feat(product-commerce): 款式主键统一 style_id 并下线评价采集

采集端与 data-hub 统一用 style_id (= dim_style.id) 关联款式，style_name 降级为显示冗余。款式主键： - 14 张采集表加 style_id BIGINT + style_match_method + *_style_id_idx 索引 - 一律逻辑外键：不建 FOREIGN KEY，早期 10 个 *_style_id_fk 在 schema.sql 里幂等 DROP - loader 下发 dim_style.id；db 层 attach_style_ids / resolve_style_ids_by_product 兜底 - 新增 backfill_style_id.py 三段回填（款名 / 日报按商品 ID / 广告表按 spu_id） - 新增视图 v_style_id_coverage、v_style_unmatched_product、v_style_unmatched_source - 实库补录 199 个商品 ID 到 67 款 spus；日报 style_id 覆盖 49.1% -> 79.0%， tm 广告 99.4% / jd 92.4% / dy 83.2%，其余 11 张 100% 评价采集下线： - product_reviews 归档为 product_reviews_archived_20261101 - 差评改由 shop.{tmall,jd,dy}_negative_reviews.daily 采集，消费方读 v_negative_reviews - 删除 import_product_reviews.py / orchestrate_review_collection.py 及相关命令与绑定 - dy_product_scraping 默认流程摘除评论采集，保留 --reviews-only 天猫广告： - 迁移到独立表 tm_product_ad_daily_metrics，不再写 product_daily_metrics.raw_data 同批附带： - 光合上传支持天猫新店 new_bag（仅黑曜石系列） - 京东商智双月日历排除 -diff-month 补位格 - 万相主体搜索词覆盖 STYLE_SEARCH_KEYWORDS - ERP 部分编码失败时用已成功编码销量兜底，避免整款归零
feat(product-commerce): 系列聚合去重与采集开关两层语义
2026-11-01 16:43:27 +08:00 · 2026-10-30 11:49:31 +08:00 · 2026-10-28 14:51:44 +08:00 · 2026-10-28 14:39:50 +08:00 · 2026-10-28 14:20:59 +08:00 · 2026-10-27 10:10:33 +08:00
194 changed files with 18104 additions and 11775 deletions
@@ -33,12 +33,15 @@ Thumbs.db
 # Agent-local learning logs
 .learnings/

+# Agent tool local state (security-scan ledger, session memory)
+/.mimosa/
+/.workbuddy/
+
 # Local tool artifacts
 /.playwright-cli/
 /output/
+
+# Local dsh workbench scaffold (generated at runtime, bundles node_modules)
+/workbench/
 /nul
 /new_all_bag.xlsx
-
-# Workbench (dsh plugin) local artifacts
-/workbench/plugin/node_modules/
-/workbench/cordis.local.yml
@@ -19,7 +19,7 @@ lint:
  stage: quality
  script:
    - uv run ruff check src tests
-    - uv run ruff check src/gyxx_flow/modules/content_marketing/daily_creator_exposure_scope.py src/gyxx_flow/modules/content_marketing/chanmama_export.py src/gyxx_flow/modules/content_marketing/self_operated_accounts.py src/gyxx_flow/modules/content_marketing/self_xhs_scraper.py src/gyxx_flow/modules/content_marketing/data/tools/sync_self_video_metrics.py src/gyxx_flow/modules/product_commerce/market_rank_limits.py src/gyxx_flow/modules/product_commerce/market_rank_product_import.py src/gyxx_flow/modules/shop_intelligence/collectors/dy_after_sales_quality.py src/gyxx_flow/modules/shop_intelligence/collectors/tmall_negative_reviews.py
+    - uv run ruff check src/gyxx_flow/modules/content_marketing/daily_creator_exposure_scope.py src/gyxx_flow/modules/content_marketing/chanmama_export.py src/gyxx_flow/modules/content_marketing/self_operated_accounts.py src/gyxx_flow/modules/content_marketing/self_xhs_scraper.py src/gyxx_flow/modules/content_marketing/data/tools/sync_self_video_metrics.py src/gyxx_flow/modules/product_commerce/market_rank_limits.py src/gyxx_flow/modules/product_commerce/market_rank_product_import.py src/gyxx_flow/modules/product_commerce/collect_douyin_qianchuan_ads.py src/gyxx_flow/modules/product_commerce/collect_erp_sales_theme_daily.py src/gyxx_flow/modules/shop_intelligence/collectors/dy_after_sales_quality.py src/gyxx_flow/modules/shop_intelligence/collectors/tmall_negative_reviews.py

 test:
  stage: test
@@ -41,6 +41,7 @@ uv run gyxx accounts login|sync [account_id]
 - Manual/backfill/maintenance commands must be declared in `config/commands.json`; never reintroduce recursive executable-script discovery.
 - Each executable `workflow_id` compiles to a LangGraph `StateGraph`. Preserve the public contracts of the workflow model, `RunJournal`, `LockManager`, shadow comparison, and `EffectLedger` when changing graph execution.
 - External systems: PostgreSQL via runtime-injected cloud DSN (`GYXX_POSTGRES_DSN`; no address/credential defaults in source, loopback rejected in cloud mode); Hermes loopback-only roles on `127.0.0.1:8642/8643` (key `GYXX_HERMES_API_KEY`); Feishu keeps its existing identity. Every browser binding keeps a unique CDP port plus isolated Profile/Cookie/storage-state; scripts sharing an account declare it under `accounts` in `runtime-bindings.json` — the vault `state/accounts/<id>/` is the single login authority (`gyxx accounts login/sync`) and its cookies merge read-only into member scripts while profiles stay isolated.
+- 登录统一管理：正式执行的工作流由 `src/gyxx_flow/cookie_gate.py` 在采集步骤前注入每账号 `cookie-gate.<id>` 步骤（blocking 账号失效→阻断并飞书通知人工重登；`login_command_id` 账号先自动账密恢复；`credential_env_names` 账号 defer 给脚本内账密登录）。阻断记录 `pending_reruns`，由调度器维护循环里的 `CookieGateRecoveryManager` 在登录态恢复后按原业务日期自动重跑并发飞书确认。采集脚本不得内嵌扫码/验证码等待——失效时抛 `CookieExpiredError`（`adapters/browser.py`）退出 75；编排器工作流在 `workflows.json` 用 `required_accounts` 声明账号依赖。不要重新引入 relogin_*/批量补采里的扫码重登子进程。
 - 启动或重启工作流控制台（`gyxx console`）必须携带 `--env-file D:\product-collector-analyze-flow\.env`（与调度器进程一致），否则 `PG_HOST/PG_PORT/PG_DB/PG_USER/PG_PASSWORD` 等云端 PostgreSQL 凭据和账号配置不会注入，「商品经营日报采集」「聚水潭全店铺款式日报」等正式执行会报“缺少 PostgreSQL 运行时凭据”。参考启动脚本：`var/tmp/start-console.ps1`。
 - Source sync (`gyxx sources status/apply`): manifest-driven three-way hash drift check; only untransformed, conflict-free one-sided changes may be auto-copied, and only with `--execute`. Credentials, cookies, logs, and collected data never enter sync manifests.

@@ -40,17 +40,6 @@ var/                            默认运行数据；不属于源码
 - `shop_intelligence`：店铺、竞店和京东自营业绩。
 - `supply_chain`：采购确认、补货、库存预警和采购单更新。

