fix(data): 收口 A 股 Baostock venue 迁移

.env.example

@@ -103,12 +103,13 @@ FRED_API_KEY=
 # 给调度器一份「每天去给哪些指数拍成分快照」的点名册：逗号分隔指数代码，留空=禁用调度
 # （后台循环不起，constituent_snapshot 表保持空，panel_score 传 index_code 一律走 non-PIT 降级）。
 #
-# 为什么要它：akshare 只回**当前**成分、免费历史拿不到 → 唯一 PIT 路径是从启用日起每日把
+# 为什么要它：免费源只回**当前**成分、历史成分拿不到 → 唯一 PIT 路径是从启用日起每日把
 # 当前成分落库、向前累积。**覆盖从配置那天起向前增长，不回填历史，越早配可用于无偏回测的历史越长。**
 #
-# 目前 connector 走 akshare 中证接口，只支持 **A股指数代码**（美股/港股成分快照需另接数据源，未做）：
+# 目前 connector 对三大核心指数优先走 Baostock，其余 A股指数回退 AkShare 中证接口；
+# 美股/港股成分快照需另接数据源，未做：
 #   000300 沪深300 / 000905 中证500 / 000016 上证50 / 000852 中证1000 / 399006 创业板指
-# 不预设市场：按你实际要回测的指数填，别贪多（akshare 是爬虫接口，抓越多越易被封）。
+# 不预设市场：按你实际要回测的指数填，别贪多（公开接口抓越多越容易限流）。
 # 手动 POST /constituents/snapshot 不受本项影响（仍可随时按需落一份）。
 CONSTITUENT_SNAPSHOT_INDICES=
 # 调度检查间隔（小时，默认 12）。幂等：每轮只补「今天还没快照」的指数，<24h 不会重复打源站；

README.md

@@ -186,7 +186,7 @@ Inalpha splits *scheduling* from *compute*. The agent runtime fans out the grid
 A deep dive doesn't hand you one "correct answer." Beyond the usual technical, fundamental, and sentiment analysts, you can convene a panel of master personas — Buffett (value / moats), Lynch (GARP growth), Wood (disruptive innovation), Burry (contrarian / bubbles), Druckenmiller (macro trends), Marks (cycles / risk): each argues in their own style, naturally forming opposing views that feed a synthesized judgment.
 
 - **Opt-in, cost-controlled.** A plain deep dive costs the same as before; you only pay for the masters you actually convene.
-- **Views grounded in data.** Each persona reads technicals / fundamentals / web intel with `as_of` pinned to *now* — no passing a stale forecast off as the present. Fundamentals are read point-in-time, filtered by report period and release lag so a backtest never sees a number before it was public (akshare today; yfinance v1 not yet PIT, flagged in place).
+- **Views grounded in data.** Each persona reads technicals / fundamentals / web intel with `as_of` pinned to *now* — no passing a stale forecast off as the present. Fundamentals are read point-in-time, filtered by report period and release lag so a backtest never sees a number before it was public (Baostock today; yfinance v1 not yet PIT, flagged in place).
 - **Structured bull / bear / risk debate — now live.** Beyond the parallel analysts, opposing-stance bull and bear researchers argue across rounds while a risk researcher stress-tests both — triggered only when the analysts genuinely disagree, with a soft early-stop when arguments stop changing and the full decision chain (why it debated, why it stopped, how it was synthesized) persisted for replay.
 
