        )
+        pipe = Pipeline(db, adapter, settings)
+        pipe.ingest_reference(TRADE_DATE)
+        return db, pipe, Scheduler(db, pipe)
+
+    def test_scheduled_refresh_runs_on_open_day(self) -> None:
+        db, pipe, sched = self._make()
+        ran = sched.tick(clock_at(TRADE_DATE, 20, 0))
+        self.assertIn("stocks_refresh", ran)
+        ran = sched.tick(clock_at(TRADE_DATE, 23, 10))
+        self.assertIn("stocks_refresh", ran)  # second slot catches late renames
+        self.assertIsNotNone(pipe.active_batch("stocks", TRADE_DATE))
+
+    def test_no_refresh_on_closed_day(self) -> None:
+        db, _pipe, sched = self._make()
+        sched.tick(clock_at("20240907", 20, 30))  # fixture: Saturday closed
+        runs = db.fetchall("SELECT * FROM job_runs WHERE job_id='stocks_refresh'")
+        self.assertEqual(runs, [])
+
+
+if __name__ == "__main__":
+    unittest.main()