-## 智能工作台（dsh-refact 分支）
-
-`workbench/` 以 [deepseek-harness](https://github.com/deepseek-ai/deepseek-harness) 为
-agent 底座，把现有工作流以 dsh 插件形式接入 Web UI：侧边栏选中任意工作流即可提问、
-诊断、修复，支持监控告警与受控启动。使用与设计见 [workbench/README.md](workbench/README.md)
-与 [docs/workbench.md](docs/workbench.md)。
-
-```powershell
-powershell -File workbench\bin\start-workbench.ps1
-```
-
 ## 开发环境

 ```powershell
@@ -50,14 +50,7 @@
      "kind": "python",
      "notification_workflow_id": "content.marketing_report.daily"
    },
-    {
-      "id": "content.login.refresh",
-      "module": "content_marketing",
-      "entry": "data/tools/friday_relogin_parallel.py",
-      "kind": "python",
-      "notification_workflow_id": "content.relogin.weekly"
-    },
-    {
+        {
      "id": "content.creator_report.generate",
      "module": "content_marketing",
      "entry": "data/tools/generate_creator_report.py",
@@ -281,6 +274,17 @@
      "kind": "python",
      "notification_workflow_id": "product.online_stock_alert.daily"
    },
+    {
+      "id": "product.points_exchange_weekly.run",
+      "module": "product_commerce",
+      "entry": "points_exchange_weekly.py",
+      "kind": "python",
+      "default_args": [
+        "--from",
+        "{business_date}"
+      ],
+      "notification_workflow_id": "product.points_exchange.weekly"
+    },
    {
      "id": "product.import.daily",
      "module": "product_commerce",
@@ -472,6 +476,23 @@
        "{business_date}"
      ]
    },
+    {
+      "id": "product.erp_sales_theme_daily.collect",
+      "module": "product_commerce",
+      "entry": "collect_erp_sales_theme_daily.py",
+      "kind": "python",
+      "default_args": [
+        "--date",
+        "{business_date}",
+        "--execute"
+      ]
+    },
+    {
+      "id": "product.hk_anpei_public_inventory.refresh",
+      "module": "product_commerce",
+      "entry": "public_sku_inventory_sheet.py",
+      "kind": "python"
+    },
    {
      "id": "product.backfill.run",
      "module": "product_commerce",
@@ -491,16 +512,6 @@
      "kind": "python",
      "notification_workflow_id": "product.market_rank"
    },
-    {
-      "id": "product.review.orchestrate",
-      "module": "product_commerce",
-      "entry": "orchestrate_review_collection.py",
-      "kind": "python",
-      "default_args": [
-        "--target-date",
-        "{business_date}"
-      ]
-    },
    {
      "id": "supply.replenishment.arrival_sync",
      "module": "supply_chain",
@@ -222,9 +222,6 @@
    "content.marketing_report.daily": {
      "condition": "正式执行且启用发送时投递最终营销日报。"
    },
-    "content.relogin.weekly": {
-      "condition": "刷新期间投递各平台扫码二维码，并在平台失败时投递失败汇总。"
-    },
    "product.alert.daily": {
      "condition": "命中销量下滑规则且尚未重复通知时投递告警。"
    },
@@ -234,6 +231,9 @@
    "product.online_stock_alert.daily": {
      "condition": "线上商品当前库存低于 3 时向配置接收人发送库存预警。"
    },
+    "product.points_exchange.weekly": {
+      "condition": "每周汇总上一周积分兑换商品排行后向配置接收人发送周报。"
+    },
    "supply.purchase_confirmation.daily": {
      "condition": "采购确认结果生成完成后投递最终业务通知。"
    },
@@ -253,6 +253,15 @@
        "wang_yunlong",
        "stock_alert_user_19832771651"
      ]
+    },
+    "product.points_exchange.weekly": {
+      "enabled": true,
+      "app_profile": "hermes-analyzer",
+      "person_ids": [
+        "online_stock_alert_recipient",
+        "wang_yunlong",
+        "stock_alert_user_19832771651"
+      ]
    }
  }
 }
@@ -88,7 +88,6 @@
        "jd.com"
      ],
      "required_cookie_names": [
-        "pin",
        "thor"
      ],
      "credential_env_names": [
@@ -116,7 +115,6 @@
        "jd.com"
      ],
      "required_cookie_names": [
-        "pin",
        "thor"
      ],
      "credential_env_names": [
@@ -145,10 +143,7 @@
      ],
      "required_cookie_names": [
        "u_id",
-        "u_co_id",
-        "ASP.NET_SessionId",
-        ".ASPXAUTH",
-        "token"
+        "u_co_id"
      ],
      "credential_env_names": [
        "ERP_USERNAME",
@@ -168,6 +163,7 @@
        "product_commerce:collect_erp_yesterday_metrics.py",
        "product_commerce:erp_login_product_analysis.py",
        "product.erp_all_shop_daily",
+        "product.erp_sales_theme_daily.collect",
        "product.sales_sheet.sync",
        "supply_chain:orchestrator/scripts/ProductReplenishment.py",
        "supply_chain:orchestrator/scripts/PurchaseConfirmation.py",
@@ -298,7 +294,8 @@
        "xiaohongshu.com"
      ],
      "required_cookie_names": [
-        "web_session",
+        "customer-sso-sid",
+        "solar.beaker.session.id",
        "a1"
      ],
      "keepalive": {
@@ -312,8 +309,6 @@
      },
      "members": [
        "content.comments.collect_xiaohongshu",
-        "content_marketing:data/tools/relogin_pgy.py",
-        "content_marketing:data/tools/relogin_xiaohongshu.py",
        "content_marketing:pgy_xhs_scraper_v2.py"
      ]
    },
@@ -340,7 +335,6 @@
      },
      "members": [
        "content.comments.collect_douyin",
-        "content_marketing:data/tools/relogin_douyin.py",
        "content_marketing:self_douyin_scraper.py"
      ]
    },
@@ -353,9 +347,6 @@
        "xingtu.cn"
      ],
      "required_cookie_names": [
-        "sessionid",
-        "sessionid_ss",
-        "sid_tt",
        "sid_guard"
      ],
      "keepalive": {
@@ -367,13 +358,15 @@
        "headless": true,
        "real_chrome": true
      },
-      "members": [
-        "content_marketing:data/tools/relogin_xingtu.py"
-      ]
+      "members": []
    },
    "chanmama": {
      "cdp_port": 22207,
      "login_mode": "D",
+      "credential_env_names": [
+        "CHANMAMA_ACCOUNT",
+        "CHANMAMA_PASSWORD"
+      ],
      "required_cookie_domains": [
        "chanmama.com"
      ],
@@ -411,8 +404,7 @@
      },
      "members": [
        "content.comments.collect_bilibili",
-        "content.metrics.collect_self_bilibili",
-        "content_marketing:data/tools/relogin_bilibili.py"
+        "content.metrics.collect_self_bilibili"
      ]
    }
  },
@@ -582,23 +574,6 @@
      ],
      "cdp_port": 22016
    },
-    "content_marketing:data/tools/friday_relogin.bat": {
-      "script_id": "content_marketing:data/tools/friday_relogin.bat",
-      "state_key": "content_marketing:data/tools/friday_relogin.bat",
-      "aliases": [
-        "content_marketing:data/tools/friday_relogin.bat"
-      ],
-      "cdp_port": 22017
-    },
-    "content.login.refresh": {
-      "script_id": "content_marketing:data/tools/friday_relogin_parallel.py",
-      "state_key": "content_marketing:data/tools/friday_relogin_parallel.py",
-      "aliases": [
-        "content_marketing:data/tools/friday_relogin_parallel.py"
-      ],
-      "cdp_port": 22018,
-      "login_mode": "D"
-    },
    "content.creator_report.generate": {
      "script_id": "content_marketing:data/tools/generate_creator_report.py",
      "state_key": "content_marketing:data/tools/generate_creator_report.py",
@@ -679,83 +654,6 @@
      ],
      "cdp_port": 22135
    },
-    "content_marketing:data/tools/relogin_bilibili.py": {
-      "script_id": "content_marketing:data/tools/relogin_bilibili.py",
-      "state_key": "content_marketing:data/tools/relogin_bilibili.py",
-      "aliases": [
-        "content_marketing:data/tools/relogin_bilibili.py"
-      ],
-      "cdp_port": 22028,
-      "login_mode": "D",
-      "required_cookie_domains": [
-        "bilibili.com"
-      ],
-      "required_cookie_names": [
-        "SESSDATA"
-      ]
-    },
-    "content_marketing:data/tools/relogin_douyin.py": {
-      "script_id": "content_marketing:data/tools/relogin_douyin.py",
-      "state_key": "content_marketing:data/tools/relogin_douyin.py",
-      "aliases": [
-        "content_marketing:data/tools/relogin_douyin.py"
-      ],
-      "cdp_port": 22029,
-      "login_mode": "D",
-      "required_cookie_domains": [
-        "douyin.com"
-      ],
-      "required_cookie_names": [
-        "sessionid"
-      ]
-    },
-    "content_marketing:data/tools/relogin_pgy.py": {
-      "script_id": "content_marketing:data/tools/relogin_pgy.py",
-      "state_key": "content_marketing:data/tools/relogin_pgy.py",
-      "aliases": [
-        "content_marketing:data/tools/relogin_pgy.py"
-      ],
-      "cdp_port": 22030,
-      "login_mode": "D",
-      "required_cookie_domains": [
-        "xiaohongshu.com"
-      ]
-    },
-    "content_marketing:data/tools/relogin_xiaohongshu.py": {
-      "script_id": "content_marketing:data/tools/relogin_xiaohongshu.py",
-      "state_key": "content_marketing:data/tools/relogin_xiaohongshu.py",
-      "aliases": [
-        "content_marketing:data/tools/relogin_xiaohongshu.py"
-      ],
-      "cdp_port": 22031,
-      "login_mode": "D",
-      "required_cookie_domains": [
-        "xiaohongshu.com"
-      ],
-      "required_cookie_names": [
-        "web_session"
-      ]
-    },
-    "content_marketing:data/tools/relogin_xingtu.py": {
-      "script_id": "content_marketing:data/tools/relogin_xingtu.py",
-      "state_key": "content_marketing:data/tools/relogin_xingtu.py",
-      "aliases": [
-        "content_marketing:data/tools/relogin_xingtu.py"
-      ],
-      "cdp_port": 22032,
-      "login_mode": "D",
-      "required_cookie_domains": [
-        "douyin.com",
-        "oceanengine.com",
-        "xingtu.cn"
-      ],
-      "required_cookie_names": [
-        "sessionid",
-        "sessionid_ss",
-        "sid_tt",
-        "sid_guard"
-      ]
-    },
    "content.failed.retry": {
      "script_id": "content_marketing:data/tools/retry_failed.py",
      "state_key": "content_marketing:data/tools/retry_failed.py",
@@ -796,8 +694,12 @@
      ],
      "cdp_port": 22165,
      "login_mode": "C",
-      "required_cookie_domains": ["xiaohongshu.com"],
-      "credential_env_names": ["XHS_ARK_ACCOUNT"]
+      "required_cookie_domains": [
+        "xiaohongshu.com"
+      ],
+      "credential_env_names": [
+        "XHS_ARK_ACCOUNT"
+      ]
    },
    "content_marketing:data/tools/sync_style_categories.py": {
      "script_id": "content_marketing:data/tools/sync_style_categories.py",
@@ -1217,21 +1119,14 @@
      ],
      "cdp_port": 22163
    },
-    "product_commerce:db/sync_dim_style.py": {
-      "script_id": "product_commerce:db/sync_dim_style.py",
-      "state_key": "product_commerce:db/sync_dim_style.py",
+    "product.points_exchange_weekly.run": {
+      "script_id": "product_commerce:points_exchange_weekly.py",
+      "state_key": "product_commerce:points_exchange_weekly.py",
      "aliases": [
-        "product_commerce:db/sync_dim_style.py"
+        "product_commerce:points_exchange_weekly.py",
+        "product.points_exchange_weekly.run"
      ],
-      "cdp_port": 22070
-    },
-    "product_commerce:db/sync_sku_master.py": {
-      "script_id": "product_commerce:db/sync_sku_master.py",
-      "state_key": "product_commerce:db/sync_sku_master.py",
-      "aliases": [
-        "product_commerce:db/sync_sku_master.py"
-      ],
-      "cdp_port": 22071
+      "cdp_port": 22168
    },
    "product_commerce:dy_audience_profile_collect.py": {
      "script_id": "product_commerce:dy_audience_profile_collect.py",
@@ -1283,14 +1178,6 @@
      ],
      "cdp_port": 22076
    },
-    "product_commerce:import_product_reviews.py": {
-      "script_id": "product_commerce:import_product_reviews.py",
-      "state_key": "product_commerce:import_product_reviews.py",
-      "aliases": [
-        "product_commerce:import_product_reviews.py"
-      ],
-      "cdp_port": 22077
-    },
    "product_commerce:insert_bitable_records.py": {
      "script_id": "product_commerce:insert_bitable_records.py",
      "state_key": "product_commerce:insert_bitable_records.py",
@@ -1361,14 +1248,6 @@
      ],
      "cdp_port": 22083
    },
-    "product.review.orchestrate": {
-      "script_id": "product_commerce:orchestrate_review_collection.py",
-      "state_key": "product_commerce:orchestrate_review_collection.py",
-      "aliases": [
-        "product_commerce:orchestrate_review_collection.py"
-      ],
-      "cdp_port": 22084
-    },
    "product_commerce:reapply_erp_override.py": {
      "script_id": "product_commerce:reapply_erp_override.py",
      "state_key": "product_commerce:reapply_erp_override.py",
@@ -1441,6 +1320,26 @@
        "erp321.com"
      ]
    },
+    "product.erp_sales_theme_daily.collect": {
+      "script_id": "product_commerce:collect_erp_sales_theme_daily.py",
+      "state_key": "product_commerce:collect_erp_sales_theme_daily.py",
+      "aliases": [
+        "product_commerce:collect_erp_sales_theme_daily.py"
+      ],
+      "cdp_port": 22167,
+      "login_mode": "C",
+      "required_cookie_domains": [
+        "erp321.com"
+      ]
+    },
+    "product.hk_anpei_public_inventory.refresh": {
+      "script_id": "product_commerce:public_sku_inventory_sheet.py",
+      "state_key": "product_commerce:public_sku_inventory_sheet.py",
+      "aliases": [
+        "product_commerce:public_sku_inventory_sheet.py"
+      ],
+      "cdp_port": 22166
+    },
    "product_commerce:scripts/insert_jd_main_image_records.py": {
      "script_id": "product_commerce:scripts/insert_jd_main_image_records.py",
      "state_key": "product_commerce:scripts/insert_jd_main_image_records.py",
@@ -1553,7 +1452,9 @@
        "GUANGHE_USERNAME",
        "GUANGHE_PASSWORD",
        "GUANGHE_LUGGAGE_USERNAME",
-        "GUANGHE_LUGGAGE_PASSWORD"
+        "GUANGHE_LUGGAGE_PASSWORD",
+        "GUANGHE_NEW_BAG_USERNAME",
+        "GUANGHE_NEW_BAG_PASSWORD"
      ]
    },
    "product.jd_video_upload.run": {
@@ -11,10 +11,7 @@
      "workflow_id": "content.metrics.daily"
    },
    {
-      "at": [
-        "01:00",
-        "05:00"
-      ],
+      "at": "06:00",
      "business_date_offset_days": 0,
      "enabled": true,
      "kind": "daily",
@@ -42,15 +39,6 @@
      "kind": "daily",
      "workflow_id": "content.marketing_report.daily"
    },
-    {
-      "at": "10:00",
-      "days": [
-        "Friday"
-      ],
-      "enabled": false,
-      "kind": "weekly",
-      "workflow_id": "content.relogin.weekly"
-    },
    {
      "at": "12:00",
      "days": [
@@ -209,6 +197,16 @@
      "kind": "daily",
      "workflow_id": "product.online_stock_alert.daily"
    },
+    {
+      "at": "09:45",
+      "business_date_offset_days": -7,
+      "days": [
+        "Monday"
+      ],
+      "enabled": true,
+      "kind": "weekly",
+      "workflow_id": "product.points_exchange.weekly"
+    },
    {
      "at": "08:40",
      "business_date_offset_days": -1,
@@ -221,6 +219,13 @@
      "kind": "daily",
      "workflow_id": "product.erp_all_shop_daily"
    },
+    {
+      "at": "09:00",
+      "business_date_offset_days": 0,
+      "enabled": true,
+      "kind": "daily",
+      "workflow_id": "product.hk_anpei_public_inventory.daily"
+    },
    {
      "at": "12:00",
      "business_date_offset_days": 0,
@@ -330,6 +335,8 @@
    },
    {
      "at": "07:00",
+      "business_date_offset_days": 0,
+      "enabled": false,
      "kind": "daily",
      "workflow_id": "supply.replenishment_alert.daily"
    },
@@ -7,7 +7,7 @@
      "source_relative_path": "bilibili_comment_scraper.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/bilibili_comment_scraper.py",
      "source_sha256": "c7f116f30889125c668e7f4dd796e7249ba7ea8621d5ab5fb518028f56c7f8dc",
-      "target_sha256": "109a01a6dc2b79261cff68ff42818a7561f3319b6be3788d9300bb5ce0217a96",
+      "target_sha256": "d3ae63aab128a3e4078cf357aa4dfcf1f2c3b36c3116989d8f2252be62544e53",
      "transformed": true,
      "category": "source"
    },
@@ -23,7 +23,7 @@
      "source_relative_path": "chanmama_scraper.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/chanmama_scraper.py",
      "source_sha256": "2edc26879b57d73ec5992c80e5ef74d76a509f09bff7284b1ae17e90df543e07",
-      "target_sha256": "8b87268e0f105cf4d349b192c7b2c24d5e318a9d88fcd14264824b083103b24e",
+      "target_sha256": "b808b412e9bb8d824c1a117cbecdd28234408a55fa60891f1ae748abeb7bc415",
      "transformed": true,
      "category": "source"
    },
@@ -111,7 +111,7 @@
      "source_relative_path": "data/tools/batch_rescrape_bilibili.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/batch_rescrape_bilibili.py",
      "source_sha256": "5b07fe27f1b76714ca3f3feae34e3d719277ed0cfe124d76cacd8e3c904b1fca",
-      "target_sha256": "10e5dee90c591201daa7a7b0eb13a40ed1c97d7d51b1b93ef8f1d5d4b8e09580",
+      "target_sha256": "8fa0e5838615f36c9841deb08f6710693e63b011f1f91138c1121d99277dd391",
      "transformed": true,
      "category": "tool"
    },
@@ -119,7 +119,7 @@
      "source_relative_path": "data/tools/batch_rescrape_douyin.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/batch_rescrape_douyin.py",
      "source_sha256": "15b1438513f8216a6c7a103744e8a1ca6bc63b32c111d7c9bce31c18be9e84a6",
-      "target_sha256": "3233a7840e07a87535a8df065562a7cb681dbc140e002d82970598de3d9768cf",
+      "target_sha256": "e0f1227a7b66d85ea65191b5f9c6c5bf588e85cc0cd66e20afae7a9dc7a2684e",
      "transformed": true,
      "category": "tool"
    },
@@ -127,7 +127,7 @@
      "source_relative_path": "data/tools/batch_rescrape_xiaohongshu.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/batch_rescrape_xiaohongshu.py",
      "source_sha256": "6d85f7dfd52c7a92b951ac29e7ee3377aebdb743af2a362d1f38366188dcad08",
-      "target_sha256": "01be5d1ba260ccd46e01b9e72b6f7dd3d08577615455fc9010de4efe8a710aea",
+      "target_sha256": "c481bd6af34040388abafed0466d5632b3892865bf192d245664627003a4c43f",
      "transformed": true,
      "category": "tool"
    },
@@ -183,7 +183,7 @@
      "source_relative_path": "data/tools/db.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/db.py",
      "source_sha256": "83f158a78841244c1f02255d1500018b5a9dbe1f282cbdb937acc344b08abc25",
-      "target_sha256": "62ffad854cc2bc3a3ae4f2a28b4ee345bc9eecf72101b647c3aa165c691b456a",
+      "target_sha256": "bfdb761761532263ae6c555ee455cad4efa300b7fa5639931813f4f95e79ff6a",
      "transformed": true,
      "category": "tool"
    },
@@ -203,14 +203,6 @@
      "transformed": true,
      "category": "tool"
    },
-    {
-      "source_relative_path": "data/tools/friday_relogin_parallel.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/friday_relogin_parallel.py",
-      "source_sha256": "affed4ab6e5e179c85dc92999502b98e007c71243ed26977b3c5eeba23b9e3dc",
-      "target_sha256": "438bedfb874c7eb566e1998815a22ca72b7d177cee221fd39625a96872bd450b",
-      "transformed": true,
-      "category": "tool"
-    },
    {
      "source_relative_path": "data/tools/generate_creator_report.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/generate_creator_report.py",
@@ -291,54 +283,6 @@
      "transformed": true,
      "category": "tool"
    },
-    {
-      "source_relative_path": "data/tools/relogin_bilibili.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/relogin_bilibili.py",
-      "source_sha256": "b8dc12539bcd6f1a368d50ead1f5d33d6577a75c92f0b7bafa75a67edd7a73fa",
-      "target_sha256": "d47f26f62fbe3262128327b916151385e527125a18f27f7455b8382aa5a77ecb",
-      "transformed": true,
-      "category": "tool"
-    },
-    {
-      "source_relative_path": "data/tools/relogin_douyin.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/relogin_douyin.py",
-      "source_sha256": "3c2017f8a801528ff17e4c4bae24615256b8788c185f9a797fe87470e35b1645",
-      "target_sha256": "82a0ecf4101f2eaded94cc4cd9054cb08e1e27519a58cfc0d3126888cc1ce5d4",
-      "transformed": true,
-      "category": "tool"
-    },
-    {
-      "source_relative_path": "data/tools/relogin_pgy.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/relogin_pgy.py",
-      "source_sha256": "759c6a68a0851b4c87d8febfc5d94925831a3b2d7cdf1e6a6e74dcd52b83ae3f",
-      "target_sha256": "d75697ae2b6ac1f23bb79077a4df0af33113813a46525a6f3980b04391857cac",
-      "transformed": true,
-      "category": "tool"
-    },
-    {
-      "source_relative_path": "data/tools/relogin_transaction.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/relogin_transaction.py",
-      "source_sha256": "57ec20de8d99290268ed3d1fc2b66bf3faeaca807d7f3744dd51de6cc93628b2",
-      "target_sha256": "57ec20de8d99290268ed3d1fc2b66bf3faeaca807d7f3744dd51de6cc93628b2",
-      "transformed": false,
-      "category": "tool"
-    },
-    {
-      "source_relative_path": "data/tools/relogin_xiaohongshu.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/relogin_xiaohongshu.py",
-      "source_sha256": "fcdea30adecb066c4f3699ded457df0bc763d978248276ddbffa9f18bc9bdf2c",
-      "target_sha256": "e700dd57e5cf87d68529bcd6c062e73d6bd9f1e6ef06ef3122e2b16c288aaf1a",
-      "transformed": true,
-      "category": "tool"
-    },
-    {
-      "source_relative_path": "data/tools/relogin_xingtu.py",
-      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/relogin_xingtu.py",
-      "source_sha256": "c9d434cb02a57ebd7b884598563077d733d56e45dad59ed30774eb7e1c668cb0",
-      "target_sha256": "06ec1088677d69511057008acb2b05cca2da015a8219aee10bf19bbae3c25d79",
-      "transformed": true,
-      "category": "tool"
-    },
    {
      "source_relative_path": "data/tools/retry_failed.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/retry_failed.py",
@@ -351,7 +295,7 @@
      "source_relative_path": "data/tools/schema_gyxx_super_data.sql",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/data/tools/schema_gyxx_super_data.sql",
      "source_sha256": "43d202d7d0452b7d0671ee9c40149c21328db4b2863fa4d030af456f89defe3c",
-      "target_sha256": "001870851dea9c373a77b87335604c4363b06c3eb0ed7c97b4a8d5e891e56758",
+      "target_sha256": "4b2eca918bb81de0cac529b786d0d37ce04af2c1bc4be81cd20c8ba9456b1348",
      "transformed": true,
      "category": "sql"
    },
@@ -415,7 +359,7 @@
      "source_relative_path": "douyin_comment_scraper.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/douyin_comment_scraper.py",
      "source_sha256": "ec785ead9c95c5de65c00b06212b1f03d0159fed475a95a2b593d3b24e486e43",
-      "target_sha256": "6e7ac4357fe466656b93ca7bae6c49adc4d59fa75974a39c6885c7433bd4d402",
+      "target_sha256": "35d3caf8bc85a82ca15c3fc9ff317c21460173996106ffee8f1d3b5f1d13b801",
      "transformed": true,
      "category": "source"
    },
@@ -455,7 +399,7 @@
      "source_relative_path": "pgy_xhs_scraper_v2.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/pgy_xhs_scraper_v2.py",
      "source_sha256": "2be78ef76df38e80da2099a76e6e3fd9dcdc2e8875afb44279f54003ccb5f563",
-      "target_sha256": "e5df4b34f67a565ec5d5ce4dd5857e80e363dd7559e7e29006dcd3e421276c8a",
+      "target_sha256": "472c13700adc98144b53f2921bbdde15ebb7dc138eaab76885b514dadd306887",
      "transformed": true,
      "category": "source"
    },
@@ -479,7 +423,7 @@
      "source_relative_path": "self_bilibili_scraper.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/self_bilibili_scraper.py",
      "source_sha256": "2ef650a16dde61697101bb219a8c3adcd682fec1529bb9c5312dc5b384d83a72",
-      "target_sha256": "6ec5ed6991a93f806fb74d0870e953e5a05d52043a83f2fa261e2ec67bee6ad7",
+      "target_sha256": "1c09171d1269a3a38a5e6f9f99b98d0da4c7ff11d075613fc288ee209e82e9d2",
      "transformed": true,
      "category": "source"
    },
@@ -487,7 +431,7 @@
      "source_relative_path": "self_douyin_scraper.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/self_douyin_scraper.py",
      "source_sha256": "4aba1264dee052521e5712d7219916e534db5705e3cec75caaa1b025a4bc2506",
-      "target_sha256": "62307df115c8d7336b470baba074163ab3f3468bae8b0af266af370e2e42d9f1",
+      "target_sha256": "945659cd80435bab08bfe1dee13c451c8c4b5fe362313ea09591525c551d3c48",
      "transformed": true,
      "category": "source"
    },
@@ -511,7 +455,7 @@
      "source_relative_path": "xiaohongshu_comment_scraper.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/xiaohongshu_comment_scraper.py",
      "source_sha256": "58bc07bb2fd67897e18a9981f931df8fadf24a6cc5f09d55bd3aaa2e3c6756a4",
-      "target_sha256": "afca1104324546da2315fdf7a427c3f34021eef31ddc2fea76877f818718ccb0",
+      "target_sha256": "fff780d9e16717e6a1ba9c5ee4db22cb1bb5a85f75807c811072c20a6bd390de",
      "transformed": true,
      "category": "source"
    },
@@ -527,7 +471,7 @@
      "source_relative_path": "xingtu_scraper_v2.py",
      "target_relative_path": "src/gyxx_flow/modules/content_marketing/xingtu_scraper_v2.py",
      "source_sha256": "0d4eb87779064e13b8d78daa6dba13698f1149a26b3a1de2e064c117a29cf49a",
-      "target_sha256": "b36339444e26048a90c8ac999a0a628bcddd4f64842924f6c60e600c7a3bec37",
+      "target_sha256": "dfdd07af20c8e2e59411fb6e32a82588db5601d6f3fba75d596e0b5ac19b2be8",
      "transformed": true,
      "category": "source"
    }
@@ -32,7 +32,7 @@
      "target_relative_path": "src/gyxx_flow/modules/shop_intelligence/collectors/dy_store_competitor_store_scraping.py",
      "category": "source",
      "source_sha256": "009879a7dd341ca6ecbc8eb66f5700114c73bcaba3a53f159014a76232d0aa6b",
-      "target_sha256": "d5e85ded1f852ba418229c75c985d0ece967dfbd5fd31697fe407b64d1c3db05",
+      "target_sha256": "9ab1328c2f1743944e8f096e548255884bc9724bf1e900c60c069d443c886902",
      "transformed": true
    },
    {
@@ -48,7 +48,7 @@
      "target_relative_path": "src/gyxx_flow/modules/shop_intelligence/collectors/jd_peer_store_data_collector.py",
      "category": "source",
      "source_sha256": "4ebdb52df7b384d2bfdd2fc840d55d14e18864002c20a36685b63cee1966cf4c",
-      "target_sha256": "39faf18e9cd5e441323a8e056e022f179018a44eac224d64383cd7ae9fa1d611",
+      "target_sha256": "f03b1a7226071f2225a2a71748ad2c27a102cde67ba02d37c63f15480dd0de32",
      "transformed": true
    },
    {
@@ -88,7 +88,7 @@
      "target_relative_path": "src/gyxx_flow/modules/shop_intelligence/db/__init__.py",
      "category": "source",
      "source_sha256": "ffc89b78ff0081a129e04eb6de4985ab0a2a4613c9e8a1d8b2750d8c17e8bc05",
-      "target_sha256": "a4bae02d0c07f3e3986896b31e62f41bf15edbf74d3776b1e050e381c15a8413",
+      "target_sha256": "08b5b8920338100339e7c716423b0e9260d4e007d0a1cc3f7510d51970a66a44",
      "transformed": true
    },
    {
@@ -96,7 +96,7 @@
      "target_relative_path": "src/gyxx_flow/modules/shop_intelligence/db/db.py",
      "category": "source",
      "source_sha256": "168b6f87c1827b267f763ddce7eb15d0c8d652c8c0c7af18dcd0c6c7da5cddd1",
-      "target_sha256": "01469f9eb80aee5e80bc28e548dbe2cd5ce861ab35cd480b33921a2adc799320",
+      "target_sha256": "4ab56df6c8ff75b1a67256ad8cdef34c58bb97e1a126707aac7ccf82980a56ec",
      "transformed": true
    },
    {
@@ -104,7 +104,7 @@
      "target_relative_path": "src/gyxx_flow/modules/shop_intelligence/db/schema.sql",
      "category": "sql",
      "source_sha256": "73230590bde7ae6359c02c82c2db096c0d3faa55b374e30348c47d7b40ea498b",
-      "target_sha256": "a69fa151f989f106d00f344bab43ec9e6a09cf404bdc581624cc0901cdc13d20",
+      "target_sha256": "9558543cf313aadb785715fca8f5cb99219488e7cb6c37cb00310bfe84f7ee74",
      "transformed": true
    },
    {
@@ -120,7 +120,7 @@
      "target_relative_path": "src/gyxx_flow/modules/supply_chain/orchestrator/scripts/ProductReplenishment.py",
      "category": "source",
      "source_sha256": "02f6345142ebd84e25dc2f3b0651bc139ef31dd3994ff739efe66344c77ee49e",
-      "target_sha256": "e74211f2fe97ac47e704a040f442f51c97fecc15ea30dd1f0dbeacaf5940fd80",
+      "target_sha256": "ae5154b1f2af1e03307c07e4188225cd9601d1e1b5316fc0f6a78d23303ba8de",
      "transformed": true
    },
    {
@@ -3,6 +3,11 @@
  "workflows": [
    {
      "id": "content.metrics.daily",
+      "required_accounts": [
+        "douyin-content",
+        "douyin-shop",
+        "xiaohongshu-content"
+      ],
      "module": "content_marketing",
      "trigger": "scheduled",
      "execution": {
@@ -302,6 +307,11 @@
    },
    {
      "id": "content.metrics.backfill",
+      "required_accounts": [
+        "douyin-content",
+        "douyin-shop",
+        "xiaohongshu-content"
+      ],
      "module": "content_marketing",
      "trigger": "scheduled",
      "execution": {
@@ -430,56 +440,7 @@
        "task_name": "YingxiaoYunying_DailyMarketingReport"
      }
    },
-    {
-      "id": "content.relogin.weekly",
-      "module": "content_marketing",
-      "trigger": "scheduled",
-      "execution": {
-        "steps": [
-          {
-            "id": "relogin",
-            "name": "刷新内容平台登录态",
-            "description": "并行启动五个平台的二维码重新登录流程，按配置轮次重试，持久化登录状态并通知结果。",
-            "data_flow": {
-              "sources": [
-                {
-                  "system": "内容平台",
-                  "label": "五个平台登录页与当前会话状态"
-                }
-              ],
-              "processing": [
-                "并行发起二维码登录并按轮次重试",
-                "校验并持久化新的登录会话"
-              ],
-              "destinations": [
-                {
-                  "system": "浏览器状态",
-                  "label": "各平台 Cookie 与 storage state"
-                },
-                {
-                  "system": "飞书",
-                  "label": "登录刷新结果通知"
-                }
-              ]
-            },
-            "entry": "data/tools/friday_relogin_parallel.py",
-            "args": [
-              "--max-attempts",
-              "3",
-              "--round-timeout",
-              "300",
-              "--screenshot-delay",
-              "60"
-            ]
-          }
-        ]
-      },
-      "provenance": {
-        "source_project": "content",
-        "task_name": "YingxiaoYunying_FridayRelogin"
-      }
-    },
-    {
+        {
      "id": "content.creator_report.monthly",
      "module": "content_marketing",
      "trigger": "scheduled",
@@ -781,6 +742,9 @@
    },
    {
      "id": "shop.metrics.weekly",
+      "required_accounts": [
+        "douyin-shop"
+      ],
      "module": "shop_intelligence",
      "trigger": "scheduled",
      "execution": {
@@ -890,6 +854,9 @@
    },
    {
      "id": "shop.competitor.weekly",
+      "required_accounts": [
+        "douyin-shop"
+      ],
      "module": "shop_intelligence",
      "trigger": "scheduled",
      "execution": {
@@ -1159,7 +1126,7 @@
              "30",
              "--headless"
            ],
-            "timeout_seconds": 14400,
+            "timeout_seconds": 28800,
            "replay_policy": "idempotent"
          }
        ]
@@ -1213,7 +1180,7 @@
              "{business_date}",
              "--end-date",
              "{business_date}",
-              "--headless"
+              "--headed"
            ],
            "timeout_seconds": 14400,
            "replay_policy": "idempotent"
@@ -1592,6 +1559,10 @@
    },
    {
      "id": "product.daily",
+      "required_accounts": [
+        "douyin-shop",
+        "erp-jushuitan"
+      ],
      "module": "product_commerce",
      "trigger": "scheduled",
      "execution": {
@@ -1734,6 +1705,51 @@
        "task_name": "OnlineItemStockAlert"
      }
    },
+    {
+      "id": "product.points_exchange.weekly",
+      "module": "product_commerce",
+      "trigger": "scheduled",
+      "execution": {
+        "steps": [
+          {
+            "id": "query_rank_notify",
+            "name": "查询积分兑换明细并按商品生成兑换排行周报",
+            "description": "调用企业数据交互积分兑换查询接口（startTime/endTime 覆盖整周自然日），按商品汇总 quantity 兑换数量并从高到低排名；默认统计上一个完整自然周（周一至周日），将排行榜以飞书卡片发送给私域库存预警的同一批接收人。",
+            "replay_policy": "idempotent",
+            "data_flow": {
+              "sources": [
+                {
+                  "system": "企业数据交互接口",
+                  "label": "积分兑换明细",
+                  "detail": "/open/data/v1/points-exchange/query，startTime 为窗口首日 00:00:00、endTime 为窗口末日 23:59:59"
+                }
+              ],
+              "processing": [
+                "以业务日期作为窗口起点，缺省统计 7 个自然日",
+                "按 itemId 汇总 quantity，并保留商品名称与 SKU 数",
+                "按兑换数量从高到低排序并分段组织飞书消息"
+              ],
+              "destinations": [
+                {
+                  "system": "飞书",
+                  "label": "积分兑换商品排行周报",
+                  "condition": "窗口内存在兑换数量大于 0 的商品"
+                }
+              ]
+            },
+            "entry": "points_exchange_weekly.py",
+            "args": [
+              "--from",
+              "{business_date}"
+            ]
+          }
+        ]
+      },
+      "provenance": {
+        "source_project": "data-hub",
+        "task_name": "PointsExchangeWeeklyRanking"
+      }
+    },
    {
      "id": "product.ecommerce_costs.daily",
      "module": "product_commerce",
@@ -1780,7 +1796,7 @@
          {
            "id": "tmall_import",
            "name": "导入万相台商品广告指标",
-            "description": "解析解压后的商品 CSV，将花费、广告归因成交金额和投产比写入 product_daily_metrics.raw_data.tm_ad_metrics；款式按商品 ID 优先、商品名称精确包含兜底，并记录匹配方式，不覆盖经营日报字段。",
+            "description": "解析解压后的商品 CSV，将花费、广告归因成交金额和投产比写入独立的 tm_product_ad_daily_metrics（每日每商品一行），报表其余指标存 metrics JSON；款式按商品 ID 优先、商品名称精确包含兜底，并记录匹配方式。不再写 product_daily_metrics.raw_data。",
            "entry": "import_tmall_product_ads.py",
            "depends_on": [
              "tmall_download"
@@ -1802,13 +1818,13 @@
              "processing": [
                "校验 CSV 编码、日期、商品主体和广告指标",
                "合并同商品同日多计划行",
-                "按平台/日期/商品幂等更新广告 JSON 节点"
+                "按日期/商品幂等写入天猫商品投放日报表"
              ],
              "destinations": [
                {
                  "system": "PostgreSQL",
-                  "label": "product_daily_metrics.raw_data.tm_ad_metrics",
-                  "detail": "由 data-hub 产品生命进程读取并计算广告成交、投产比、电商费比；节点同时保存 style_name/style_match_method"
+                  "label": "tm_product_ad_daily_metrics",
+                  "detail": "由 data-hub 产品生命进程读取 spend/attributed_deal_amount 计算广告成交、投产比、电商费比；style_name/style_match_method 记录款式匹配，metrics 保留报表全部指标"
                }
              ]
            }
@@ -1916,6 +1932,7 @@
              ],
              "processing": [
                "校验广告主 ID、店铺标识和商品数据明细",
+                "任务中心被重定向到账户选择页时自动点击目标广告主后重试",
                "按业务日期下载原始 Excel",
                "重置千川错误的 worksheet A1 dimension 并校验日期覆盖"
              ],
@@ -2231,6 +2248,54 @@
      },
      "note": "每天自动采集昨天数据；控制台可手动选择任意补采日期并幂等覆盖该日。全选聚水潭全部店铺，不拆京东、抖音、天猫；每次只采集一个业务日期。"
    },
+    {
+      "id": "product.hk_anpei_public_inventory.daily",
+      "module": "product_commerce",
+      "trigger": "scheduled",
+      "execution": {
+        "steps": [
+          {
+            "id": "query_and_refresh_sheet",
+            "name": "刷新香港安沛 SKU 公有可用数",
+            "description": "校验计划书固定 SKU 与现有库存工作表中的商品编码对应关系，调用聚水潭 /open/inventory/query 查询库存及锁定量，并将公有可用数写入仓库现有库存列。",
+            "entry": "public_sku_inventory_sheet.py",
+            "timeout_seconds": 900,
+            "replay_policy": "idempotent",
+            "data_flow": {
+              "sources": [
+                {
+                  "system": "光影行星香港安沛2026Q2计划书",
+                  "label": "现有库存表的 51 个唯一 SKU"
+                },
+                {
+                  "system": "聚水潭开放 API",
+                  "label": "/open/inventory/query",
+                  "detail": "请求 has_lock_qty=true 以取得库存锁定量"
+                }
+              ],
+              "processing": [
+                "校验整批 SKU 库存响应后再写入飞书",
+                "按商品编码（光影）匹配行；公有可用数 = qty - order_lock - lock_qty + virtual_qty",
+                "没有库存记录的 SKU 清空仓库现有库存单元格",
+                "仅写入仓库现有库存列，不覆盖图片、备注、在途生产或其它字段"
+              ],
+              "destinations": [
+                {
+                  "system": "飞书电子表格",
+                  "label": "光影行星香港安沛2026Q2計劃書",
+                  "detail": "現有庫存工作表；按商品编码（光影）更新 D 列仓库现有库存"
+                }
+              ]
+            }
+          }
+        ]
+      },
+      "provenance": {
+        "source_project": "data-hub",
+        "task_name": "HongKongAnpeiPublicSkuInventoryDaily"
+      },
+      "note": "每天 Asia/Shanghai 09:00 执行。聚水潭应用凭据从 GYXX 调度进程环境读取；不得使用开发配置默认值。"
+    },
    {
      "id": "product.video_upload",
      "module": "product_commerce",
@@ -2240,7 +2305,7 @@
          {
            "id": "upload",
            "name": "上传所选天猫店铺短视频",
-            "description": "先按前端所选店铺扫描飞书并筛选符合路由的视频附件，再按视频 file_token、目标账号和飞书记录写入 PostgreSQL 附件级发布账本；仅在存在可领取附件时登录对应光合账号上传，已发布附件自动跳过。",
+            "description": "先按前端所选店铺扫描飞书并筛选符合路由的视频附件，再按视频 file_token、目标账号和飞书记录写入 PostgreSQL 附件级发布账本；仅在存在可领取附件时登录对应光合账号上传，已发布附件自动跳过。支持旗舰店、鑫华达专卖店和天猫新店（天猫新店只发黑曜石系列，视频源与其他店铺为同一张飞书表）。",
            "replay_policy": "idempotent",
            "data_flow": {
              "sources": [
@@ -2254,11 +2319,11 @@
                },
                {
                  "system": "淘宝光合浏览器状态",
-                  "label": "旗舰店与箱包店独立账号和 Profile"
+                  "label": "旗舰店、箱包店与天猫新店独立账号和 Profile"
                }
              ],
              "processing": [
-                "按前端所选旗舰店或箱包店规则筛选飞书记录并拆分视频附件",
+                "按前端所选旗舰店、箱包店或天猫新店规则筛选飞书记录并拆分视频附件；天猫新店仅保留全部款式都属于黑曜石系列的记录",
                "按视频 file_token、目标账号和飞书记录原子领取未上传附件",
                "确认存在可上传附件后登录所选店铺账号",
                "生成标题、关联天猫商品并发布；明确成功后写 published，提交后结果不明写 ambiguous"
@@ -2299,7 +2364,7 @@
        "label": "天猫话题关键词",
        "hint": "输入完整话题或关键词；例如输入“秋上新”也会匹配并选中“秋上新开拍了”。"
      },
-      "note": "默认由常驻调度器按 config/schedules.json 每 3 天 14:00 执行；前端仍可手动选择旗舰店或箱包店后执行。脚本先扫描飞书并按所选店铺规则筛选，只有数据库存在可领取附件时才登录对应账号上传。允许同一业务日期重复扫描新增附件，PostgreSQL 附件级账本阻止已发布或结果不明的附件自动重发；框架安全预演不会启动上传脚本。"
+      "note": "默认由常驻调度器按 config/schedules.json 每 3 天 14:00 执行；前端仍可手动选择旗舰店、箱包店或天猫新店后执行。天猫新店账号为gyxx箱包旗舰店:龙虾仔（凭据 GUANGHE_NEW_BAG_USERNAME/GUANGHE_NEW_BAG_PASSWORD），只上传黑曜石系列，视频来源与其他店铺相同。脚本先扫描飞书并按所选店铺规则筛选，只有数据库存在可领取附件时才登录对应账号上传。允许同一业务日期重复扫描新增附件，PostgreSQL 附件级账本阻止已发布或结果不明的附件自动重发；框架安全预演不会启动上传脚本。"
    },
    {
      "id": "product.jd_video_upload",
@@ -2350,7 +2415,6 @@
            "args": [
              "--anchor-record-id",
              "recvrXTRWWFXeJ",
-              "--manual-login",
              "--execute"
            ],
            "timeout_seconds": 21600
@@ -2607,6 +2671,9 @@
    },
    {
      "id": "supply.purchase_confirmation.daily",
+      "required_accounts": [
+        "erp-jushuitan"
+      ],
      "module": "supply_chain",
      "trigger": "scheduled",
      "execution": {
@@ -2657,6 +2724,9 @@
    },
    {
      "id": "supply.replenishment.weekly",
+      "required_accounts": [
+        "erp-jushuitan"
+      ],
      "module": "supply_chain",
      "trigger": "scheduled",
      "execution": {
@@ -2712,6 +2782,9 @@
    },
    {
      "id": "supply.replenishment_alert.daily",
+      "required_accounts": [
+        "erp-jushuitan"
+      ],
      "module": "supply_chain",
      "trigger": "scheduled",
      "execution": {
@@ -2807,10 +2880,13 @@
        "source_project": "supply",
        "task_name": "auto-flow-replenishment-arrival-sync"
      },
-      "note": "队列表为空时成功结束；失败行保留 failed 状态，下一次调度自动重试。默认租户为 1，可用 GYXX_ERP_TENANT_ID 覆盖。"
+      "note": "队列表为空时成功结束；失败行保留 failed 状态，下一次调度自动重试。data-hub 单公司化后队列表已无 tenant_id：脚本会先探测实际表结构，有该列仍按 GYXX_ERP_TENANT_ID（默认 1）过滤，没有则按单公司口径直接认领。"
    },
    {
      "id": "product.market_rank",
+      "required_accounts": [
+        "douyin-shop"
+      ],
      "module": "product_commerce",
      "trigger": "scheduled",
      "execution": {
@@ -0,0 +1,47 @@
+# GYXX Flow 开机自启：启用/关闭 Windows 自动登录
+#
+# 背景：控制台+调度器的自启链是
+#   开机 → 自动登录进桌面 → Startup\gyxx-keepalive.vbs → keepalive → 控制台(8765)+调度器
+# 唯一缺口是 AutoAdminLogon（默认 0，重启后停在登录界面，Startup 不触发）。
+# 采集依赖浏览器 CDP，必须跑在交互式用户会话，因此不能用服务/Task Scheduler SYSTEM 替代。
+#
+# 启用（交互输入密码，不会留在命令历史）：
+#   powershell -ExecutionPolicy Bypass -File D:\gyxx-flow\deploy\watchdog\autologon.ps1
+# 撤销：
+#   powershell -ExecutionPolicy Bypass -File D:\gyxx-flow\deploy\watchdog\autologon.ps1 -Disable
+#
+# 注意：DefaultPassword 以明文存于 HKLM Winlogon（仅 Administrators 可读），
+# 内网采集机可接受；介意安全性请改用 Sysinternals Autologon.exe（LSA secret 存储）。
+param(
+    [switch]$Disable
+)
+
+$winlogon = 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Winlogon'
+
+if ($Disable) {
+    Set-ItemProperty $winlogon -Name AutoAdminLogon -Value '0'
+    Remove-ItemProperty $winlogon -Name DefaultPassword -ErrorAction SilentlyContinue
+    Write-Host '已关闭自动登录（DefaultPassword 已清除）。' -ForegroundColor Yellow
+    exit 0
+}
+
+$pwd1 = Read-Host -AsSecureString "输入 $env:USERNAME 的 Windows 登录密码"
+$pwd2 = Read-Host -AsSecureString '再输一次确认'
+$plain1 = [Net.NetworkCredential]::new('', $pwd1).Password
+$plain2 = [Net.NetworkCredential]::new('', $pwd2).Password
+if ($plain1 -ne $plain2) { throw '两次输入不一致，未做任何修改。' }
+if ([string]::IsNullOrEmpty($plain1)) {
+    Write-Warning '密码为空：策略 LimitBlankPasswordUse=1 会拒绝空密码自动登录，请先给账户设置密码。'
+    exit 1
+}
+
+Set-ItemProperty $winlogon -Name AutoAdminLogon -Value '1'
+Set-ItemProperty $winlogon -Name DefaultUserName -Value $env:USERNAME
+Set-ItemProperty $winlogon -Name DefaultDomainName -Value $env:COMPUTERNAME
+Set-ItemProperty $winlogon -Name DefaultPassword -Value $plain1
+# AutoLogonCount 每次自动登录会递减，删掉表示不限次数
+Remove-ItemProperty $winlogon -Name AutoLogonCount -ErrorAction SilentlyContinue
+
+Write-Host '自动登录已启用。' -ForegroundColor Green
+Write-Host '验证方式：挑空闲时间重启一次，开机应自动进桌面，约 2 分钟内控制台(127.0.0.1:8765)与调度器就位；'
+Write-Host '也可查看 D:\gyxx-flow\var\logs\keepalive\keepalive.log 出现 keepalive-start / scheduler-restarted / console-restarted。'
@@ -0,0 +1,336 @@
+# -*- coding: utf-8 -*-
+"""控制台与调度器的进程外保活（登录自启动，随用户会话常驻）。
+
+职责（不承载任何业务调度规则，业务调度仍由 `gyxx schedule run` 唯一负责）：
+1. 控制台：探测 http://127.0.0.1:8765/api/health——进程不在则拉起；
+   进程在但探测连续失败（wedged）则结束该进程树后拉起。控制台进程内
+   另有看门狗（console.py `_console_watchdog`）自检失败会主动退出，此处
+   负责把退出后的空窗补上，两层互为补充。
+2. 调度器：读 `var/state/scheduler/service.lock` 的 PID 判断存活，锁主
+   已死则按 install.ps1 同款参数拉起 `python -m gyxx_flow schedule run`；
+   调度器自身单实例锁保证不会双跑。
+3. 节流：每个目标有冷却间隔；短时间反复重启触发熔断，只记录不再拉起，
+   避免故障风暴。
+
+用法：
+  gyxx_keepalive.py            # 常驻循环（Startup 的 vbs 以 pythonw 启动）
+  gyxx_keepalive.py --once     # 单轮探测+处置后退出，供验收/手动巡检
+
+日志：var/logs/keepalive/keepalive.log（追加）。
+自身单实例：var/state/keepalive.lock（PID 存活检测，重复启动直接退出）。
+"""
+
+from __future__ import annotations
+
+import argparse
+import ctypes
+import http.client
+import json
+import os
+import subprocess
+import sys
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+ENV_FILES = [
+    Path(os.getenv("GYXX_RUNTIME_ENV_FILE", r"D:\product-collector-analyze-flow\.env")),
+    Path(
+        os.getenv(
+            "GYXX_RUNTIME_JUSHUITAN_ENV_FILE",
+            r"D:\product-collector-analyze-flow\private-secrets\jushuitan.env",
+        )
+    ),
+]
+CONSOLE_HOST = "127.0.0.1"
+CONSOLE_PORT = int(os.getenv("GYXX_CONSOLE_PORT", "8765"))
+LOOP_INTERVAL_SECONDS = float(os.getenv("GYXX_KEEPALIVE_INTERVAL_SECONDS", "30"))
+PROBE_TIMEOUT_SECONDS = 5.0
+PROBE_FAILURE_LIMIT = 3
+RESTART_COOLDOWN_SECONDS = float(os.getenv("GYXX_KEEPALIVE_COOLDOWN_SECONDS", "120"))
+TRIP_MAX_RESTARTS = 5
+TRIP_WINDOW_SECONDS = 30 * 60
+TRIP_PAUSE_SECONDS = 60 * 60
+
+DATA_ROOT = PROJECT_ROOT / "var"
+LOG_PATH = DATA_ROOT / "logs" / "keepalive" / "keepalive.log"
+LOCK_PATH = DATA_ROOT / "state" / "keepalive.lock"
+CONSOLE_LOG_DIR = DATA_ROOT / "logs" / "console-restarts"
+SCHEDULER_LOG_DIR = DATA_ROOT / "logs" / "scheduler-service"
+SCHEDULER_SERVICE_LOCK = DATA_ROOT / "state" / "scheduler" / "service.lock"
+
+
+def _log(event: str, detail: str = "") -> None:
+    stamp = datetime.now(timezone.utc).astimezone().strftime("%Y-%m-%d %H:%M:%S")
+    line = f"{stamp} {event} {detail}".rstrip()
+    LOG_PATH.parent.mkdir(parents=True, exist_ok=True)
+    with LOG_PATH.open("a", encoding="utf-8", newline="\n") as stream:
+        stream.write(line + "\n")
+    print(line, flush=True)
+
+
+def _pid_alive(pid: int) -> bool:
+    if pid <= 0:
+        return False
+    kernel32 = ctypes.windll.kernel32
+    handle = kernel32.OpenProcess(0x1000, False, pid)  # PROCESS_QUERY_LIMITED_INFORMATION
+    if not handle:
+        return False
+    try:
+        exit_code = ctypes.c_ulong()
+        if kernel32.GetExitCodeProcess(handle, ctypes.byref(exit_code)):
+            return exit_code.value == 259  # STILL_ACTIVE
+        return False
+    finally:
+        kernel32.CloseHandle(handle)
+
+
+def _acquire_self_lock() -> bool:
+    LOCK_PATH.parent.mkdir(parents=True, exist_ok=True)
+    if LOCK_PATH.is_file():
+        try:
+            existing = json.loads(LOCK_PATH.read_text(encoding="utf-8"))
+            if _pid_alive(int(existing.get("pid", 0))):
+                return False
+        except Exception:
+            pass
+    LOCK_PATH.write_text(
+        json.dumps({"pid": os.getpid(), "started_at": datetime.now(timezone.utc).isoformat()}),
+        encoding="utf-8",
+    )
+    return True
+
+
+def _console_env_files() -> list[Path]:
+    return [path for path in ENV_FILES if path.is_file()]
+
+
+def _spawn(argv: list[str], log_dir: Path, log_prefix: str) -> int:
+    log_dir.mkdir(parents=True, exist_ok=True)
+    stamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+    log_path = log_dir / f"{log_prefix}-{stamp}.log"
+    kwargs: dict[str, object] = {}
+    if os.name == "nt":
+        kwargs["creationflags"] = (
+            subprocess.CREATE_NO_WINDOW | subprocess.CREATE_NEW_PROCESS_GROUP
+        )
+    else:
+        kwargs["start_new_session"] = True
+    with log_path.open("a", encoding="utf-8", newline="\n") as stream:
+        process = subprocess.Popen(
+            argv,
+            cwd=str(PROJECT_ROOT),
+            env=os.environ,
+            stdin=subprocess.DEVNULL,
+            stdout=stream,
+            stderr=subprocess.STDOUT,
+            shell=False,
+            **kwargs,
+        )
+    return process.pid
+
+
+def _console_argv() -> list[str]:
+    argv = [
+        str(PROJECT_ROOT / ".venv" / "Scripts" / "python.exe"),
+        "-m",
+        "gyxx_flow",
+        "console",
+        "--host",
+        CONSOLE_HOST,
+        "--port",
+        str(CONSOLE_PORT),
+    ]
+    for env_file in _console_env_files():
+        argv.extend(("--env-file", str(env_file)))
+    return argv
+
+
+def _scheduler_argv() -> list[str]:
+    argv = [
+        str(PROJECT_ROOT / ".venv" / "Scripts" / "python.exe"),
+        "-m",
+        "gyxx_flow",
+        "schedule",
+        "run",
+    ]
+    for env_file in _console_env_files():
+        argv.extend(("--env-file", str(env_file)))
+    return argv
+
+
+def _probe_console() -> bool:
+    try:
+        connection = http.client.HTTPConnection(CONSOLE_HOST, CONSOLE_PORT, timeout=PROBE_TIMEOUT_SECONDS)
+        try:
+            connection.request("GET", "/api/health")
+            response = connection.getresponse()
+            response.read()
+            return response.status == 200
+        finally:
+            connection.close()
+    except Exception:
+        return False
+
+
+def _port_owner_pid(port: int) -> int | None:
+    try:
+        output = subprocess.run(
+            ["netstat", "-ano", "-p", "tcp"],
+            capture_output=True,
+            text=True,
+            timeout=10,
+        ).stdout
+    except Exception:
+        return None
+    needle = f"{CONSOLE_HOST}:{port}"
+    for line in output.splitlines():
+        parts = line.split()
+        if (
+            len(parts) >= 5
+            and parts[0].upper() == "TCP"
+            and parts[1] == needle
+            and parts[3].upper() == "LISTENING"
+        ):
+            try:
+                return int(parts[4])
+            except ValueError:
+                return None
+    return None
+
+
+def _kill_pid_tree(pid: int) -> None:
+    if pid <= 0 or pid == os.getpid():
+        return
+    subprocess.run(
+        ["taskkill", "/PID", str(pid), "/T", "/F"],
+        capture_output=True,
+        timeout=30,
+    )
+
+
+def _scheduler_lock_pid() -> int | None:
+    try:
+        payload = json.loads(SCHEDULER_SERVICE_LOCK.read_text(encoding="utf-8"))
+        return int(payload.get("pid", 0)) or None
+    except Exception:
+        return None
+
+
+class _TargetState:
+    """记录单目标的连续失败计数、冷却与熔断状态。"""
+
+    def __init__(self, name: str) -> None:
+        self.name = name
+        self.failures = 0
+        self.last_restart_at = 0.0
+        self.recent_restarts: list[float] = []
+        self.tripped_at: float | None = None
+
+    def tripped(self, now: float) -> bool:
+        if self.tripped_at is None:
+            return False
+        if now - self.tripped_at >= TRIP_PAUSE_SECONDS:
+            self.tripped_at = None
+            self.recent_restarts.clear()
+            _log("keepalive-trip-reset", f"target={self.name} 熔断窗口结束恢复守护")
+            return False
+        return True
+
+    def may_restart(self, now: float) -> bool:
+        if self.tripped(now):
+            return False
+        if now - self.last_restart_at < RESTART_COOLDOWN_SECONDS:
+            return False
+        self.recent_restarts = [
+            stamp for stamp in self.recent_restarts if now - stamp < TRIP_WINDOW_SECONDS
+        ]
+        if len(self.recent_restarts) >= TRIP_MAX_RESTARTS:
+            self.tripped_at = now
+            _log(
+                "keepalive-tripped",
+                f"target={self.name} {TRIP_MAX_RESTARTS} 次重启/{TRIP_WINDOW_SECONDS // 60}分钟 "
+                f"暂停拉起{TRIP_PAUSE_SECONDS // 60}分钟，请人工检查日志",
+            )
+            return False
+        return True
+
+    def record_restart(self, now: float) -> None:
+        self.last_restart_at = now
+        self.recent_restarts.append(now)
+
+
+def run_cycle(console: _TargetState, scheduler: _TargetState) -> None:
+    now = time.time()
+
+    # ---- 控制台 ----
+    if _probe_console():
+        console.failures = 0
+    else:
+        console.failures += 1
+        owner = _port_owner_pid(CONSOLE_PORT)
+        _log(
+            "console-probe-failed",
+            f"{console.failures}/{PROBE_FAILURE_LIMIT} port_owner_pid={owner}",
+        )
+        if console.failures >= PROBE_FAILURE_LIMIT and console.may_restart(now):
+            if owner is not None and _pid_alive(owner):
+                _log("console-kill-wedged", f"pid={owner} 探测持续失败，结束进程树")
+                _kill_pid_tree(owner)
+                time.sleep(3)
+            pid = _spawn(_console_argv(), CONSOLE_LOG_DIR, "keepalive-console")
+            console.record_restart(time.time())
+            console.failures = 0
+            _log("console-restarted", f"pid={pid}")
+
+    # ---- 调度器 ----
+    scheduler_pid = _scheduler_lock_pid()
+    scheduler_alive = scheduler_pid is not None and _pid_alive(scheduler_pid)
+    if not scheduler_alive and scheduler.may_restart(now):
+        pid = _spawn(_scheduler_argv(), SCHEDULER_LOG_DIR, "keepalive-start")
+        scheduler.record_restart(time.time())
+        _log(
+            "scheduler-restarted",
+            f"pid={pid} prev_lock_pid={scheduler_pid}",
+        )
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="GYXX Flow 控制台/调度器保活")
+    parser.add_argument("--once", action="store_true", help="单轮探测+处置后退出")
+    arguments = parser.parse_args(argv)
+
+    if not _acquire_self_lock():
+        _log("keepalive-duplicate-exit", "已有保活进程在运行")
+        return 0
+
+    console = _TargetState("console")
+    scheduler = _TargetState("scheduler")
+    missing_env = [str(path) for path in ENV_FILES if not path.is_file()]
+    _log(
+        "keepalive-start",
+        f"pid={os.getpid()} once={arguments.once}"
+        + (f" missing_env={missing_env}" if missing_env else ""),
+    )
+    try:
+        while True:
+            run_cycle(console, scheduler)
+            if arguments.once:
+                _log("keepalive-once-done", "")
+                return 0
+            time.sleep(LOOP_INTERVAL_SECONDS)
+    except KeyboardInterrupt:
+        return 0
+    finally:
+        try:
+            if LOCK_PATH.is_file():
+                payload = json.loads(LOCK_PATH.read_text(encoding="utf-8"))
+                if int(payload.get("pid", 0)) == os.getpid():
+                    LOCK_PATH.unlink()
+        except Exception:
+            pass
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,5 @@
+' GYXX Flow 保活登录自启动：无窗口启动 deploy/watchdog/gyxx_keepalive.py，
+' 由它负责拉起控制台（含进程内自检看门狗）与调度器。
+' 安装：复制到 %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup\
+CreateObject("WScript.Shell").Run _
+    """D:\gyxx-flow\.venv\Scripts\pythonw.exe"" ""D:\gyxx-flow\deploy\watchdog\gyxx_keepalive.py""", 0, False
@@ -4,6 +4,8 @@
 param(
    [string]$ProjectRoot = (Resolve-Path (Join-Path $PSScriptRoot "..\..")).Path,
    [string]$DataRoot = "",
+    [string]$RuntimeEnvFile = "D:\product-collector-analyze-flow\.env",
+    [string]$JushuitanEnvFile = "D:\product-collector-analyze-flow\private-secrets\jushuitan.env",
    [string]$ServiceName = "gyxx-flow-scheduler",
    [string]$NssmPath = "nssm.exe"
 )
@@ -12,6 +14,13 @@ $ErrorActionPreference = "Stop"
 Set-StrictMode -Version Latest

 $project = (Resolve-Path -LiteralPath $ProjectRoot).Path
+$runtimeEnvPath = (Resolve-Path -LiteralPath $RuntimeEnvFile -ErrorAction Stop).Path
+$runtimeEnvPaths = @($runtimeEnvPath)
+if (Test-Path -LiteralPath $JushuitanEnvFile -PathType Leaf) {
+    $runtimeEnvPaths += (Resolve-Path -LiteralPath $JushuitanEnvFile -ErrorAction Stop).Path
+} else {
+    Write-Warning "Jushuitan environment file not found; Jushuitan workflows will fail safely until it is provisioned."
+}
 if ([string]::IsNullOrWhiteSpace($DataRoot)) {
    $DataRoot = Join-Path $project "var"
 }
@@ -33,7 +42,9 @@ function Invoke-Nssm([string[]]$Arguments) {

 Invoke-Nssm -Arguments @("install", $ServiceName, $python)
 Invoke-Nssm -Arguments @("set", $ServiceName, "AppDirectory", $project)
-Invoke-Nssm -Arguments @("set", $ServiceName, "AppParameters", "-m gyxx_flow schedule run")
+$envFileArguments = $runtimeEnvPaths | ForEach-Object { '--env-file "' + $_ + '"' }
+$schedulerArguments = '-m gyxx_flow schedule run ' + ($envFileArguments -join ' ')
+Invoke-Nssm -Arguments @("set", $ServiceName, "AppParameters", $schedulerArguments)
 Invoke-Nssm -Arguments @(
    "set", $ServiceName, "AppEnvironmentExtra",
    "GYXX_PROJECT_ROOT=$project", "GYXX_DATA_ROOT=$data", "PYTHONUTF8=1"
@@ -46,6 +46,27 @@ Scrapling 安装和管理。具体边界与验证命令见
 `EnvironmentFile=` 或当前进程环境优先于文件中的同名值，避免本地文件意外覆盖密钥系统
 注入值。

+## 香港安沛 SKU 公有库存同步
+
+`product.hk_anpei_public_inventory.daily` 每天按 `Asia/Shanghai` 09:00 查询计划书中的固定 SKU 清单，
+调用聚水潭 `/open/inventory/query`，并按商品编码（光影）更新飞书电子表格「現有庫存」工作表的
+「仓库现有库存」列。它要求调度进程环境提供
+`JUSHUITAN_APP_KEY` 和 `JUSHUITAN_APP_SECRET`；`JUSHUITAN_BASE_URL` 可选，默认使用聚水潭官方 OpenAPI
+地址。没有密钥时工作流会在查询前失败，不会把库存留成 0 或改写多维表格。
+
+Windows NSSM 调度服务从受保护的运行时环境文件读取凭据。安装器默认读取通用运行环境文件
+`D:\product-collector-analyze-flow\.env`，以及独立的聚水潭密钥文件
+`D:\product-collector-analyze-flow\private-secrets\jushuitan.env`。两个文件都按服务账户实际身份限制 ACL；
+可分别通过 `-RuntimeEnvFile` 和 `-JushuitanEnvFile` 指定其他路径：
+
+```powershell
+.\deploy\windows-service\install.ps1 -ProjectRoot D:\gyxx-flow -DataRoot D:\gyxx-flow-data -RuntimeEnvFile D:\product-collector-analyze-flow\.env -JushuitanEnvFile D:\product-collector-analyze-flow\private-secrets\jushuitan.env
+```
+
+仅将这两个变量的正式聚水潭应用值写入服务账户可读的环境文件，并限制文件 ACL；不要使用开发配置中的默认值，
+也不要把真实值写进 Git、PowerShell 命令参数、服务日志或文档。已安装的服务需要让调度进程重新读取环境文件后，
+新计划才会使用这些值。Linux 部署则将相同变量放入权限为 `0640` 的 `EnvironmentFile`。
+
 ## 三平台电商费用日报部署边界

 `product.ecommerce_costs.daily` 是包含天猫万相台、京东京准通和抖音千川分支的电商费用工作流；三个
@@ -197,3 +218,12 @@ sudo systemctl start gyxx-flow.service
 ## Windows 兼容入口

 `deploy/windows-service/` 暂时保留为 legacy NSSM 兼容入口。NSSM 只守护同一个 `gyxx schedule run` 进程，不注册 Windows Task Scheduler，也不保存业务时间规则。新服务器部署以 Linux systemd 为准。
+
+## Windows 工作站保活（控制台 + 调度器自检自启动）
+
+交互式 Windows 工作站（无法用 systemd/NSSM 常驻服务时）使用 `deploy/watchdog/gyxx_keepalive.py` 做进程外保活：
+
+- **控制台自检**：控制台进程内置看门狗线程（`console.py` `_console_watchdog`），周期探测自身 `GET /api/health`；连续失败（默认 3 次，`GYXX_CONSOLE_WATCHDOG_FAILURES` / `GYXX_CONSOLE_WATCHDOG_INTERVAL_SECONDS` 可调，`GYXX_CONSOLE_WATCHDOG=0` 关闭）即以退出码 70 自杀，交给保活拉起。`/api/health` 不鉴权、只报进程 PID、uptime 与调度锁存活，供探针使用；API 500 现在会把完整堆栈落到 stderr（随启动重定向进 `var/logs/console-restarts/`）。
+- **保活脚本**：登录自启动（Startup 目录 `gyxx-keepalive.vbs` → `pythonw gyxx_keepalive.py`）后常驻，每 30 秒一轮：控制台探测连续失败或端口空置则结束僵死进程树并按 `--env-file` 同款参数拉起；读 `var/state/scheduler/service.lock` 判定调度器存活，锁主死亡即拉起 `gyxx schedule run`（调度器单实例锁保证不双跑）。单目标冷却 120 秒；30 分钟内 5 次重启触发熔断，暂停拉起 1 小时并记日志待人工检查。日志在 `var/logs/keepalive/keepalive.log`。
+- **验收**：`python deploy/watchdog/gyxx_keepalive.py --once` 单轮探测+处置；健康态应只输出 `keepalive-start`/`keepalive-once-done`。
+- 该脚本只做进程可用性守护，不承载任何业务时间规则（唯一调度路径仍是 `gyxx schedule run`）；未注册 Windows Task Scheduler。注意：Startup 方式要求用户已登录——机器重启后需人工登录一次，之后控制台与调度器会在约 2 分钟内自动就位。
@@ -33,8 +33,10 @@ PostgreSQL 的动态配置表。生命进程负责款式名、品牌、ERP 款
 | `product.sales_sheet.daily` | 款式与 ERP 编码 | `sync_monthly_sales_sheet.py` |
 | `product.erp_all_shop_daily` | 全部款式与 ERP 编码 | `backfill_erp_all_shop_daily.py` |

-共 12 个定时工作流依赖这套动态配置。维护脚本 `feishu_comment_batch.py` 和
-`db/sync_sku_master.py` 也已改为读取同一项目数据库，避免从非定时入口绕回旧主表。
+共 12 个定时工作流依赖这套动态配置。维护脚本 `feishu_comment_batch.py` 也已改为读取同一项目数据库，
+避免从非定时入口绕回旧主表。原先承担「动态配置 → 款式身份」投影的 `db/sync_dim_style.py` 与
+`db/sync_sku_master.py` 均已废弃：运行时身份已直读 `dim_style`，这两个脚本跑一次等于把自己写回自己，
+款式新增与改款统一走 data-hub「产品生命进程」建档页。

 `product.alert.daily`、商品导入、万相台广告、视频发布和市场排行使用各自数据库或专用配置，
 不依赖旧配置主表，因此不在本次切换范围内。
@@ -56,14 +58,15 @@ PostgreSQL 的动态配置表。生命进程负责款式名、品牌、ERP 款

 数据库使用父子表：

- `cmt_styles`：产品生命进程主表，一个款式一行，保存 ERP 编码、四个平台商品编码和 NFC 数量。
+- `dim_style`：产品生命进程主表，一个款式一行，保存 ERP 款式编码、四个平台商品编码、系列归属和销量飞书表；采集端运行时身份的唯一权威来源。
+- `cmt_styles`：笔记分析用的款式挂载表（`cmt_notes_master.style_id` 引用其 id），已退出 product 采集身份，仅在 `dim_style` 不可用时作兜底。
 - `workflow_dynamic_styles`：一个款式一行，保存共享飞书目标及内容工作流参数。
 - `workflow_dynamic_style_platforms`：一个款式可有多个平台子行，保留平台画像表、启停状态和旧商品 ID 兼容快照。
 - 旧 `workflow_dynamic_configs` 保留为原始迁移审计，不再参与运行时读取。
 - `erp_codes`、`item_ids` 使用 PostgreSQL 数组，页面接受中英文逗号或换行并自动去重。
 - `destinations` 使用通用 JSONB 目标列表，每项包含 `key`、`label`、`url`、`description`、`enabled`；六个旧目标字段仍同步保留，保证已有工作流兼容。后续新增分析逻辑只需增加一个稳定的 `key` 和对应飞书表地址。
 - 款式带递增 `revision`；整页保存和删除必须提交 `If-Match`，避免并发覆盖。
- `cmt_styles.platform_product_ids` 的四个键固定为 `tm`（天猫）、`jd`（京东旗舰店）、`jd_self`（京东自营）、`dy`（抖音）；空数组表示该平台尚未补齐。
+- `dim_style` 的四个平台列固定为 `tm_spus`（天猫）、`jd_spus`（京东旗舰店）、`jd_self_skus`（京东自营）、`dy_spus`（抖音）；空数组表示该平台尚未补齐。同系列只需主款填写，子款靠系列归属继承。

 ## 首次迁移

@@ -98,31 +101,78 @@ uv run --env-file 'D:/product-collector-analyze-flow/.env' `
 uv run --env-file 'D:/product-collector-analyze-flow/.env' `
  --project 'D:/gyxx-flow' `
  python -m gyxx_flow.migration.sync_feishu_style_identity --apply
-
-uv run --env-file 'D:/product-collector-analyze-flow/.env' `
-  --project 'D:/gyxx-flow' `
-  python 'src/gyxx_flow/modules/product_commerce/db/sync_dim_style.py'
 ```

 脚本按款式合并旧表的多平台行，只写入 `cmt_styles` 的品牌、ERP 编码和四个平台商品编码，
-不会覆盖 NFC、图片、类目或市场标签；空源字段不会清空生命进程已有身份。拼多多等当前四键
+不会覆盖 NFC、图片、类目或市场标签；空源字段不会清空已有身份。拼多多等当前四键
 模型之外的平台只保留在动态平台配置/审计中，不会伪装成天猫、京东或抖音编码。默认是预览，
 必须显式传 `--apply` 才会写数据库。

+> **`db/sync_dim_style.py` 已废弃并删除（2026-10-28）。**
+> 采集端运行时身份改为直接读 `dim_style`，该脚本再跑就是把 `dim_style` 投影回它自己，
+> 既补不了旧台账数据，也可能把空值写回。款式新增与改款统一走 data-hub「产品生命进程」
+> 建档页，下一轮调度自动生效，不需要再手工同步。
+
 服务器必须注入 `GYXX_POSTGRES_DSN`，或完整的 `PG_HOST/PG_PORT/PG_DB/PG_USER/PG_PASSWORD`。
 没有数据库且没有数据库生成的运行缓存时，相关工作流会明确失败，不会静默回读旧飞书主表。

 ## 运行时路径

 ```text
-产品生命进程 cmt_styles (商品身份)
+产品生命进程 dim_style (商品身份：ERP 编码 / 四平台编码 / 系列归属 / 销量表)
    +
 workflow_dynamic_styles + workflow_dynamic_style_platforms (目标/画像/启停)
    ↓
-统一运行时聚合
+统一运行时聚合（dim_style 优先，cmt_styles 仅在 dim_style 缺失时兜底）
    ├─ 完整身份门 → StyleConfigLoader → 商品经营工作流
    └─ 目标兼容输出 → feishu_mapping    → 内容营销工作流 → 各业务飞书目标表
 ```