 ### 6. Skills — absorb outside research playbooks
@@ -234,7 +234,7 @@ Where each capability stands today. Live module inventory and the end-to-end dec
 | ✅ Shipped | Bull / bear researcher debate | D-9 | opposing-stance researchers under `services/research` |
 | ✅ Shipped | Scheduler / cron agent mode | D-9 | `scheduler_jobs` + advisory lock + `/api/scheduler/*` management plane |
 | ✅ Shipped | RiskGuard per-account isolation | D-9.1a | `RiskGuardFactory` removes cross-account state bleed |
-| ✅ Shipped | Multi-market data sources — web search + financial fundamentals | D-10 | zero-key DDGS web search · akshare (A-shares/HK) + yfinance (global) fundamentals · analyst integration + fallback · per-market lookbackDays |
+| ✅ Shipped | Multi-market data sources — web search + financial fundamentals | D-10 | zero-key DDGS web search · `baostock` is the logical A-share venue: Tencent HTTPS bars/ticker + Baostock fundamentals/calendar/constituents; yfinance covers global markets including HK · analyst integration + fallback · per-market lookbackDays |
 | ✅ Shipped | Risk engine — all 5 rules live in HTTP path | D-9 closed | `closed_trades` writes from HTTP order flow; `RoutingCalendar` for US equity + crypto; all trade-based rules trigger on real data |
 | ✅ Shipped | `askUserChoice` — `ask` permission path | D-11 (issue #2) | pending-permission flow resolves the `ask` state (no longer a workaround) |
 | ✅ Shipped | `permissions.yaml` configuration | D-11 (issue #4) | `config/permissions.default.yaml` + `yaml_loader.ts` replace the hard-coded `defaults.ts` |
@@ -249,7 +249,7 @@ Where each capability stands today. Live module inventory and the end-to-end dec
 | ✅ Shipped | Factor discovery — L1 | D-12 | restricted qlib-style DSL (zero eval/exec) · `factor_candidates` pool · multiple-testing correction + null-IC benchmark · discovery workflow → propose; register is dashboard-only |
 | ✅ Shipped | Cross-sectional factor scoring | D-12 | `factor.panel_score` · `POST /panel/score` · cross-sectional Rank IC (rank the pool each period vs forward cross-sectional return) · native Alpha101 a1/a3 · orthogonal to single-name timing |
 | ✅ Shipped | Time-series cross-validation — anti-overfitting | D-12 | WalkForward / PurgedKFold / Combinatorial Purged CV + Deflated Sharpe · `POST /backtest/cv` · test fold always includes the latest bar · auto-fallback to walk-forward when samples are short |
-| ✅ Shipped | Point-in-time fundamentals | D-12 | akshare financials filtered by report-period + release lag · `GET /fundamentals?as_of=` · prevents look-ahead (yfinance v1 not yet PIT, explicitly flagged) |
+| ✅ Shipped | Point-in-time fundamentals | D-12 | Baostock financials filtered by actual publication date · `GET /fundamentals?as_of=` · prevents look-ahead (yfinance v1 not yet PIT, explicitly flagged) |
 | 🗓️ Planned | Strategy evolution — E2 | E2 | multi-generation loop + MAP-Elites + Island Model + `unified-diff` mutations (E1 single-generation closed loop already shipped in D-9) |
 | 🗓️ Planned | Factor discovery — L2 / L3 | L2 / L3 | multi-agent factor crew (L2) + weekly automated scans (L3), on top of the L1 DSL pipeline already shipped |
 | 🗓️ Planned | Automated decay handling | TBD | reflection-driven backtest + auto-trim of decaying factors — today the decay patrol only alerts, never moves the book |

README.zh-CN.md

@@ -184,7 +184,7 @@ Inalpha 把*调度*和*算力*分开：agent runtime 负责扇出网格、聚合
 一次深度研究不只给你一个"标准答案"。除了技术面、基本面、情绪面的常规 analyst，你还能叫上一支"大师团"——巴菲特（价值 / 护城河）、林奇（GARP 成长）、伍德（颠覆创新）、伯里（逆向 / 泡沫）、德鲁肯米勒（宏观趋势）、马克斯（周期 / 风险）：各按自己的风格给观点，天然形成对立视角，再汇进综合判断。
 
 - **按需启用，成本可控。** 不点名时普通研究成本不变；点了哪几位大师，才多跑那几次。
-- **观点要落到数据。** 大师视角接技术面 / 基本面 / web 情报，`as_of` 是真现在，不许拿过时预测当现在。财报按 point-in-time 读取，依报告期 + 发布滞后过滤，回测看不到尚未公开的数字（akshare 已生效；yfinance v1 尚未 PIT，已就地标注）。
+- **观点要落到数据。** 大师视角接技术面 / 基本面 / web 情报，`as_of` 是真现在，不许拿过时预测当现在。财报按 point-in-time 读取，依实际披露日期过滤，回测看不到尚未公开的数字（Baostock 已生效；yfinance v1 尚未 PIT，已就地标注）。
 - **bull / bear / risk 结构化辩论——已上线。** 在并行 analyst 之外，立场对抗的多头与空头研究员多轮交锋，风险官同时压测双方——只在 analyst 真出现分歧时才触发，论点不再变化时软早停，完整决策链路（为什么辩、为什么停、如何综合）全程落盘可复盘。
 
 ### 6. Skills · 吸收外部投研方法论
@@ -232,7 +232,7 @@ Inalpha 把*调度*和*算力*分开：agent runtime 负责扇出网格、聚合
 | ✅ 已上线 | Bull / Bear 研究员辩论 | D-9 | `services/research` 立场对抗研究员 |
 | ✅ 已上线 | Scheduler / cron agent 模式 | D-9 | `scheduler_jobs` + advisory lock + `/api/scheduler/*` 管理面 |
 | ✅ 已上线 | RiskGuard 账户级隔离 | D-9.1a | `RiskGuardFactory` 去除跨账户状态串联 |
-| ✅ 已上线 | 多市场数据源 — web 搜索 + 财报基本面 | D-10 | DDGS 零 key web search · akshare（A股/港股）+ yfinance（全球）基本面 · analyst 接入 + 兜底 · lookbackDays 按市场分化 |
+| ✅ 已上线 | 多市场数据源 — web 搜索 + 财报基本面 | D-10 | DDGS 零 key web search · `baostock` 是 A 股逻辑 venue：腾讯 HTTPS 行情/最新价 + Baostock 基本面/日历/成分；yfinance 覆盖全球（含港股）· analyst 接入 + 兜底 · lookbackDays 按市场分化 |
 | ✅ 已上线 | 风控引擎 — 5 条规则全在 HTTP 路径激活 | D-9 收口 | `closed_trades` 由 HTTP 订单流写入；`RoutingCalendar` 覆盖美股 + crypto；trade-based 规则全在真实数据上触发 |
 | ✅ 已上线 | `askUserChoice` — `ask` 权限路径 | D-11（issue #2） | pending-permission 流程把 `ask` 状态从 workaround 救回（已收口） |
 | ✅ 已上线 | `permissions.yaml` 配置化 | D-11（issue #4） | `config/permissions.default.yaml` + `yaml_loader.ts` 替代 `defaults.ts` 硬编码 |
@@ -247,7 +247,7 @@ Inalpha 把*调度*和*算力*分开：agent runtime 负责扇出网格、聚合
 | ✅ 已上线 | 因子发现 — L1 | D-12 | 受限 qlib 风格 DSL（零 eval/exec）· `factor_candidates` 候选池 · 多重检验校正 + null IC 基准 · discovery workflow → propose；register 仅 dashboard 人工 |
 | ✅ 已上线 | 横截面因子打分 | D-12 | `factor.panel_score` · `POST /panel/score` · 横截面 Rank IC（每期对全池按因子排序 vs 跨标的前瞻收益）· Alpha101 a1/a3 原生 · 与单标的择时口径正交 |
 | ✅ 已上线 | 时序交叉验证 — 防过拟合 | D-12 | WalkForward / PurgedKFold / Combinatorial Purged CV + Deflated Sharpe · `POST /backtest/cv` · test 段始终含最新 bar · 样本不足自动回落 walk-forward |
-| ✅ 已上线 | 财报 point-in-time | D-12 | akshare 财报按报告期 + 发布滞后过滤 · `GET /fundamentals?as_of=` · 防前视（yfinance v1 尚未 PIT，已显式标注） |
+| ✅ 已上线 | 财报 point-in-time | D-12 | Baostock 财报按实际披露日期过滤 · `GET /fundamentals?as_of=` · 防前视（yfinance v1 尚未 PIT，已显式标注） |
 | 🗓️ 已规划 | 策略进化 — E2 | E2 | 多代演化 + MAP-Elites + Island Model + `unified-diff` 变异（E1 单代闭环已在 D-9 上线） |
 | 🗓️ 已规划 | 因子发现 — L2 / L3 | L2 / L3 | 多 agent 因子小组（L2）+ 每周自动扫描（L3），建立在已上线的 L1 DSL pipeline 之上 |
 | 🗓️ 已规划 | 因子衰减自动处置 | 待定 | 反思驱动回测 + 衰减因子自动剔除——当前衰减巡检只告警、绝不替你动仓 |