+系列共享规则：平台商品编码、ERP 编码与 9 类飞书地址都以「系列」为一个采集单元，
+子款留空即可；子款误填也不会丢——loader 会把整条系列聚合并去重后挂到锚点。
+
+- **平台编码聚合去重到锚点**：整条系列各平台的商品 ID 先并集去重，再挂到锚点，由锚点
+  一个采集单元去抓。子款的原始编码保留在 `series.platform_ids` 供诊断，采集口判定让出
+  （「缺 tm商品编码（同系列「X」已在采）」），同一个商品 ID 不会被采两次、访客与加购
+  不会重复入账。锚点不在采集范围内时子款仍可回退自采。
+- **ERP 编码先聚合再采集**：正常建档形态是锚点填一条款式编码 + 各平台商品编码，子款只填
+  自己的款式编码；也可能出现锚点 `erp_style_codes` 为空、编码全散落在子款上的形态。
+  loader 一律把整条系列的 ERP 编码聚合成并集挂到锚点（`series.erp_owner`），锚点作为唯一
+  ERP 采集单元去聚水潭取数，子款让出并记录「ERP 已由同系列 X 归集采集」，但仍保留自己的
+  原始编码（`series.own_erp_codes`）供按款查询与月度销量表使用。
+  **例外（2026-10-30）**：子款自身 `collect_enabled = FALSE` 时，它的 ERP 编码不并入聚合
+  （`series.erp_excluded` 可见），锚点不代采；锚点自身 `FALSE` 仍是整系列停采。
+  平台商品编码不受子款开关影响，仍按整系列聚合。
+- **飞书地址聚合去重下发**：`workflow_config._propagate_series_targets()` 与
+  `_apply_series_sharing()` 两处都按「锚点优先、其余按款名」收集整系列非空地址，去重后
+  下发到每一行。一个表位只能落一张表，同字段出现多个不同地址时保留优先级最高的一条并打
+  `[WARN] 系列「X」<字段> 有 N 个不同表`。子款想用独立一套表，必须拆成独立系列。
+- **月度销量表按款明细**：`product.sales_sheet.daily` 的报表口径**不用**锚点上的聚合并集，
+  而是每个款式只查自己建档时填的编码。采集端的并集是给聚水潭当唯一取数单元用的，报表
+  照搬会让同系列每个成员行各查一遍并集、销量重复计入。锚点自身没编码时该行按 0 写入
+  （`--allow-missing-erp-as-zero`），同系列销量落在各子款行上，合计与采集总量一致。
+
+采集门只校验身份：ERP 日采要求 ERP 款式编码，平台日采再叠加对应平台商品编码。
+**销量飞书表地址不再是采集前置条件**——缺表地址的款式照常采集并落库，只是写表环节按
+地址缺失跳过（`get_bitable_style_map` 不生成该款目标），不影响数据入库与后续分析。
+主图采集同理：`get_main_image_styles` 只要求身份齐全，**主图飞书表地址不参与采集门**，
+缺地址的款照常抓图并写 `main_image_creatives`，`get_bitable_main_image_map` 只返回有地址
+的款，写表脚本对没有地址的款打 `[NO-TABLE]` 跳过飞书、仍入库。
+
+**写表脚本对目标地址 fail closed（2026-10-30）**：`insert_bitable_records.py`、
+`scripts/insert_main_image_records.py`、`scripts/insert_jd_main_image_records.py` 读取
+`get_bitable_style_map()` / `get_bitable_main_image_map()` 失败或返回空时，直接打
+`[FAIL] …拒绝回退本地 JSON` 并退出码 1，**不再回退** `bitable_style_map.json` /
+`bitable_main_image_map.json`。库与本地文件不一致时，回退会让数据静默写到错误的表，
+比直接失败更难发现。`styles_input.json` 同理由 `collect_erp_yesterday_metrics.load_erp_styles`
+与 `export_bitable_records.load_style_meta` 兜住：身份源不可用以 `lifecycle_unavailable` 失败。
+
+**旧表只保留「行集合 + 启停」职责**：`workflow_dynamic_styles` / `_platforms` 仍决定有哪些
+款式、`enabled` 启停、`note`、`style_content`、`destinations`，但 ERP 编码、平台商品编码与
+9 类地址全部由 `dim_style` 覆盖（`merge_lifecycle_style_records` 中 `managed` 由
+`jsonb_build_object` 恒生成四个键而恒为真）。旧表里**不在 dim_style 的残留行**（孤儿款）
+地址会被清空、不再采集，属于建档改名/合并后的历史垃圾，应直接删除
+（2026-10-30 已清理 `拓界mode`、`逐星evo`，备份 `var/tmp/workflow_dynamic_orphans_20261030.json`）。
+
 `StyleConfigLoader` 只在数据库短暂不可用时使用最近一次数据库成功读取后生成的本地缓存；
 `feishu_mapping` 的目标表字段缓存仍保留，用来减少对各业务飞书表的字段查询。
@@ -43,7 +43,7 @@ uv run gyxx scripts run <command_id> --date 2026-08-01 --execute
 | 京东自营品牌单日回采 | `shop.jd_self_operated.collect_brand` | `--date` 自动成为起止日期 |
 | 商品历史补采 | `product.backfill.run` | `--date` 自动成为 `--from/--to` |
 | 商品评价补采 | `product.review.orchestrate` | `--date` 自动成为目标日期 |
-| 天猫旗舰店和箱包店视频上传 | `product.video_upload.run` | 业务日期用于运行审计；默认顺序扫描两店，由 PostgreSQL 附件级账本防重 |
+| 天猫旗舰店、箱包店与天猫新店视频上传 | `product.video_upload.run` | 业务日期用于运行审计；`--store all` 顺序扫描三家，由 PostgreSQL 附件级账本防重 |
 | 京东旗舰店视频上传 | `product.jd_video_upload.run` | 业务日期只用于运行审计；从固定凌云air记录锚点按飞书视图顺序扫描，日期可为空 |
 | 采购单更新 | `supply.workflow.run` | 自动选择 `mcp-run purchase-order-update` |

@@ -51,16 +51,16 @@ uv run gyxx scripts run <command_id> --date 2026-08-01 --execute

 ### 天猫视频上传工作流

-天猫视频上传注册为幂等定时工作流 `product.video_upload`，按 `config/schedules.json` 以 2026-09-04 为起始日每 3 天 14:00（Asia/Shanghai）执行；前端仍可手动选择“光影行星旗舰店 / 光影行星鑫华达专卖店:龙虾仔”并立即运行，也可填写天猫话题完整名称或关键词。正式执行只处理所选店铺，话题搜索结果按包含关系选中。默认 dry-run 只生成运行记录并跳过浏览器节点；确认飞书待传记录、对应淘宝光合店铺的登录态和 PostgreSQL 后，再显式正式执行：
+天猫视频上传注册为幂等定时工作流 `product.video_upload`，按 `config/schedules.json` 以 2026-09-04 为起始日每 3 天 14:00（Asia/Shanghai）执行；前端仍可手动选择“光影行星旗舰店 / 光影行星鑫华达专卖店:龙虾仔 / 天猫新店（gyxx箱包旗舰店:龙虾仔）”并立即运行，也可填写天猫话题完整名称或关键词。正式执行只处理所选店铺，话题搜索结果按包含关系选中。默认 dry-run 只生成运行记录并跳过浏览器节点；确认飞书待传记录、对应淘宝光合店铺的登录态和 PostgreSQL 后，再显式正式执行：

 ```bash
 uv run gyxx run product.video_upload --date 2026-08-07
 uv run gyxx run product.video_upload --date 2026-08-07 --execute
 ```

-前端把店铺选择作为受限运行参数传给脚本，只允许 `flagship` 或 `luggage`。脚本先读取飞书并应用所选店铺路由，再写入/读取附件级账本；只有存在可领取的未上传附件时才启动有头 Chrome、登录对应账号并上传。旗舰店只选择“天猫上传”未勾选且有附件的光影行星男/女记录；箱包店只选择女性、全部款式命中箱包店白名单且有附件的记录。两店分别使用 `GUANGHE_USERNAME` / `GUANGHE_PASSWORD` 和 `GUANGHE_LUGGAGE_USERNAME` / `GUANGHE_LUGGAGE_PASSWORD`，箱包店当前账号为 `光影行星鑫华达专卖店:龙虾仔`，密码配置保持不变，并保留独立浏览器 Profile。发生登录或页面错误时保留窗口供人工定位，排查完成后手动关闭。
+前端把店铺选择作为受限运行参数传给脚本，只允许 `flagship`、`luggage` 或 `new_bag`。脚本先读取飞书并应用所选店铺路由，再写入/读取附件级账本；只有存在可领取的未上传附件时才启动有头 Chrome、登录对应账号并上传。旗舰店只选择“天猫上传”未勾选且有附件的光影行星男/女记录；箱包店只选择女性、全部款式命中箱包店白名单且有附件的记录；天猫新店不区分性别，只要求一条记录的**全部**「代拍产品」款式都属黑曜石系列（关键词 `黑曜石`，如黑曜石托特、黑曜石吐司、黑曜石电脑包、黑曜石饺子），组合款里掺了其它系列整条记录都不进新店，视频源与其它店铺为同一张飞书表。三店分别使用 `GUANGHE_USERNAME` / `GUANGHE_PASSWORD`、`GUANGHE_LUGGAGE_USERNAME` / `GUANGHE_LUGGAGE_PASSWORD` 和 `GUANGHE_NEW_BAG_USERNAME` / `GUANGHE_NEW_BAG_PASSWORD`，箱包店当前账号为 `光影行星鑫华达专卖店:龙虾仔`，天猫新店当前账号为 `gyxx箱包旗舰店:龙虾仔`，密码配置保持不变，各自保留独立浏览器 Profile（`-luggage` / `-new-bag`）。发生登录或页面错误时保留窗口供人工定位，排查完成后手动关闭。

-每次正式扫描都会把视频附件写入 PostgreSQL 附件级发布账本，防重身份由视频 `file_token + 目标账号 + 飞书记录` 构成。账本状态为 `published` 的附件在下次执行时自动跳过，因此同一业务日期可以安全重跑并发现新增附件。旗舰店仍只在一条飞书记录的全部附件发布成功后回写“天猫上传”；箱包店不借用该勾选状态，是否已上传完全以自己的目标账号账本为准。
+每次正式扫描都会把视频附件写入 PostgreSQL 附件级发布账本，防重身份由视频 `file_token + 目标账号 + 飞书记录` 构成。账本状态为 `published` 的附件在下次执行时自动跳过，因此同一业务日期可以安全重跑并发现新增附件。旗舰店仍只在一条飞书记录的全部附件发布成功后回写“天猫上传”；箱包店和天猫新店不借用该勾选状态，是否已上传完全以自己的目标账号账本（`guanghe-luggage` / `guanghe-new-bag`）为准，同一视频可以分别发到多家店铺。

 首次正式执行会读取 `config/tmall-video-publication-backfill.json`，将 2026-07-26 箱包店旧日志中可按 `record_id + file_token + 文件名` 再次核验的 25 个明确成功附件回填为 `published`。配置与当前飞书任一身份不一致都会停止，不会静默猜测。该回填不依赖部署机保留旧 `var/` 日志文件；证据路径与成功行号已固化在配置中供审计。

@@ -103,11 +103,17 @@ uv run gyxx scripts run product.jd_video_upload.run --date 2026-08-07 --arg=--re

 唯一工作流 ID 是 `product.ecommerce_costs.daily`，每天 10:00 由项目内 Python 调度器运行，
 业务日期为前一天；天猫、京东、抖音三个下载分支并行启动，各自完成下载后再导入。天猫流程先进入万相台商品报表选择昨日，提交下载任务；任务在下载列表显示“生成成功”
-后下载 ZIP、做路径安全校验并解压 CSV，随后将商品级广告指标幂等写入
-`product_daily_metrics.raw_data.tm_ad_metrics`。商品款式先按 `dim_style.tm_spus` 的商品 ID 匹配，
-匹配不到再按商品名称做精确包含匹配，并在节点内保留 `style_name` 与
-`style_match_method`（`product_id`/`product_name`/`ambiguous`/`unmatched`）。它不会覆盖同一商品日期
-已有的访客、成交等经营日报字段；data-hub 产品生命进程从该节点读取广告成交金额和花费并计算投产比、电商费比。
+后下载 ZIP、做路径安全校验并解压 CSV，随后将商品级广告指标幂等写入独立表
+`tm_product_ad_daily_metrics`（每日每商品一行，主键 `(stat_date, product_id)`）。花费、直接/间接/归因
+成交金额和投产比走标量列，报表其余指标（展现、点击、点击率、转化率、预售成交等）原样存在 `metrics`
+JSONB。商品款式先按 `dim_style.tm_spus` 的商品 ID 匹配，匹配不到再按商品名称做精确包含匹配，结果
+写进 `style_name` 与 `style_match_method`（`product_id`/`product_name`/`ambiguous`/`unmatched`）。
+天猫广告数据**不再写入** `product_daily_metrics.raw_data.tm_ad_metrics`，投放与经营日报分表，经营数据
+重新同步不会覆盖投放数据；data-hub 产品生命进程从新表读取花费和广告归因成交金额并计算投产比、电商费比。
+
+历史数据迁移（2026-11-01 一次性执行）：`src/gyxx_flow/modules/product_commerce/scripts/
+migrate_tm_ad_metrics_to_table.py`，默认 dry-run，`--execute` 先备份旧节点到运行时 TMP_ROOT 再回填，
+校验行数与花费合计一致后才允许 `--clear-source` 清除 `raw_data` 里的旧节点。

 正式服务必须注入 `WANXIANG_ACCOUNT`、`WANXIANG_PASSWORD`、`GYXX_DATA_ROOT` 和
 `GYXX_POSTGRES_DSN`。默认浏览器 Profile 是
@@ -139,8 +145,9 @@ uv run gyxx scripts run product.import.tmall_ads --date 2026-08-17 --execute
 1. 检查 `gyxx schedule status`、服务日志和对应 `run_id`，确认失败发生在申请、下载、解压还是入库。
 2. 检查同一 Profile 是否仍能打开万相台商品报表；登录失效时在有头命令中人工完成验证。
 3. 检查日期目录中的 `manifest.json`、CSV 表头和 `日期/主体ID/主体名称`；ZIP 存在不代表数据库已导入。
-4. 检查 PostgreSQL 的 `product_daily_metrics` 是否存在 `platform='tm'`、目标日期和商品 ID，且
-   `raw_data->'tm_ad_metrics'` 已更新；确认经营日报其他 JSON 节点仍保留。
+4. 检查 PostgreSQL 的 `tm_product_ad_daily_metrics` 是否存在目标日期与商品 ID 的行，`spend`、
+   `attributed_deal_amount` 是否为目标值，`style_name` 是否已匹配；确认 `product_daily_metrics`
+   同日期经营数据（访客、成交）仍完整。
 5. 只有在外部状态核对清楚后才按同一业务日期重跑，不能用换日期或重复申请掩盖下载结果不明。

 ## 调度操作
@@ -161,6 +168,19 @@ uv run gyxx schedule run --env-file /etc/gyxx-flow/gyxx-flow.env

 需要停用单个工作流时，优先修改该工作流的调度启用状态，先 dry-run，再重启服务。其他模块继续运行。

+### 开机自启（Windows 生产机）
+
+重启/关机开机后的自动恢复链路：**Windows 自动登录 → `Startup\gyxx-keepalive.vbs` → `deploy/watchdog/gyxx_keepalive.py` → 控制台(8765) + `gyxx schedule run` 调度器**。keepalive 每 30 秒巡检并负责拉起死掉的控制台/调度器（单实例锁防双跑，冷却 120s、熔断保护），日志在 `var/logs/keepalive/keepalive.log`，验收用 `python deploy/watchdog/gyxx_keepalive.py --once`。
+
+采集依赖浏览器 CDP，必须运行在交互式用户会话中，因此不能改成 Windows 服务或 Task Scheduler SYSTEM 任务（也不允许注册 Windows 任务计划）。免登录自启的唯一环节是自动登录，用脚本启用（密码交互输入、写入 `HKLM\...\Winlogon`，仅 Administrators 可读）：
+
+```powershell
+powershell -ExecutionPolicy Bypass -File D:\gyxx-flow\deploy\watchdog\autologon.ps1           # 启用
+powershell -ExecutionPolicy Bypass -File D:\gyxx-flow\deploy\watchdog\autologon.ps1 -Disable # 撤销
+```
+
+启用后重启验证：开机自动进桌面，约 2 分钟内控制台与调度器就位；停机窗内到期任务由调度器 catch-up 自动补跑。
+
 ## 数据和浏览器状态

 生产 `GYXX_DATA_ROOT` 必须位于源码目录之外，Linux 推荐 `/var/lib/gyxx-flow`：
@@ -190,16 +210,38 @@ uv run gyxx schedule run --env-file /etc/gyxx-flow/gyxx-flow.env
 - `tmall-ozko`：商品经营日报 ozko 账号，900 秒；
 - `xiaohongshu-content`：小红书，900 秒；
 - `douyin-content`：抖音内容侧，900 秒；
- `xingtu`：星图，900 秒；
- `chanmama`：蝉妈妈，900 秒；
+- `chanmama`：蝉妈妈，900 秒（脚本自带账密登录）；
 - `bilibili-content`：B 站，900 秒。

+星图/百应采集（`xingtu_scraper_v2.py`）的登录态挂在 `douyin-shop` 账号下；`xingtu` 账号当前无成员脚本，仅保留 vault。
+
 商品经营日报 `product_commerce:taobao_sycm_collect.py` 会在同一次运行中按品牌路由账号：光影行星使用 `tmall-shop`，ozko 使用 `tmall-ozko`，两边各自使用独立 Profile 和账号 vault。`tmall-ozko` 没有静态成员脚本是有意设计，因为它由这个按品牌拆分的包装工作流动态选择；登录态成功后仍会回写 `state/accounts/tmall-ozko/`，由同一套 keepalive 负责续期。

 账号需要在 `config/runtime-bindings.json` 的 `accounts.<account_id>.keepalive` 中显式开启续期，并配置一个不会产生业务副作用的已登录页面 URL。所有账号由唯一的 `gyxx schedule run` 常驻进程在后台维护；每个账号使用独立 CDP 端口、独立 Profile 和账号级 Cookie vault，启动后按 `initial_delay_seconds` 错峰访问，避免同时打开多个登录页面。访问成功且没有跳转到登录页时，Scrapling 才会把最新 Cookie/storage state 原子写回 `state/accounts/<account_id>/`，并同步给成员脚本；不需要另建 Windows 任务计划、cron 或 timer。

 Cookie 续期只能延长站点支持滑动续期的会话，不能突破服务端硬过期、主动退出、风控或验证码。keepalive 检测到登录页/无效 Cookie 时会停止发布，不会用失效状态覆盖最后一份有效 vault；此时按 `gyxx accounts list` 检查 `keepalive.status`，必要时重新执行 `gyxx accounts login <account_id>`。

+### 工作流执行前的 Cookie 健康网关
+
+每个正式执行（非 dry-run）的工作流在采集步骤前会自动注入"每账号一个"的 `cookie-gate.<account_id>` 关键步骤，由 `src/gyxx_flow/cookie_gate.py` 实现按账号分三级的状态机：
+
+- **blocking**（纯扫码/验证码账号：`douyin-shop`、`xiaohongshu-content`、`douyin-content`、`bilibili-content` 等）：先看 keepalive 近期成功状态（默认 1 小时内视为新鲜，可用 `GYXX_COOKIE_GATE_MAX_STATUS_AGE_SECONDS` 调整），新鲜且 vault 静态有效则直接放行；否则用 keepalive 同款无头浏览器真实探活。探活失败即判定过期：gate 步骤失败，依赖它的采集步骤记 `dependency-failed` 跳过、工作流失败退出，并通过飞书通知登录负责人（`config/notification-routing.json` 中手机号为 `+8613663785145` 的人员，默认 30 分钟节流，可用 `GYXX_LOGIN_ALERT_OPEN_IDS` 覆盖收件人）执行 `uv run gyxx accounts login <account_id>` 完成扫码/验证码重登。
+- **recover**（配置了 `login_command_id` 的账密自动重登账号：`erp-jushuitan`）：过期时先自动跑账密登录恢复，恢复成功放行；失败才阻断并飞书通知。
+- **defer**（成员脚本自带账号密码登录的账号：京东系、天猫系、`chanmama`）：不注入 gate，保持脚本内账密自动登录流程不变。
+
+按账号划分的持久状态在 `state/accounts/<id>/cookie_state.json`（`healthy` ↔ `awaiting_login`）。编排器型工作流（入口脚本本身无账号绑定、由子采集脚本使用账号）在 `config/workflows.json` 的 `required_accounts` 字段声明依赖账号。`--rerun-step`/`--resume-from` 切片和 shadow 运行不发送通知；dry-run 完全不注入 gate。设置 `GYXX_COOKIE_GATE=0` 可临时关闭整网关。
+
+采集脚本不再内嵌扫码/验证码登录：运行中检测到登录态失效会以 `CookieExpiredError` 快速失败并按退出码 75 记为 cookie 跳过；`self_xhs_scraper.py --login-only` 是唯一保留的人工登录引导（该脚本不在任何账号成员内，自管绑定状态）。
+
+### 登录恢复后的自动重跑
+
+gate 阻断工作流时会把待重跑清单（工作流 + 业务日期，shadow/验收运行不记录，同工作流同日期去重，上限 20 条）写进 `state/accounts/<id>/cookie_state.json` 的 `pending_reruns`。常驻调度器（`gyxx schedule run`）的维护循环里挂着 `CookieGateRecoveryManager`：
+
+- 每个心跳检查有待重跑记录的账号：keepalive 近期成功 + vault 静态有效 → 立即视为恢复；否则每 5 分钟做一次真实探活（mode C 账号会自动账密重登，重登成功同样触发恢复）。
+- 恢复确认后，用调度器同款启动器按**原业务日期**自动重跑全部待重跑工作流（日志在 `data/logs/scheduler/`），清空清单并飞书通知"登录态已恢复 + 已自动重跑清单"。
+- 启动失败的任务保留在清单里下一轮重试；同一账号两次重跑启动之间有 5 分钟冷却；控制台手动脚本类临时运行 id（`script.*`）不会自动重跑。
+- 与正常调度并发安全：工作流级锁 + EffectLedger 幂等保证同一天不会重复产生副作用。
+
 项目不会自动搬动或删除已有 `var/`。迁移数据根时应停止调度器，完整备份和复制原目录，修改环境变量后依次运行 doctor、dry-run 和一次指定日期的对账执行。

 ## 清理和保留
@@ -251,6 +293,215 @@ uv run gyxx effects reconcile <workflow_id> \

 浏览器状态异常时，只备份并处理目标脚本的 `state/browser/<module>/<script>`，不要清理其他脚本状态。调度状态异常时先停止唯一调度进程并备份 `state/scheduler`，不要同时启动第二实例。

+## 商品日报采集开关（product.daily）
+
+采集开关是 `dim_style.collect_enabled`（2026-10-29 起），**两层语义**（2026-10-30 起）：
+
+1. **锚点是系列总开关**：以该系列 `is_series_anchor = TRUE` 那一行为准；系列没有锚点时按款名
+   排序第一个。锚点 `FALSE` → **整条系列一款都不采**，同系列其余款式跟随锚点。
+2. **锚点 `TRUE` 时，子款开关只决定自己的 ERP 编码是否参与聚合**：子款 `TRUE` 的 ERP 款式编码
+   并入锚点一起去聚水潭取数；子款 `FALSE` 的编码不并入（既不由它自己采，也不让锚点代采）。
+   被排除的子款列在 `series.erp_excluded`，运行时打
+   `[INFO] 系列「X」子款 Y 已关闭采集，ERP 编码不纳入「Z」聚合`。平台商品编码仍按整系列聚合，
+   不受子款开关影响。
+
+旧的独立表 `product_daily_collect_scope` 已迁移并删除，备份在
+`var/tmp/collect_scope_before_drop_20261029.json`。
+
+默认值是 `TRUE`（采集），因此产品生命进程在 `dim_style` 新增的款式**无需任何操作**就会进入日报；
+配件、包装、耗材、停售款才需要显式关掉。
+
+关掉整条系列（改锚点那一行即可，子款自动跟随）：
+
+```sql
+-- 单系列
+UPDATE dim_style SET collect_enabled = FALSE WHERE style_name = '黑曜石托特包2';
+
+-- 批量：把若干系列的所有行一起关（无锚点的系列也覆盖到）
+UPDATE dim_style SET collect_enabled = FALSE
+WHERE COALESCE(NULLIF(btrim(series_name), ''), style_name)
+      IN ('飞机盒', '香片', '雨衣', '鞋袋');
+```
+
+恢复采集把 `FALSE` 改成 `TRUE`。改完无需重启控制台或调度器：`StyleConfigLoader.load()`
+每次运行都直接读库，下一轮 `product.daily`（每天 08:40，业务日 T-1）即生效。
+
+两点注意：
+
+- **开关落在锚点行上才控制整条系列。** ERP 编码与三平台商品编码都由
+  `_apply_series_sharing()` 按整系列聚合并去重后挂到锚点（子款误填的编码也会并入，但
+  `collect_enabled = FALSE` 的子款 ERP 编码除外），锚点是系列唯一的采集单元；子款自己的
+  编码保留在 `series.own_erp_codes` 与 `series.platform_ids` 备查。
+- 关掉只影响**商品日报**（tm/jd/dy 平台段 + ERP 段）。天猫/京东主图周采集、聚水潭全店铺日报、
+  三平台电商费用**不受此开关影响**。
+
+放行不等于一定会采：采集门仍要求「ERP 款式编码 + 各平台商品编码」齐备；飞书目标表地址
+（下一节的 9 类 `bitable_*` 列）只决定能否写飞书，缺地址的款照常采集入库。缺口写在
+`var/logs/product_commerce/<date>_skipped_styles.json` 的 `skipped.reason`。
+
+改款式名时要同步改 `series_name`／锚点标记，否则旧开关会变成孤儿行、新名字按默认 TRUE 采集。
+
+## 飞书表地址维护（dim_style）
+
+2026-10-29 起，9 类飞书多维表格地址**全部维护在 `dim_style`**，而且只写在系列锚点行上。
+`workflow_dynamic_styles` / `workflow_dynamic_style_platforms` 里的同名字段已废弃，
+代码不再读取；旧值备份在 `var/tmp/bitable_urls_before_move_20261029.json`。
+
+| 用途 | `dim_style` 列 | 原动态配置字段 |
+|---|---|---|
+| 销量表 | `bitable_weekly_url` | `sales_bitable_url` |
+| 主图表 | `bitable_main_image_url` | `main_image_bitable_url` |
+| 合作达人表 | `bitable_creator_url` | `creator_bitable_url` |
+| 自营合作达人表 | `bitable_self_creator_url` | `self_creator_bitable_url` |
+| 每周/每月笔记分析 | `bitable_weekly_note_url` | `weekly_note_analysis_url` |
+| 平台单品分析 | `bitable_style_analysis_url` | `style_analysis_bitable_url` |
+| 人群画像（天猫） | `bitable_persona_tm_url` | `tm_persona_bitable_url` |
+| 人群画像（京东） | `bitable_persona_jd_url` | `jd_persona_bitable_url` |
+| 人群画像（抖音） | `bitable_persona_dy_url` | `dy_persona_bitable_url` |
+
+子款留空即可：两处代码会把整条系列的地址**聚合去重**后统一下发——
+`workflow_config._propagate_series_targets()` 在合并建档数据时按「锚点优先、其余按款名」
+收集整系列的非空地址，去重后写回每一行；`style_config_loader._apply_series_sharing()`
+在聚合配置时再做一次同样的兜底。因此子款误填的地址不会丢、也不会让同系列出现两套表。
+一个表位只能落一张表：同一字段填了多个不同地址时，保留优先级最高的那条，并在日志里打
+`[WARN] 系列「X」<字段> 有 N 个不同表`，提示到锚点手动裁决。
+
+```sql
+-- 改某个系列的主图表（只改锚点一行，子款自动跟随）
+UPDATE dim_style
+SET bitable_main_image_url = 'https://xxx.feishu.cn/base/<base>?table=<tbl>'
+WHERE style_name = '逐星';   -- 必须是该系列的锚点行
+
+-- 查看哪些锚点还缺地址
+SELECT style_name, series_name,
+       bitable_weekly_url, bitable_main_image_url, bitable_creator_url
+FROM dim_style
+WHERE is_series_anchor AND COALESCE(bitable_weekly_url, '') = ''
+ORDER BY style_name;
+```
+
+改完不需要重启：`StyleConfigLoader.load()` 每次运行都读库，下一轮工作流即生效。
+若某类地址为空，对应环节会跳过写表但照常采集入库（例如主图缺表地址打 `[NO-TABLE]`）。
+
+**地址读取是 fail-closed 的**（2026-10-30）：写表脚本拿不到 `dim_style` 地址时会打
+`[FAIL] … 拒绝回退本地 JSON` 并以退出码 1 结束，**不会**改用 `bitable_style_map.json` /
+`bitable_main_image_map.json` 里的旧地址。所以看到这类 `[FAIL]` 先去查库里的地址，
+不要去改本地 JSON（这两个文件已无读取方）。
+
+新增系列时，地址写在锚点行。若某款业务上确实是独立一套表（历史上 `逐星GT` 这种情形），
+必须把它拆成独立系列（单独的 `series_name` + 锚点标记）并把地址写在自己那行；只写地址
+而不拆系列的话，聚合会把它并入原系列，只在日志里留一条告警，不会真的生效。
+
+## 款式主键统一（style_id = dim_style.id，2026-11-01）
+
+采集端不再只用 `style_name` 关联款式。**唯一关联键是 `style_id`**（= `dim_style.id`，
+与 data-hub 的 22 张表同一套主键），`style_name` 降级为显示冗余。这样在 dim_style
+改款名时，历史指标表不会变成孤儿。
+
+### 数据从哪里来
+
+| 环节 | 机制 |
+|---|---|
+| loader 下发 | `merge_lifecycle_style_records()` 输出 `style_id` + `style_match_method`，经 `style_config_loader` 进入每条款式配置 |
+| db 层兜底 | `db.attach_style_ids()`（按款名）/ `db.resolve_style_ids_by_product()`（按平台商品 ID）在 upsert 内补齐 |
+| 历史回填 | `backfill_style_id.py` |
+
+### 回填与验收
+
+```powershell
+# 只报数，不写库
+.venv\Scripts\python.exe backfill_style_id.py --dry-run
+# 执行（幂等，只补 style_id IS NULL 的行）
+.venv\Scripts\python.exe backfill_style_id.py
+# 只补款名表，跳过按商品 ID 的两步
+.venv\Scripts\python.exe backfill_style_id.py --skip-product-table --skip-ad-tables
+```
+
+回填分三段，缺一段覆盖率就上不去：
+
+| 段 | 表 | 依据 |
+|---|---|---|
+| 按款名 | 13 张（`daily_style_metrics` / `weekly_style_metrics` / `erp_all_shop_style_daily_metrics` 等） | `t.style_name = dim_style.style_name` |
+| 按商品 ID | `product_daily_metrics` | `platform + product_id` 反查 `*_spus` |
+| 按商品 ID（二次） | 三张广告表 | 同上；**顺带把空的 `style_name` 补成 dim_style 的权威款名** |
+
+广告表必须走第三段：平台的报表不给款式名，`jd_product_ad_daily_metrics` 有 27%、
+`dy_product_ad_daily_metrics` 有 40% 的行 `style_name` 为空，只按款名回填完全无效。
+补完后 data-hub 现有按款名的 join 立刻多认回一批行，不必等主键切换。
+
+```sql
+-- 各表回填进度；除 product_daily_metrics 与三张广告表外都应 100%
+SELECT * FROM v_style_id_coverage ORDER BY tbl;
+-- 未建档的平台商品 ID，仅统计经营日报口径
+SELECT * FROM v_style_unmatched_product ORDER BY platform, rows DESC;
+-- 同上，但把三张广告表也算进来（纯投流商品只在这里出现）
+SELECT * FROM v_style_unmatched_source ORDER BY platform, rows DESC;
+```
+
+`style_match_method` 取值：`dim_style`（按款名精确）、`product_id`（按平台商品 ID）、
+`product_name`（按商品名包含）、`unmatched`（确实没建档）。
+
+### 补 spus 的正确做法
+
+`product_daily_metrics.product_no` 存的是**款名**（不是 ERP 编码），天猫填充率 100%、
+京东 97.8%、抖音 0%。补 spus 时按这个优先级，命中率远高于猜商品标题：
+
+1. `product_no` 精确等于 `dim_style.style_name`；
+2. `product_no` 与款名互为前后缀（长度差 ≤ 3，如「蓝鹊pro」→「蓝鹊」、「知更大号」→「知更」）；
+3. 商品标题包含完整款名（抖音只能用这一条），并用「命中位置后的版本后缀」排除串款
+   —— 例如标题「盖亚斜挎包**3**」不能归到「盖亚斜挎」；
+4. 编码与标题互相矛盾时**不要自动写**，列出来人工定（实测有天猫单品编码写「理岩」、
+   标题写「阿波罗 air」，支付额 50 万，猜错代价极大）。
+
+物料类（数据线、内胆包、补邮费、配件包、礼品袋、定制服务）不该进 `spus`，应留
+`unmatched`。`dim_style` 里本身建成款的物料（`配件包`、`飞机盒`、`电泵` 等）也不写。
+
+### 必须知道的边界
+
+- `product_daily_metrics` 表里**没有款名列**，只能拿 `platform + product_id` 去
+  `tm_spus / jd_spus / dy_spus` 反查。未建档的商品 ID 回填不了，要靠建档页补 `spus`
+  才能涨，不是脚本 bug。
+- 2026-11-01 补过一轮：`product_daily_metrics` 49.1% → **86.0%（jd）/ 77.3%（tm）/
+  76.7%（dy）**，整体 79.0%；广告表 tm 99.4% / jd 92.4% / dy 83.2%。
+  剩余 259 个商品 ID（dy 107 / jd 52 / tm 100）未建档，清单见 `v_style_unmatched_source`。
+- **不建物理外键**：14 张表的 `style_id` 都只做**逻辑外键**（与 data-hub 侧一致）。
+  库里没有任何指向 `dim_style(id)` 的 FK 约束，完整性靠写入端（loader 下发 +
+  `attach_style_ids` 兜底）、`backfill_style_id.py --dry-run`、`v_style_id_coverage`
+  三道保证。理由：采集语义是「宁可留 NULL 待回填，也不要整批写入失败」，
+  物理外键会在传错 id 时直接中断整天采集；此外拆系列/合并同款要改数据时，
+  还得先删约束。早期 P4 建过的 10 个 `<表名>_style_id_fk` 已在 schema.sql 里
+  幂等 `DROP CONSTRAINT IF EXISTS` 清掉，老库重跑 schema 即自动回到无 FK 状态。
+- 主键一律没动：`daily_style_metrics` 仍是 `(metric_date, style_name, platform)`。
+  data-hub 有 6 处按 `style_name` join，改主键收益小、回归面大。
+
+## 商品评价采集下线（product_reviews，2026-11-01）
+
+`product_reviews` 已**停止写入并归档**为 `product_reviews_archived_20261101`。
+原因是 JD 侧 `style_name` 全靠 `sku_id` 反查，而映射表不存在，214 行全部为 NULL，
+两个消费方都是 `WHERE style_name = ?`，等于入库了读不到。
+
+差评统一由三个独立工作流采集：`shop.tmall_negative_reviews.daily`、
+`shop.jd_negative_reviews.daily`、`shop.dy_negative_reviews.daily`。
+
+### 消费方改读视图
+
+```sql
+SELECT platform, comment_date, rating, content, style_id, style_name, style_match_method
+FROM v_negative_reviews;   -- tm + jd + dy 三张差评表 UNION，款式按 ID 优先、商品名兜底
+```
+
+改动点：`analyze_style.py:pull_reviews`、`content_marketing/daily_marketing_report.py:_review_context`。
+
+### 口径变化（必须知道）
+
+- 数据源**只有差评一档**，不再有好评/中评。风格分析的「评价等级分布」因此只剩差评。
+- 营销日报的 `sample_size` = 差评条数，`negative_rate` 恒为 1.0；门槛从「至少 5 条评价」
+  改成「至少 1 条差评」（旧门槛在纯差评源上会导致永不产出）。
+- 覆盖面反而更全：天猫差评 97 条（旧表天猫为 0）、京东 32 条（旧表 8 月断更）。
+- 若产品侧确实需要好评样本，应把差评采集器扩成「按等级采集」，**不要恢复 product_reviews**。
+- `dy_product_scraping.py` 默认流程里的评论采集已摘除（只产出无人读取的本地 JSON）；
+  需要历史样本时用 `--reviews-only` 显式触发。
+
 ## 业务源码同步