docs/04-current-state.md

@@ -430,8 +430,8 @@ manager 综合已覆盖）、debate 轮内并行化、reasoning 进活动流前
   Sharpe 分布 + DSR；`POST /backtest/cv`（cpcv 数据不足自动回落 walk_forward，整体甩
   ProcessPool 不阻塞事件循环）；`paper.cv_backtest` tool。**单策略报 DSR、不报 PBO**（CSCV-PBO
   需多配置作输入，留 grid 层）。
-- **财报 / bar Point-in-Time（阶段 A）**（ADR-0053 · #100）：akshare 财报按"报告期末 + 120 天
-  发布滞后 ≤ as_of"过滤未披露报告期（防未来函数）；`GET /fundamentals?as_of=`；paper
+- **财报 / bar Point-in-Time（阶段 A）**（ADR-0053 · #100）：Baostock 财报按实际公告日
+  `pubDate ≤ as_of` 过滤未披露报告期（防未来函数）；`GET /fundamentals?as_of=`；paper
   `DataClient.get_bars_pit` 把 bar 截断到 as_of。yfinance v1 不做 PIT 但响应显式标注；与 FRED
   宏观 PIT（ADR-0044）口径对齐。
 

infra/migrations/versions/0036_bars_a_share_symbol_canonical.py

@@ -0,0 +1,93 @@
+"""规范化 A 股 bars 的历史 symbol 格式。
+
+兼容期曾接受 ``SH.600519``、``600519.SH`` 等形式。venue 已统一为 ``baostock`` 后，
+这些 symbol 仍会形成不同的持久化 namespace；本迁移将其合并为 ``sh.600519`` / ``sz.000001``。
+"""
+
+from __future__ import annotations
+
+from alembic import op
+
+revision: str = "0036"
+down_revision: str | None = "0035"
+branch_labels: str | tuple[str, ...] | None = None
+depends_on: str | tuple[str, ...] | None = None
+
+_PREFIX_SYMBOL_PATTERN = "^(sh|sz)[.][0-9][0-9][0-9][0-9][0-9][0-9]$"
+_SUFFIX_SYMBOL_PATTERN = "^[0-9][0-9][0-9][0-9][0-9][0-9][.](sh|sz)$"
+
+_CANONICAL_SYMBOL_SQL = f"""
+CASE
+    WHEN btrim(symbol) ~* '{_PREFIX_SYMBOL_PATTERN}' THEN
+        lower(split_part(btrim(symbol), '.', 1)) || '.' ||
+        split_part(btrim(symbol), '.', 2)
+    WHEN btrim(symbol) ~* '{_SUFFIX_SYMBOL_PATTERN}' THEN
+        lower(substring(btrim(symbol) from '\\.([^.]+)$')) || '.' ||
+        regexp_replace(btrim(symbol), '\\.(sh|sz)$', '', 'i')
+END
+"""
+
+
+def upgrade() -> None:
+    """把旧 venue / symbol 组合合并到 canonical Baostock identity。"""
+    op.execute(
+        f"""
+        INSERT INTO bars (ts, venue, symbol, timeframe, open, high, low, close, volume)
+        SELECT DISTINCT ON (ts, canonical_symbol, timeframe)
+            ts,
+            'baostock',
+            canonical_symbol,
+            timeframe,
+            open,
+            high,
+            low,
+            close,
+            volume
+        FROM (
+            SELECT
+                ts,
+                venue,
+                symbol,
+                timeframe,
+                open,
+                high,
+                low,
+                close,
+                volume,
+                {_CANONICAL_SYMBOL_SQL} AS canonical_symbol
+            FROM bars
+            WHERE venue IN ('akshare', 'baostock')
+              AND (
+                  btrim(symbol) ~* '{_PREFIX_SYMBOL_PATTERN}'
+                  OR btrim(symbol) ~* '{_SUFFIX_SYMBOL_PATTERN}'
+              )
+        ) AS candidates
+        WHERE canonical_symbol IS NOT NULL
+          AND NOT (venue = 'baostock' AND symbol = canonical_symbol)
+        ORDER BY
+            ts,
+            canonical_symbol,
+            timeframe,
+            CASE WHEN venue = 'baostock' THEN 0 ELSE 1 END,
+            symbol
+        ON CONFLICT (ts, venue, symbol, timeframe) DO NOTHING
+        """
+    )
+    op.execute(
+        f"""
+        DELETE FROM bars
+        WHERE venue IN ('akshare', 'baostock')
+          AND (
+              btrim(symbol) ~* '{_PREFIX_SYMBOL_PATTERN}'
+              OR btrim(symbol) ~* '{_SUFFIX_SYMBOL_PATTERN}'
+          )
+          AND NOT (
+              venue = 'baostock'
+              AND symbol = ({_CANONICAL_SYMBOL_SQL})
+          )
+        """
+    )
+
+
+def downgrade() -> None:
+    """symbol 原始大小写/前后缀不可无损恢复，降级保持 canonical 数据。"""

packages/orchestration/skills/cn-equity-research/SKILL.md

@@ -24,7 +24,7 @@ fail 就 stop，不跨层补救**。同一结论要多源交叉验证，不信
 | 公司名 → 可交易代码（一切个股研究的第一步） | `data.search_symbol`（A股返 sh./sz. 格式） |
 | 实时价格 / 当日涨跌 | `data.get_ticker` |
 | 历史 K 线（趋势 / 区间定位） | `data.get_bars`（保持默认 fresh） |
-| PE_TTM / PB / ROE / 营收与利润增速 | `data.get_fundamentals`（venue=akshare） |
+| PE_TTM / PB / ROE / 营收与利润增速 | `data.get_fundamentals`（venue=baostock） |
 | 市场级财经快讯（盘面消息） | `data.get_market_news` |
 | 行业板块涨跌幅榜（顺风 / 逆风） | `data.get_market_sectors` |
 | 当日强势股 + 题材标签（主线确认） | `data.get_market_movers` |

packages/orchestration/src/tools/data.ts

@@ -266,7 +266,7 @@ export const dataGetTickerTool = createTool({
     取 venue/symbol 的"最新价"（单值，不是 K 线）。
 
     - fresh=true（**默认**）：直接调外部市场实时报价，绕过 DB 缓存。
-      **支持 venue**：binance / yfinance / alpaca；baostock / fred 不支持（返
+      **支持 venue**：binance / yfinance / alpaca / baostock；fred 不支持（返
       FRESH_NOT_SUPPORTED_FOR_VENUE，hint 提示切 fresh=false）。
       网络抖动 ~200-800ms；不要高频循环调（rate-limit）。
     - fresh=false：从 DB 拿最新 1m → fallback 1h，任意 venue 都支持。
@@ -281,8 +281,8 @@ export const dataGetTickerTool = createTool({
     何时不用：
     - 要历史走势 / 做技术分析 → 用 data.get_bars
     - 要 N 根 K 线 → 用 data.get_bars
-    - A 股 / FRED 想要"实时"价 → baostock/fred 没这能力，改 fresh=false 走 DB
-      cache（先 backfill_bars 灌一遍最新数据再调）；港股走 yfinance 有实时 ticker
+    - A 股现价可用 baostock + fresh=true；FRED 没有盘中报价能力，改用 data.get_bars
+      或显式 fresh=false 读 DB cache。港股走 yfinance
 
     坑：
     - 默认 fresh=true！想要 DB cache 必须显式传 fresh:false