 外部业务源码目录只用于发现更新，GYXX Flow 运行时不导入或挂载这些目录。只读检查：
@@ -0,0 +1,386 @@
+# 款式主键统一 + 评论采集下线 方案
+
+> 状态：**已实施（2026-11-01）**，见文末「八、实施记录」
+> 涉及项目：`gyxx-flow`（采集端）、`data-hub`（消费端，共用同一个 PostgreSQL）
+> 编写日期：2026-11-01
+
+## 一、三条结论
+
+1. **`product_reviews` 可以下线**，但不能直接删——它有两个消费方（风格分析、营销日报）要改成读三张差评表。
+2. **`style_id` 就是 `dim_style.id`**，与 data-hub 现有 `style_id` 同一体系，不会产生第二套 ID。已实测：`erp_product.style_id` 的 135 个值 **135/135** 命中 `dim_style.id`，`own_brand_product.style_id` 31/34 命中（3 条为历史脏数据）。
+3. **改造入口只有一处**：`DIM_STYLE_RUNTIME_SQL` 早已 `SELECT s.id`，`merge_lifecycle_style_records()` 也已经拿到这个 id——只是被编码进 `source_record_id = "lifecycle_style:<id>"` 字符串里，没有作为独立字段下发给写入脚本。把它提成正式字段，全链路就都有了。
+
+## 二、现状盘点
+
+### 2.1 两套引用方式并存
+
+| 侧 | 关联方式 | 表 |
+|---|---|---|
+| data-hub（产品生命、营销中台、客服、内容） | `style_id BIGINT` → `dim_style.id` | `erp_product`、`own_brand_product`、`erp_marketing_*`、`erp_cs_*`、`cmt_*`、`market_product_classification` 等 22 张 |
+| gyxx-flow（采集端） | `style_name TEXT` | 15 张，见 2.3 |
+
+### 2.2 实测匹配率（2026-11-01）
+
+| 表 | 总行数 | 按 style_name 命中 dim_style |
+|---|---|---|
+| `daily_style_metrics` | 17,352 | 17,352（100%） |
+| `erp_all_shop_style_daily_metrics` | 9,399 | 9,399（100%） |
+| `product_reviews` | 1,936 | 1,434（74%，全部来自抖音；京东 214 行 style_name 全 NULL） |
+| `product_daily_metrics` | 75,568 | **无款式列**，只能靠 `tm_spus/jd_spus/dy_spus @> to_jsonb(product_id)` 现算 |
+
+`product_daily_metrics` 按商品 ID 反查的现状：
+
+| 平台 | 行数 | 可匹配 | 去重商品 ID | 可匹配 ID |
+|---|---|---|---|---|
+| dy | 28,972 | 13,615（47%） | 253 | 83 |
+| jd | 17,044 | 9,394（55%） | 160 | 61 |
+| tm | 29,552 | 14,089（48%） | 281 | 92 |
+
+一对多歧义 **0 条**。约一半行匹配不上，原因是 **458 个商品 ID 从未写进 dim_style 的 spus 数组**（dy 170 / jd 99 / tm 189），不是字段设计问题。
+
+### 2.3 采集端 15 张只存 style_name 的表
+
+`daily_style_metrics`、`weekly_style_metrics`、`erp_all_shop_style_daily_metrics`、`daily_persona_metrics`、`main_image_creatives`、`fact_style_analysis_reports`、`decline_events`、`fact_alert_history`、`jd_self_weekly_sales`、`persona_screenshots`、`product_reviews`、`jd_product_ad_daily_metrics`、`dy_product_ad_daily_metrics`、`workflow_dynamic_styles`、`product_daily_metrics`（无列，靠 ID 反查）。
+
+## 三、目标模型
+
+- **唯一关联键 = `style_id`（`dim_style.id`）**，所有采集表落库时必须带上。
+- `style_name` 保留为**显示冗余**，不再作为 join 键；等下游切完 join 后可以只在视图里出现。
+- 匹配不上的行（未建档商品、广告未匹配 SKU、差评）**允许 `style_id IS NULL`**，但必须同时写 `style_match_method`（`product_id` / `product_name` / `anchor` / `unmatched`），便于统计漏维护。
+- **不动现有主键**。`daily_style_metrics` 的 PK 仍是 `(metric_date, style_name, platform)`——data-hub 有 6 处 `metric.style_name = style.name` 的 join，改 PK 收益小、回归面大。加列、加索引、切 join 即可。
+
+## 四、实施分期
+
+### P0 观测层（只读，不动数据）
+
+新增两个视图，先把问题量化出来，方案落地后靠它们验收：
+
+```sql
+-- 未建档的平台商品 ID（推动在 dim_style 里补 spus）
+CREATE OR REPLACE VIEW v_style_unmatched_product AS
+SELECT m.platform, m.product_id,
+       MAX(m.product_name) AS product_name,
+       MIN(m.stat_date) AS first_date, MAX(m.stat_date) AS last_date,
+       COUNT(*) AS rows
+FROM product_daily_metrics m
+WHERE NOT EXISTS (
+        SELECT 1 FROM dim_style d
+        WHERE (m.platform = 'tm' AND d.tm_spus @> to_jsonb(m.product_id))
+           OR (m.platform = 'jd' AND d.jd_spus @> to_jsonb(m.product_id))
+           OR (m.platform = 'dy' AND d.dy_spus @> to_jsonb(m.product_id)))
+GROUP BY m.platform, m.product_id;
+
+-- 各表 style_id 回填进度
+CREATE OR REPLACE VIEW v_style_id_coverage AS
+SELECT 'daily_style_metrics' AS tbl, COUNT(*) AS rows,
+       COUNT(style_id) AS with_style_id FROM daily_style_metrics
+UNION ALL SELECT 'erp_all_shop_style_daily_metrics', COUNT(*), COUNT(style_id)
+       FROM erp_all_shop_style_daily_metrics
+UNION ALL SELECT 'product_daily_metrics', COUNT(*), COUNT(style_id)
+       FROM product_daily_metrics;
+```
+
+验收锚点：`v_style_id_coverage` 里前三张表 `with_style_id` 开始上涨到预期比例；`v_style_unmatched_product` 的条数随建档推进下降。
+
+### P1 评论采集链路下线
+
+**要删的东西**
+
+| 位置 | 动作 |
+|---|---|
+| `orchestrate_daily_collection.py:437-452` | 移除 `import_product_reviews` 步骤 |
+| `import_product_reviews.py` | 删除 |
+| `orchestrate_review_collection.py` | 删除 |
+| `config/commands.json:516` `product.review.orchestrate` | 移除命令 |
+| `db/__init__.py` `upsert_product_reviews` | 删除 |
+| `db/schema.sql:813-870` `product_reviews` 建表段 | 移除（库里的表归档不删） |
+
+**要补的东西**
+
+1. 建统一差评视图，字段对齐两个消费方现在的用法（`platform` / `rating` / `content` / `style_id` / `comment_date`）：
+
+```sql
+CREATE OR REPLACE VIEW v_negative_reviews AS
+SELECT 'tm' AS platform, n.business_date AS comment_date, n.review_level AS rating,
+       n.content, NULL::text AS order_id, NULL::text AS product_id,
+       n.product_name, n.review_id, n.review_at, n.raw_review
+FROM tmall_negative_reviews n
+UNION ALL
+SELECT 'jd', n.business_date, n.review_level, n.content, n.order_id, NULL,
+       n.product_name, n.review_id, n.review_at, n.raw_review
+FROM jd_negative_reviews n
+UNION ALL
+SELECT 'dy', n.business_date, n.rating_level, n.content, n.order_id, n.product_id,
+       n.product_name, n.review_id, n.review_at, n.raw_review
+FROM dy_negative_reviews n;
+```
+
+2. 改两个消费方：
+   - `analyze_style.py:112 pull_reviews()` → 改查 `v_negative_reviews`（按 `style_id` 或 `product_id` 关联）
+   - `content_marketing/daily_marketing_report.py:1020 _review_context()` → 同上
+
+3. `product_reviews` 归档而不是删除：
+
+```sql
+ALTER TABLE product_reviews RENAME TO product_reviews_archived_20261101;
+```
+
+**必须明确的功能退化**：`product_reviews` 存的是好/中/差全量，差评三表只有差评。下线后：
+- 风格分析里的「评价等级分布」从三档变成只有差评一档（差评样本反而更完整：天猫 97 + 京东 32 + 抖音 14，而 `product_reviews` 里天猫是 0、京东 8 月就断了）。
+- 营销日报的「近 7 天差评归因」不受影响，反而多了天猫来源。
+- 若产品侧确实需要好评样本沉淀，需要把三个差评采集器扩成「按等级采集」，不要恢复 `product_reviews`。
+
+另外 `dy_product_scraping.py` 主流程里的评论采集动作只产出本地 JSON，`product_reviews` 停写后就没有消费方，建议同步关掉（或保留但不再进 `product.daily`）。
+
+### P2 loader 下发 style_id（1 处改动，收益最大）
+
+`src/gyxx_flow/adapters/workflow_config.py:322 merge_lifecycle_style_records()`：
+
+```python
+output["style_id"] = lifecycle.get("id")          # dim_style.id
+output["style_match_method"] = "dim_style"
+```
+
+同时 `source_record_id` 保留（控制台/账本还在用），只是不再承担传 ID 的职责。
+
+影响面：所有通过 `runtime_workflow_configs()` 取款的脚本自动获得 `style_id`，包括日报聚合、周报、ERP 全店铺、人像、主图、声明式分析。这一步不改表、不改写入逻辑，可以先合入验证 loader 输出。
+
+### P3 采集表加列 + 回填 + 写入端填充
+
+**统一 DDL 模板**（以 `daily_style_metrics` 为例，其余同构）：
+
+```sql
+ALTER TABLE daily_style_metrics ADD COLUMN IF NOT EXISTS style_id BIGINT;
+UPDATE daily_style_metrics d SET style_id = s.id
+FROM dim_style s WHERE s.style_name = d.style_name AND d.style_id IS NULL;
+CREATE INDEX IF NOT EXISTS daily_style_metrics_style_date_idx
+    ON daily_style_metrics (style_id, metric_date DESC);
+```
+
+按表处理（**必须匹配**的加 `NOT NULL`，**允许未匹配**的保持可空）：
+
+| 表 | 款式来源 | style_id | 写入端改动 |
+|---|---|---|---|
+| `daily_style_metrics` | loader 款式清单 | NOT NULL | `aggregate_daily_final.py` + `db.upsert_daily_metrics` |
+| `weekly_style_metrics` | 从 daily 聚合 | NOT NULL | `weekly_aggregate.py` |
+| `erp_all_shop_style_daily_metrics` | loader ERP 款 | NOT NULL | `backfill_erp_all_shop_daily.py` |
+| `daily_persona_metrics` | loader | NOT NULL | `collect_*_persona_to_bitable.py` |
+| `main_image_creatives` | loader | NOT NULL | `insert_main_image_records.py` / `insert_jd_main_image_records.py` |
+| `fact_style_analysis_reports` | loader | NOT NULL | `analyze_style.py` |
+| `decline_events` / `fact_alert_history` | daily 派生 | NOT NULL | `check_nine_day_decline.py` |
+| `jd_self_weekly_sales` | loader | NOT NULL | `jd_self_inventory_sales_collector.py` |
+| `persona_screenshots` | loader | NOT NULL | `taobao_dmp_item_crowd_insight_screenshots.py` |
+| `product_daily_metrics` | 平台商品 ID 反查 | **可空** | `import_product_daily.py` |
+| `jd_product_ad_daily_metrics` / `dy_product_ad_daily_metrics` | 名称/ID 反查 | **可空**（已有 `match_status`） | 两个 `import_*_ad_*.py` |
+| 差评三表 | 反查 | **可空** | `collectors/*_negative_reviews.py` 入库侧 |
+
+**`product_daily_metrics` 的回填要处理系列聚合带来的歧义**（当前 0 条歧义，但要按锚点优先写死优先级）：
+
+```sql
+UPDATE product_daily_metrics m
+SET style_id = pick.style_id
+FROM LATERAL (
+    SELECT d.id AS style_id
+    FROM dim_style d
+    WHERE (m.platform = 'tm' AND d.tm_spus @> to_jsonb(m.product_id))
+       OR (m.platform = 'jd' AND d.jd_spus @> to_jsonb(m.product_id))
+       OR (m.platform = 'dy' AND d.dy_spus @> to_jsonb(m.product_id))
+    ORDER BY d.is_series_anchor DESC, LENGTH(d.style_name), d.id
+    LIMIT 1
+) pick
+WHERE m.style_id IS NULL;
+```
+
+（真实脚本里用 Python 侧 `_resolve_style_match()` 复用同一套优先级，SQL 只用于一次性回填。）
+
+回填后预期的非空比例 = 2.2 节的实测值：**dy 47% / jd 55% / tm 48%**。这不是 bug——剩下的是 458 个未建档商品 ID，要靠 P5 的人补 spus 才能涨。
+
+### P4 约束 + 下游切 join
+
+1. ~~对「必须匹配」的表加外键~~ → **改为不建物理外键（2026-11-01 定案）**。
+
+原始设想是给 10 张「必须匹配」的表加 `FOREIGN KEY (style_id) REFERENCES dim_style(id) ON UPDATE CASCADE`，
+实施后又按业主意见拆掉了。理由是采集链路的语义与硬约束冲突：
+
+- 采集要求 **fail-soft**：一个商品 ID 没建档，正确行为是把该行 `style_id` 留 NULL 并标
+  `unmatched`，继续写其它行。物理外键会让整批 INSERT/UPDATE 直接报错，等于「一个商品
+  未建档 → 当天数据全丢」，比留 NULL 更糟。
+- 款式会动：拆系列、合并同款、重建 dim_style 行都要先卸约束再动数据，运维成本不小。
+- data-hub 侧 22 张表本来就用 `style_id` 关联、**不带物理外键**，两边保持一致才好维护。
+
+改为**逻辑外键**，完整性由三处保证：
+
+| 环节 | 手段 |
+|---|---|
+| 写入 | loader 下发 `style_id`（源自 dim_style 本身）；缺失时 `attach_style_ids` / `resolve_style_ids_by_product` 批量兜底 |
+| 校验 | `backfill_style_id.py --dry-run` 预演；`v_style_id_coverage` 报每张表覆盖率 |
+| 观测 | `v_style_unmatched_product`（日报口径）/ `v_style_unmatched_source`（全源口径）列未建档清单 |
+
+老库里早期建过的 10 个 `<表名>_style_id_fk` 由 `schema.sql` 里的 `DROP CONSTRAINT IF EXISTS`
+幂等清理；只保留 `*_style_id_idx` 索引（join 性能靠索引，不靠约束）。
+
+2. data-hub 侧把 `metric.style_name = style.name` 换成 `metric.style_id = style.id`。涉及 `ProductLifecycleMapper.xml` 第 120、155 行（`daily_style_metrics`），以及 `JOIN dim_style dim ON dim.style_name = style.name`（第 15 行）。改完后 text join 变 bigint join，是纯性能收益；`style_name` 列仍在，改错了可以立即回切。
+
+3. 加测试：
+   - `tests/.../test_daily_style_metrics_style_id.py`：写入必须带 `style_id`，且等于 loader 输出的 id
+   - `tests/.../test_style_id_coverage.py`：`v_style_id_coverage` 里「必须匹配」的表 `with_style_id == rows`
+   - 现有 `test_import_tmall_product_ads.py` 的风格（断言 SQL 形状）继续用在新增字段上
+
+### P5 数据治理（不是代码活，但决定 P3 的收益）
+
+`v_style_unmatched_product` 出来的 458 个商品 ID 需要在 dim_style 建档页补 `tm_spus/jd_spus/dy_spus`。分批推进：先补最近 90 天有销量的，历史行允许永久 NULL。
+
+## 五、风险与取舍
+
+| 风险 | 影响 | 应对 |
+|---|---|---|
+| 款名在 dim_style 被改，历史表 `style_name` 变孤儿 | 现在 join 直接丢数据，这正是要 style_id 的原因 | P3 之后 join 走 id，款名改动不再影响 |
+| 加 `NOT NULL` 时某天有未建档款写入 | 写入失败，工作流中断 | 先在写入端做「未匹配记 NULL + 打 WARN」，P5 补齐后再收紧为 NOT NULL |
+| `product_reviews` 下线后没有好评样本 | 风格分析评价分布退化 | 明确接受；需要时扩差评采集器等级，不恢复旧表 |
+| 改 PK（有人会想顺手做） | data-hub 6 处 join + 飞书脚本全回归 | **本次明确不做**，PK 保持 style_name |
+| 差评表天猫侧没有 product_id | 只能靠商品名反查款式 | 采集器补 product_id（页面有商品链接可解析），或退化为名称匹配并标记 `unmatched` |
+
+## 六、验收标准
+
+1. `v_style_id_coverage`：`daily_style_metrics`、`erp_all_shop_style_daily_metrics`、`weekly_style_metrics` 的 `with_style_id = rows`（100%）。
+2. `product_daily_metrics` 的 `style_id` 非空比例不低于 2.2 节的实测基线（47%/55%/48%），且逐月上升。
+3. `product_reviews` 无新增行；两个消费方在 `v_negative_reviews` 上跑通，风格分析与营销日报输出结构不变。
+4. data-hub 产品生命页、营销中台预算页的款式指标与改造前同月数值一致（对账 2026-10 全月）。
+
+## 七、建议执行顺序
+
+P0 → P1（下线评论链路，独立可交付）→ P2（loader 一行改动，验证输出）→ P3（分表灰度，先做 `daily_style_metrics`）→ P4（约束 + 下游切换）→ P5（补建档）。
+
+P1 与 P2/P3 没有依赖关系，可以先做 P1 拿到收益（少一条采集链路、少一张表），再从容做款式主键统一。
+
+## 八、实施记录（2026-11-01）
+
+### 已落地
+
+| 阶段 | 成果 |
+|---|---|
+| P0 | `v_style_unmatched_product`、`v_style_id_coverage` 两个观测视图已建并验证 |
+| P1 | 删 `import_product_reviews.py` / `orchestrate_review_collection.py`；移除 `product.review.orchestrate` 命令与 runtime binding；`run_collect` 不再调评论入库；建 `v_negative_reviews`；两个消费方改读视图；`product_reviews` 归档为 `product_reviews_archived_20261101`（1936 行已备份到 `var/tmp/product_reviews_archive_20261101.json`） |
+| P2 | `merge_lifecycle_style_records` 输出 `style_id` + `style_match_method`；loader 透传；实库验证 **204/204 款全部拿到 style_id** |
+| P3 | 14 张采集表加 `style_id`（+ `style_match_method`）+ 索引；db 层 `attach_style_ids` / `resolve_style_ids_by_product` 统一兜底；`backfill_style_id.py` 回填历史 |
+| P4 | ~~10 张「必须匹配」表加 `style_id → dim_style(id)` 物理外键~~ → 已按要求**全部拆除**，改逻辑外键（库里 0 个 FK，只留 14 个 `style_id` 索引）；新增 `tests/modules/product_commerce/test_style_id_identity.py`（28 条契约测试） |
+| P5 | 提供 `v_style_unmatched_product` 清单（dy 170 / jd 99 / tm 189），需业务侧补 spus |
+
+### 回填结果（实库）
+
+| 表 | 行数 | 有 style_id | 比例 |
+|---|---|---|---|
+| daily_style_metrics | 17,352 | 17,352 | 100% |
+| weekly_style_metrics | 532 | 532 | 100% |
+| erp_all_shop_style_daily_metrics | 9,399 | 9,399 | 100% |
+| daily_persona_metrics | 2,357 | 2,357 | 100% |
+| main_image_creatives | 9,094 | 9,094 | 100% |
+| fact_style_analysis_reports | 465 | 465 | 100% |
+| decline_events / fact_alert_history | 14 / 14 | 14 / 14 | 100% |
+| persona_screenshots | 25 | 25 | 100% |
+| jd_self_weekly_sales | 117 | 111 | 94.9%（6 款已不在 dim_style） |
+| tm_product_ad_daily_metrics | 11,727 | 11,108 | 94.7% |
+| jd_product_ad_daily_metrics | 33,808 | 24,707 | 73.1% |
+| dy_product_ad_daily_metrics | 18,202 | 10,875 | 59.7% |
+| product_daily_metrics | 75,568 | 37,098 | 49.1%（未建档商品 ID 所致） |
+
+### 与方案不同之处
+
+1. **天猫广告表已先行迁到独立表**：方案编写时天猫指标还写在
+   `product_daily_metrics.raw_data.tm_ad_metrics`，实施期间该迁移已完成，因此改为给
+   `tm_product_ad_daily_metrics` 加 `style_id`（并纳入 `v_style_id_coverage`），
+   并未改动 `raw_data` 节点。
+2. **写入端没有逐个脚本改**：方案原计划在 13 个写入脚本里各自填充，实际收敛到 db 层
+   的 `attach_style_ids` / `resolve_style_ids_by_product`（一次批量查询），再加 loader
+   下发。少改 13 处、不会漏，代价是 upsert 多一次按款名/商品 ID 的查询。
+3. **`dy_product_scraping.py` 默认流程的评论采集直接摘除**（仍保留 `--reviews-only`
+   显式入口），比方案建议的「保留但不再进 product.daily」更彻底。
+4. **物理外键最终没有保留**：P4 建过的 10 个 `<表名>_style_id_fk` 已全部 DROP，
+   `schema.sql` 改为「幂等清理 + 只建索引」。`style_id` 与 `dim_style.id` 是纯逻辑
+   外键关系，理由见 P4 段（fail-soft 采集语义、款式可重建、与 data-hub 对齐）。
+
+### 仍待处理
+
+- **data-hub 侧 join 未切**（`ProductLifecycleMapper.xml` 6 处仍是 `metric.style_name = style.name`）。
+  现在两边都有 id，可以随时切；切换后是 bigint join 替代 text join。
+- P5 补建档：458 个商品 ID（dy 170 / jd 99 / tm 189）写进 `dim_style.spus`。
+- 既有红灯（与本次无关）：`test_tm_brand_profiles` 的 backfill 用例、`shop_intelligence`
+  的 12 个 jd 竞店用例，改动前即失败。
+
+## 九、P5 补 spus 实施记录（2026-11-01）
+
+### 依据的权威字段
+
+补 spus 的匹配依据不是商品标题，而是 **`product_daily_metrics.product_no` —— 它存的就是款名**：
+
+| 平台 | `product_no` 填充率 | 可用性 |
+|---|---:|---|
+| 天猫 | 100%（29552/29552） | 权威 |
+| 京东 | 97.8%（16664/17044） | 权威 |
+| 抖音 | 0%（0/28972） | 只能靠标题 |
+
+这一条此前被忽略，导致「458 个未建档」被误判为只能猜标题。实际天猫/京东 136 个商品
+靠 `product_no` 就精确命中，另有 31 个靠前后缀近似命中（「蓝鹊pro」→「蓝鹊」、
+「知更大号」→「知更」、「星迹」→「星迹2」）。
+
+注意 `product_no` **不是** ERP 编码：`dim_style.erp_style_codes` 存的是 `["10510"]`
+这种数字码，两者用 `product_no = erp_style_codes` 只能对上 2 个，别混淆。
+
+### 匹配算法（四路，按优先级）
+
+1. `product_no` 精确 = `style_name` → 直接写；
+2. `product_no` 与款名互为前后缀且长度差 ≤ 3，且候选唯一 → 直接写；
+3. 标题包含完整款名（抖音只能走这条）→ 用「命中位置后的版本后缀」防串款：
+   标题「盖亚斜挎包**3**」、`dim_style` 有「盖亚斜挎3」时必须指后者而非「盖亚斜挎」；
+   若该版本款没建档，判「版本款未建档」而不是硬归到基础款；
+4. 编码与标题互相矛盾 → 进人工清单，**不自动写**。
+
+实测第 4 类里最贵的一条：天猫 `649545919334` 商品编码写「理岩」、标题写「阿波罗air」，
+支付额 **50.2 万**。猜错就是 50 万挂错款，只能人工定。
+
+### 写入结果
+
+| 项 | 数值 |
+|---|---:|
+| 写入商品 ID | 199 个 |
+| 涉及款式 | 67 个 |
+| 覆盖 `product_daily_metrics` 行 | 22,636 |
+| 未写（物料 / 冲突待定） | 87 个 |
+
+合并策略：`tm_spus / jd_spus / dy_spus` 取**并集追加**，不覆盖已有值；写前备份到
+`var/tmp/dim_style_spus_backup_20261101.json`（204 行）。
+
+### 覆盖率变化
+
+| 表 | 补前 | 补后 |
+|---|---:|---:|
+| product_daily_metrics | 49.1% | **79.0%** |
+| tm_product_ad_daily_metrics | 94.7% | **99.4%** |
+| jd_product_ad_daily_metrics | 73.1% | **92.4%** |
+| dy_product_ad_daily_metrics | 59.7% | **83.2%** |
+
+广告表的提升来自新增的第三段回填（按商品 ID + 补空 `style_name`）：京东广告里
+32 个商品 ID 就占了 9,101 行空值，补完 `style_name` 空白从 9,095 降到 2,554，
+抖音从 7,327 降到 3,062。这意味着 **data-hub 侧不改任何 SQL，按款名的 join 已经多认回
+11,359 行**。
+
+### 剩余 259 个未建档 ID 的构成
+
+| 分类 | 商品数 | 支付额 | 处置 |
+|---|---:|---:|---|
+| 待建档-真款 | 128 | 6.9 万 | 需在 dim_style 新建款（星云系列链接、拾影单肩包、幻月、纪元、极地冰冠、阿瑞斯、盖亚S1、漫游者、光/暗系列等） |
+| 待建档-ozko 线 | 20 | 0.2 万 | ozko 子品牌款（逐云、微笑系列、甜酷系列、银鸥、独角兽、HB、cdbb、761X…） |
+| 白名单-物料配件 | 30 | 5.6 万 | 不该进 spus（内胆包、数据线、风扇、防丢器…） |
+| 白名单-服务/券 | 4 | 0 | 补邮费、兑换券、补贴 |
+| 冲突待定 | 10 | 62.9 万 | 编码与标题矛盾，需人工定 |
+
+清单：`var/tmp/spus_todo.csv`（按分档 + 金额排序）。
+
+### 新增的长期机制
+
+- `backfill_style_id.py` 增加 `AD_PRODUCT_TABLES` 与 `_backfill_ad_tables_by_product`，
+  广告表按 `spu_id / product_id` 二次回填，并只在 `style_name` 为空时补权威款名
+  （`COALESCE(NULLIF(BTRIM(style_name), ''), %s)`，不覆盖平台给出的值）。
+- 新增视图 `v_style_unmatched_source`：`product_daily_metrics` + 三张广告表统一口径，
+  补建档只看这一个视图就够（实测与原 `v_style_unmatched_product` 数值一致，
+  说明没有纯投流商品只出现在广告表）。
@@ -1,89 +0,0 @@
-# 智能工作台设计（dsh-refact 分支）
-
-> 状态：v1 已实现。本文档记录「以 deepseek-harness 为 agent 底座、gyxx-flow 工作流插件化」
-> 的架构决策与边界。使用说明见 [../workbench/README.md](../workbench/README.md)。
-
-## 目标与约束
-
- **底座**：deepseek-harness（`dsh`）Web UI + DeepSeek 模型 = 智能体运行时；
-  gyxx-flow 保持唯一的工作流执行/调度事实来源（LangGraph 引擎、`RunJournal`、
-  `LockManager`、`EffectLedger` 契约不变）。
- **插件化**：不重写工作流；通过 dsh「everything-is-a-plugin」扩展点把现有
-  控制台能力投影为智能体工具与 UI 面板。
- **UI**：在原 dsh Web UI 侧边栏追加工作流入口（`sidebar.footer.action` +
-  `shell.overlay` 两个**追加型**插槽，不替换任何内置区域），选中工作流即可
-  提问 / 诊断 / 修复。
- **生产约束**（AGENTS.md）：`gyxx schedule run` 仍是唯一生产调度路径；
-  正式执行保持 `execute + confirmed` 双确认；控制台仍以 `--env-file` 注入云端凭据
-  （启动脚本默认 `D:\product-collector-analyze-flow\.env`）。
-
-## 为什么不把 gyxx-flow 改写进 dsh monorepo
-
-dsh 是 pnpm monorepo（host/client 双聚合、Typert 远程契约、自有构建链）。
-把 Python 工作流引擎迁进去既不可能也无必要。dsh 的外部插件机制
-（`--patch` cordis.yml + 绝对路径插件 + `dsh.client` 包声明）就是为这种
-「外部系统接入」设计的。我们选择**进程外集成**：
-
- gyxx-flow 控制台 HTTP API 是唯一集成面（已含脱敏、并发守卫、写操作令牌）；
- 工作台插件只是控制台的客户端 + 智能体能力注册器；
- 控制台挂了，dsh 照常可用；dsh 挂了，调度器照常跑。
-
-## 组件
-
-### 1. 控制台诊断端点（Python，`src/gyxx_flow/console.py`）
-
-| 端点 | 说明 |
-| --- | --- |
-| `GET /api/workflows/{id}` | 单工作流详情（overview 投影 + `schedule_revision`） |
-| `GET /api/workflows/{id}/runs/{run_id}` | 单次运行 + 步骤明细 |
-| `GET /api/workflows/{id}/runs/{run_id}/diagnosis` | 诊断包：运行 + journal trace + 脱敏日志尾部 |
-
-诊断包日志来源：`run.json` 的 `trace.paths.log` 目录 + 运行时间窗内的控制台日志
-（`logs/console/{workflow_id}-*.log`），每个文件限读尾部 256KB，经 `_sanitize_error`
-脱敏并截断，单响应最多 5 个文件。`run_id` 与 `workflow_id` 强制配对（404 不泄露
-跨工作流记录）。
-
-### 2. 宿主插件（`workbench/plugin/gyxx-workbench.mjs`，零运行时依赖）
-
- **7 个工具**（原始 JSON-Schema `ToolDefinition`，不 import 任何 dsh 包，
-  保证以绝对路径加载时无解析风险）；
- **系统提示词段**（`gyxx-workbench`，order 700）：中文运维契约与安全规则；
- **失败监控器**：`ctx.effect` 轮询，`workflow_id+run_id` 去重，自定义事件
-  `gyxx-workbench/alert` + 可选自动诊断会话；
- **桥接服务**（127.0.0.1:8790）：为浏览器面板代理控制台 API（控制台禁 CORS，
-  面板无法直连），并承载 `/bridge/ask` 会话创建与降级页面。
-  写操作校验 `x-gyxx-workbench: 1` + JSON content-type；CORS 仅回环来源。
-
-### 3. 侧边栏面板（`workbench/plugin/client/`）
-
- React 源码经 esbuild 打包为 CJS，平台模块（react 等）按 dsh 模块表协议外置，
-  包装为 `window.__ModuleLoader__.load({id, factory})`——与官方 tsdown 产物同协议；
- `package.json` 的 `dsh.client` 声明让 `client-modules` 扫描器（支持路径型
-  Loader 条目，向上找最近 package.json）发现并提供该浏览器包；
- 插槽：`sidebar.footer.action`（「工作流」开关 + 聚合状态点）、`shell.overlay`
-  （左抽屉面板 + 失败 toast）。均为 list 型追加插槽，不替换内置区域；
- 面板动作：提问/诊断/修复（创建带工作流上下文的智能体会话）、试运行/正式运行
-  （浏览器端 confirm）、停止、运行历史查看、单运行诊断视图。
-
-### 4. 会话创建（`/bridge/ask`）
-
-`ctx.agents.create({ sessionId, meta: { cwd, origin } })` → `agent.followup()` 注入
-带工作流上下文的首条用户消息。新会话自动出现在 dsh 会话列表（host 侧创建即入册）。
-文案模板区分 ask/diagnose/repair 三种动作。
-
-## 已知限制（v1）
-
- 会话创建后不能自动聚焦（dsh 客户端无公开的「选中会话」运行时 API）；面板以
-  toast 提示用户在会话列表中点开。
- `sidebar.workspaces` 为 single 型插槽（替换即失去会话树），故面板走 overlay
-  抽屉而非内嵌会话树；若未来 dsh 提供追加型侧栏区块插槽，可平移。
- 客户端包协议依赖 dsh 未冻结的 developer-preview 契约；升级 dsh 需回归验证
-  （README「依赖的 dsh 扩展点」一节列出了核对清单）。
- `autoDiagnose` 默认关闭：自动开会话会产生模型调用成本。
-
-## 测试
-
- Python：`tests/test_console.py` 新增 5 用例（详情/运行详情/诊断包/越权配对/HTTP 路由）。
- 插件：`workbench/plugin/tests/`（node:test）——工具注册、离线错误结构化、
-  confirmed 守卫、桥接代理。
- 端到端手工验证：`start-workbench.ps1` → 面板选中工作流 → 提问/诊断/修复/试运行。
@@ -17,6 +17,15 @@ uv run gyxx console
 uv run gyxx console --env-file D:\secure\gyxx-flow.env
 ```

+香港客服库存工作流还需要聚水潭应用密钥。Windows 生产控制台必须额外加载受保护的聚水潭环境文件，
+这样控制台发起的手动执行也能取得凭据；调度器和控制台应加载相同的环境文件集合：
+
+```powershell
+uv run gyxx console `
+  --env-file D:\product-collector-analyze-flow\.env `
+  --env-file D:\product-collector-analyze-flow\private-secrets\jushuitan.env
+```
+
 商品经营正式执行会在启动子进程前校验云端 PostgreSQL 和该工作流实际需要的运行时配置。
 需要 Hermes 分析或业务通知的工作流仍会校验 `hermes-analyzer` lark-cli user 身份和负责人
 通知路由；天猫/京东视频上传则校验 `STYLE_ANALYSIS_LLM_BASE_URL`（或
@@ -144,3 +153,14 @@ uv run gyxx console --host 0.0.0.0 --port 8765

 控制台不提供工作流入口、参数、运行时绑定或凭据的在线编辑能力；这些仍由受版本控制的
 项目配置和代码维护。
+
+## 自检与保活
+
+- `GET /api/health` 返回进程 PID、uptime 与调度器锁存活状态，不鉴权、不触外部依赖，专供探针使用。
+- 控制台进程内置看门狗线程：周期自探测 `/api/health`，连续失败（默认 3 次）即以退出码 70 退出，
+  由 `deploy/watchdog/gyxx_keepalive.py`（登录自启动常驻）负责拉起；保活同时守护调度器进程，
+  详见 `docs/deployment.md` 的「Windows 工作站保活」章节。环境开关：
+  `GYXX_CONSOLE_WATCHDOG=0` 关闭、`GYXX_CONSOLE_WATCHDOG_INTERVAL_SECONDS`（默认 30）、
+  `GYXX_CONSOLE_WATCHDOG_FAILURES`（默认 3）。
+- API 返回「控制台处理请求失败」时，完整堆栈在控制台 stderr 日志
+  （`var/logs/console-restarts/console-*.stderr.log`）中可查。
@@ -69,7 +69,7 @@
 | `08:30` | 2 | 1 | 1 | 达人月报生成 [`content.creator_report.monthly`]（暂停）；京东与天猫主图周采集 [`product.main_image.weekly`] |
 | `08:40` | 1 | 1 | 0 | 商品经营日报采集 [`product.daily`] |
 | `09:00` | 6 | 6 | 0 | 笔记清单与合作同步 [`content.notes_master.daily`]；京东自营业绩日采集 [`shop.jd_self_operated.daily`]；抖音待改价申诉；聚水潭全店铺款式日报 [`product.erp_all_shop_daily`]；天猫百亿补贴批量报名；补货建议预计到货同步 [`supply.replenishment_arrival_sync.daily`] |
-| `10:00` | 8 | 5 | 3 | 营销日报生成与发送 [`content.marketing_report.daily`]（暂停）；内容平台登录态刷新 [`content.relogin.weekly`]（暂停）；内容周度汇总 [`content.summary.weekly`]（暂停）；抖音待改价申诉；三平台电商费用采集 [`product.ecommerce_costs.daily`]；商品月度销量表 [`product.sales_sheet.daily`]；天猫百亿补贴批量报名；市场排行周采集 [`product.market_rank`] |
+| `10:00` | 8 | 5 | 3 | 营销日报生成与发送 [`content.marketing_report.daily`]（暂停）；内容周度汇总 [`content.summary.weekly`]（暂停）；抖音待改价申诉；三平台电商费用采集 [`product.ecommerce_costs.daily`]；商品月度销量表 [`product.sales_sheet.daily`]；天猫百亿补贴批量报名；市场排行周采集 [`product.market_rank`] |
 | `11:00` | 3 | 3 | 0 | 抖音待改价申诉；款式周期分析 [`product.style_analysis.interval`]；天猫百亿补贴批量报名 |
 | `12:00` | 5 | 5 | 0 | 多平台评论周采集 [`content.comments.weekly`]；店铺经营指标周采集 [`shop.metrics.weekly`]；抖音待改价申诉；商品人群画像日采集 [`product.persona.daily`]；天猫百亿补贴批量报名 |
 | `12:30` | 1 | 1 | 0 | 竞店指标周采集 [`shop.competitor.weekly`] |
@@ -98,7 +98,7 @@

 - 指标采集：`content.metrics.daily`（内容指标日报采集）、`content.metrics.backfill`（内容指标补采）。
 - 报告与汇总：`content.marketing_report.daily`（营销日报生成与发送）、`content.creator_report.monthly`（达人月报生成）、`content.summary.monthly`（内容月度汇总）、`content.summary.weekly`（内容周度汇总）。
- 内容同步与互动：`content.notes_master.daily`（笔记清单与合作同步）、`content.comments.weekly`（多平台评论周采集）、`content.relogin.weekly`（内容平台登录态刷新）。
+- 内容同步与互动：`content.notes_master.daily`（笔记清单与合作同步）、`content.comments.weekly`（多平台评论周采集）。

 ### 4.2 商品经营（`product_commerce`，12 条）

@@ -124,7 +124,6 @@
 | 1 | `content.metrics.daily` | 内容指标日报采集 | 内容营销 | 每天 `01:00`、`05:00` | 0 | 启用 | 5 |
 | 2 | `content.metrics.backfill` | 内容指标补采 | 内容营销 | 每周一 `14:00` | 0 | 暂停 | 2 |
 | 3 | `content.marketing_report.daily` | 营销日报生成与发送 | 内容营销 | 每天 `10:00` | 0 | 暂停 | 1 |
-| 4 | `content.relogin.weekly` | 内容平台登录态刷新 | 内容营销 | 每周五 `10:00` | 0 | 暂停 | 1 |
 | 5 | `content.creator_report.monthly` | 达人月报生成 | 内容营销 | 每月 1 日 `08:30` | 0 | 启用 | 1 |
 | 6 | `content.summary.monthly` | 内容月度汇总 | 内容营销 | 每月 1 日 `08:00` | 0 | 启用 | 1 |
 | 7 | `content.notes_master.daily` | 笔记清单与合作同步 | 内容营销 | 每天 `09:00` | 0 | 启用 | 1 |
@@ -0,0 +1,66 @@
+# 工作流总览（未隐藏）
+
+> 生成日期：2026-10-08。数据来源：`config/workflows.json`、`config/schedules.json`、控制台显示名（`var/state/console/workflow-display-names.json`）；数据库表提取自各入口脚本及共享 DB 辅助模块的内联 SQL。
+> 范围：控制台中未被隐藏的工作流共 **31 个**（全部 40 个中剔除浏览器端隐藏的 10 个，其中 `content.relogin.weekly` 已不在当前配置中）。
+> 定时时间均为 **Asia/Shanghai**，由常驻调度器 `gyxx schedule run` 按 `config/schedules.json` 触发；标注「未启用」的仅支持控制台手动运行。
+> 启动脚本为仓库根目录相对路径，由工作流引擎相对 `src/gyxx_flow/modules/<模块>/` 解析执行；`×N` 表示同一脚本被 N 个步骤以不同参数复用。
+>
+> **关联数据库表说明**：均指云端 PostgreSQL（`GYXX_POSTGRES_DSN`），（写）= INSERT/UPDATE/DELETE，（读）= SELECT，（建）= CREATE TABLE IF NOT EXISTS。三点公共约定：
+> - 内容营销用到款式映射的脚本会经 `feishu_mapping` → `adapters/workflow_config.py` 间接读动态配置表 `workflow_dynamic_styles`、`workflow_dynamic_style_platforms`（并合并读 `cmt_styles`，首跑建 `workflow_dynamic_*` 表），下表不逐一列出。
+> - 店铺情报各入库脚本启动时执行 `shop_intelligence/db/schema.sql`、商品电商多数入库脚本执行 `product_commerce/db/schema.sql`，幂等建该模块全量表（21 / 26 张）；下表只列该工作流**实际读写**的表。
+> - 读写 `information_schema` / `pg_catalog` 系统目录的通用校验不列出。
+
+## 内容营销（content_marketing，4 个）
+
+| 工作流名称 | ID | 功能 | 定时执行 | 启动脚本 | 关联数据库表 |
+|---|---|---|---|---|---|
+| 合作达人内容指标日报采集 | `content.metrics.daily` | 按日采集合作达人在 B站、小红书蒲公英、星图（未命中用抖店达人广场补采）的曝光互动指标并同步 PostgreSQL；05:00 档补采 01:00 未完成任务 | 每日 01:00、05:00 | `src/gyxx_flow/modules/content_marketing/run_all.py`<br>`src/gyxx_flow/modules/content_marketing/data/tools/sync_metrics_to_cmt_notes.py` | `cmt_note_metric_snapshots`（写）<br>`cmt_notes_master`（读写）<br>`cmt_creators`（读写）<br>`cmt_styles`（读） |
+| 自营内容指标日报采集 | `content.metrics.self_operated.daily` | 刷新自营款式映射，采集自营 B站、蝉妈妈、小红书 ARK 指标；飞书曝光回填并将三平台全量内容指标按日快照入库 | 每日 06:00 | `…/content_marketing/data/tools/refresh_self_mapping.py`<br>`…/content_marketing/self_bilibili_scraper.py`<br>`…/content_marketing/chanmama_scraper.py`<br>`…/content_marketing/self_xhs_scraper.py`<br>`…/content_marketing/data/tools/sync_metrics_to_cmt_notes.py`<br>`…/content_marketing/data/tools/sync_self_video_metrics.py` | `cmt_self_video_metrics`（写）<br>`cmt_note_metric_snapshots`（写）<br>`cmt_notes_master`（读写）<br>`cmt_creators`（读写）<br>`cmt_styles`（读） |
+| 笔记清单与合作同步 | `content.notes_master.daily` | 一次飞书拉取同时落笔记身份与合作字段；笔记数量按去重链接口径维护主表 | 每日 09:00 | `src/gyxx_flow/modules/content_marketing/data/tools/sync_notes_master.py` | `cmt_notes_master`（建+写）<br>`cmt_note_metric_snapshots`（建）<br>`cmt_styles`（读写）<br>`cmt_creators`（读写） |
+| 多平台评论周采集 | `content.comments.weekly` | B站、小红书、抖音三平台评论并行采集，任一平台失败则工作流失败 | 每周日 12:00 | `…/content_marketing/data/tools/batch_rescrape_bilibili.py`<br>`…/content_marketing/data/tools/batch_rescrape_xiaohongshu.py`<br>`…/content_marketing/data/tools/batch_rescrape_douyin.py` | `cmt_comments`（写）<br>`cmt_notes_master`（读写） |
+
+## 店铺情报（shop_intelligence，12 个）
+
+| 工作流名称 | ID | 功能 | 定时执行 | 启动脚本 | 关联数据库表 |
+|---|---|---|---|---|---|
+| 店铺经营指标周采集 | `shop.metrics.weekly` | 三平台作为独立节点并行采集店铺经营指标并写 PostgreSQL（不写飞书 Base） | 每周一 12:00 | `src/gyxx_flow/modules/shop_intelligence/runners/run_shop.py` ×3 | `jd_shop_weekly`、`dy_shop_weekly`、`tm_shop_weekly`（各平台步骤分别写） |
+| 竞店指标周采集 | `shop.competitor.weekly` | 京东、抖音并行采集竞店指标并入库，单平台失败不阻断另一平台 | 每周一 12:30 | `src/gyxx_flow/modules/shop_intelligence/runners/run_peer_store.py` ×2 | `jd_peer_store_ranking`、`dy_peer_store_ranking`（各平台步骤分别写） |
+| 京东自营业绩日采集 | `shop.jd_self_operated.daily` | 采集京东自营品牌与商品业绩（业务日期 = 运行日 −2） | 每日 09:00 | `…/shop_intelligence/collectors/jd_self_operated_brand_daily.py`<br>`…/shop_intelligence/collectors/jd_self_operated_product_daily.py` | `jd_self_operated_brand_daily`（写）<br>`jd_self_operated_product_daily`（写） |
+| 京东客服消息采集 | `shop.jd_customer_service.daily` | 按业务日在商智消息视图采集顾客会话，单日须完成全部顾客，失败保留进度可重试 | 每月 21 日 03:00 | `src/gyxx_flow/modules/shop_intelligence/collectors/jd_customer_service_messages.py` | `jd_customer_service_conversations`、`jd_customer_service_messages`（写） |
+| 抖音客服消息采集 | `shop.douyin_customer_service.daily` | 按业务日采集抖音客服会话，失败保留进度可安全重试 | 每月 21 日 03:20 | `src/gyxx_flow/modules/shop_intelligence/collectors/dy_customer_service_messages.py` | `dy_customer_service_conversations`、`dy_customer_service_messages`、`dy_customer_service_conversation_orders`（写）<br>`dy_customer_service_products`（读写） |
+| 京东差评采集 | `shop.jd_negative_reviews.daily` | 按店铺选择日期范围逐日采集京东负面评价，按评价 ID 幂等 | 每月 22 日 03:00 | `src/gyxx_flow/modules/shop_intelligence/collectors/jd_negative_reviews.py` | `jd_negative_reviews`（写） |
+| 抖音差评采集 | `shop.dy_negative_reviews.daily` | 逐日采集抖音负面评价（评价历史仅保留近 30 天），按订单 ID 幂等 | 每月 22 日 03:20 | `src/gyxx_flow/modules/shop_intelligence/collectors/dy_negative_reviews.py` | `dy_negative_reviews`（写） |
+| 京东品退问题采集 | `shop.jd_after_sales_issues.daily` | 在商智 31 天限制下按连续区间采集售后服务单，按申请日期写业务日，幂等更新 | 每月 23 日 03:00 | `src/gyxx_flow/modules/shop_intelligence/collectors/jd_after_sales_issues.py` | `jd_after_sales_issues`（写，含范围外残留 DELETE 清理） |
+| 抖音品退采集 | `shop.douyin_after_sales_quality.daily` | 采集抖音售后单，按申请时间归属业务日，账号 + 售后编号幂等更新 | 每月 23 日 03:20 | `src/gyxx_flow/modules/shop_intelligence/collectors/dy_after_sales_quality.py` | `dy_after_sales_quality_orders`、`dy_after_sales_quality_negotiations`（写） |
+| 天猫客服消息采集 | `shop.tmall_customer_service.daily` | 按每次最多 31 天的窗口顺序采集天猫客服聊天记录，失败窗口保留数据 | 每月 24 日 03:40（未启用，仅手动） | `src/gyxx_flow/modules/shop_intelligence/collectors/tmall_customer_service_messages.py` | `tmall_customer_service_conversations`、`tmall_customer_service_messages`（写） |
+| 天猫差评采集 | `shop.tmall_negative_reviews.daily` | 固定筛选负面评价并按评价时间范围采集后入库 | 每月 25 日 03:50（未启用，仅手动） | `src/gyxx_flow/modules/shop_intelligence/collectors/tmall_negative_reviews.py` | `tmall_negative_reviews`（写） |
+| 抖音待改价申诉 | `shop.douyin_price_appeal` | 每个时点重新扫描可申诉并发起价格申诉；已申诉商品不重复出现 | 每日每小时整点（00:00–23:00） | `src/gyxx_flow/modules/shop_intelligence/collectors/dy_store_competitor_store_scraping.py` | 无（浏览器采集 + 本地 JSON） |
+
+## 商品电商（product_commerce，13 个）
+
+| 工作流名称 | ID | 功能 | 定时执行 | 启动脚本 | 关联数据库表 |
+|---|---|---|---|---|---|
+| 商品人群画像周采集 | `product.persona.daily` | 天猫、抖音、京东人群画像并行采集并写入对应飞书多维表 | 每周二 12:00 | `src/gyxx_flow/modules/product_commerce/run_daily_persona.py` | `daily_persona_metrics`（写） |
+| 商品经营日报采集 | `product.daily` | 按业务日期（运行日 −1）采集 ERP 与三平台数据，幂等写 PostgreSQL，汇总分析后导出并写飞书日报 | 每日 08:40 | `src/gyxx_flow/modules/product_commerce/orchestrate_daily_collection.py` | 写：`product_reviews`、`product_daily_metrics`、`daily_style_metrics`、`fact_style_analysis_reports`、`decline_events`、`fact_alert_history`<br>读：`weekly_style_metrics`、`daily_persona_metrics`、`main_image_creatives`、`cmt_styles`、`cmt_notes_master`、`cmt_comments`、`cmt_creators`（及上述写表） |
+| 私域库存预警 | `product.online_stock_alert.daily` | 查询线上渠道全部商品，库存低于 3 时汇总 SKU 与库存发飞书预警 | 每日 09:30 | `src/gyxx_flow/modules/product_commerce/online_stock_alert.py` | 无（企业数据交互 HTTP API + 飞书） |
+| 私域商品兑换排行 | `product.points_exchange.weekly` | 按自然周查询私域积分兑换明细，按商品汇总 `quantity` 兑换数量并从高到低排名，以飞书卡片发送周报 | 每周一 09:45（统计上一个完整自然周） | `src/gyxx_flow/modules/product_commerce/points_exchange_weekly.py` | 无（企业数据交互 HTTP API + 飞书） |
+| 三平台电商费用采集 | `product.ecommerce_costs.daily` | 并行采集天猫万相台、京东、抖音千川业务日前一天的费用，各平台先下载再导入，单平台失败不阻断其他 | 每日 10:00 | `…/product_commerce/collect_tmall_wanxiang_ads.py`<br>`…/product_commerce/import_tmall_product_ads.py`<br>`…/product_commerce/collect_jd_ad_costs.py`<br>`…/product_commerce/import_jd_ad_costs.py`<br>`…/product_commerce/collect_douyin_qianchuan_ads.py`<br>`…/product_commerce/import_douyin_qianchuan_ads.py` | `product_daily_metrics`（写，天猫原始费用并入 raw_data JSONB）<br>`jd_product_ad_daily_metrics`、`dy_product_ad_daily_metrics`（建+写）<br>`dim_style`（读） |
+| 京东与天猫主图周采集 | `product.main_image.weekly` | 京东与天猫万相台主图并行采集，各自完成 PG 入库和飞书插入，互不影响 | 每周日 08:30 | `…/product_commerce/run_weekly_jd_main_image.py`<br>`…/product_commerce/run_weekly_main_image.py` | `main_image_creatives`（写） |
+| 飞书商品月度销量表填写 | `product.sales_sheet.daily` | 抓取并校验全部 ERP 报表后填写飞书月度销量表，同名月份页幂等覆盖，无 ERP 编码产品按 0 写入 | 每月 1 日 10:00 | `src/gyxx_flow/modules/product_commerce/sync_monthly_sales_sheet.py` | 无（聚水潭浏览器采集 + 飞书写入） |
+| 聚水潭全店铺款式日报 | `product.erp_all_shop_daily` | 聚水潭商品主题分析全选全部店铺，采集业务日前一天所有 ERP 款式的自定义销量与退货量并幂等写 PostgreSQL | 每日 09:00 | `src/gyxx_flow/modules/product_commerce/backfill_erp_all_shop_daily.py` | `erp_all_shop_style_daily_metrics`（读写） |
+| 客服香港库存工作流 | `product.hk_anpei_public_inventory.daily` | 校验计划书 SKU 与库存表商品编码对应关系，调聚水潭库存接口查询并将公有可用数写入仓库现有库存列 | 每日 09:00 | `src/gyxx_flow/modules/product_commerce/public_sku_inventory_sheet.py` | 无（聚水潭 API + 飞书表格） |
+| 天猫视频上传 | `product.video_upload` | 按店铺规则扫描飞书视频附件，凭附件级发布账本防重，仅在存在可领取附件时登录光合账号上传 | 每 3 天 14:00 | `src/gyxx_flow/modules/product_commerce/upload_video_to_guanghe.py` | `fact_video_publications`、`fact_video_publication_attempts`（建+读写，附件级发布账本）<br>`dim_style`、`product_daily_metrics`（读） |
+| 京东视频上传 | `product.jd_video_upload` | 从凌云air基线扫描到视图末尾识别新增视频，以 record_id + file_token 幂等发布到京东 | 每 3 天 14:00 | `src/gyxx_flow/modules/product_commerce/upload_video_to_jd.py` | `fact_video_publications`、`fact_video_publication_attempts`（建+读写，发布账本）<br>`dim_style`、`product_daily_metrics`（读） |
+| 天猫百亿补贴批量报名 | `product.tmall_baibu_apply` | 每次从飞书下载报名模板逐入口提交；默认仅导入 3 个不带超链入口（含老天猫全渠道 3C 后置完善草稿），可切换全部 6 个入口 | 每日每小时整点（00:00–23:00） | `…/product_commerce/tmall_baibu_apply_old_chain_bag.py`<br>`…/product_commerce/tmall_baibu_apply_old_all_bag.py`<br>`…/product_commerce/tmall_baibu_apply_old_all_3c.py`<br>`…/product_commerce/tmall_baibu_apply_old_chain_3c.py`<br>`…/product_commerce/tmall_baibu_apply_new_chain_bag.py`<br>`…/product_commerce/tmall_baibu_apply_new_all_bag.py` | 无（浏览器操作 + 飞书） |
+| 三平台市场排行采集（用于行军图） | `product.market_rank` | 并行采集天猫、京东、抖音市场排行，按月份写入品牌行军图数据库并复用已有人工分类 | 每月 1 日 10:00 | `src/gyxx_flow/modules/product_commerce/orchestrate_market_rank_collection.py` | `market_product`、`market_product_classification`（读写）<br>`fact_market_product_snapshot`（写，重跑先 DELETE）<br>`fact_market_rank_reports`、`fact_market_rank_notifications`（建+读写）<br>`dim_market_style`、`dim_market_scene`（读） |
+
+## 供应链（supply_chain，2 个）
+
+| 工作流名称 | ID | 功能 | 定时执行 | 启动脚本 | 关联数据库表 |
+|---|---|---|---|---|---|
+| 补货预警日处理 | `supply.replenishment_alert.daily` | 依据配置 SKU 与库存阈值生成预警并写 PostgreSQL，仅存在预警时发送业务通知 | 每日 07:00 | `src/gyxx_flow/modules/supply_chain/run.py` | `fact_threshold_alert`（读写）<br>`dim_product`（写）<br>`workflow_runs`（读写，运行账本）<br>`bitable_insert_log`（读写，通知去重账本） |
+| 同步聚水潭到货时间 | `supply.replenishment_arrival_sync.daily` | 读取到货同步待处理队列，按商品编码在聚水潭采购单页面更新预计到货时间并回写状态，失败行下次自动重试 | 每日 09:00 | `src/gyxx_flow/modules/supply_chain/orchestrator/scripts/collect_replenishment_arrival_sync.py` | `erp_replenishment_arrival_sync`（读写，队列表由 ERP 应用写入） |
+
+## 附：已隐藏的工作流（未收录）
+
+浏览器端隐藏的 10 个 ID：`content.creator_report.monthly`、`content.marketing_report.daily`、`content.metrics.backfill`、`content.relogin.weekly`（配置中已不存在）、`content.summary.monthly`、`content.summary.weekly`、`product.alert.daily`、`product.style_analysis.interval`、`supply.purchase_confirmation.daily`、`supply.replenishment.weekly`。
@@ -223,6 +223,8 @@ class AccountKeepaliveManager:
        self._next_due: dict[str, float] = {}
        self._locks = LockManager(settings.data_root / "state" / "locks")
        self._login_runner = login_runner or self._run_configured_login
+        self._audit_store: Any | None = None
+        self._audit_seen: dict[str, tuple[str, float]] = {}

    def tick(self) -> list[dict[str, object]]:
        """Run due account refreshes once; never starts work in dry-run mode."""
@@ -479,6 +481,9 @@ class AccountKeepaliveManager:
            real_chrome=account.keepalive_real_chrome,
            retries=1,
            persist_state_on_close=False,
+            # 目标站点均为国内站，代理出口反被部分 WAF 拉黑(如蝉妈妈
+            # TLS reset)，账号保活一律直连。
+            extra_flags=["--no-proxy-server"],
        )
        try:
            browser.start()
@@ -583,8 +588,63 @@ class AccountKeepaliveManager:
        if result.get("status") == "success":
            payload["last_success_at"] = timestamp
        atomic_write_json(path, payload)
+        self._audit_keepalive(account, result)
        return dict(result)

+    _AUDIT_STATE_BY_PROBE = {
+        "success": "healthy",
+        "expired": "expired",
+        "not_logged_in": "not_logged_in",
+        "busy": "busy",
+        "disabled": "disabled",
+    }
+
+    def _audit_keepalive(self, account: Any, result: dict[str, object]) -> None:
+        """Record probe outcomes so the real cookie lifetime becomes measurable.
+
+        The resident scheduler probes every ``keepalive_interval_seconds``
+        (10 minutes by default), which is what turns "logins seem to last about
+        a week" into a measured distribution.  Repeated identical outcomes
+        within an hour collapse into one row, so the event table tracks state
+        changes rather than polling noise.
+        """
+
+        try:
+            from gyxx_flow.ops.audit import snapshot_for_account
+
+            probe = str(result.get("status", "") or "error")
+            state = self._AUDIT_STATE_BY_PROBE.get(probe, "probe_error")
+            key = str(account.account_id)
+            previous = self._audit_seen.get(key)
+            if previous is not None:
+                previous_state, previous_at = previous
+                if previous_state == state and (self._clock() - previous_at) < 3600:
+                    return
+            store = self._audit()
+            if store is None or not store.enabled:
+                return
+            if store.record_cookie_event(
+                account_id=key,
+                state=state,
+                snapshot=snapshot_for_account(account),
+                probe_status=probe,
+                probe_source="keepalive",
+                detail=str(result.get("error_type", "") or ""),
+            ):
+                self._audit_seen[key] = (state, self._clock())
+        except Exception:  # noqa: BLE001 - telemetry must never break keepalive
+            return
+
+    def _audit(self) -> Any | None:
+        if self._audit_store is None:
+            try:
+                from gyxx_flow.ops.audit import AuditStore
+
+                self._audit_store = AuditStore(self.settings.data_root)
+            except Exception:  # noqa: BLE001
+                return None
+        return self._audit_store
+

 def _vault_cookie_valid(account: Any, store: BrowserCookieStore) -> bool:
    if not account.cookie_file.exists():
@@ -761,6 +821,8 @@ def login_account(
            "--remote-debugging-address=127.0.0.1",
            "--disable-session-crashed-bubble",
            "--disable-features=InfiniteSessionRestore",
+            # 国内站直连，避免代理出口被 WAF 拉黑导致登录/探测失败
+            "--no-proxy-server",
        ],
    )
    try:
@@ -168,3 +168,22 @@ def looks_like_login_url(url: str) -> bool:
    path = parsed.path.casefold()
    markers = ("login", "passport", "signin", "sign-in")
    return any(marker in hostname or marker in path for marker in markers)
+
+
+class CookieExpiredError(RuntimeError):
+    """A platform login state expired and no automatic recovery exists.
+
+    Collectors raise this instead of waiting for a human QR/SMS login: the
+    account cookie gate has already blocked unverified accounts before the
+    run, so reaching this point means the session died mid-run.  The CLI maps
+    it to ``COOKIE_SKIP_EXIT_CODE`` so the engine records a cookie skip.
+    """
+
+
+def cookie_expired_message(account_id: str, platform: str) -> str:
+    """One standard, operator-actionable expiry message across platforms."""
+
+    return (
+        f"{platform} 登录态已失效：请执行 `uv run gyxx accounts login {account_id}` "
+        "完成扫码/验证码重新登录（或等待飞书通知后处理）"
+    )
@@ -6,16 +6,20 @@ import hashlib
 import hmac
 import json
 import os
+import re
 import time
 from collections.abc import Mapping
+from datetime import datetime
 from typing import Any
 from urllib.error import HTTPError, URLError
 from urllib.parse import urlsplit
 from urllib.request import Request, urlopen

 ITEMS_QUERY_URI = "/open/data/v1/items/query"
+POINTS_EXCHANGE_QUERY_URI = "/open/data/v1/points-exchange/query"
 DEFAULT_DATA_INTERACTION_BASE_URL = "https://test.gyxx.cn/api"
 _MAX_RESPONSE_BYTES = 16 * 1024 * 1024
+_API_DATETIME = re.compile(r"^\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}$")


 class DataInteractionError(RuntimeError):
@@ -33,6 +37,71 @@ def query_items(
 ) -> list[dict[str, Any]]:
    """Query the item list and return every item row from the API response."""

+    return _signed_query(
+        ITEMS_QUERY_URI,
+        {"channel": channel, "keyword": keyword},
+        label="items query",
+        base_url=base_url,
+        app_key=app_key,
+        app_secret=app_secret,
+        timeout=timeout,
+    )
+
+
+def query_points_exchange(
+    *,
+    start_time: str | datetime,
+    end_time: str | datetime,
+    base_url: str | None = None,
+    app_key: str | None = None,
+    app_secret: str | None = None,
+    timeout: float = 60,
+) -> list[dict[str, Any]]:
+    """Query points-exchange consumption rows for one closed time window."""
+
+    start = _api_datetime(start_time, field="startTime")
+    end = _api_datetime(end_time, field="endTime")
+    if start > end:
+        raise DataInteractionError("points exchange window start is after end")
+    return _signed_query(
+        POINTS_EXCHANGE_QUERY_URI,
+        {"startTime": start, "endTime": end},
+        label="points exchange query",
+        base_url=base_url,
+        app_key=app_key,
+        app_secret=app_secret,
+        timeout=timeout,
+    )
+
+
+def _api_datetime(value: str | datetime, *, field: str) -> str:
+    if isinstance(value, datetime):
+        if value.tzinfo is not None:
+            raise DataInteractionError(
+                f"{field} must be a naive Asia/Shanghai datetime"
+            )
+        return value.strftime("%Y-%m-%d %H:%M:%S")
+    if not isinstance(value, str) or not _API_DATETIME.fullmatch(value.strip()):
+        raise DataInteractionError(
+            f"{field} must use the 'YYYY-MM-DD HH:MM:SS' format"
+        )
+    try:
+        datetime.strptime(value.strip(), "%Y-%m-%d %H:%M:%S")
+    except ValueError as exc:
+        raise DataInteractionError(f"{field} is not a valid datetime") from exc
+    return value.strip()
+
+
+def _signed_query(
+    uri: str,
+    payload: Mapping[str, Any],
+    *,
+    label: str,
+    base_url: str | None,
+    app_key: str | None,
+    app_secret: str | None,
+    timeout: float,
+) -> list[dict[str, Any]]:
    secret = app_secret or os.environ.get("GYXX_DATA_INTERACTION_APP_SECRET", "")
    if not secret.strip():
        raise DataInteractionError(
@@ -50,19 +119,19 @@ def query_items(
        raise DataInteractionError("GYXX_DATA_INTERACTION_BASE_URL must be an HTTP(S) URL")

    body = json.dumps(
-        {"channel": channel, "keyword": keyword},
+        dict(payload),
        ensure_ascii=False,
        separators=(",", ":"),
    ).encode("utf-8")
    timestamp = str(int(time.time() * 1000))
    body_sha = hashlib.sha256(body).hexdigest()
-    canonical = f"POST\n{ITEMS_QUERY_URI}\n{timestamp}\n{body_sha}"
+    canonical = f"POST\n{uri}\n{timestamp}\n{body_sha}"
    signature = hmac.new(
        secret.encode("utf-8"), canonical.encode("utf-8"), hashlib.sha256
    ).hexdigest()

    request = Request(
-        base.rstrip("/") + ITEMS_QUERY_URI,
+        base.rstrip("/") + uri,
        data=body,
        headers={
            "Content-Type": "application/json; charset=utf-8",
@@ -78,31 +147,31 @@ def query_items(
    except HTTPError as exc:
        detail = exc.read(2_000).decode("utf-8", errors="replace").strip()
        suffix = f": {detail}" if detail else ""
-        raise DataInteractionError(f"items query returned HTTP {exc.code}{suffix}") from exc
+        raise DataInteractionError(f"{label} returned HTTP {exc.code}{suffix}") from exc
    except URLError as exc:
-        raise DataInteractionError(f"items query request failed: {exc.reason}") from exc
+        raise DataInteractionError(f"{label} request failed: {exc.reason}") from exc

    if len(raw) > _MAX_RESPONSE_BYTES:
-        raise DataInteractionError("items query response exceeds the size limit")
+        raise DataInteractionError(f"{label} response exceeds the size limit")
    try:
        payload = json.loads(raw.decode("utf-8"))
    except (UnicodeDecodeError, json.JSONDecodeError) as exc:
-        raise DataInteractionError("items query returned invalid JSON") from exc
+        raise DataInteractionError(f"{label} returned invalid JSON") from exc

-    return _extract_item_rows(payload)
+    return _extract_rows(payload, label=label)


-def _extract_item_rows(payload: object) -> list[dict[str, Any]]:
+def _extract_rows(payload: object, *, label: str) -> list[dict[str, Any]]:
    if isinstance(payload, Mapping):
        code = payload.get("code")
        if code not in (None, 0, 200, "0", "200") and payload.get("success") is not True:
            message = str(payload.get("message") or payload.get("msg") or "")[:300]
            suffix = f": {message}" if message else ""
-            raise DataInteractionError(f"items query API reported code {code}{suffix}")
+            raise DataInteractionError(f"{label} API reported code {code}{suffix}")
        if payload.get("success") is False:
            message = str(payload.get("message") or payload.get("msg") or "")[:300]
            suffix = f": {message}" if message else ""
-            raise DataInteractionError(f"items query API reported failure{suffix}")
+            raise DataInteractionError(f"{label} API reported failure{suffix}")
        data: object = payload.get("data", payload.get("result"))
    else:
        data = payload
@@ -114,9 +183,9 @@ def _extract_item_rows(payload: object) -> list[dict[str, Any]]:
                data = candidate
                break
    if not isinstance(data, list):
-        raise DataInteractionError("items query response does not contain an item array")
+        raise DataInteractionError(f"{label} response does not contain an item array")
    if not all(isinstance(row, Mapping) for row in data):
-        raise DataInteractionError("items query response contains a non-object item")
+        raise DataInteractionError(f"{label} response contains a non-object item")
    return [dict(row) for row in data]


@@ -124,5 +193,7 @@ __all__ = [
    "DEFAULT_DATA_INTERACTION_BASE_URL",
    "DataInteractionError",
    "ITEMS_QUERY_URI",
+    "POINTS_EXCHANGE_QUERY_URI",
    "query_items",
+    "query_points_exchange",
 ]
@@ -766,6 +766,11 @@ class RuntimeIntegrationCatalog:
                        f"{binding.profile_dir.name}-luggage"
                    )
                ),
+                "GUANGHE_NEW_BAG_USER_DATA_DIR": str(
+                    binding.profile_dir.with_name(
+                        f"{binding.profile_dir.name}-new-bag"
+                    )
+                ),
                "DY_COOKIES_FILE": str(binding.cookie_file),
                "DY_STORAGE_STATE_FILE": str(binding.storage_state_file),
            }
@@ -1210,29 +1215,49 @@ def _merge_storage_state(


 def _account_cookie_valid(account: AccountBinding, cookies: object) -> bool:
-    """A vault cookie set is valid when a live account cookie is present."""
+    """A vault cookie set is valid when a live account cookie is present.
+
+    With ``required_cookie_names`` configured, every required name must be
+    present as a live, domain-matched cookie; a single anonymous cookie that
+    happens to share a required name (e.g. ``a1``) must not validate a guest
+    session.  Without required names, any live cookie on a required domain
+    keeps accepting the set.
+    """
    if not isinstance(cookies, list):
        return False
    required_names = account.required_cookie_names
    required_domains = account.required_cookie_domains
-    for item in cookies:
-        if not isinstance(item, dict):
-            continue
-        if _cookie_expired(item):
-            continue
-        if not str(item.get("value", "")):
-            continue
-        name = str(item.get("name", ""))
-        domain = str(item.get("domain", "")).casefold().lstrip(".")
-        name_matches = not required_names or name in required_names
-        domain_matches = not required_domains or any(
+
+    def _domain_matches(domain: str) -> bool:
+        return not required_domains or any(
            domain == required.casefold()
            or domain.endswith("." + required.casefold())
            for required in required_domains
        )
-        if name_matches and domain_matches:
-            return True
-    return False
+
+    def _live(item: object) -> bool:
+        return (
+            isinstance(item, dict)
+            and not _cookie_expired(item)
+            and bool(str(item.get("value", "")))
+        )
+
+    if not required_names:
+        return any(
+            _live(item)
+            and _domain_matches(
+                str(item.get("domain", "")).casefold().lstrip(".")
+            )
+            for item in cookies
+        )
+
+    outstanding = set(required_names)
+    for item in cookies:
+        if not _live(item):
+            continue
+        if _domain_matches(str(item.get("domain", "")).casefold().lstrip(".")):
+            outstanding.discard(str(item.get("name", "")))
+    return not outstanding


 def _cookie_expired(cookie: Mapping[str, object], *, now: float | None = None) -> bool:
@@ -0,0 +1,338 @@
+"""Minimal, signed Jushuitan inventory-query client for workflow adapters."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import os
+import time
+import uuid
+from collections.abc import Mapping
+from dataclasses import dataclass
+from decimal import Decimal, InvalidOperation
+from typing import Any
+from urllib.parse import urlparse
+
+import requests
+
+_TOKEN_PATH = "/openWeb/auth/getInitToken"
+_INVENTORY_PATH = "/open/inventory/query"
+_RATE_LIMIT_CODES = {199, 200}
+_ACCESS_TOKEN_INVALID_CODE = 100
+_MAX_RESPONSE_BYTES = 2 * 1024 * 1024
+_MAX_PAGES = 100
+_REQUEST_INTERVAL_SECONDS = 1.1
+
+
+class JushuitanInventoryError(RuntimeError):
+    """Safe-to-log error raised for Jushuitan query or response failures."""
+
+
+@dataclass(frozen=True, slots=True)
+class InventoryRecord:
+    sku: str
+    qty: int
+    order_lock: int
+    lock_qty: int
+    virtual_qty: int
+
+    @property
+    def public_available_qty(self) -> int:
+        return self.qty - self.order_lock - self.lock_qty + self.virtual_qty
+
+
+def _sign(app_secret: str, params: Mapping[str, object]) -> str:
+    source = app_secret + "".join(
+        f"{key}{value}"
+        for key, value in sorted(params.items())
+        if key.lower() != "sign" and value is not None and str(value) != ""
+    )
+    return hashlib.md5(source.encode("utf-8")).hexdigest()
+
+
+def _response_code(value: object) -> int:
+    if isinstance(value, bool):
+        raise JushuitanInventoryError("Jushuitan response has an invalid code")
+    try:
+        return int(value)
+    except (TypeError, ValueError) as exc:
+        raise JushuitanInventoryError(
+            "Jushuitan response has an invalid code"
+        ) from exc
+
+
+def _quantity(value: object, *, field: str, sku: str) -> int:
+    if value is None:
+        return 0
+    if isinstance(value, bool):
+        raise JushuitanInventoryError(
+            f"Jushuitan inventory has an invalid {field} for SKU {sku}"
+        )
+    try:
+        parsed = Decimal(str(value).strip())
+    except (InvalidOperation, ValueError) as exc:
+        raise JushuitanInventoryError(
+            f"Jushuitan inventory has an invalid {field} for SKU {sku}"
+        ) from exc
+    if not parsed.is_finite() or parsed != parsed.to_integral_value():
+        raise JushuitanInventoryError(
+            f"Jushuitan inventory has a non-integer {field} for SKU {sku}"
+        )
+    return int(parsed)
+
+
+def _value(item: Mapping[str, Any], snake_name: str, camel_name: str) -> object:
+    if snake_name in item:
+        return item[snake_name]
+    return item.get(camel_name)
+
+
+class JushuitanInventoryClient:
+    """Signed form client matching data-hub's Jushuitan API contract."""
+
+    def __init__(
+        self,
+        *,
+        app_key: str | None = None,
+        app_secret: str | None = None,
+        base_url: str | None = None,
+        timeout: tuple[float, float] = (10.0, 45.0),
+    ) -> None:
+        self.app_key = (app_key or os.environ.get("JUSHUITAN_APP_KEY", "")).strip()
+        self.app_secret = (
+            app_secret or os.environ.get("JUSHUITAN_APP_SECRET", "")
+        ).strip()
+        configured_url = (
+            base_url
+            or os.environ.get("JUSHUITAN_BASE_URL", "")
+            or "https://openapi.jushuitan.com"
+        ).strip()
+        parsed_url = urlparse(configured_url)
+        if (
+            parsed_url.scheme.lower() != "https"
+            or not parsed_url.netloc
+            or parsed_url.username
+            or parsed_url.password
+            or parsed_url.query
+            or parsed_url.fragment
+        ):
+            raise JushuitanInventoryError(
+                "JUSHUITAN_BASE_URL must be an HTTPS base URL without credentials"
+            )
+        if not self.app_key or not self.app_secret:
+            raise JushuitanInventoryError(
+                "JUSHUITAN_APP_KEY and JUSHUITAN_APP_SECRET must be injected "
+                "into the scheduler environment"
+            )
+        self.base_url = configured_url.rstrip("/")
+        self.timeout = timeout
+        self.session = requests.Session()
+        self._access_token = ""
+        self._last_request_at = 0.0
+
+    def query(self, skus: list[str]) -> dict[str, InventoryRecord]:
+        normalized_skus = list(dict.fromkeys(sku.strip() for sku in skus if sku.strip()))
+        if not normalized_skus:
+            raise JushuitanInventoryError("No SKU values were supplied")
+        if len(normalized_skus) > 100:
+            raise JushuitanInventoryError(
+                "A single inventory query may include at most 100 SKUs"
+            )
+
+        self._access_token = self._request_access_token()
+        page_index = 1
+        page_count = 1
+        records: dict[str, InventoryRecord] = {}
+        requested = set(normalized_skus)
+        while page_index <= page_count:
+            if page_index > _MAX_PAGES:
+                raise JushuitanInventoryError(
+                    "Jushuitan inventory response exceeded the page safety limit"
+                )
+            response = self._query_page(normalized_skus, page_index)
+            data = response.get("data")
+            if not isinstance(data, Mapping):
+                raise JushuitanInventoryError(
+                    "Jushuitan inventory response is missing its data object"
+                )
+            raw_rows = data.get("inventorys")
+            if raw_rows is None:
+                raw_rows = data.get("datas")
+            if raw_rows is None:
+                raw_rows = []
+            if not isinstance(raw_rows, list):
+                raise JushuitanInventoryError(
+                    "Jushuitan inventory response has an invalid inventory list"
+                )
+            for index, raw_item in enumerate(raw_rows, start=1):
+                if raw_item is None:
+                    continue
+                if not isinstance(raw_item, Mapping):
+                    raise JushuitanInventoryError(
+                        f"Jushuitan inventory row {index} is invalid"
+                    )
+                sku_value = _value(raw_item, "sku_id", "skuId")
+                sku = str(sku_value or "").strip()
+                if not sku:
+                    raise JushuitanInventoryError(
+                        f"Jushuitan inventory row {index} has no SKU"
+                    )
+                if sku not in requested:
+                    raise JushuitanInventoryError(
+                        "Jushuitan returned a SKU that was not requested"
+                    )
+                record = InventoryRecord(
+                    sku=sku,
+                    qty=_quantity(
+                        _value(raw_item, "qty", "qty"), field="qty", sku=sku
+                    ),
+                    order_lock=_quantity(
+                        _value(raw_item, "order_lock", "orderLock"),
+                        field="order_lock",
+                        sku=sku,
+                    ),
+                    lock_qty=_quantity(
+                        _value(raw_item, "lock_qty", "lockQty"),
+                        field="lock_qty",
+                        sku=sku,
+                    ),
+                    virtual_qty=_quantity(
+                        _value(raw_item, "virtual_qty", "virtualQty"),
+                        field="virtual_qty",
+                        sku=sku,
+                    ),
+                )
+                if sku in records:
+                    raise JushuitanInventoryError(
+                        f"Jushuitan returned duplicate inventory rows for SKU {sku}"
+                    )
+                records[sku] = record
+
+            raw_page_count = data.get("page_count", data.get("pageCount", 1))
+            try:
+                page_count = max(1, int(raw_page_count or 1))
+            except (TypeError, ValueError) as exc:
+                raise JushuitanInventoryError(
+                    "Jushuitan inventory response has an invalid page count"
+                ) from exc
+            if page_count > _MAX_PAGES:
+                raise JushuitanInventoryError(
+                    "Jushuitan inventory response exceeded the page safety limit"
+                )
+            page_index += 1
+        return records
+
+    def _request_access_token(self) -> str:
+        params: dict[str, object] = {
+            "app_key": self.app_key,
+            "charset": "utf-8",
+            "code": uuid.uuid4().hex[:6],
+            "grant_type": "authorization_code",
+            "timestamp": str(int(time.time())),
+        }
+        response = self._post_form(_TOKEN_PATH, params)
+        data = response.get("data")
+        if not isinstance(data, Mapping):
+            raise JushuitanInventoryError(
+                "Jushuitan token response is missing its data object"
+            )
+        token = data.get("access_token") or data.get("accessToken")
+        if not isinstance(token, str) or not token.strip():
+            raise JushuitanInventoryError(
+                "Jushuitan token response has no access token"
+            )
+        return token.strip()
+
+    def _query_page(self, skus: list[str], page_index: int) -> dict[str, Any]:
+        biz = {
+            "sku_ids": ",".join(skus),
+            "has_lock_qty": True,
+            "page_index": page_index,
+            "page_size": 100,
+        }
+        params: dict[str, object] = {
+            "access_token": self._access_token,
+            "app_key": self.app_key,
+            "charset": "utf-8",
+            "timestamp": str(int(time.time())),
+            "version": "2",
+            "biz": json.dumps(biz, ensure_ascii=False, separators=(",", ":")),
+        }
+        for token_attempt in range(2):
+            response = self._post_form(
+                _INVENTORY_PATH,
+                params,
+                allowed_api_codes=frozenset({_ACCESS_TOKEN_INVALID_CODE}),
+            )
+            code = _response_code(response.get("code"))
+            if code == _ACCESS_TOKEN_INVALID_CODE and token_attempt == 0:
+                self._access_token = self._request_access_token()
+                params["access_token"] = self._access_token
+                params["timestamp"] = str(int(time.time()))
+                continue
+            return response
+        raise JushuitanInventoryError("Jushuitan rejected the access token")
+
+    def _post_form(
+        self,
+        path: str,
+        params: dict[str, object],
+        *,
+        allowed_api_codes: frozenset[int] = frozenset(),
+    ) -> dict[str, Any]:
+        rate_limit_attempt = 0
+        while True:
+            now = time.monotonic()
+            wait_seconds = _REQUEST_INTERVAL_SECONDS - (now - self._last_request_at)
+            if wait_seconds > 0:
+                time.sleep(wait_seconds)
+            self._last_request_at = time.monotonic()
+            signed_params = dict(params)
+            signed_params["sign"] = _sign(self.app_secret, signed_params)
+            try:
+                response = self.session.post(
+                    f"{self.base_url}{path}",
+                    data=signed_params,
+                    timeout=self.timeout,
+                    headers={"Accept": "application/json"},
+                )
+            except requests.RequestException as exc:
+                raise JushuitanInventoryError(
+                    f"Jushuitan request failed for {path}"
+                ) from exc
+            if response.status_code < 200 or response.status_code >= 300:
+                raise JushuitanInventoryError(
+                    f"Jushuitan endpoint {path} returned HTTP {response.status_code}"
+                )
+            if len(response.content) > _MAX_RESPONSE_BYTES:
+                raise JushuitanInventoryError(
+                    "Jushuitan response exceeded the 2 MiB safety limit"
+                )
+            try:
+                payload = response.json()
+            except ValueError as exc:
+                raise JushuitanInventoryError(
+                    f"Jushuitan endpoint {path} returned invalid JSON"
+                ) from exc
+            if not isinstance(payload, dict):
+                raise JushuitanInventoryError(
+                    f"Jushuitan endpoint {path} returned an invalid response"
+                )
+            code = _response_code(payload.get("code"))
+            if code in _RATE_LIMIT_CODES and rate_limit_attempt < 4:
+                time.sleep(0.5 * (2**rate_limit_attempt))
+                rate_limit_attempt += 1
+                params["timestamp"] = str(int(time.time()))
+                continue
+            if code != 0 and code not in allowed_api_codes:
+                raise JushuitanInventoryError(
+                    f"Jushuitan endpoint {path} returned API code {code}"
+                )
+            return payload
+
+
+__all__ = [
+    "InventoryRecord",
+    "JushuitanInventoryClient",
+    "JushuitanInventoryError",
+]
@@ -37,8 +37,8 @@ def is_browser_timeout_error(error: BaseException) -> bool:
    )


-def _is_browser_closed_error(error: BaseException) -> bool:
-    """Recognize a browser target that disappeared during cleanup."""
+def is_browser_closed_error(error: BaseException) -> bool:
+    """Recognize a browser target that disappeared (page/context/browser)."""

    if any(
        error_type.__name__ == "TargetClosedError"
@@ -155,7 +155,7 @@ class ScraplingBrowser:
                    # and context-manager cleanup.  The current state cannot
                    # be exported then, but that must not turn a successful
                    # download into a false failure or mask the action error.
-                    if not _is_browser_closed_error(exc):
+                    if not is_browser_closed_error(exc):
                        raise
        finally:
            try:
@@ -33,6 +33,34 @@ DEFAULT_SEED_PATH = (
    / "workflow-dynamic-config-seed.json"
 )

+DIM_STYLE_RUNTIME_SQL = """
+    SELECT s.id,
+           s.style_name AS name,
+           COALESCE(s.brand, '') AS brand,
+           COALESCE(s.erp_style_codes, '[]'::jsonb) AS erp_style_codes,
+           jsonb_build_object(
+               'tm', COALESCE(s.tm_spus, '[]'::jsonb),
+               'jd', COALESCE(s.jd_spus, '[]'::jsonb),
+               'dy', COALESCE(s.dy_spus, '[]'::jsonb),
+               'jd_self', COALESCE(s.jd_self_skus, '[]'::jsonb)
+           ) AS platform_product_ids,
+           COALESCE(s.bitable_weekly_url, '') AS sales_bitable_url,
+           COALESCE(s.bitable_main_image_url, '') AS main_image_bitable_url,
+           COALESCE(s.bitable_creator_url, '') AS creator_bitable_url,
+           COALESCE(s.bitable_self_creator_url, '') AS self_creator_bitable_url,
+           COALESCE(s.bitable_weekly_note_url, '') AS weekly_note_analysis_url,
+           COALESCE(s.bitable_style_analysis_url, '') AS style_analysis_bitable_url,
+           COALESCE(s.bitable_persona_tm_url, '') AS tm_persona_bitable_url,
+           COALESCE(s.bitable_persona_jd_url, '') AS jd_persona_bitable_url,
+           COALESCE(s.bitable_persona_dy_url, '') AS dy_persona_bitable_url,
+           COALESCE(s.series_name, '') AS series_name,
+           COALESCE(s.is_series_anchor, FALSE) AS is_series_anchor,
+           COALESCE(s.collect_enabled, TRUE) AS collect_enabled
+    FROM dim_style s
+    WHERE COALESCE(s.style_name, '') <> ''
+    ORDER BY s.style_name, s.id
+"""
+
 CONFIG_FIELDS = (
    "platform",
    "erp_codes",
@@ -298,8 +326,9 @@ def merge_lifecycle_style_records(
    """Overlay lifecycle product identity on legacy target configuration rows.

    ``workflow_dynamic_*`` still owns Feishu target URLs and persona targets, while
-    ``cmt_styles`` owns ERP codes and product IDs.  The result intentionally keeps
-    the old flat runtime contract so existing collectors need no per-workflow fork.
+    ``dim_style`` (产品生命进程建档) owns ERP codes, product IDs and the sales sheet.
+    The result intentionally keeps the old flat runtime contract so existing
+    collectors need no per-workflow fork.
    """

    dynamic_by_style: dict[str, list[Mapping[str, Any]]] = {}
@@ -311,7 +340,7 @@ def merge_lifecycle_style_records(
    lifecycle_by_style: dict[str, Mapping[str, Any]] = {}
    lifecycle_order: list[str] = []
    for row in lifecycle_rows:
-        style_name = " ".join(str(row.get("name") or "").split())
+        style_name = " ".join(str(row.get("name") or row.get("style_name") or "").split())
        if not style_name or style_name in lifecycle_by_style:
            continue
        lifecycle_by_style[style_name] = row
@@ -325,7 +354,17 @@ def merge_lifecycle_style_records(
        legacy_rows = dynamic_by_style.get(style_name, [])
        lifecycle = lifecycle_by_style.get(style_name)
        if lifecycle is None:
-            records.extend(dict(row) for row in legacy_rows)
+            # 不在 dim_style 的款式（改名/删除后的残留行）不再继承旧表地址：
+            # 飞书地址已统一由 dim_style 锚点持有，旧表列只是历史数据。
+            for row in legacy_rows:
+                record = dict(row)
+                for field in URL_FIELDS:
+                    record[field] = ""
+                # 已不在 dim_style 的残留款：没有主键可下发，显式标记未匹配，
+                # 避免下游把 None 当成「还没回填」而不是「确实没建档」。
+                record["style_id"] = None
+                record["style_match_method"] = "unmatched"
+                records.append(record)
            continue

        managed, lifecycle_ids = _decode_lifecycle_product_ids(
@@ -379,6 +418,33 @@ def merge_lifecycle_style_records(
                    if item_id not in dynamic_unknown_ids.setdefault(label, []):
                        dynamic_unknown_ids[label].append(item_id)

+        # 飞书地址统一以 dim_style 为准（2026-10-29）：9 类地址全部维护在系列
+        # 锚点行，旧表同名字段彻底不参与，缺失就是缺失、不再回退。
+        sales_bitable_url = str(lifecycle.get("sales_bitable_url") or "").strip()
+        series_name = str(lifecycle.get("series_name") or "").strip()
+        is_series_anchor = bool(lifecycle.get("is_series_anchor"))
+
+        template["sales_bitable_url"] = sales_bitable_url
+        template["series_name"] = series_name
+        template["is_series_anchor"] = is_series_anchor
+        # 款式主键统一 (2026-11-01): dim_style.id 作为正式字段下发。
+        # 原先它只被编码进 source_record_id="lifecycle_style:<id>" 字符串，写入脚本
+        # 拿不到，只能靠 style_name 关联——改款名历史行就成孤儿。这里提成字段后，
+        # 所有经 runtime_workflow_configs() 取款的脚本自动获得 style_id。
+        style_id_value = lifecycle.get("id")
+        template["style_id"] = (
+            int(style_id_value) if style_id_value is not None else None
+        )
+        template["style_match_method"] = (
+            "dim_style" if style_id_value is not None else "unmatched"
+        )
+        for field in URL_FIELDS:
+            template[field] = str(lifecycle.get(field) or "").strip()
+        # 采集开关由产品生命进程在 dim_style 上维护（系列级，读锚点那行）。
+        # 旧 cmt_styles 回退路径没有该列，按默认 TRUE 处理，保持历史口径。
+        enabled_flag = lifecycle.get("collect_enabled")
+        template["collect_enabled"] = True if enabled_flag is None else bool(enabled_flag)
+
        if managed:
            labels = []
            for label in LIFECYCLE_PLATFORM_LABELS.values():
@@ -413,13 +479,70 @@ def merge_lifecycle_style_records(
                else list(dynamic_ids.get(key, dynamic_unknown_ids.get(label, [])))
            )
            output["source_record_id"] = (
-                f"cmt_style:{lifecycle.get('id')}" if key and managed else output.get("source_record_id")
+                f"lifecycle_style:{lifecycle.get('id')}" if key and managed else output.get("source_record_id")
            )
            records.append(output)

+    _propagate_series_targets(records)
    return records


+def _propagate_series_targets(records: list[dict[str, Any]]) -> None:
+    """把系列内 9 类飞书地址聚合去重后统一下发到整条系列（就地修改）。
+
+    约定是地址只维护在 ``dim_style`` 的系列锚点行，子款留空。但建档时子款误填
+    无法从表结构上禁止，所以这里不「只认锚点」，而是把整条系列的非空地址按
+    「锚点优先、其余按款名」收集起来，按 URL 去重后取优先级最高的一条下发到
+    该系列每一行——子款写的地址不会因为不在锚点上而被静默丢弃。
+
+    一个表位只能落一张表：同一字段出现多个不同地址时保留优先级最高的那条并
+    打印告警，提示到锚点手动裁决，避免同系列数据被拆到两个表。
+
+    没有锚点时按款名排序取第一个，与采集开关、ERP 归集的判定口径保持一致。
+    """
+
+    by_series: dict[str, list[dict[str, Any]]] = {}
+    for record in records:
+        series = " ".join(str(record.get("series_name") or "").split())
+        if series:
+            by_series.setdefault(series, []).append(record)
+
+    for series, rows in by_series.items():
+        # merge 会把同一款展开成多条平台记录，聚合前先按款归并，避免重复计数
+        by_style: dict[str, list[dict[str, Any]]] = {}
+        for row in rows:
+            by_style.setdefault(str(row.get("style_name") or ""), []).append(row)
+        ordered = sorted(
+            by_style,
+            key=lambda name: (
+                not any(row.get("is_series_anchor") for row in by_style[name]),
+                name,
+            ),
+        )
+
+        chosen: dict[str, str] = {}
+        for field in URL_FIELDS:
+            candidates: list[str] = []
+            for name in ordered:
+                for row in by_style[name]:
+                    value = _plain_url(row.get(field))
+                    if value and value not in candidates:
+                        candidates.append(value)
+            if not candidates:
+                continue
+            if len(candidates) > 1:
+                print(
+                    f"[WARN] 系列「{series}」{field} 有 {len(candidates)} 个不同地址，"
+                    f"按「{ordered[0]}」下发并忽略其余: {candidates[1:]}"
+                )
+            chosen[field] = candidates[0]
+
+        if not chosen:
+            continue
+        for row in rows:
+            row.update(chosen)
+
+
 def _plain_url(value: Any) -> str:
    text = str(value or "").strip()
    match = re.search(r"\]\((https?://[^)]+)\)", text)
@@ -1158,16 +1281,24 @@ class PostgresWorkflowConfigStore:
                    "WHERE s.enabled ORDER BY s.style_name, p.id"
                )
                rows = cursor.fetchall()
-                cursor.execute("SELECT to_regclass('public.cmt_styles') AS table_name")
-                cmt_table = cursor.fetchone() or {}
-                if cmt_table.get("table_name"):
-                    cursor.execute(
-                        "SELECT id, name, brand, erp_style_codes, platform_product_ids "
-                        "FROM cmt_styles ORDER BY name, id"
-                    )
+                cursor.execute("SELECT to_regclass('public.dim_style') AS table_name")
+                dim_table = cursor.fetchone() or {}
+                if dim_table.get("table_name"):
+                    # 产品生命进程（dim_style）是款式身份的唯一权威来源：
+                    # ERP 款式编码逐款填写，平台商品编码与销量表允许整系列只填主款。
+                    cursor.execute(DIM_STYLE_RUNTIME_SQL)
                    lifecycle_rows = cursor.fetchall()
                else:
-                    lifecycle_rows = []
+                    cursor.execute("SELECT to_regclass('public.cmt_styles') AS table_name")
+                    cmt_table = cursor.fetchone() or {}
+                    if cmt_table.get("table_name"):
+                        cursor.execute(
+                            "SELECT id, name, brand, erp_style_codes, platform_product_ids "
+                            "FROM cmt_styles ORDER BY name, id"
+                        )
+                        lifecycle_rows = cursor.fetchall()
+                    else:
+                        lifecycle_rows = []
        personas: dict[int, dict[str, str]] = {}
        for row in rows:
            if row.get("platform_enabled") is False:
@@ -79,6 +79,10 @@ class WorkflowEntry:
    steps: tuple[WorkflowStepEntry, ...] = ()
    notification_workflow_id: str | None = None
    topic_config: WorkflowTopicConfig | None = None
+    # Orchestrator entries whose own binding carries no runtime account while
+    # the spawned child collectors do; the cookie gate probes these before
+    # the workflow runs.
+    required_accounts: tuple[str, ...] = ()


@dataclass(frozen=True, slots=True)
@@ -218,6 +222,9 @@ def _parse_workflow(item: Any) -> WorkflowEntry:
    if trigger == "scheduled" and (not isinstance(task_name, str) or not task_name):
        raise CatalogError(f"scheduled workflow requires source task name: {workflow_id}")
    topic_config = _parse_topic_config(item.get("topic_config"), workflow_id)
+    required_accounts = _parse_required_accounts(
+        item.get("required_accounts", []), workflow_id
+    )
    return WorkflowEntry(
        workflow_id=workflow_id,
        module=module,
@@ -229,9 +236,18 @@ def _parse_workflow(item: Any) -> WorkflowEntry:
        note=item.get("note"),
        steps=steps,
        topic_config=topic_config,
+        required_accounts=required_accounts,
    )


+def _parse_required_accounts(value: Any, workflow_id: str) -> tuple[str, ...]:
+    if not isinstance(value, list) or not all(
+        isinstance(item, str) and _WORKFLOW_ID.fullmatch(item) for item in value
+    ):
+        raise CatalogError(f"invalid required_accounts for {workflow_id}")
+    return tuple(dict.fromkeys(value))
+
+
 def _parse_topic_config(value: Any, workflow_id: str) -> WorkflowTopicConfig | None:
    if value is None:
        return None
@@ -229,6 +229,36 @@ def build_parser() -> argparse.ArgumentParser:
    status_parser = acceptance_commands.add_parser("status")
    status_parser.add_argument("--json", action="store_true")
    status_parser.add_argument("--output", type=Path)
+
+    audit_parser = commands.add_parser(
+        "audit", help="query the PostgreSQL operational audit tables"
+    )
+    audit_commands = audit_parser.add_subparsers(dest="audit_command", required=True)
+
+    audit_runs = audit_commands.add_parser("runs", help="recent workflow runs")
+    audit_runs.add_argument("--workflow")
+    audit_runs.add_argument("--status")
+    audit_runs.add_argument("--date", dest="business_date")
+    audit_runs.add_argument("--limit", type=int, default=30)
+    audit_runs.add_argument("--json", action="store_true")
+
+    audit_cookies = audit_commands.add_parser(
+        "cookies", help="recent cookie observations"
+    )
+    audit_cookies.add_argument("--account")
+    audit_cookies.add_argument("--state")
+    audit_cookies.add_argument("--limit", type=int, default=30)
+    audit_cookies.add_argument("--json", action="store_true")
+
+    audit_stats = audit_commands.add_parser(
+        "cookie-stats", help="measured cookie lifetime per account"
+    )
+    audit_stats.add_argument("--account")
+    audit_stats.add_argument("--json", action="store_true")
+
+    audit_commands.add_parser(
+        "schema", help="print the DDL that creates the audit tables"
+    )
    return parser


@@ -304,6 +334,8 @@ def main(
            and arguments.acceptance_command == "status"
        ):
            return _acceptance_status_command(arguments, resolved_settings, output)
+        if arguments.command == "audit":
+            return _audit_command(arguments, resolved_settings, output)
        if arguments.command == "scripts" and arguments.scripts_command == "list":
            return _list_scripts(arguments, output)
        if arguments.command == "scripts" and arguments.scripts_command == "run":
@@ -549,6 +581,152 @@ def _acceptance_status_command(
    return EXIT_SUCCESS if report.is_complete else EXIT_ACCEPTANCE_INCOMPLETE


+def _format_seconds(value: object) -> str:
+    """Human-readable duration for the audit tables (None -> '-')."""
+
+    if value is None:
+        return "-"
+    try:
+        seconds = int(value)
+    except (TypeError, ValueError):
+        return "-"
+    if seconds >= 86400:
+        return f"{seconds / 86400:.1f}d"
+    if seconds >= 3600:
+        return f"{seconds / 3600:.1f}h"
+    if seconds >= 60:
+        return f"{seconds / 60:.0f}m"
+    return f"{seconds}s"
+
+
+def _format_timestamp(value: object) -> str:
+    if value is None:
+        return "-"
+    text = value.isoformat() if hasattr(value, "isoformat") else str(value)
+    return text.replace("T", " ")[:19]
+
+
+def _audit_command(
+    arguments: argparse.Namespace, settings: Settings, output: TextIO
+) -> int:
+    from gyxx_flow.ops.audit import SCHEMA_SQL, AuditStore
+
+    command = arguments.audit_command
+    if command == "schema":
+        output.write(SCHEMA_SQL.strip() + "\n")
+        return EXIT_SUCCESS
+
+    store = AuditStore(settings.data_root)
+    if not store.enabled:
+        output.write("审计已关闭（GYXX_OPS_AUDIT=off）\n")
+        return EXIT_CONFIGURATION
+
+    if command == "runs":
+        rows = store.query_runs(
+            workflow_id=arguments.workflow,
+            status=arguments.status,
+            business_date=arguments.business_date,
+            limit=arguments.limit,
+        )
+    elif command == "cookies":
+        rows = store.query_cookie_events(
+            account_id=arguments.account,
+            state=arguments.state,
+            limit=arguments.limit,
+        )
+    elif command == "cookie-stats":
+        rows = (
+            store.query_cookie_lifetimes(account_id=arguments.account)
+            if arguments.account
+            else store.query_cookie_stats()
+        )
+    else:  # pragma: no cover - argparse 已限制取值
+        raise CliConfigurationError(f"unsupported audit command: {command}")
+
+    if store.last_error and not rows:
+        output.write(f"审计库不可用：{store.last_error}\n")
+        return EXIT_CONFIGURATION
+
+    if getattr(arguments, "json", False):
+        _write_json(output, {"rows": rows})
+        return EXIT_SUCCESS
+
+    if not rows:
+        output.write("暂无审计记录\n")
+        return EXIT_SUCCESS
+
+    if command == "runs":
+        output.write(
+            f"{'started':19s}  {'workflow':40s} {'date':10s} {'mode':7s} "
+            f"{'status':8s} {'dur':>6s}  steps\n"
+        )
+        for row in rows:
+            steps = (
+                f"{row.get('step_success', 0)}/{row.get('step_total', 0)}"
+                f" failed={row.get('step_failed', 0)}"
+            )
+            output.write(
+                f"{_format_timestamp(row.get('started_at')):19s}  "
+                f"{str(row.get('workflow_id', ''))[:40]:40s} "
+                f"{_format_timestamp(row.get('business_date'))[:10]:10s} "
+                f"{str(row.get('mode', '')):7s} "
+                f"{str(row.get('status', '')):8s} "
+                f"{_format_seconds(row.get('duration_seconds')):>6s}  {steps}\n"
+            )
+            if row.get("error"):
+                output.write(f"{'':21s}error: {str(row['error'])[:160]}\n")
+        return EXIT_SUCCESS
+
+    if command == "cookies":
+        output.write(
+            f"{'observed':19s}  {'account':24s} {'state':14s} {'probe':14s} "
+            f"{'fingerprint':12s}  expires_at\n"
+        )
+        for row in rows:
+            output.write(
+                f"{_format_timestamp(row.get('observed_at')):19s}  "
+                f"{str(row.get('account_id', ''))[:24]:24s} "
+                f"{str(row.get('state', '')):14s} "
+                f"{str(row.get('probe_status', '')):14s} "
+                f"{str(row.get('fingerprint', ''))[:12]:12s}  "
+                f"{_format_timestamp(row.get('declared_expires_at'))}\n"
+            )
+        return EXIT_SUCCESS
+
+    # cookie-stats / 单账号 lifetime
+    if arguments.account:
+        output.write(
+            f"{'fingerprint':12s} {'first_seen':19s} {'last_healthy':19s} "
+            f"{'failed_at':19s} {'lifetime':>9s} {'declared':19s}\n"
+        )
+        for row in rows:
+            output.write(
+                f"{str(row.get('fingerprint', ''))[:12]:12s} "
+                f"{_format_timestamp(row.get('first_seen_at')):19s} "
+                f"{_format_timestamp(row.get('last_healthy_at')):19s} "
+                f"{_format_timestamp(row.get('first_failed_at')):19s} "
+                f"{_format_seconds(row.get('lifetime_seconds')):>9s} "
+                f"{_format_timestamp(row.get('declared_expires_at')):19s}\n"
+            )
+        return EXIT_SUCCESS
+
+    output.write(
+        f"{'account':24s} {'sessions':>8s} {'expired':>7s} {'avg':>7s} "
+        f"{'median':>7s} {'min':>7s} {'max':>7s}  last_failed\n"
+    )
+    for row in rows:
+        output.write(
+            f"{str(row.get('account_id', ''))[:24]:24s} "
+            f"{row.get('sessions', 0):>8d} {row.get('expired_sessions', 0):>7d} "
+            f"{_format_seconds(row.get('avg_lifetime_seconds')):>7s} "
+            f"{_format_seconds(row.get('median_lifetime_seconds')):>7s} "
+            f"{_format_seconds(row.get('min_lifetime_seconds')):>7s} "
+            f"{_format_seconds(row.get('max_lifetime_seconds')):>7s}  "
+            f"{_format_timestamp(row.get('last_failed_at'))}\n"
+        )
+    return EXIT_SUCCESS
+
+
 def _list_workflows(
    arguments: argparse.Namespace,
    registry: WorkflowRegistry,
@@ -703,7 +881,19 @@ def _python_scheduler(
    if enable_account_keepalive and not dry_run:
        from gyxx_flow.accounts import AccountKeepaliveManager

-        maintenance = AccountKeepaliveManager(settings).tick
+        keepalive_manager = AccountKeepaliveManager(settings)
+        recovery_manager = _build_cookie_recovery_manager(
+            settings, launcher=launcher, keepalive=keepalive_manager
+        )
+
+        def maintenance() -> None:  # noqa: D107 - scheduler maintenance chain
+            keepalive_manager.tick()
+            if recovery_manager is not None:
+                try:
+                    recovery_manager.tick()
+                except Exception:  # noqa: BLE001 - 自动重跑不得影响调度心跳
+                    pass
+
    return PythonScheduler(
        catalog,
        settings.data_root,
@@ -717,6 +907,29 @@ def _python_scheduler(
    )


+def _build_cookie_recovery_manager(
+    settings: Settings,
+    *,
+    launcher: object,
+    keepalive: object,
+):
+    """Build the cookie-gate recovery rerun manager; None when disabled."""
+
+    from gyxx_flow.cookie_gate import (
+        CookieGate,
+        CookieGateRecoveryManager,
+        gate_enabled,
+    )
+
+    if not gate_enabled():
+        return None
+    try:
+        gate = CookieGate(settings, keepalive=keepalive)  # type: ignore[arg-type]
+    except Exception:  # noqa: BLE001 - 无集成配置时没有可恢复的账号
+        return None
+    return CookieGateRecoveryManager(gate, launcher=launcher)
+
+
 def _schedule_run_command(
    arguments: argparse.Namespace,
    catalog: WorkflowCatalog,
@@ -885,6 +1098,11 @@ def _execute_once(
    dry_run: bool,
    settings: Settings,
 ) -> tuple[int, dict[str, object]]:
+    if not dry_run:
+        # 每账号 Cookie 健康门控：dry-run 不注入，也不做任何探测/通知。
+        from gyxx_flow.cookie_gate import apply_cookie_gates
+
+        workflow = apply_cookie_gates(workflow, settings=settings)
    context = RunContext.create(
        workflow.workflow_id,
        business_date.isoformat(),
@@ -5,6 +5,7 @@ from __future__ import annotations
 import calendar
 import hashlib
 import hmac
+import http.client
 import ipaddress
 import json
 import os
@@ -15,6 +16,7 @@ import sys
 import tempfile
 import threading
 import time
+import traceback
 import unicodedata
 import uuid
 from contextlib import suppress
@@ -24,7 +26,7 @@ from http import HTTPStatus
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from importlib import resources
 from pathlib import Path
-from typing import Any, Mapping, Protocol, Sequence, TextIO, cast
+from typing import Any, Callable, Mapping, Protocol, Sequence, TextIO, cast
 from urllib.parse import parse_qs, unquote, urlsplit
 from zoneinfo import ZoneInfo

@@ -171,7 +173,6 @@ _WORKFLOW_NAMES = {
    "content.metrics.daily": "合作达人内容指标日报采集",
    "content.metrics.self_operated.daily": "自营内容指标日报采集",
    "content.marketing_report.daily": "营销日报生成与发送",
-    "content.relogin.weekly": "内容平台登录态刷新",
    "content.creator_report.monthly": "达人月报生成",
    "content.summary.monthly": "内容月度汇总",
    "content.comments.weekly": "多平台评论周采集",
@@ -190,7 +191,6 @@ _WORKFLOW_NAMES = {
    "product.backfill": "商品数据补采",
    "product.market_rank": "市场排行周采集",
    "product.tmall_baibu_apply": "天猫百亿补贴批量报名",
-    "product.review_collection": "商品评价采集",
    "product.weekly_aggregate.documented_missing": "商品周聚合（待接入）",
    "shop.metrics.weekly": "店铺经营指标周采集",
    "shop.competitor.weekly": "竞店指标周采集",
@@ -2246,7 +2246,7 @@ class WorkflowConsoleService:
        jd_customer_service_store: str | None = None
        if workflow_id == "product.video_upload":
            raw_store = payload.get("store", "flagship")
-            if raw_store not in {"flagship", "luggage"}:
+            if raw_store not in {"flagship", "luggage", "new_bag"}:
                raise ConsoleRequestError("天猫视频上传店铺参数无效")
            video_upload_store = str(raw_store)
            tmall_topic_keyword = _normalize_tmall_topic_keyword(
@@ -3736,6 +3736,7 @@ class WorkflowConsoleHTTPServer(ThreadingHTTPServer):
        self.console_service = service
        self.console_token = token
        self.bind_host = bind_host
+        self.started_at = time.time()
        super().__init__(server_address, WorkflowConsoleRequestHandler)

    def handle_error(
@@ -3796,6 +3797,29 @@ class WorkflowConsoleRequestHandler(BaseHTTPRequestHandler):
            if method in {"GET", "HEAD"} and path == "/assets/app.js":
                self._static("app.js", "text/javascript; charset=utf-8", head=method == "HEAD")
                return
+            if method == "GET" and path == "/api/health":
+                # 自检探针端点：不鉴权、不触外部依赖，只反映进程与调度锁存活，
+                # 供进程内看门狗与 deploy/watchdog 保活脚本探测。
+                try:
+                    scheduler = _scheduler_process_from_lock(
+                        self.console_server.console_service.settings
+                    )
+                except Exception:
+                    scheduler = None
+                self._json(
+                    HTTPStatus.OK,
+                    {
+                        "ok": True,
+                        "pid": os.getpid(),
+                        "started_at": self.console_server.started_at,
+                        "uptime_seconds": round(
+                            time.time() - self.console_server.started_at, 1
+                        ),
+                        "scheduler_running": scheduler is not None,
+                        "scheduler_pid": scheduler.pid if scheduler else None,
+                    },
+                )
+                return
            if method == "GET" and path == "/api/overview":
                self._require_api_access(mutation=False)
                self._json(HTTPStatus.OK, self.console_server.console_service.overview())
@@ -4020,6 +4044,9 @@ class WorkflowConsoleRequestHandler(BaseHTTPRequestHandler):
        except (BrokenPipeError, ConnectionResetError):
            return
        except Exception:
+            # 前端只拿到固定文案；完整堆栈落到 stderr（随启动重定向进
+            # var/logs/console-restarts 或服务日志），否则故障无法定位。
+            traceback.print_exc()
            self._json_error(HTTPStatus.INTERNAL_SERVER_ERROR, "控制台处理请求失败")

    def _require_api_access(self, *, mutation: bool) -> None:
@@ -4190,6 +4217,92 @@ def create_console_server(
    )


+CONSOLE_WATCHDOG_EXIT_CODE = 70
+
+
+def _probe_console_health(port: int, timeout_seconds: float) -> bool:
+    # 只连本机回环端口的固定路径 /api/health，探测自身 HTTP 服务往返。
+    connection = http.client.HTTPConnection("127.0.0.1", port, timeout=timeout_seconds)
+    try:
+        connection.request("GET", "/api/health")
+        response = connection.getresponse()
+        response.read()
+        return response.status == HTTPStatus.OK
+    except Exception:
+        return False
+    finally:
+        connection.close()
+
+
+def _console_watchdog(
+    port: int,
+    *,
+    interval_seconds: float,
+    failure_limit: int,
+    probe_timeout_seconds: float,
+    stopped: threading.Event,
+    probe: Callable[[int, float], bool] | None = None,
+    exit_process: Callable[[int], None] = os._exit,
+    sleep: Callable[[float], None] = time.sleep,
+) -> None:
+    """进程内自检：周期探测自身 /api/health，连续失败则退出交给守护拉起。
+
+    HTTP accept 线程活着但工作线程全部卡死（锁竞争/句柄耗尽）时，外部
+    静态页能打开而 API 全挂——正是"控制台暂时不可用"的形态；只有真实
+   走一遍 HTTP 往返才能暴露，所以探针必须打 socket 而不是查内存状态。
+    """
+    run_probe = probe or _probe_console_health
+    failures = 0
+    while not stopped.is_set():
+        if run_probe(port, probe_timeout_seconds):
+            failures = 0
+        else:
+            failures += 1
+            print(
+                f"⚠️ 控制台自检失败 {failures}/{failure_limit}: 127.0.0.1:{port}/api/health",
+                file=sys.stderr,
+                flush=True,
+            )
+            if failures >= failure_limit:
+                print(
+                    "❌ 控制台自检连续失败，进程退出等待守护脚本拉起",
+                    file=sys.stderr,
+                    flush=True,
+                )
+                exit_process(CONSOLE_WATCHDOG_EXIT_CODE)
+                return
+        sleep(interval_seconds)
+
+
+def _start_console_watchdog(server: WorkflowConsoleHTTPServer) -> threading.Event:
+    interval = max(
+        5.0, float(os.environ.get("GYXX_CONSOLE_WATCHDOG_INTERVAL_SECONDS", "30"))
+    )
+    limit = int(os.environ.get("GYXX_CONSOLE_WATCHDOG_FAILURES", "3"))
+    enabled = (
+        os.environ.get("GYXX_CONSOLE_WATCHDOG", "1").strip().lower()
+        not in {"0", "false", "no", "off"}
+        and limit > 0
+    )
+    stopped = threading.Event()
+    if not enabled:
+        return stopped
+    thread = threading.Thread(
+        target=_console_watchdog,
+        args=(server.server_port,),
+        kwargs={
+            "interval_seconds": interval,
+            "failure_limit": limit,
+            "probe_timeout_seconds": min(5.0, interval),
+            "stopped": stopped,
+        },
+        name="gyxx-console-watchdog",
+        daemon=True,
+    )
+    thread.start()
+    return stopped
+
+
 def serve_console(
    settings: Settings,
    *,
@@ -4211,11 +4324,13 @@ def serve_console(
    display_host = "127.0.0.1" if host in {"0.0.0.0", "::"} else host
    output.write(f"GYXX Flow console: http://{display_host}:{server.server_port}\n")
    output.flush()
+    watchdog_stopped = _start_console_watchdog(server)
    try:
        server.serve_forever(poll_interval=0.5)
    except KeyboardInterrupt:
        pass
    finally:
+        watchdog_stopped.set()
        server.server_close()
    return 0

@@ -0,0 +1,966 @@
+"""Per-account cookie health gate injected before workflow collection steps.
+
+Every runtime account a workflow depends on is checked before the collection
+steps that use it may run.  Accounts fall into three gate modes derived from
+``config/runtime-bindings.json``:
+
+* ``blocking`` -- QR-code / SMS-only accounts (no credential environment and
+  no configured login command).  The gate verifies the login state and, when
+  it has expired, fails the gate step so every dependent step is skipped with
+  ``dependency-failed`` and sends a Feishu alert asking the login owner to
+  re-login interactively via ``gyxx accounts login <account_id>``.
+* ``recover`` -- accounts with a vault-managed credential login command
+  (``login_mode`` ``C``).  The keepalive manager re-runs the credential login
+  automatically; only a failed recovery blocks the workflow and alerts.
+* ``defer`` -- accounts whose member scripts embed username/password login
+  flows.  These keep their script-level login behaviour unchanged and are not
+  gated at all.
+
+The per-account decision is persisted as a small state machine under
+``state/accounts/<id>/cookie_state.json`` (``healthy`` <-> ``awaiting_login``)
+so repeat failures alert at most once per notify interval instead of once per
+workflow run.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from collections.abc import Callable
+from dataclasses import dataclass, replace
+from datetime import date, datetime, timezone
+from pathlib import Path
+from typing import Any, Mapping
+
+from gyxx_flow.accounts import (
+    AccountKeepaliveManager,
+    _keepalive_requires_relogin,
+    _read_keepalive_status,
+    _vault_cookie_valid,
+)
+from gyxx_flow.adapters.browser import BrowserCookieStore
+from gyxx_flow.adapters.feishu_messages import send_lark_bot_message
+from gyxx_flow.adapters.integration import RuntimeIntegrationCatalog
+from gyxx_flow.catalog import WorkflowCatalog
+from gyxx_flow.core.artifacts import atomic_write_json
+from gyxx_flow.core.config import Settings
+from gyxx_flow.core.context import RunContext
+from gyxx_flow.notification_routing import (
+    ANALYZER_ACCEPTANCE_RECIPIENT_OPEN_ID,
+    ANALYZER_NOTIFICATION_APP_PROFILE,
+    NotificationRoutingStore,
+)
+from gyxx_flow.workflow.model import StepDefinition, WorkflowDefinition
+from gyxx_flow.workflow.steps import StepExecution
+
+GATE_MODE_BLOCKING = "blocking"
+GATE_MODE_RECOVER = "recover"
+GATE_MODE_DEFER = "defer"
+
+# 飞书通知的默认接收人：按手机号在 config/notification-routing.json 的
+# people 段解析 open_id；解析失败时回退到同名 analyzer 常量收件人。
+DEFAULT_LOGIN_ALERT_MOBILE = "+8613663785145"
+
+GYXX_COOKIE_GATE = "GYXX_COOKIE_GATE"
+GYXX_COOKIE_GATE_MAX_STATUS_AGE_SECONDS = "GYXX_COOKIE_GATE_MAX_STATUS_AGE_SECONDS"
+GYXX_LOGIN_ALERT_OPEN_IDS = "GYXX_LOGIN_ALERT_OPEN_IDS"
+GYXX_LOGIN_ALERT_APP_PROFILE = "GYXX_LOGIN_ALERT_APP_PROFILE"
+GYXX_LOGIN_ALERT_NOTIFY_INTERVAL_SECONDS = "GYXX_LOGIN_ALERT_NOTIFY_INTERVAL_SECONDS"
+
+_GATE_STEP_PREFIX = "cookie-gate"
+_GATE_STATE_FILENAME = "cookie_state.json"
+_GATE_TIMEOUT_SECONDS = 1800.0
+_DEFAULT_MAX_STATUS_AGE_SECONDS = 3600.0
+_MIN_MAX_STATUS_AGE_SECONDS = 300.0
+_DEFAULT_NOTIFY_INTERVAL_SECONDS = 1800.0
+_DISABLED_VALUES = frozenset({"0", "false", "off", "no", "disabled"})
+# 自动重跑：待重跑清单上限、健康探活节流、同账号重跑启动冷却。
+_MAX_PENDING_RERUNS = 20
+_RECOVERY_PROBE_INTERVAL_SECONDS = 300.0
+_RECOVERY_LAUNCH_COOLDOWN_SECONDS = 300.0
+_RECOVERY_HISTORY_LIMIT = 20
+
+STATE_HEALTHY = "healthy"
+STATE_AWAITING_LOGIN = "awaiting_login"
+
+
+@dataclass(frozen=True, slots=True)
+class GateVerdict:
+    """One account's gate decision, recorded in the run journal on failure."""
+
+    account_id: str
+    mode: str
+    blocked: bool
+    message: str
+    recovered: bool = False
+    notified: bool = False
+
+
+@dataclass(frozen=True, slots=True)
+class CookieGateStepAction:
+    """ExecutableStep that turns one account verdict into a step outcome."""
+
+    gate: "CookieGate"
+    account_id: str
+
+    def execute(
+        self,
+        *,
+        context: RunContext,
+        timeout_seconds: float | None,
+        dry_run: bool,
+    ) -> StepExecution:
+        if dry_run:  # pragma: no cover - the engine skips dry-run steps first
+            return StepExecution(exit_code=0, skipped=True, reason="dry-run")
+        verdict = self.gate.check(self.account_id, context=context)
+        if verdict.blocked:
+            return StepExecution(
+                exit_code=1,
+                error=verdict.message,
+                reason="cookie-expired",
+            )
+        return StepExecution(exit_code=0)
+
+
+class CookieGate:
+    """Resolve, probe, and gate the accounts one workflow depends on."""
+
+    def __init__(
+        self,
+        settings: Settings,
+        *,
+        catalog: RuntimeIntegrationCatalog | None = None,
+        keepalive: AccountKeepaliveManager | None = None,
+        environment: Mapping[str, str] | None = None,
+        now: Callable[[], datetime] | None = None,
+        sender: Callable[..., dict[str, Any]] | None = None,
+    ) -> None:  # noqa: D107 - simple wiring
+        self.settings = settings
+        self.catalog = catalog or RuntimeIntegrationCatalog.load_default(
+            project_root=settings.project_root,
+            data_root=settings.data_root,
+        )
+        self.keepalive = keepalive or AccountKeepaliveManager(
+            settings, catalog=self.catalog
+        )
+        self._environment = dict(os.environ if environment is None else environment)
+        self._now = now or (lambda: datetime.now(timezone.utc))
+        self._sender = sender or send_lark_bot_message
+        self._workflow_catalog: WorkflowCatalog | None = None
+        self._audit_store: Any | None = None
+
+    # ------------------------------------------------------------------
+    # Account resolution
+    # ------------------------------------------------------------------
+
+    def gate_mode(self, account_id: str) -> str:
+        account = self.catalog.account_for(account_id)
+        if str(account.login_command_id or "").strip():
+            return GATE_MODE_RECOVER
+        if account.credential_env_names or account.login_mode == "B":
+            return GATE_MODE_DEFER
+        return GATE_MODE_BLOCKING
+
+    def collect_workflow_accounts(
+        self,
+        workflow: WorkflowDefinition,
+    ) -> dict[str, set[str]]:
+        """Map each gated account to the step ids that depend on it.
+
+        Accounts resolve two ways: directly, from each step binding's
+        ``account``; and declaratively, from the catalog workflow's
+        ``required_accounts`` (used by orchestrator entries whose own binding
+        has no account while the spawned child collectors do).
+        """
+
+        accounts: dict[str, set[str]] = {}
+        for step in workflow.steps:
+            account_id = self._account_for_step(step)
+            if account_id:
+                accounts.setdefault(account_id, set()).add(step.step_id)
+        for account_id in self._declared_required_accounts(workflow.workflow_id):
+            self.catalog.account_for(account_id)  # unknown id: loud config error
+            accounts.setdefault(account_id, set()).update(
+                step.step_id for step in workflow.steps
+            )
+        # Defer-mode accounts keep their script-level credential login flows
+        # and are intentionally not gated.
+        return {
+            account_id: step_ids
+            for account_id, step_ids in accounts.items()
+            if self.gate_mode(account_id) != GATE_MODE_DEFER
+        }
+
+    def _account_for_step(self, step: StepDefinition) -> str | None:
+        action = step.action
+        entry = getattr(action, "entry", None)
+        module_id = getattr(entry, "module", "")
+        command_entry = getattr(action, "command_entry", None) or getattr(
+            entry, "entry", ""
+        )
+        if not module_id or not command_entry:
+            return None
+        try:
+            binding = self.catalog.binding_for(f"{module_id}:{command_entry}")
+        except Exception:  # noqa: BLE001 - unbound entries run ungated
+            return None
+        return binding.account or None
+
+    def workflow_catalog(self) -> WorkflowCatalog:
+        """Load (once) the catalog used for declared account and rerun lookups."""
+
+        if self._workflow_catalog is None:
+            try:
+                self._workflow_catalog = WorkflowCatalog.load(
+                    self.settings.project_root / "config"
+                )
+            except Exception:  # noqa: BLE001 - catalog issues surface elsewhere
+                self._workflow_catalog = WorkflowCatalog(
+                    timezone="UTC", workflows=(), schedules=()
+                )
+        return self._workflow_catalog
+
+    def _declared_required_accounts(self, workflow_id: str) -> tuple[str, ...]:
+        for entry in self.workflow_catalog().workflows:
+            if entry.workflow_id == workflow_id:
+                return tuple(entry.required_accounts)
+        return ()
+
+    # ------------------------------------------------------------------
+    # State machine
+    # ------------------------------------------------------------------
+
+    def check(self, account_id: str, *, context: RunContext) -> GateVerdict:
+        account = self.catalog.account_for(account_id)
+        mode = self.gate_mode(account_id)
+        store = BrowserCookieStore(account.cookie_file, account.storage_state_file)
+        static_valid = _vault_cookie_valid(account, store)
+        status = _read_keepalive_status(account)
+        if static_valid and self._keepalive_success_is_fresh(status):
+            self._write_state(account, STATE_HEALTHY, probe="fresh-keepalive", context=context)
+            return GateVerdict(
+                account_id=account_id,
+                mode=mode,
+                blocked=False,
+                message=f"账号 {account_id} 登录态有效（keepalive 近期验证通过）",
+            )
+        if self._acceptance_enabled():
+            # 验收模式不开浏览器：按静态证据判定，且不投递飞书通知。
+            if static_valid and not _keepalive_requires_relogin(status):
+                return GateVerdict(
+                    account_id=account_id,
+                    mode=mode,
+                    blocked=False,
+                    message=f"账号 {account_id} 静态 cookie 校验通过（验收模式）",
+                )
+            return self._expire(
+                account,
+                mode,
+                context,
+                detail="验收模式下静态 cookie 校验未通过",
+                notify=False,
+            )
+
+        outcome = self.keepalive.refresh_account(account_id)
+        probe_status = str(outcome.get("status", "error"))
+        if probe_status == "success":
+            recovered = bool(outcome.get("login_recovered"))
+            self._write_state(
+                account,
+                STATE_HEALTHY,
+                probe="success",
+                recovered=recovered,
+                context=context,
+            )
+            message = (
+                f"账号 {account_id} 登录态有效（自动账密重登成功）"
+                if recovered
+                else f"账号 {account_id} 登录态有效（探活通过）"
+            )
+            return GateVerdict(
+                account_id=account_id,
+                mode=mode,
+                blocked=False,
+                message=message,
+                recovered=recovered,
+            )
+        if probe_status == "busy":
+            if static_valid and not _keepalive_requires_relogin(status):
+                return GateVerdict(
+                    account_id=account_id,
+                    mode=mode,
+                    blocked=False,
+                    message=f"账号 {account_id} 探活锁忙，静态 cookie 校验通过",
+                )
+            return self._expire(
+                account,
+                mode,
+                context,
+                detail="探活锁被占用且静态 cookie 校验未通过",
+            )
+        if probe_status in {"expired", "not_logged_in"}:
+            evidence = _recovery_evidence(_read_keepalive_status(account))
+            detail = "cookie 已过期，无法正常采集"
+            if probe_status == "not_logged_in" and evidence:
+                detail = f"登录态缺失；{evidence}"
+            elif evidence:
+                detail = f"cookie 已过期；{evidence}"
+            return self._expire(account, mode, context, detail=detail)
+        # timeout / error / http_error / disabled：探测结果不可信时回退静态证据。
+        if static_valid and not _keepalive_requires_relogin(status):
+            return GateVerdict(
+                account_id=account_id,
+                mode=mode,
+                blocked=False,
+                message=f"账号 {account_id} 探活异常（{probe_status}），静态 cookie 校验通过",
+            )
+        return self._expire(
+            account,
+            mode,
+            context,
+            detail=f"探活异常（{probe_status}）且静态 cookie 校验未通过",
+        )
+
+    # ------------------------------------------------------------------
+    # Expiry, notification, persisted state
+    # ------------------------------------------------------------------
+
+    def _expire(
+        self,
+        account: Any,
+        mode: str,
+        context: RunContext,
+        *,
+        detail: str,
+        notify: bool | None = None,
+    ) -> GateVerdict:
+        account_id = str(account.account_id)
+        if notify is None:
+            notify = not context.shadow and not self._acceptance_enabled()
+        state = self._read_state(account)
+        notified = False
+        notification_note = ""
+        if notify and not self._within_notify_interval(state):
+            recipients = self._alert_recipients()
+            message = _expiry_message(account_id, mode, context, detail)
+            for open_id, profile in recipients:
+                try:
+                    self._sender(
+                        user_id=open_id,
+                        text=message,
+                        idempotency_key=f"cookie-gate:{account_id}:{context.run_id}",
+                        profile=profile,
+                    )
+                    notified = True
+                except Exception as exc:  # noqa: BLE001 - alerting must not mask the verdict
+                    notification_note = f"{type(exc).__name__}: {exc}"[:300]
+            if recipients and notified:
+                state["last_notified_at"] = _now_iso(self._now)
+                state.pop("notification_error", None)
+            elif notification_note:
+                state["notification_error"] = notification_note
+        state["state"] = STATE_AWAITING_LOGIN
+        state["updated_at"] = _now_iso(self._now)
+        state["last_probe_status"] = "expired"
+        state["last_blocked_run"] = context.run_id
+        if not context.shadow and not self._acceptance_enabled():
+            # 记录待自动重跑清单：登录态恢复后由调度器维护循环按原业务日期重跑。
+            state["pending_reruns"] = _record_pending_rerun(
+                state.get("pending_reruns"), context, _now_iso(self._now)
+            )
+        self._save_state(account, state)
+        try:
+            keepalive_status = _read_keepalive_status(account) or {}
+        except Exception:  # noqa: BLE001 - state file is advisory
+            keepalive_status = {}
+        self._audit_cookie(
+            account,
+            "not_logged_in"
+            if str(keepalive_status.get("status", "")) == "not_logged_in"
+            else "expired",
+            probe_status=str(keepalive_status.get("status", "expired")),
+            probe_source="gate",
+            detail=detail,
+            context=context,
+        )
+        message = _expiry_message(account_id, mode, context, detail)
+        if notified:
+            message += " 已通过飞书通知登录负责人。"
+        elif notification_note:
+            message += f" 飞书通知发送失败（{notification_note}）。"
+        return GateVerdict(
+            account_id=account_id,
+            mode=mode,
+            blocked=True,
+            message=message,
+            notified=notified,
+        )
+
+    def _write_state(
+        self,
+        account: Any,
+        state_name: str,
+        *,
+        probe: str,
+        recovered: bool = False,
+        context: RunContext | None = None,
+    ) -> None:
+        if state_name == STATE_HEALTHY:
+            # 健康心跳要落在每一次判定上，才能算准「最后一次可用」的时间。
+            self._audit_cookie(
+                account,
+                "healthy",
+                probe_status=probe,
+                probe_source="gate",
+                context=context,
+            )
+        state = self._read_state(account)
+        if (
+            state_name == STATE_HEALTHY
+            and state.get("state") == STATE_HEALTHY
+            and state.get("last_probe_status") in {"fresh-keepalive", "success"}
+            and not recovered
+        ):
+            return  # 持续健康时避免每次运行都重写状态文件
+        state["state"] = state_name
+        state["updated_at"] = _now_iso(self._now)
+        state["last_probe_status"] = probe
+        if state_name == STATE_HEALTHY:
+            state["last_healthy_at"] = _now_iso(self._now)
+            if recovered:
+                state["last_recovered_at"] = _now_iso(self._now)
+            state.pop("notification_error", None)
+        self._save_state(account, state)
+
+    def _state_path(self, account: Any) -> Path:
+        return account.cookie_file.with_name(_GATE_STATE_FILENAME)
+
+    def _read_state(self, account: Any) -> dict[str, object]:
+        path = self._state_path(account)
+        if not path.exists():
+            return {"account_id": str(account.account_id)}
+        try:
+            payload = json.loads(path.read_text(encoding="utf-8"))
+        except (OSError, json.JSONDecodeError):
+            return {"account_id": str(account.account_id)}
+        if not isinstance(payload, dict):
+            return {"account_id": str(account.account_id)}
+        return dict(payload)
+
+    def _save_state(self, account: Any, state: Mapping[str, object]) -> None:
+        payload = {
+            "account_id": str(account.account_id),
+            **state,
+        }
+        try:
+            atomic_write_json(self._state_path(account), payload)
+        except OSError:  # pragma: no cover - state file is advisory only
+            pass
+
+    # ------------------------------------------------------------------
+    # Cookie telemetry (best-effort; never blocks or fails the gate)
+    # ------------------------------------------------------------------
+
+    def _audit(self) -> Any | None:
+        if self._audit_store is None:
+            try:
+                from gyxx_flow.ops.audit import AuditStore
+
+                self._audit_store = AuditStore(
+                    self.settings.data_root,
+                    environment=self._environment,
+                )
+            except Exception:  # noqa: BLE001 - telemetry is optional
+                return None
+        return self._audit_store
+
+    def _audit_cookie(
+        self,
+        account: Any,
+        state: str,
+        *,
+        probe_status: str,
+        probe_source: str,
+        detail: str = "",
+        context: RunContext | None = None,
+    ) -> None:
+        """Append one cookie observation.
+
+        Stores a value-free fingerprint plus the earliest declared expiry of the
+        required cookies, so failure patterns (how long a login really lasts,
+        and whether it dies before or after the declared expiry) can be derived
+        from history instead of guessed.  Errors are swallowed: gating must
+        never depend on the audit database.
+        """
+
+        try:
+            store = self._audit()
+            if store is None or not store.enabled:
+                return
+            store.record_cookie_event(
+                account_id=str(getattr(account, "account_id", "") or ""),
+                state=state,
+                snapshot=self._cookie_snapshot(account),
+                probe_status=probe_status,
+                probe_source=probe_source,
+                detail=detail,
+                workflow_id=str(getattr(context, "workflow_id", "") or ""),
+                run_id=str(getattr(context, "run_id", "") or ""),
+            )
+        except Exception:  # noqa: BLE001 - telemetry is advisory only
+            return
+
+    def _cookie_snapshot(self, account: Any) -> Any | None:
+        from gyxx_flow.ops.audit import snapshot_for_account
+
+        return snapshot_for_account(account)
+
+    def _within_notify_interval(self, state: Mapping[str, object]) -> bool:
+        raw = state.get("last_notified_at")
+        if not isinstance(raw, str):
+            return False
+        try:
+            last = datetime.fromisoformat(raw)
+        except ValueError:
+            return False
+        now = self._now()
+        if last.tzinfo is None:
+            last = last.replace(tzinfo=timezone.utc)
+        interval = _env_int(
+            self._environment,
+            GYXX_LOGIN_ALERT_NOTIFY_INTERVAL_SECONDS,
+            _DEFAULT_NOTIFY_INTERVAL_SECONDS,
+            minimum=0,
+        )
+        return (now - last).total_seconds() < interval
+
+    def _keepalive_success_is_fresh(self, status: Mapping[str, object] | None) -> bool:
+        if not isinstance(status, Mapping) or status.get("status") != "success":
+            return False
+        raw = status.get("last_success_at")
+        if not isinstance(raw, str):
+            return False
+        try:
+            last = datetime.fromisoformat(raw)
+        except ValueError:
+            return False
+        now = self._now()
+        if last.tzinfo is None:
+            last = last.replace(tzinfo=timezone.utc)
+        max_age = _env_float(
+            self._environment,
+            GYXX_COOKIE_GATE_MAX_STATUS_AGE_SECONDS,
+            _DEFAULT_MAX_STATUS_AGE_SECONDS,
+            minimum=_MIN_MAX_STATUS_AGE_SECONDS,
+        )
+        return (now - last).total_seconds() <= max_age
+
+    def _alert_recipients(self) -> list[tuple[str, str]]:
+        profile = (
+            self._environment.get(GYXX_LOGIN_ALERT_APP_PROFILE, "").strip()
+            or ANALYZER_NOTIFICATION_APP_PROFILE
+        )
+        override = [
+            value.strip()
+            for value in self._environment.get(GYXX_LOGIN_ALERT_OPEN_IDS, "").split(",")
+            if value.strip()
+        ]
+        if override:
+            return [(open_id, profile) for open_id in override]
+        try:
+            payload, _revision = NotificationRoutingStore(self.settings).snapshot()
+            for person in payload["people"].values():
+                if person.get("mobile") == DEFAULT_LOGIN_ALERT_MOBILE:
+                    binding = person.get("bindings", {}).get(profile)
+                    if binding:
+                        return [(str(binding["open_id"]), profile)]
+        except Exception:  # noqa: BLE001 - fall back to the constant recipient
+            pass
+        return [(ANALYZER_ACCEPTANCE_RECIPIENT_OPEN_ID, profile)]
+
+    def _acceptance_enabled(self) -> bool:
+        value = self._environment.get("GYXX_WORKFLOW_ACCEPTANCE", "")
+        return bool(value and value.strip().casefold() in {"1", "true", "yes", "on"})
+
+
+def gate_enabled(environment: Mapping[str, str] | None = None) -> bool:
+    values = os.environ if environment is None else environment
+    return values.get(GYXX_COOKIE_GATE, "").strip().casefold() not in _DISABLED_VALUES
+
+
+def apply_cookie_gates(
+    workflow: WorkflowDefinition,
+    *,
+    settings: Settings,
+    environment: Mapping[str, str] | None = None,
+    gate: CookieGate | None = None,
+) -> WorkflowDefinition:
+    """Inject one critical gate step per gated account in front of its users.
+
+    Gate steps are graph sources; each step that uses an account gains the
+    account's gate as an extra dependency.  A failed gate therefore skips its
+    dependents with ``dependency-failed`` through the regular engine, while
+    unrelated branches keep running.  Operator slices (``--rerun-step`` /
+    ``--resume-from``) are produced before this injection and stay gate-free.
+    """
+
+    if not workflow.steps:
+        return workflow
+    if gate is None:
+        try:
+            gate = CookieGate(
+                settings,
+                environment=os.environ if environment is None else environment,
+            )
+        except Exception:  # noqa: BLE001 - 无集成配置时不门控；步骤执行仍会显式报错
+            return workflow
+    account_steps = gate.collect_workflow_accounts(workflow)
+    if not account_steps:
+        return workflow
+    existing_ids = {step.step_id for step in workflow.steps}
+    gate_steps: list[StepDefinition] = []
+    for account_id in sorted(account_steps):
+        step_id = f"{_GATE_STEP_PREFIX}.{account_id}"
+        if step_id in existing_ids:
+            continue
+        gate_steps.append(
+            StepDefinition(
+                step_id,
+                CookieGateStepAction(gate, account_id),
+                timeout_seconds=_GATE_TIMEOUT_SECONDS,
+                max_attempts=1,
+            )
+        )
+    if not gate_steps:
+        return workflow
+    patched: list[StepDefinition] = []
+    for step in workflow.steps:
+        dependencies = list(dict.fromkeys(step.depends_on))
+        changed = False
+        for account_id, step_ids in account_steps.items():
+            gate_id = f"{_GATE_STEP_PREFIX}.{account_id}"
+            if step.step_id in step_ids and gate_id not in dependencies:
+                dependencies.append(gate_id)
+                changed = True
+        if changed:
+            step = replace(step, depends_on=tuple(dependencies))
+        patched.append(step)
+    return WorkflowDefinition(
+        workflow.workflow_id,
+        (*gate_steps, *patched),
+    )
+
+
+class CookieGateRecoveryManager:
+    """Re-run workflows blocked by expired cookies once the login recovers.
+
+    Runs inside the resident scheduler's maintenance loop.  For every account
+    with a non-empty ``pending_reruns`` list, it re-verifies the login state
+    (fresh keepalive success first, then a throttled live probe); on recovery
+    it relaunches each recorded workflow with its original business date
+    through the scheduler's own launcher, clears the pending list, and sends a
+    Feishu confirmation to the login owner.
+    """
+
+    def __init__(
+        self,
+        gate: CookieGate,
+        *,
+        launcher: Any,
+        probe_interval_seconds: float = _RECOVERY_PROBE_INTERVAL_SECONDS,
+        launch_cooldown_seconds: float = _RECOVERY_LAUNCH_COOLDOWN_SECONDS,
+    ) -> None:
+        self.gate = gate
+        self._launcher = launcher
+        self._probe_interval = probe_interval_seconds
+        self._launch_cooldown = launch_cooldown_seconds
+
+    def tick(self) -> list[dict[str, object]]:
+        """Attempt recovery reruns once; returns redacted outcomes per account."""
+
+        outcomes: list[dict[str, object]] = []
+        for account in self.gate.catalog.accounts:
+            state = self.gate._read_state(account)
+            pending = state.get("pending_reruns")
+            if not isinstance(pending, list) or not pending:
+                continue
+            now = self.gate._now()
+            if not self._launch_cooldown_passed(state, now):
+                continue
+            if not self._account_recovered(account, state, now):
+                continue
+            launched, remaining = self._launch_pending(account, pending, now)
+            state["pending_reruns"] = remaining
+            state["updated_at"] = _now_iso(self.gate._now)
+            state["last_recovered_at"] = state["updated_at"]
+            if not remaining:
+                state["state"] = STATE_HEALTHY
+                state["last_probe_status"] = "recovered-rerun"
+            if launched:
+                state["last_rerun_launch_at"] = state["updated_at"]
+                history = [
+                    item
+                    for item in state.get("rerun_history", [])
+                    if isinstance(item, dict)
+                ]
+                history.extend(launched)
+                state["rerun_history"] = history[-_RECOVERY_HISTORY_LIMIT:]
+            self.gate._save_state(account, state)
+            if launched:
+                self._notify_recovered(str(account.account_id), launched)
+            outcomes.append(
+                {
+                    "account_id": str(account.account_id),
+                    "launched": launched,
+                    "pending": len(remaining),
+                }
+            )
+        return outcomes
+
+    # ------------------------------------------------------------------
+    # internals
+    # ------------------------------------------------------------------
+
+    def _launch_cooldown_passed(
+        self, state: Mapping[str, object], now: datetime
+    ) -> bool:
+        raw = state.get("last_rerun_launch_at")
+        if not isinstance(raw, str):
+            return True
+        try:
+            last = datetime.fromisoformat(raw)
+        except ValueError:
+            return True
+        if last.tzinfo is None:
+            last = last.replace(tzinfo=timezone.utc)
+        return (now - last).total_seconds() >= self._launch_cooldown
+
+    def _account_recovered(
+        self,
+        account: Any,
+        state: dict[str, object],
+        now: datetime,
+    ) -> bool:
+        """Fresh keepalive success first; otherwise one throttled live probe."""
+
+        store = BrowserCookieStore(account.cookie_file, account.storage_state_file)
+        if _vault_cookie_valid(account, store) and self.gate._keepalive_success_is_fresh(
+            _read_keepalive_status(account)
+        ):
+            return True
+        raw = state.get("last_recovery_probe_at")
+        if isinstance(raw, str):
+            try:
+                last_probe = datetime.fromisoformat(raw)
+            except ValueError:
+                last_probe = None
+            if last_probe is not None:
+                if last_probe.tzinfo is None:
+                    last_probe = last_probe.replace(tzinfo=timezone.utc)
+                if (now - last_probe).total_seconds() < self._probe_interval:
+                    return False
+        state["last_recovery_probe_at"] = _now_iso(self.gate._now)
+        self.gate._save_state(account, state)
+        try:
+            outcome = self.gate.keepalive.refresh_account(str(account.account_id))
+        except Exception:  # noqa: BLE001 - 恢复探活失败留给下一轮
+            return False
+        return str(outcome.get("status", "error")) == "success"
+
+    def _launch_pending(
+        self,
+        account: Any,
+        pending: list[object],
+        now: datetime,
+    ) -> tuple[list[dict[str, object]], list[dict[str, object]]]:
+        known_ids = {
+            entry.workflow_id for entry in self.gate.workflow_catalog().workflows
+        }
+        launched: list[dict[str, object]] = []
+        remaining: list[dict[str, object]] = []
+        for item in pending:
+            if not isinstance(item, dict):
+                continue  # 控制台脚本等临时运行 id：不在 catalog 中，直接丢弃
+            workflow_id = item.get("workflow_id")
+            business_date = item.get("business_date")
+            if not isinstance(workflow_id, str) or not isinstance(business_date, str):
+                continue
+            if workflow_id not in known_ids:
+                continue
+            try:
+                process = self._launcher.launch(
+                    workflow_id, now, date.fromisoformat(business_date)
+                )
+            except Exception as exc:  # noqa: BLE001 - 保留 pending，下一轮重试
+                remaining.append(
+                    {
+                        "workflow_id": workflow_id,
+                        "business_date": business_date,
+                        "error": f"{type(exc).__name__}: {exc}"[:200],
+                    }
+                )
+                continue
+            close = getattr(self._launcher, "close", None)
+            if callable(close):
+                try:
+                    close(process)
+                except Exception:  # noqa: BLE001 - 关闭日志句柄失败不影响重跑
+                    pass
+            launched.append(
+                {
+                    "workflow_id": workflow_id,
+                    "business_date": business_date,
+                    "pid": int(getattr(process, "pid", 0) or 0),
+                    "launched_at": _now_iso(self.gate._now),
+                }
+            )
+        return launched, remaining
+
+    def _notify_recovered(
+        self,
+        account_id: str,
+        launched: list[dict[str, object]],
+    ) -> None:
+        lines = [
+            f"- {item['workflow_id']}（业务日期 {item['business_date']}）"
+            for item in launched
+        ]
+        message = (
+            "【GYXX 登录态已恢复】账号 "
+            f"{account_id} 登录态已恢复，已自动重跑被阻断的工作流：\n"
+            + "\n".join(lines)
+            + "\n如本次重跑仍有失败，请查看运行日志或工作流控制台。"
+        )
+        key = "-".join(
+            [account_id, *(str(item["workflow_id"]) for item in launched)]
+        )
+        for open_id, profile in self.gate._alert_recipients():
+            try:
+                self.gate._sender(
+                    user_id=open_id,
+                    text=message,
+                    idempotency_key=f"cookie-recovery:{key}",
+                    profile=profile,
+                )
+            except Exception:  # noqa: BLE001 - 确认通知失败不影响已发起的重跑
+                continue
+
+
+def _record_pending_rerun(
+    current: object,
+    context: RunContext,
+    blocked_at: str,
+) -> list[dict[str, object]]:
+    """Append one blocked run to the pending list, deduped and bounded."""
+
+    pending: list[dict[str, object]] = [
+        dict(item)
+        for item in (current if isinstance(current, list) else [])
+        if isinstance(item, dict)
+    ]
+    entry = {
+        "workflow_id": context.workflow_id,
+        "business_date": context.business_date.isoformat(),
+        "blocked_run_id": context.run_id,
+        "blocked_at": blocked_at,
+    }
+    pending = [
+        item
+        for item in pending
+        if not (
+            item.get("workflow_id") == entry["workflow_id"]
+            and item.get("business_date") == entry["business_date"]
+        )
+    ]
+    pending.append(entry)
+    return pending[-_MAX_PENDING_RERUNS:]
+
+
+def _recovery_evidence(status: Mapping[str, object] | None) -> str:
+    if not isinstance(status, Mapping):
+        return ""
+    if not status.get("login_attempted"):
+        return ""
+    reason = status.get("login_reason")
+    error_type = status.get("login_error_type")
+    exit_code = status.get("login_exit_code")
+    parts = ["自动账密重登已尝试但未成功"]
+    if isinstance(reason, str) and reason:
+        parts.append(reason)
+    if isinstance(error_type, str) and error_type:
+        parts.append(f"错误={error_type}")
+    if isinstance(exit_code, int) and not isinstance(exit_code, bool):
+        parts.append(f"退出码={exit_code}")
+    return "（" + "，".join(parts) + "）"
+
+
+def _expiry_message(
+    account_id: str,
+    mode: str,
+    context: RunContext,
+    detail: str,
+) -> str:
+    mode_note = (
+        "账号密码自动重登失败，需人工处理（如验证码）"
+        if mode == GATE_MODE_RECOVER
+        else "该平台仅支持扫码/验证码登录，无法自动恢复"
+    )
+    return (
+        "【GYXX 登录态失效】工作流被阻断，需要人工重新登录\n"
+        f"工作流: {context.workflow_id}（业务日期 {context.business_date.isoformat()}）\n"
+        f"账号: {account_id}\n"
+        f"原因: {detail}；{mode_note}\n"
+        f"请在服务器上执行: uv run gyxx accounts login {account_id}\n"
+        "（将打开浏览器窗口，完成扫码或验证码登录后自动保存并同步到全部成员脚本；"
+        "受影响的采集步骤本次已跳过，登录完成后下次运行自动恢复）"
+    )
+
+
+def _now_iso(now: Callable[[], datetime]) -> str:
+    return now().astimezone(timezone.utc).isoformat(timespec="seconds")
+
+
+def _env_int(
+    environment: Mapping[str, str],
+    name: str,
+    default: int,
+    *,
+    minimum: int,
+) -> int:
+    raw = environment.get(name, "").strip()
+    try:
+        value = int(raw)
+    except (TypeError, ValueError):
+        return default
+    return max(minimum, value)
+
+
+def _env_float(
+    environment: Mapping[str, str],
+    name: str,
+    default: float,
+    *,
+    minimum: float,
+) -> float:
+    raw = environment.get(name, "").strip()
+    try:
+        value = float(raw)
+    except (TypeError, ValueError):
+        return default
+    return max(minimum, value)
+
+
+__all__ = [
+    "DEFAULT_LOGIN_ALERT_MOBILE",
+    "GATE_MODE_BLOCKING",
+    "GATE_MODE_DEFER",
+    "GATE_MODE_RECOVER",
+    "GYXX_COOKIE_GATE",
+    "GYXX_LOGIN_ALERT_OPEN_IDS",
+    "CookieGate",
+    "CookieGateRecoveryManager",
+    "CookieGateStepAction",
+    "GateVerdict",
+    "apply_cookie_gates",
+    "gate_enabled",
+]
@@ -1,4 +1,4 @@
-"""Content metrics, creator, comment, report, and login workflows."""
+"""Content metrics, creator, comment, and report workflows."""

 from __future__ import annotations

@@ -14,7 +14,6 @@ CONTENT_SCHEDULED_WORKFLOW_IDS = (
    "content.metrics.self_operated.daily",
    "content.metrics.backfill",
    "content.marketing_report.daily",
-    "content.relogin.weekly",
    "content.creator_report.monthly",
    "content.summary.monthly",
    "content.notes_master.daily",
@@ -28,7 +27,6 @@ CONTENT_COMMENT_TIMEOUT_SECONDS = 6 * 60 * 60
 # weakening the module-wide write lock.
 CONTENT_NOTES_MASTER_LOCK_TIMEOUT_SECONDS = 6 * 60 * 60
 CONTENT_RESOURCE = "module:content_marketing"
-_RELOGIN_WORKFLOW_ID = "content.relogin.weekly"
 _PARALLEL_COMMENT_WORKFLOW_ID = "content.comments.weekly"
 # Keep both long-running daily metrics graphs on workflow-scoped resources so
 # their collection steps do not block unrelated content workflows.
@@ -94,7 +92,6 @@ class ContentMarketingModule:
                    ),
                    resource=workflow_resource,
                    command_factory=command_factory,
-                    official_notification=workflow_id == _RELOGIN_WORKFLOW_ID,
                    independent_step_resources=(
                        workflow_id in {
                            _PARALLEL_COMMENT_WORKFLOW_ID,
@@ -77,23 +77,15 @@ def is_logged_in(page: Page) -> bool:


 def maybe_wait_for_login(page: Page, login_timeout: int) -> None:
-    """检测登录态,未登录就等用户扫码。B 站登录页 URL 含 passport.bilibili.com/login。"""
+    """检测登录态；未登录立即降级为游客态，不再等待人工扫码。
+
+    登录统一由账号 vault 管理（gyxx accounts login bilibili-content），
+    由工作流执行前的 cookie 网关保证登录态。
+    """
    if is_logged_in(page):
        return
-    log("B 站未登录 (无 SESSDATA cookie)。请在弹出的浏览器窗口扫码登录。")
-    # 跳到登录页,确保用户能看到二维码
-    try:
-        page.goto("https://passport.bilibili.com/login", wait_until="domcontentloaded", timeout=30000)
-    except Exception as exc:
-        log(f"  跳登录页失败 (继续等当前页扫码): {exc}")
-    deadline = time.time() + login_timeout
-    while time.time() < deadline:
-        if is_logged_in(page):
-            log("登录成功,SESSDATA 已写入")
-            save_state(page)
-            return
-        page.wait_for_timeout(2000)
-    log(f"  [WARN] 等待 {login_timeout}s 仍未登录,继续以游客态采集 (评论只会拿到热门 3-5 条)")
+    log("  [WARN] B 站未登录 (无 SESSDATA cookie)，降级为游客态采集 (评论只会拿到热门 3-5 条)。"
+        "如需全量评论请执行 gyxx accounts login bilibili-content")


 def safe_filename(value: str) -> str:
@@ -437,48 +429,17 @@ def write_outputs(url: str, comments: list[dict[str, Any]], result: dict[str, An
    return json_path, csv_path


-def login_only(login_timeout: int) -> int:
-    """只触发登录(headed),登录完成立刻退出。给 relogin_bilibili.py 用。"""
-    def action(page: Page) -> None:
-        page.goto("https://passport.bilibili.com/login", wait_until="domcontentloaded", timeout=30000)
-        page.wait_for_timeout(1500)
-        maybe_wait_for_login(page, login_timeout)
-
-    PROFILE_DIR.mkdir(parents=True, exist_ok=True)
-    with DynamicSession(
-        headless=False,
-        real_chrome=True,
-        user_data_dir=str(PROFILE_DIR),
-        locale="zh-CN",
-        timezone_id="Asia/Shanghai",
-        timeout=90000,
-        network_idle=False,
-        disable_resources=False,
-        google_search=False,
-        page_setup=restore_cookies,
-        max_pages=1,
-    ) as session:
-        session.fetch("https://passport.bilibili.com/login", page_action=action, wait=1000)
-    log("[login-only] 完成")
-    return 0
-
-
 def main() -> int:
    parser = argparse.ArgumentParser(description="Scrape Bilibili video comments with a real browser and API.")
-    parser.add_argument("url", help="Bilibili video URL (传 dummy 'login' 配合 --login-only 也可)")
+    parser.add_argument("url", help="Bilibili video URL")
    parser.add_argument("--max-reply-pages", type=int, default=200, help="max sub-reply pages per top-level comment")
    parser.add_argument("--max-pages", type=int, default=200, help="最多抓取一级评论页数")
    parser.add_argument("--delay", type=float, default=0.25, help="接口请求间隔秒数")
    parser.add_argument("--headless", action="store_true", help="无头模式")
    parser.add_argument("--login-timeout", type=int, default=0,
-                        help="未登录时等待扫码秒数(>0 启用登录检测;默认 0 沿用游客态)")
-    parser.add_argument("--login-only", action="store_true",
-                        help="只触发登录(headed)然后退出,给 relogin_bilibili 用")
+                        help="保留参数（登录等待已移除；未登录将直接以游客态采集）")
    args = parser.parse_args()

-    if args.login_only:
-        return login_only(args.login_timeout if args.login_timeout > 0 else 300)
-
    comments, result = scrape_comments(
        args.url, args.max_pages, args.delay, args.headless,
        args.max_reply_pages, args.login_timeout,
@@ -16,6 +16,7 @@ import time
 from datetime import datetime
 from pathlib import Path

+from gyxx_flow.adapters.scrapling import is_browser_closed_error
 from gyxx_flow.core.exit_codes import COOKIE_SKIP_EXIT_CODE
 from gyxx_flow.modules.content_marketing import chanmama_export
 from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
@@ -52,6 +53,11 @@ REFRESH_WAIT_SECONDS = int(os.getenv("CHANMAMA_REFRESH_WAIT_SECONDS", "600"))
 CHANMAMA_LIST_READY_TIMEOUT_SECONDS = int(
    os.getenv("CHANMAMA_LIST_READY_TIMEOUT_SECONDS", "45")
 )
+# 公告/活动浮层可能盖住视频列表导致渲染判定一直不满足；渲染等待期间按
+# 该间隔无交互清理浮层（派发 ESC + 移除 fixed/absolute 浮层节点）。
+CHANMAMA_OVERLAY_DISMISS_INTERVAL_SECONDS = float(
+    os.getenv("CHANMAMA_OVERLAY_DISMISS_INTERVAL_SECONDS", "10")
+)
 # 蝉妈妈导出接口实测可能在 15 分钟边界后才把 Excel 写入下载目录；把轮询
 # 上限做成可配置，并默认留出 20 分钟，避免文件已生成却被脚本判定为失败。
 CHANMAMA_EXPORT_WAIT_SECONDS = int(
@@ -99,16 +105,20 @@ def create_driver():
    profile_dir = str(PATHS.browser_profile_dir)
    os.makedirs(profile_dir, exist_ok=True)
    cdp_url = os.getenv("GYXX_BROWSER_CDP_URL") or None
+    # 蝉妈妈是国内站且其 WAF 会拉黑代理出口(Clash 节点 TLS 直接 reset，
+    # 导出 API 全断、确认弹窗永远不弹)；强制浏览器直连。
+    direct_flags = ["--no-proxy-server"]
+    # StealthySession(隐藏指纹补丁)下蝉妈妈前端会静默压制功能: 页面能开
+    # 但导出确认弹窗不渲染(只剩空壳遮罩)。真 Chrome + 非 stealthy + 直连
+    # 已实测弹窗完整渲染(含取消/导出按钮)，不再使用 stealthy。
    try:
        return ScraplingDriver.start(
            profile_dir=profile_dir,
            download_dir=DOWNLOAD_DIR,
            cdp_url=cdp_url,
-            stealthy=True,
-            # 蝉妈妈新版 Vue 页面与 StealthySession + 系统 Chrome 组合
-            # 不兼容：HTML 能返回，但 #app 不会挂载。使用 Scrapling
-            # 自带 Chromium 仍保留隐藏指纹会话，且不影响已连接的 CDP。
-            **({"real_chrome": False} if cdp_url is None else {}),
+            stealthy=False,
+            real_chrome=True,
+            extra_flags=direct_flags,
        )
    except Exception as exc:
        if (
@@ -125,23 +135,25 @@ def create_driver():
            profile_dir=profile_dir,
            download_dir=DOWNLOAD_DIR,
            cdp_url=None,
-            stealthy=True,
-            real_chrome=False,
+            stealthy=False,
+            real_chrome=True,
+            extra_flags=direct_flags,
        )


 def save_cookies(driver):
-    """保存 cookie 到文件"""
+    """保存 cookie 到文件。
+
+    写裸列表格式,与账号 vault 同步(sync_binding_state)和适配器
+    BrowserCookieStore 的绑定 cookie 文件格式保持一致 —— 同一文件
+    被三方读写,字典格式会让适配器下次启动/重建时直接拒绝加载。
+    """
    try:
        cookies = driver.get_cookies()
-        data = {
-            "cookies": cookies,
-            "timestamp": time.time(),
-        }
        cookie_path = Path(COOKIE_FILE)
        cookie_path.parent.mkdir(parents=True, exist_ok=True)
        temp_path = cookie_path.with_name(f".{cookie_path.name}.tmp")
-        temp_path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+        temp_path.write_text(json.dumps(cookies, ensure_ascii=False, indent=2), encoding="utf-8")
        os.replace(temp_path, cookie_path)
        print(f"💾 已保存 cookie ({len(cookies)} 条) 到: {COOKIE_FILE}")
        return True
@@ -153,6 +165,9 @@ def save_cookies(driver):
 def load_cookies(driver):
    """从文件加载 cookie 到浏览器

+    兼容两种格式: 裸列表(vault/适配器格式,按文件 mtime 判断时效)与
+    旧字典格式({"cookies": [...], "timestamp": ...})。
+
    Returns:
        bool: cookie 是否有效（有效期内）
    """
@@ -161,16 +176,20 @@ def load_cookies(driver):
        return False

    try:
-        data = json.loads(Path(COOKIE_FILE).read_text(encoding="utf-8"))
+        payload = json.loads(Path(COOKIE_FILE).read_text(encoding="utf-8"))
+        if isinstance(payload, list):
+            cookies = payload
+            timestamp = os.path.getmtime(COOKIE_FILE)
+        else:
+            cookies = payload.get("cookies", [])
+            timestamp = payload.get("timestamp", 0)

-        timestamp = data.get("timestamp", 0)
        age = time.time() - timestamp

        if age > COOKIE_MAX_AGE:
            print(f"⏰ cookie 已过期 ({int(age // 86400)} 天前)，需要重新登录")
            return False

-        cookies = data.get("cookies", [])
        if not cookies:
            print("📭 cookie 为空，需要重新登录")
            return False
@@ -297,7 +316,10 @@ def handle_captcha(driver) -> bool:
 def verify_login(driver):
    """验证是否已登录

-    通过访问目标页面，检查是否出现登录相关的元素
+    通过访问目标页面，检查是否出现登录相关的元素。
+    cookie 失效时蝉妈妈会把目标页重定向到 login.html,但重定向常在
+    首屏渲染后数秒才发生——等待 URL 稳定后再判定,避免误判已登录
+    (误判会跳过重新登录,导致后续导航一直超时)。
    """
    try:
        # 跳转到目标页面
@@ -305,10 +327,12 @@ def verify_login(driver):
        WebDriverWait(driver, 12).until(
            EC.presence_of_element_located((By.TAG_NAME, "body"))
        )
-        time.sleep(3)
+        # cookie 失效时的重定向在首屏渲染后 3~20 秒才发生,固定等满 10 秒
+        # 再判定,不能因 URL 短暂未变而提前判为已登录
+        time.sleep(10)
+        cur_lower = driver.current_url.lower()

        # 检查是否跳转回登录页 / 注册页（cookie 失效时蝉妈妈会重定向到 login.html 或 register.html）
-        cur_lower = driver.current_url.lower()
        if "login" in cur_lower or "register" in cur_lower:
            print(f"🔒 检测到被重定向到登录/注册页 ({driver.current_url})，cookie 已失效")
            return False
@@ -458,6 +482,180 @@ def login(driver):
        return False


+def _dismiss_overlays(driver) -> int:
+    """无交互清理遮挡浮层：派发 ESC 并移除 fixed/absolute 定位的弹层。"""
+    try:
+        removed = int(driver.execute_script(_OVERLAY_DISMISS_SCRIPT) or 0)
+    except Exception as error:
+        print(f"⚠️ 浮层清理脚本执行失败(忽略): {error}")
+        return 0
+    if removed:
+        print(f"🧹 已清理 {removed} 个遮挡浮层")
+    return removed
+
+
+def _dismiss_export_stage_overlays(driver) -> int:
+    """导出阶段清理弹窗(ESC + 移除弹层节点)。
+
+    只在弹窗不可确认时调用；返回移除的弹窗数。
+    """
+    try:
+        removed = int(driver.execute_script(_EXPORT_STAGE_DISMISS_SCRIPT) or 0)
+    except Exception as error:
+        print(f"⚠️ 导出阶段弹窗清理脚本执行失败(忽略): {error}")
+        return 0
+    if removed > 0:
+        print(f"🧹 导出阶段清理 {removed} 个遮挡弹窗")
+    return removed
+
+
+def _export_dialog_state(driver) -> str:
+    """探测导出阶段弹窗状态: confirmable / present / none。"""
+    try:
+        state = driver.execute_script(_EXPORT_DIALOG_STATE_SCRIPT)
+    except Exception:
+        return "none"
+    return state if state in {"confirmable", "present", "none"} else "none"
+
+
+def _log_export_dialog_texts(driver) -> None:
+    """把当前可见弹窗的类名与文字打进日志，用于定位弹窗身份。"""
+    try:
+        texts = driver.execute_script(_EXPORT_DIALOG_TEXT_SCRIPT) or []
+    except Exception:
+        return
+    for i, text in enumerate(texts[:4], 1):
+        print(f"🪟 弹窗{i}内容: {text}")
+
+
+def _click_export_confirm(driver) -> bool:
+    """点击导出确认弹窗中的【导出】按钮，成功返回 True。"""
+    # 关键：用 contains(.,'导出') 匹配后代文本（"导出"在 <span> 内）
+    confirm_selectors = [
+        # 精确：主按钮(primary) + 含"导出"文字
+        "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(.,'导出')]",
+        # 弹窗内任意含"导出"的按钮
+        "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(@class,'el-button--primary')][contains(.,'导出')]",
+        # 兼容"确定"/"确认"
+        "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(.,'确定')]",
+        "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(.,'确认')]",
+    ]
+    for sel in confirm_selectors:
+        try:
+            btn = driver.find_element(By.XPATH, sel)
+            if btn.is_displayed():
+                # 原生点击优先(可触发内层元素的 Vue @click)，失败再退 JS 合成点击
+                try:
+                    btn.click()
+                except Exception:
+                    driver.execute_script("arguments[0].click();", btn)
+                print(f"✅ 已在弹窗中点击导出按钮: {sel}")
+                return True
+        except Exception:
+            continue
+    return False
+
+
+def _reclick_export_control(driver) -> bool:
+    """重新触发一次导出按钮(第一次点击可能已被弹窗吞掉)。
+
+    优先 Playwright 原生点击(在元素命中点派发可信鼠标事件，能触发绑定
+    在按钮内层 span/div 上的 Vue @click)；JS 合成 click 的目标是按钮自身，
+    内层监听器不会触发——蝉妈妈改版后导出点击无声失效就是这个原因。
+    """
+    if _native_click_export_button(driver):
+        return True
+    try:
+        return bool(driver.execute_script(_CLICK_EXPORT_CONTROL_SCRIPT))
+    except Exception:
+        return False
+
+
+_EXPORT_BUTTON_XPATHS = [
+    "//button[contains(.,'导出数据')]",
+    "//button[contains(.,'批量导出')]",
+    "//button[contains(.,'导出视频')]",
+    "//*[contains(@class,'export') and (self::button or self::a or @role='button')]",
+]
+
+
+def _native_click_export_button(driver) -> bool:
+    """用原生可信点击触发导出按钮，成功返回 True。"""
+    for xp in _EXPORT_BUTTON_XPATHS:
+        try:
+            elements = driver.find_elements(By.XPATH, xp)
+        except Exception:
+            continue
+        for el in elements:
+            try:
+                if el.is_displayed():
+                    el.click()
+                    return True
+            except Exception:
+                continue
+    return False
+
+
+def _wait_video_list_ready(driver, timeout_seconds: float) -> None:
+    """等待视频列表渲染就绪，期间周期性清理遮挡浮层。
+
+    语义与 WebDriverWait.until 保持一致：浏览器侧的暂时性异常只记录
+    不中断，到达 deadline 后抛 TimeoutException。
+    """
+    timeout = max(0.0, float(timeout_seconds))
+    deadline = time.monotonic() + timeout
+    next_dismiss_at = 0.0
+    last_error: Exception | None = None
+    while True:
+        try:
+            if driver.execute_script(_VIDEO_LIST_READY_SCRIPT):
+                return
+        except Exception as error:
+            last_error = error
+        now = time.monotonic()
+        if now >= deadline:
+            detail = f": {last_error}" if last_error else ""
+            raise TimeoutException(
+                f"condition was not met within {timeout}s{detail}"
+            ) from last_error
+        if now >= next_dismiss_at:
+            _dismiss_overlays(driver)
+            next_dismiss_at = now + CHANMAMA_OVERLAY_DISMISS_INTERVAL_SECONDS
+        time.sleep(0.5)
+
+
+def _dump_render_failure(driver, url: str) -> None:
+    """渲染失败时落截图与页面 HTML，供事后定位是浮层/验证码还是改版。"""
+    try:
+        stamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+        diag_dir = Path(DOWNLOAD_DIR) / "_diag"
+        diag_dir.mkdir(parents=True, exist_ok=True)
+        screenshot = diag_dir / f"render_failure_{stamp}.png"
+        driver.save_screenshot(str(screenshot))
+        (diag_dir / f"render_failure_{stamp}.html").write_text(
+            driver.page_source, encoding="utf-8"
+        )
+        print(f"🩺 渲染失败诊断已保存: {screenshot}")
+    except Exception as error:
+        print(f"⚠️ 渲染失败诊断保存失败: {error}")
+
+
+def _dump_export_failure(driver) -> None:
+    """导出失败时落截图与页面 HTML，供事后定位弹窗/按钮的实际形态。"""
+    try:
+        stamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+        diag_dir = Path(DOWNLOAD_DIR) / "_diag"
+        diag_dir.mkdir(parents=True, exist_ok=True)
+        screenshot = diag_dir / f"export_failure_{stamp}.png"
+        driver.save_screenshot(str(screenshot))
+        (diag_dir / f"export_failure_{stamp}.html").write_text(
+            driver.page_source, encoding="utf-8"
+        )
+        print(f"🩺 导出失败诊断已保存: {screenshot}")
+    except Exception as error:
+        print(f"⚠️ 导出失败诊断保存失败: {error}")
+
+
 def navigate_to_target(driver, url):
    """跳转到目标博主视频页面"""
    last_error = None
@@ -475,12 +673,11 @@ def navigate_to_target(driver, url):
                )
            time.sleep(3)  # 等待页面数据加载
            # 截图中的页面虽然 URL 正确，但正文尚未渲染；在交给导出逻辑
-            # 前确认虚拟视频列表已经出现，空白页则有限重载一次。
-            WebDriverWait(driver, CHANMAMA_LIST_READY_TIMEOUT_SECONDS).until(
-                lambda current: bool(
-                    current.execute_script(_VIDEO_LIST_READY_SCRIPT)
-                )
-            )
+            # 前确认虚拟视频列表已经出现，空白页则有限重载一次。浮层多在
+            # 首屏挂载后才弹出：进等待前先清一轮，等待期间按间隔再清，
+            # 避免延迟弹出的公告/活动浮层永久挡住渲染判定。
+            _dismiss_overlays(driver)
+            _wait_video_list_ready(driver, CHANMAMA_LIST_READY_TIMEOUT_SECONDS)
            print(f"✅ 已到达并完成渲染目标页面: {current_url}")
            return True
        except Exception as exc:
@@ -499,6 +696,7 @@ def navigate_to_target(driver, url):
            except Exception as reset_exc:
                print(f"⚠️ 蝉妈妈页面重置失败，继续重试目标页: {reset_exc}")
            time.sleep(1)
+    _dump_render_failure(driver, url)
    raise RuntimeError(f"蝉妈妈目标页加载失败: {url}; {last_error}") from last_error


@@ -539,6 +737,206 @@ def _find_video_table(driver):
    return WebDriverWait(driver, 15).until(locate)


+# 蝉妈妈公告/活动弹窗以 fixed/absolute 浮层盖住页面，且类名不稳定、未必
+# 有关闭按钮。这里不点叉：派发 ESC 键事件（多数组件库监听 document 关闭
+# 弹窗），再移除定位浮层节点并解除其对 html/body 的滚动锁；已知类名未
+# 命中时兜底移除 z-index 高且覆盖大半屏的 fixed 层。列表区
+# (.aweme-record-wrapper/.list-box/.list-auth-wrap-block) 内的元素永不命中。
+# .v-modal 是 ElementUI 直接挂在 <body> 下的灰色遮罩，与弹窗本体分离，
+# 必须显式清理，否则弹窗删掉后页面仍然灰屏、无法滚动；滚动锁 class
+# (el-popup-parent--hidden 等)同理。
+# 关键约束：只移除"可见"的浮层。页面常驻 90+ 个 display:none 的隐藏弹窗
+# 模板(含导出确认框的模板)——删掉它们后 Vue 只会在游离节点上"打开"弹窗，
+# 页面上永远渲染不出来(导出只剩灰遮罩)，因此 display:none/零尺寸节点
+# 一律跳过。
+_OVERLAY_DISMISS_SCRIPT = r"""
+() => {
+  const esc = new KeyboardEvent('keydown', {
+    key: 'Escape', code: 'Escape', keyCode: 27, which: 27, bubbles: true,
+  });
+  document.dispatchEvent(esc);
+  window.dispatchEvent(esc);
+  let removed = 0;
+  const seen = new Set();
+  const isOverlay = (el) => {
+    const style = window.getComputedStyle(el);
+    return style.position === 'fixed' || style.position === 'absolute';
+  };
+  const drop = (el) => {
+    if (seen.has(el) || !isOverlay(el)) return;
+    const style = window.getComputedStyle(el);
+    if (style.display === 'none') return;
+    const rect = el.getBoundingClientRect();
+    if (rect.width === 0 || rect.height === 0) return;
+    if (el.closest('.aweme-record-wrapper,.list-box,.list-auth-wrap-block')) return;
+    seen.add(el);
+    el.remove();
+    removed += 1;
+  };
+  const selectors = [
+    '.el-dialog__wrapper', '.el-overlay', '.el-message-box__wrapper',
+    '.ant-modal-mask', '.ant-modal-wrap', '.modal-mask', '.modal-dialog',
+    '.v-modal', '.v-modal-inner',
+    '[class*="popup"]', '[class*="dialog"]', '[class*="overlay"]', '[class*="mask"]',
+  ];
+  for (const selector of selectors) {
+    document.querySelectorAll(selector).forEach(drop);
+  }
+  if (removed === 0) {
+    document.querySelectorAll('body *').forEach((el) => {
+      const style = window.getComputedStyle(el);
+      const zIndex = Number.parseInt(style.zIndex, 10);
+      const rect = el.getBoundingClientRect();
+      if (Number.isFinite(zIndex) && zIndex > 500
+          && rect.height > window.innerHeight * 0.5) {
+        drop(el);
+      }
+    });
+  }
+  document.body.classList.remove(
+    'el-popup-parent--hidden', 'el-popup-parent--overflow-hidden', 'overflow-hidden'
+  );
+  document.documentElement.classList.remove(
+    'el-popup-parent--hidden', 'el-popup-parent--overflow-hidden', 'overflow-hidden'
+  );
+  document.documentElement.style.overflow = '';
+  document.body.style.overflow = '';
+  document.body.style.paddingRight = '';
+  return removed;
+}
+"""
+
+
+# 导出阶段专用清理：与 _OVERLAY_DISMISS_SCRIPT 同一套清理思路，但
+# 调用时机由 _export_dialog_state 决定——只在弹窗"没有可点的确认按钮"时
+# 才执行(确认框可点时直接点导出，绝不清)。三件事：
+# 1) 先 ESC + 点弹窗自带的关闭(×)按钮，走框架的正常关闭路径；
+# 2) 再移除残留的弹层节点——必须包含 .v-modal：ElementUI 的灰色遮罩直接
+#    挂在 <body> 下、与弹窗本体分离，漏掉它就是"删了弹窗页面仍灰屏"；
+# 3) 最后解除滚动锁：清掉 body/html 上的 el-popup-parent--hidden 等
+#    框架加的 overflow 锁 class 与内联样式，否则页面删完遮罩也不能滚动。
+_EXPORT_STAGE_DISMISS_SCRIPT = r"""
+() => {
+  const visible = (el) => {
+    const style = window.getComputedStyle(el);
+    const rect = el.getBoundingClientRect();
+    return style.display !== 'none' && style.visibility !== 'hidden'
+      && style.opacity !== '0' && rect.width > 0 && rect.height > 0;
+  };
+  const esc = new KeyboardEvent('keydown', {
+    key: 'Escape', code: 'Escape', keyCode: 27, which: 27, bubbles: true,
+  });
+  document.dispatchEvent(esc);
+  window.dispatchEvent(esc);
+  // 先走正常关闭：点弹窗自己的 × / 关闭按钮
+  let closed = 0;
+  document.querySelectorAll(
+    '.el-dialog__wrapper [class*="close"],.el-message-box__wrapper [class*="close"],'
+    + '.ant-modal-wrap [class*="close"],[class*="dialog"] [class*="close"],'
+    + '[class*="modal"] [class*="close"],[class*="popup"] [class*="close"]'
+  ).forEach((el) => {
+    if (visible(el)) { el.click(); closed += 1; }
+  });
+  let removed = 0;
+  const seen = new Set();
+  const drop = (el) => {
+    if (seen.has(el)) return;
+    const style = window.getComputedStyle(el);
+    if (style.position !== 'fixed' && style.position !== 'absolute') return;
+    if (style.display === 'none') return;
+    const rect = el.getBoundingClientRect();
+    if (rect.width === 0 || rect.height === 0) return;
+    if (el.closest('.aweme-record-wrapper,.list-box,.list-auth-wrap-block')) return;
+    seen.add(el);
+    el.remove();
+    removed += 1;
+  };
+  const selectors = [
+    '.el-dialog__wrapper', '.el-overlay', '.el-message-box__wrapper',
+    '.ant-modal-mask', '.ant-modal-wrap', '.modal-mask', '.modal-dialog',
+    '.v-modal', '.v-modal-inner',
+    '[class*="popup"]', '[class*="dialog"]', '[class*="overlay"]', '[class*="mask"]',
+  ];
+  for (const selector of selectors) {
+    document.querySelectorAll(selector).forEach(drop);
+  }
+  if (removed === 0) {
+    document.querySelectorAll('body *').forEach((el) => {
+      const style = window.getComputedStyle(el);
+      const zIndex = Number.parseInt(style.zIndex, 10);
+      const rect = el.getBoundingClientRect();
+      if (Number.isFinite(zIndex) && zIndex > 500
+          && rect.height > window.innerHeight * 0.5) {
+        drop(el);
+      }
+    });
+  }
+  // 解除滚动锁：框架加的锁 class + 内联 overflow 一起清
+  document.body.classList.remove(
+    'el-popup-parent--hidden', 'el-popup-parent--overflow-hidden', 'overflow-hidden'
+  );
+  document.documentElement.classList.remove(
+    'el-popup-parent--hidden', 'el-popup-parent--overflow-hidden', 'overflow-hidden'
+  );
+  document.documentElement.style.overflow = '';
+  document.body.style.overflow = '';
+  document.body.style.paddingRight = '';
+  return removed + closed;
+}
+"""
+
+
+# 导出阶段弹窗状态机：'confirmable'=有可见弹窗且弹窗里有可点的
+# 导出/确定/确认按钮(直接点，不能清)；'present'=弹窗可见但没有可点
+# 的确认按钮(空壳遮罩/公告，ESC 关不掉就强删)；'none'=没有弹窗。
+_EXPORT_DIALOG_STATE_SCRIPT = r"""
+() => {
+  const visible = (el) => {
+    const style = window.getComputedStyle(el);
+    const rect = el.getBoundingClientRect();
+    return style.display !== 'none' && style.visibility !== 'hidden'
+      && style.opacity !== '0' && rect.width > 0 && rect.height > 0;
+  };
+  const roots = [...document.querySelectorAll(
+    '.el-dialog__wrapper,.el-message-box__wrapper,.el-overlay,'
+    + '.ant-modal-wrap,.ant-modal-mask,[role="dialog"],'
+    + '[class*="dialog"],[class*="modal"],[class*="popup"]'
+  )].filter(visible);
+  for (const root of roots) {
+    const confirmable = [...root.querySelectorAll(
+      'button,a,[role="button"],[class*="confirm"],[class*="primary"]'
+    )].some((el) => {
+      if (!visible(el)) return false;
+      const text = (el.innerText || el.value || '').replace(/\s+/g, '');
+      return /导出|确定|确认/.test(text);
+    });
+    if (confirmable) return 'confirmable';
+  }
+  return roots.length ? 'present' : 'none';
+}
+"""
+
+
+# 抓取当前可见弹窗的文字内容(每窗截前 160 字)，写进日志用于定位弹窗身份。
+_EXPORT_DIALOG_TEXT_SCRIPT = r"""
+() => {
+  const visible = (el) => {
+    const style = window.getComputedStyle(el);
+    const rect = el.getBoundingClientRect();
+    return style.display !== 'none' && style.visibility !== 'hidden'
+      && style.opacity !== '0' && rect.width > 0 && rect.height > 0;
+  };
+  return [...document.querySelectorAll(
+    '.el-dialog__wrapper,.el-message-box__wrapper,.el-overlay,'
+    + '.ant-modal-wrap,[role="dialog"],[class*="dialog"],[class*="modal"],[class*="popup"]'
+  )].filter(visible).map((el) => {
+    const text = (el.innerText || '').replace(/\s+/g, ' ').trim();
+    const classes = String(el.className || '').slice(0, 80);
+    return `[${classes}] ${text.slice(0, 160)}`;
+  });
+}
+"""
+
 _VIDEO_LIST_READY_SCRIPT = r"""
 () => {
  const visible = (el) => {
@@ -909,25 +1307,107 @@ def _max_excel_mtime():


 def refresh_then_export_accounts(driver, urls, wait_seconds=REFRESH_WAIT_SECONDS):
-    """两阶段执行：先刷新所有账号全部分页，再等待并重新导出。"""
+    """两阶段执行：先刷新所有账号全部分页，再等待并重新导出。
+
+    运行中途登录态失效(目标页被重定向到登录页)时,自动用账号密码重新
+    登录并重试当前账号一次;浏览器/页面被外部关闭(如进程树被清理)时,
+    自动重建浏览器并恢复会话后重试当前账号一次。两种自愈整套运行各最多
+    一次,避免死循环。返回 (记录列表, 刷新摘要, 最终生效的驱动) —— 驱动
+    可能在重建后已不是调用方传入的那个,调用方需用返回值收尾关闭。
+    """
    refresh_summaries = []
+    relogin_state = {"attempted": False}
+    rebuild_state = {"attempted": False}
+    driver_box = {"driver": driver}
+
+    def _relogin_if_expired(exc: BaseException) -> bool:
+        """登录态失效类异常时自动重登一次;返回是否已恢复登录。"""
+        if relogin_state["attempted"] or "登录态失效" not in str(exc):
+            return False
+        relogin_state["attempted"] = True
+        print("🔐 检测到登录态失效,尝试账号密码自动重新登录...")
+        try:
+            ok = login(driver_box["driver"])
+        except Exception as login_exc:
+            print(f"⚠️ 自动重新登录异常: {login_exc}")
+            return False
+        if ok:
+            print("✅ 自动重新登录成功,重试当前账号")
+        return bool(ok)
+
+    def _restore_session(current) -> bool:
+        """重建浏览器后按 profile → cookie 文件 → 账密的顺序恢复登录态。"""
+        try:
+            if verify_login(current):
+                return True
+            if load_cookies(current) and verify_login(current):
+                return True
+            return bool(login(current))
+        except Exception as exc:
+            print(f"⚠️ 重建后会话恢复异常: {exc}")
+            return False
+
+    def _rebuild_if_browser_closed(exc: BaseException) -> bool:
+        """浏览器/页面被外部关闭类异常时自愈一次;返回是否已恢复。
+
+        先尝试同一 context 重开新页(页面级死亡,登录态保留、代价最小);
+        context 也死了才整套重建驱动并按 profile → cookie → 账密恢复会话。
+        """
+        if rebuild_state["attempted"] or not is_browser_closed_error(exc):
+            return False
+        rebuild_state["attempted"] = True
+        print("🌐 检测到浏览器页面被关闭,尝试自愈...")
+        current = driver_box["driver"]
+        try:
+            current.reopen_page()
+            print("✅ 浏览器 context 存活,已重开新页面,重试当前账号")
+            return True
+        except Exception as reopen_exc:
+            print(f"ℹ️ 页面重开失败({reopen_exc}),改为整套重建浏览器...")
+        try:
+            current.quit()
+        except Exception:
+            # 旧驱动随浏览器一起死亡,清理失败不影响重建。
+            pass
+        try:
+            driver_box["driver"] = create_driver()
+        except Exception as create_exc:
+            print(f"⚠️ 浏览器重建失败: {create_exc}")
+            return False
+        if not _restore_session(driver_box["driver"]):
+            print("⚠️ 重建后会话恢复失败,放弃自动重试")
+            return False
+        print("✅ 浏览器重建完成,重试当前账号")
+        return True
+
+    def _empty_summary(exc: BaseException) -> dict:
+        return {
+            "clicked": 0,
+            "pending": 0,
+            "zero_without_button": 0,
+            "unconfirmed": 0,
+            "pages": 0,
+            "error": str(exc),
+        }
+
    for index, url in enumerate(urls, 1):
        print(f"\n{'=' * 50}")
        print(f"🔄 刷新阶段 博主 {index}/{len(urls)}: {url}")
        print(f"{'=' * 50}")
        try:
-            navigate_to_target(driver, url)
-            summary = refresh_zero_exposure_videos(driver)
+            navigate_to_target(driver_box["driver"], url)
+            summary = refresh_zero_exposure_videos(driver_box["driver"])
        except Exception as exc:
-            print(f"⚠️ 博主 {index} 刷新扫描异常: {exc},继续下一个")
-            summary = {
-                "clicked": 0,
-                "pending": 0,
-                "zero_without_button": 0,
-                "unconfirmed": 0,
-                "pages": 0,
-                "error": str(exc),
-            }
+            if _relogin_if_expired(exc) or _rebuild_if_browser_closed(exc):
+                try:
+                    navigate_to_target(driver_box["driver"], url)
+                    summary = refresh_zero_exposure_videos(driver_box["driver"])
+                except Exception as retry_exc:
+                    print(f"⚠️ 博主 {index} 自愈后刷新仍失败: {retry_exc},继续下一个")
+                    summary = _empty_summary(retry_exc)
+            else:
+                print(f"⚠️ 博主 {index} 刷新扫描异常: {exc},继续下一个")
+                summary = _empty_summary(exc)
        refresh_summaries.append(summary)

    should_wait = any(
@@ -946,9 +1426,9 @@ def refresh_then_export_accounts(driver, urls, wait_seconds=REFRESH_WAIT_SECONDS
        print(f"📌 导出阶段 博主 {index}/{len(urls)}: {url}")
        print(f"{'=' * 50}")
        try:
-            navigate_to_target(driver, url)
+            navigate_to_target(driver_box["driver"], url)
            pre_mtime = _max_excel_mtime()
-            excel_path = export_video_data(driver, pre_mtime)
+            excel_path = export_video_data(driver_box["driver"], pre_mtime)
            if not excel_path:
                print(f"⚠️ 博主 {index} 导出失败,跳过")
                continue
@@ -956,9 +1436,23 @@ def refresh_then_export_accounts(driver, urls, wait_seconds=REFRESH_WAIT_SECONDS
            print(f"  解析出 {len(records)} 条视频记录")
            all_records.extend(records)
        except Exception as exc:
-            print(f"⚠️ 博主 {index} 导出异常: {exc},跳过")
+            if _relogin_if_expired(exc) or _rebuild_if_browser_closed(exc):
+                try:
+                    navigate_to_target(driver_box["driver"], url)
+                    pre_mtime = _max_excel_mtime()
+                    excel_path = export_video_data(driver_box["driver"], pre_mtime)
+                    if not excel_path:
+                        print(f"⚠️ 博主 {index} 自愈后导出失败,跳过")
+                        continue
+                    records, _ = parse_chanmama_export(excel_path)
+                    print(f"  解析出 {len(records)} 条视频记录(自愈后)")
+                    all_records.extend(records)
+                except Exception as retry_exc:
+                    print(f"⚠️ 博主 {index} 自愈后导出仍失败: {retry_exc},跳过")
+            else:
+                print(f"⚠️ 博主 {index} 导出异常: {exc},跳过")

-    return all_records, refresh_summaries
+    return all_records, refresh_summaries, driver_box["driver"]


 def export_video_data(driver, pre_mtime: float = 0):
@@ -978,6 +1472,10 @@ def export_video_data(driver, pre_mtime: float = 0):
    except TimeoutException:
        print("⚠️ 视频列表加载超时，继续尝试导出...")

+    # 刷新/翻页阶段可能延迟弹出公告/活动弹窗，吞掉导出点击或盖住确认框；
+    # 点导出按钮前先清一次(确认框受保护)。
+    _dismiss_export_stage_overlays(driver)
+
    # 方法1: 先在可见 DOM 中找真正的交互节点。新版按钮文字常在嵌套
    # span/div 内，旧的 contains(text()) 和第一个隐藏 locator 都会漏掉它。
    export_clicked = False
@@ -989,8 +1487,11 @@ def export_video_data(driver, pre_mtime: float = 0):
    except TimeoutException:
        control = None
    if control:
-        clicked = driver.execute_script(_CLICK_EXPORT_CONTROL_SCRIPT)
-        export_clicked = bool(clicked)
+        # 原生点击优先(可触发内层元素的 Vue @click)，JS 合成点击兜底
+        export_clicked = _native_click_export_button(driver)
+        if not export_clicked:
+            clicked = driver.execute_script(_CLICK_EXPORT_CONTROL_SCRIPT)
+            export_clicked = bool(clicked)
        if export_clicked:
            print(f"✅ 已点击导出按钮: {control.get('text') or control.get('className')}")

@@ -1032,36 +1533,38 @@ def export_video_data(driver, pre_mtime: float = 0):
        return scrape_table_data(driver)

    # 等待导出确认弹窗出现并点击【导出】按钮
-    # 蝉妈妈弹窗标题"您即将进行数据导出"，按钮为 橙色"导出"（文字在 <span> 内）
+    # 蝉妈妈弹窗标题"您即将进行数据导出"，按钮为 橙色"导出"（文字在 <span> 内）。
+    # 逐秒探测弹窗状态：可确认→立即点导出；挂着但没有可点的确认按钮
+    # (空壳遮罩/公告弹窗)→先 ESC+移除正常关，持续 5 秒以上就补点导出按钮
+    # (第一次点击可能已被弹窗吞掉)；没有弹窗→常规清理并定期补点导出。
    print("⏳ 等待导出确认弹窗...")
    export_confirmed = False
-    for wait_sec in range(20):
+    unconfirmable_since: float | None = None
+    for wait_sec in range(36):
        time.sleep(1)
-        try:
-            # 关键：用 contains(.,'导出') 匹配后代文本（"导出"在 <span> 内）
-            confirm_selectors = [
-                # 精确：主按钮(primary) + 含"导出"文字
-                "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(.,'导出')]",
-                # 弹窗内任意含"导出"的按钮
-                "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(@class,'el-button--primary')][contains(.,'导出')]",
-                # 兼容"确定"/"确认"
-                "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(.,'确定')]",
-                "//div[contains(@class,'el-dialog') or @role='dialog']//button[contains(.,'确认')]",
-            ]
-            for sel in confirm_selectors:
-                try:
-                    btn = driver.find_element(By.XPATH, sel)
-                    if btn.is_displayed():
-                        driver.execute_script("arguments[0].click();", btn)
-                        export_confirmed = True
-                        print(f"✅ 已在弹窗中点击导出按钮: {sel}")
-                        break
-                except NoSuchElementException:
-                    continue
-            if export_confirmed:
+        state = _export_dialog_state(driver)
+        if state == "confirmable":
+            unconfirmable_since = None
+            if _click_export_confirm(driver):
+                export_confirmed = True
                break
-        except Exception:
            continue
+        if state == "present":
+            now = time.monotonic()
+            if unconfirmable_since is None:
+                _log_export_dialog_texts(driver)
+                print("ℹ️ 弹窗可见但没有可点的确认按钮，尝试关闭")
+                unconfirmable_since = now
+            elif now - unconfirmable_since > 5:
+                if _reclick_export_control(driver):
+                    print("🔁 清掉卡死弹窗后补点导出按钮")
+                unconfirmable_since = None
+            _dismiss_export_stage_overlays(driver)
+            continue
+        unconfirmable_since = None
+        _dismiss_export_stage_overlays(driver)
+        if wait_sec in (10, 20, 30) and _reclick_export_control(driver):
+            print("🔁 确认框迟迟未出现，补点导出按钮")

    if not export_confirmed:
        try:
@@ -1093,9 +1596,20 @@ def export_video_data(driver, pre_mtime: float = 0):
        "⚠️ 未找到本次下载的 Excel文件(等待 "
        f"{CHANMAMA_EXPORT_WAIT_SECONDS + max(CHANMAMA_EXPORT_GRACE_SECONDS, 0)} 秒超时)"
    )
+    _dump_export_failure(driver)
    return None


+def _download_complete(path: str) -> bool:
+    """判断下载文件是否已写完整(zip EOCD 在文件尾,截断时 is_zipfile 为假)。"""
+    import zipfile
+
+    try:
+        return zipfile.is_zipfile(path)
+    except OSError:
+        return False
+
+
 def _wait_for_exported_excel(
    pre_mtime: float,
    *,
@@ -1114,6 +1628,8 @@ def _wait_for_exported_excel(
    deadline 之后落盘的竞态；两个窗口内都未出现则返回 None。
    传入 driver 时每轮执行一次无害的 evaluate 以泵 Playwright 事件循环，
    确保 download 事件处理器(_save_download)在 sleep 轮询期间也能执行。
+    泵调用发现浏览器/页面已被外部关闭时，若文件已完整落盘则照常返回，
+    否则抛出原始关闭异常交由上层重建浏览器重试，不再空等满超时。
    """
    deadline = time.time() + max(timeout_seconds, 0)
    grace_deadline = deadline + max(grace_seconds, 0)
@@ -1122,10 +1638,25 @@ def _wait_for_exported_excel(
        if driver is not None:
            try:
                driver.execute_script("1")
-            except Exception:
-                pass
+            except Exception as pump_exc:
+                if is_browser_closed_error(pump_exc):
+                    # 浏览器/页面被外部关闭(进程树清理等): 先看文件是否
+                    # 已在关闭前完整落盘，再决定放弃，避免误杀已完成下载。
+                    settled = _find_new_excel_after(pre_mtime)
+                    if settled is not None and (
+                        not os.path.exists(settled) or _download_complete(settled)
+                    ):
+                        return settled
+                    raise
+                # 其余瞬态异常(页面导航中等)不致命，继续轮询文件系统。
        new_path = _find_new_excel_after(pre_mtime)
-        if new_path is not None:
+        # 文件出现 ≠ 下载完成: 当前浏览器可能以最终名边下边写,截断的
+        # zip 会让解析报 "File is not a zip file"。zip 的 EOCD 记录在
+        # 文件末尾,is_zipfile 通过即认为下载已完整;路径不存在(测试桩)
+        # 时保持旧行为直接返回。
+        if new_path is not None and (
+            not os.path.exists(new_path) or _download_complete(new_path)
+        ):
            return new_path
        now = time.time()
        if now >= deadline and not grace_logged and grace_deadline > deadline:
@@ -1829,7 +2360,8 @@ def main():
                print("⚠️ 登录可能未成功,但继续尝试后续步骤...")

        # 4. 先刷新两个账号的全部分页，再统一等待并重新进入页面导出。
-        all_records, _refresh_summaries = refresh_then_export_accounts(
+        # 返回的驱动可能已在浏览器被外部关闭后重建,后续收尾必须用它。
+        all_records, _refresh_summaries, driver = refresh_then_export_accounts(
            driver, TARGET_URLS, wait_seconds=REFRESH_WAIT_SECONDS
        )

@@ -1019,18 +1019,28 @@ def _creative_context(metric_date: date, style_names: list[str]) -> list[dict[st


 def _review_context(metric_date: date, style_names: list[str]) -> list[dict[str, Any]]:
+    """近 7 天差评归因（读 v_negative_reviews）。
+
+    2026-11-01 起 ``product_reviews`` 下线，改读三平台差评归一视图。口径随之变化：
+
+    * 数据源只剩差评，因此 ``sample_size`` = 差评条数、``negative_rate`` 恒为 1.0，
+      不再表示「差评/全部评价」；真正要看比例的场合请取平台评价总数另算。
+    * 输出门槛从「窗口内至少 5 条评价」改为「至少 1 条差评」——旧门槛在只有差评
+      的数据源上会导致几乎永不产出。
+    * 覆盖天猫（旧表天猫为 0 行）、京东（旧表 8 月断更），负向样本反而更全。
+    """
+
    if not style_names:
        return []
-    negative_terms = ("差", "坏", "退", "重", "小", "掉色", "线头", "不值", "失望")
    with get_conn() as conn:
        with conn.cursor() as cur:
            cur.execute(
                """
-                SELECT style_name, platform, content, score, tags
-                FROM product_reviews
+                SELECT style_name, platform, content
+                FROM v_negative_reviews
                WHERE style_name = ANY(%s) AND comment_date BETWEEN %s AND %s
                  AND content IS NOT NULL AND BTRIM(content) <> ''
-                ORDER BY style_name, comment_date DESC, id DESC
+                ORDER BY style_name, comment_date DESC, review_at DESC
                """,
                (style_names, metric_date - timedelta(days=6), metric_date),
            )
@@ -1040,17 +1050,15 @@ def _review_context(metric_date: date, style_names: list[str]) -> list[dict[str,
        grouped.setdefault(row[0], []).append(row)
    output = []
    for style_name, items in grouped.items():
-        if len(items) < 5:
-            continue
-        negative = [item for item in items if (item[3] is not None and item[3] <= 2) or any(term in item[2] for term in negative_terms)]
        output.append({
            "style_name": style_name,
            "window_start": metric_date - timedelta(days=6),
            "window_end": metric_date,
            "sample_size": len(items),
-            "negative_count": len(negative),
-            "negative_rate": round(len(negative) / len(items), 4),
-            "negative_examples": [item[2][:80] for item in negative[:2]],
+            "negative_count": len(items),
+            "negative_rate": 1.0,
+            "negative_examples": [item[2][:80] for item in items[:2]],
+            "source": "v_negative_reviews",
        })
    return output

@@ -18,7 +18,6 @@ from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
 _TOOLS_DIR = PATHS.tools_root
 _PROJECT_ROOT = PATHS.module_root

-RELOGIN_SCRIPT = _TOOLS_DIR / "relogin_bilibili.py"

 try:
    sys.stdout.reconfigure(encoding="utf-8", errors="replace")  # type: ignore[attr-defined]
@@ -81,39 +80,6 @@ def comment_replace_policy(comments: list[dict], stats: dict, metrics: dict) ->
    return False, False, "空评论且无法确认平台为 0 评论"


-def do_relogin() -> bool:
-    """Run relogin_bilibili.py (headed, waits for QR scan). Returns True if successful."""
-    acceptance_policy = current_acceptance_policy()
-    if acceptance_policy.enabled:
-        acceptance_policy.record(
-            "cookie_skipped",
-            operation="content.batch_rescrape_bilibili.relogin",
-            details={
-                "status": "SKIPPED_COOKIE",
-                "reason": "cookie is invalid and QR relogin is disabled for acceptance",
-            },
-        )
-        log(
-            "  [SKIPPED_COOKIE] Cookie invalid; acceptance mode will not "
-            "launch relogin or wait for a QR scan"
-        )
-        return False
-    log("  Cookie 可能过期,触发自动重登 (请扫码)...")
-    try:
-        rc = subprocess.call(
-            [sys.executable, str(RELOGIN_SCRIPT)],
-            cwd=str(_PROJECT_ROOT),
-        )
-        if rc == 0:
-            log("  重登成功,继续采集")
-            return True
-        else:
-            log(f"  重登失败 (exit={rc})")
-            return False
-    except Exception as exc:
-        log(f"  重登脚本执行异常: {exc}")
-        return False
-

 def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:
    rows = collect_urls()
@@ -126,8 +92,6 @@ def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:
    total_comments = 0
    consecutive_fail = 0
    MAX_CONSECUTIVE_FAIL = 3
-    relogin_used = 0
-    MAX_RELOGINS = 1
    t0 = time.time()
    for i, row in enumerate(rows, 1):
        note_id = row["id"]
@@ -152,11 +116,12 @@ def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:
                    break
        if comments is None:
            consecutive_fail += 1
-            if consecutive_fail >= MAX_CONSECUTIVE_FAIL and relogin_used < MAX_RELOGINS:
-                if do_relogin():
-                    relogin_used += 1
-                    consecutive_fail = 0
-                    continue
+            if consecutive_fail >= MAX_CONSECUTIVE_FAIL:
+                # 登录统一由账号 vault 管理；连续失败只提示，不再触发扫码重登。
+                log(
+                    "  连续失败达到上限；若为登录态失效请执行 "
+                    "uv run gyxx accounts login bilibili-content"
+                )
                if current_acceptance_policy().enabled:
                    return COOKIE_SKIP_EXIT_CODE
            continue
@@ -26,7 +26,6 @@ from gyxx_flow.modules.content_marketing.serialized_page_action import (
 _TOOLS_DIR = PATHS.tools_root
 _PROJECT_ROOT = PATHS.module_root

-RELOGIN_SCRIPT = _TOOLS_DIR / "relogin_douyin.py"

 try:
    sys.stdout.reconfigure(encoding="utf-8", errors="replace")  # type: ignore[attr-defined]
@@ -104,38 +103,6 @@ def result_logged_in(result: dict | None) -> bool:
    return bool(nested_stats.get("logged_in")) if isinstance(nested_stats, dict) else False


-def do_relogin() -> bool:
-    """Run relogin_douyin.py (headed, waits for QR scan). Returns True if successful."""
-    acceptance_policy = current_acceptance_policy()
-    if acceptance_policy.enabled:
-        acceptance_policy.record(
-            "cookie_skipped",
-            operation="content.batch_rescrape_douyin.relogin",
-            details={
-                "status": "SKIPPED_COOKIE",
-                "reason": "cookie is invalid and QR relogin is disabled for acceptance",
-            },
-        )
-        log(
-            "  [SKIPPED_COOKIE] Cookie invalid; acceptance mode will not "
-            "launch relogin or wait for a QR scan"
-        )
-        return False
-    log("  Cookie 可能过期,触发自动重登 (请扫码)...")
-    try:
-        rc = subprocess.call(
-            [sys.executable, str(RELOGIN_SCRIPT)],
-            cwd=str(_PROJECT_ROOT),
-        )
-        if rc == 0:
-            log("  重登成功,继续采集")
-            return True
-        log(f"  重登失败 (exit={rc})")
-        return False
-    except Exception as exc:
-        log(f"  重登脚本执行异常: {exc}")
-        return False
-

 def scrape_with_fresh_browser(
    url: str,
@@ -203,8 +170,6 @@ def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:
    COOLDOWN_SECONDS = 300
    cooldown_used = 0
    MAX_COOLDOWNS = 3
-    relogin_used = 0
-    MAX_RELOGINS = 1  # 每轮 batch 最多自动重登1次
    t0 = time.time()
    for i, row in enumerate(rows, 1):
        note_id = row["id"]
@@ -214,16 +179,10 @@ def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:

        if comments is None:
            consecutive_fail += 1
-            # 连续失败 + 可能 cookie 过期 → 尝试自动重登
-            if consecutive_fail >= MAX_CONSECUTIVE_FAIL and relogin_used < MAX_RELOGINS:
-                if do_relogin():
-                    relogin_used += 1
-                    consecutive_fail = 0
-                    comments, stats, scrape_error = scrape_with_fresh_browser(url)
-                else:
-                    # 重登失败,走冷却逻辑；验收模式必须直接跳过，不能等待。
-                    if current_acceptance_policy().enabled:
-                        return COOKIE_SKIP_EXIT_CODE
+            # 登录统一由账号 vault 管理；不再触发扫码重登，验收模式直接跳过。
+            if consecutive_fail >= MAX_CONSECUTIVE_FAIL:
+                if current_acceptance_policy().enabled:
+                    return COOKIE_SKIP_EXIT_CODE

            if comments is None and consecutive_fail >= MAX_CONSECUTIVE_FAIL:
                if cooldown_used < MAX_COOLDOWNS:
@@ -25,7 +25,6 @@ from gyxx_flow.modules.content_marketing.serialized_page_action import (
 _TOOLS_DIR = PATHS.tools_root
 _PROJECT_ROOT = PATHS.module_root

-RELOGIN_SCRIPT = _TOOLS_DIR / "relogin_xiaohongshu.py"

 try:
    sys.stdout.reconfigure(encoding="utf-8", errors="replace")  # type: ignore[attr-defined]
@@ -87,38 +86,6 @@ def comment_replace_policy(comments: list[dict], stats: dict, metrics: dict) ->
    return False, False, "空评论且无法确认平台为 0 评论"


-def do_relogin() -> bool:
-    """Run relogin_xiaohongshu.py (headed, waits for QR scan). Returns True if successful."""
-    acceptance_policy = current_acceptance_policy()
-    if acceptance_policy.enabled:
-        acceptance_policy.record(
-            "cookie_skipped",
-            operation="content.batch_rescrape_xiaohongshu.relogin",
-            details={
-                "status": "SKIPPED_COOKIE",
-                "reason": "cookie is invalid and QR relogin is disabled for acceptance",
-            },
-        )
-        log(
-            "  [SKIPPED_COOKIE] Cookie invalid; acceptance mode will not "
-            "launch relogin or wait for a QR scan"
-        )
-        return False
-    log("  Cookie 可能过期,触发自动重登 (请扫码)...")
-    try:
-        rc = subprocess.call(
-            [sys.executable, str(RELOGIN_SCRIPT)],
-            cwd=str(_PROJECT_ROOT),
-        )
-        if rc == 0:
-            log("  重登成功,继续采集")
-            return True
-        log(f"  重登失败 (exit={rc})")
-        return False
-    except Exception as exc:
-        log(f"  重登脚本执行异常: {exc}")
-        return False
-

 def scrape_with_fresh_browser(
    url: str,
@@ -173,8 +140,6 @@ def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:
    total_comments = 0
    consecutive_fail = 0
    MAX_CONSECUTIVE_FAIL = 3
-    relogin_used = 0
-    MAX_RELOGINS = 1
    t0 = time.time()
    for i, row in enumerate(rows, 1):
        note_id = row["id"]
@@ -183,13 +148,13 @@ def main(only_ids: list[int] | None = None, dry_run: bool = False) -> int:
        comments, result, scrape_error = scrape_with_fresh_browser(url)
        if comments is None:
            consecutive_fail += 1
-            if consecutive_fail >= MAX_CONSECUTIVE_FAIL and relogin_used < MAX_RELOGINS:
-                if do_relogin():
-                    relogin_used += 1
-                    consecutive_fail = 0
-                    comments, result, scrape_error = scrape_with_fresh_browser(url)
+            if consecutive_fail >= MAX_CONSECUTIVE_FAIL:
                if current_acceptance_policy().enabled:
                    return COOKIE_SKIP_EXIT_CODE
+                log(
+                    "  连续失败达到上限；若为登录态失效请执行 "
+                    "uv run gyxx accounts login xiaohongshu-content"
+                )
            if comments is None:
                log(f"  ERROR: {scrape_error}")
                fail += 1
@@ -166,6 +166,23 @@ def get_style_id(name: str) -> int | None:
            return row["id"] if row else None


+def get_dim_style_id(name: str) -> int | None:
+    """按款式名解析 dim_style 主档 id（2026-10-27 新增）。
+
+    data-hub 消费 cmt_notes_master.style_id 时按 dim_style（系列锚点款）口径，
+    cmt_styles.id 在款式主档重排后会变成悬挂 id；写笔记前用本函数归一，
+    解析不到返回 None（新款尚未收录款式库，由调用方决定回退策略）。
+    """
+    with get_dict_conn() as conn:
+        with conn.cursor(row_factory=dict_row) as cur:
+            cur.execute(
+                "SELECT id FROM dim_style WHERE style_name = %s ORDER BY id LIMIT 1",
+                (name,),
+            )
+            row = cur.fetchone()
+            return row["id"] if row else None
+
+
 # ---------------------------------------------------------------------------
 # Creators
 # ---------------------------------------------------------------------------
@@ -362,9 +362,14 @@ def sync_notes_to_db(note_tasks: list[NoteTask], style_ids: dict[str, int]) -> d
        creator_id = creator_ids[creator_name]
        db.link_style_creator(style_id, creator_id)

+        # 2026-10-27 起笔记归属改挂 dim_style 主档 id（data-hub 消费口径），
+        # cmt_styles.id 仅供 cmt_style_creators 等采集端内部关系表使用；
+        # dim_style 缺同名款时回退 cmt_styles.id，待款式收录后归一
+        dim_style_id = db.get_dim_style_id(task.style_name) or style_id
+
        platform = detect_platform(task.note_url, task.platform_hint)
        note_id = db.upsert_note(
-            style_id=style_id,
+            style_id=dim_style_id,
            creator_id=creator_id,
            platform=platform,
            url=task.note_url,
@@ -1,666 +0,0 @@
-"""Launch five QR relogins in parallel, notify via analyzer Lark, retry 3 times."""
-
-from __future__ import annotations
-
-import argparse
-import ctypes
-import datetime as dt
-import os
-import subprocess
-import sys
-import time
-from ctypes import wintypes
-from dataclasses import dataclass
-from pathlib import Path
-from typing import Callable
-
-import psutil
-
-from gyxx_flow.adapters import (
-    ANALYZER_NOTIFICATION_APP_PROFILE,
-    COOKIE_SKIP_EXIT_CODE,
-    ResolvedNotificationRoute,
-    binding_from_environment,
-    current_acceptance_policy,
-    environment_for_child_script,
-    resolve_notification_route,
-    send_lark_bot_message,
-)
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-PROJECT_DIR = PATHS.module_root
-ANALYZER_DIR = PATHS.tmp_root / "relogin"
-WORKFLOW_ID = "content.relogin.weekly"
-MARKETING_RECIPIENT = "ou_24cc944d6e43c69c59d6560ad4e2ae6e"
-TECHNICAL_RECIPIENT = "ou_7ad5fc8012e2f741afc5346e05ffd447"
-
-
-@dataclass(frozen=True)
-class Platform:
-    label: str
-    command: tuple[str, ...]
-    required_cookie: str
-
-
-@dataclass(frozen=True, slots=True)
-class NotificationDeliveryFailure:
-    """One failed recipient delivery that must affect the workflow outcome."""
-
-    stage: str
-    platform: str
-    recipient: str
-    detail: str
-
-
-PLATFORMS = {
-    "douyin": Platform(
-        "\u6296\u97f3", ("data/tools/relogin_douyin.py",),
-        "sessionid",
-    ),
-    "bilibili": Platform(
-        "B\u7ad9", ("data/tools/relogin_bilibili.py",),
-        "SESSDATA",
-    ),
-    "pgy": Platform(
-        "\u84b2\u516c\u82f1", ("data/tools/relogin_pgy.py",),
-        "",
-    ),
-    "xingtu": Platform(
-        "\u661f\u56fe", ("data/tools/relogin_xingtu.py", "--force"),
-        "sessionid",
-    ),
-    "xiaohongshu": Platform(
-        "\u5c0f\u7ea2\u4e66", ("data/tools/relogin_xiaohongshu.py",),
-        "web_session",
-    ),
-}
-
-DEFAULT_RECIPIENTS = {
-    "pgy": MARKETING_RECIPIENT,
-    "xingtu": MARKETING_RECIPIENT,
-    "douyin": TECHNICAL_RECIPIENT,
-    "bilibili": TECHNICAL_RECIPIENT,
-    "xiaohongshu": TECHNICAL_RECIPIENT,
-}
-
-
-def resolve_platform_names(requested: list[str] | None) -> list[str]:
-    """Resolve an optional manual platform subset while preserving order."""
-    if not requested:
-        return list(PLATFORMS)
-    names: list[str] = []
-    for name in requested:
-        if name not in PLATFORMS:
-            raise ValueError(f"unknown platform: {name}")
-        if name not in names:
-            names.append(name)
-    return names
-
-
-def recipient_for_platform(platform_name: str, override: str | None = None) -> str:
-    """Return the responsible recipient, unless a manual all-platform override is set."""
-    from gyxx_flow.adapters.acceptance_policy import resolve_notification_recipients
-
-    if override:
-        recipient = override
-    else:
-        try:
-            recipient = DEFAULT_RECIPIENTS[platform_name]
-        except KeyError as exc:
-            raise ValueError(
-                f"no Lark recipient configured for platform: {platform_name}"
-            ) from exc
-    return resolve_notification_recipients(
-        (recipient,),
-        app_profile=ANALYZER_NOTIFICATION_APP_PROFILE,
-    )[0]
-
-
-def resolve_weekly_notification_route(
-    override: str | None = None,
-) -> ResolvedNotificationRoute:
-    """Resolve the scheduled workflow route while retaining legacy defaults."""
-    defaults = (
-        (override,)
-        if override
-        else tuple(dict.fromkeys(DEFAULT_RECIPIENTS.values()))
-    )
-    return resolve_notification_route(WORKFLOW_ID, defaults)
-
-
-def recipients_for_platform(
-    platform_name: str,
-    route: ResolvedNotificationRoute,
-    override: str | None = None,
-) -> tuple[str, ...]:
-    """Return dynamic recipients, or the original per-platform recipient."""
-    if not route.enabled:
-        return ()
-    if route.configured:
-        return route.open_ids
-    return (recipient_for_platform(platform_name, override),)
-
-
-def group_platforms_by_recipient(
-    platform_names: list[str], override: str | None = None,
-) -> dict[str, list[str]]:
-    grouped: dict[str, list[str]] = {}
-    for name in platform_names:
-        grouped.setdefault(recipient_for_platform(name, override), []).append(name)
-    return grouped
-
-
-def group_platforms_by_notification_recipient(
-    platform_names: list[str],
-    route: ResolvedNotificationRoute,
-    override: str | None = None,
-) -> dict[str, list[str]]:
-    """Group failed platforms for legacy or configured multi-recipient delivery."""
-    if not route.enabled:
-        return {}
-    if route.configured:
-        return {recipient: list(platform_names) for recipient in route.open_ids}
-    return group_platforms_by_recipient(platform_names, override)
-
-
-def run_with_retries(
-    platform_names: list[str],
-    *,
-    max_attempts: int,
-    run_round: Callable[[list[str], int], dict[str, bool]],
-) -> dict[str, bool]:
-    status = {name: False for name in platform_names}
-    pending = list(platform_names)
-    for attempt in range(1, max_attempts + 1):
-        if not pending:
-            break
-        round_result = run_round(pending, attempt)
-        for name in pending:
-            status[name] = bool(round_result.get(name))
-        pending = [name for name in pending if not status[name]]
-    return status
-
-
-def _valid_cookie(platform: Platform, cookie_file: Path) -> bool:
-    if not cookie_file.is_file():
-        return False
-    if not platform.required_cookie:
-        return cookie_file.stat().st_size > 10
-    try:
-        import json
-        values = json.loads(cookie_file.read_text(encoding="utf-8"))
-        return any(
-            item.get("name") == platform.required_cookie and item.get("value")
-            for item in values if isinstance(item, dict)
-        )
-    except Exception:
-        return False
-
-
-def _descendant_pids(pid: int) -> set[int]:
-    try:
-        return {child.pid for child in psutil.Process(pid).children(recursive=True)}
-    except (psutil.Error, OSError):
-        return set()
-
-
-def _terminate_process_tree(process: subprocess.Popen) -> None:
-    """Terminate the whole Windows subprocess tree, including Chrome/Scrapling."""
-    try:
-        subprocess.run(
-            ["taskkill", "/PID", str(process.pid), "/T", "/F"],
-            capture_output=True,
-            text=True,
-            check=False,
-        )
-    except OSError:
-        process.terminate()
-    try:
-        process.wait(timeout=10)
-    except subprocess.TimeoutExpired:
-        process.kill()
-
-
-def _window_for_process(pid: int) -> int | None:
-    pids = _descendant_pids(pid) | {pid}
-    handles: list[int] = []
-    user32 = ctypes.windll.user32
-    callback_type = ctypes.WINFUNCTYPE(ctypes.c_bool, ctypes.c_void_p, ctypes.c_void_p)
-
-    def callback(hwnd, _lparam):
-        process_id = ctypes.c_ulong()
-        user32.GetWindowThreadProcessId(hwnd, ctypes.byref(process_id))
-        if process_id.value in pids and user32.IsWindowVisible(hwnd):
-            handles.append(int(hwnd))
-        return True
-
-    user32.EnumWindows(callback_type(callback), 0)
-    return handles[0] if handles else None
-
-
-def _platform_from_window_title(title: str) -> str | None:
-    """Identify a login window without relying on Chrome's process tree."""
-    if "\u84b2\u516c\u82f1" in title:
-        return "pgy"
-    if "\u6296\u97f3\u6388\u6743" in title or "\u5de8\u91cf\u661f\u56fe" in title:
-        return "xingtu"
-    if "\u6296\u97f3" in title:
-        return "douyin"
-    if "bilibili" in title.lower() or "\u8d26\u53f7\u767b\u5f55" in title:
-        return "bilibili"
-    if "\u5c0f\u7ea2\u4e66" in title:
-        return "xiaohongshu"
-    return None
-
-
-def _login_windows() -> dict[str, int]:
-    """Map visible Chrome login windows to platforms by their page titles."""
-    user32 = ctypes.windll.user32
-    found: dict[str, int] = {}
-    callback_type = ctypes.WINFUNCTYPE(ctypes.c_bool, ctypes.c_void_p, ctypes.c_void_p)
-
-    def callback(hwnd, _lparam):
-        if not user32.IsWindowVisible(hwnd):
-            return True
-        length = user32.GetWindowTextLengthW(hwnd)
-        if length <= 0:
-            return True
-        buffer = ctypes.create_unicode_buffer(length + 1)
-        user32.GetWindowTextW(hwnd, buffer, length + 1)
-        platform = _platform_from_window_title(buffer.value)
-        if platform:
-            found[platform] = int(hwnd)
-        return True
-
-    user32.EnumWindows(callback_type(callback), 0)
-    return found
-
-
-def _capture_window(hwnd: int, output: Path) -> None:
-    """Maximize and render one window directly, even when it is not foreground."""
-    from PIL import Image
-
-    user32 = ctypes.windll.user32
-    gdi32 = ctypes.windll.gdi32
-    user32.ShowWindow(hwnd, 3)  # SW_MAXIMIZE
-    time.sleep(1)
-    rect = wintypes.RECT()
-    user32.GetWindowRect(hwnd, ctypes.byref(rect))
-    width, height = rect.right - rect.left, rect.bottom - rect.top
-    if width <= 0 or height <= 0:
-        raise RuntimeError(f"invalid window size: {width}x{height}")
-
-    window_dc = user32.GetWindowDC(hwnd)
-    memory_dc = gdi32.CreateCompatibleDC(window_dc)
-    bitmap = gdi32.CreateCompatibleBitmap(window_dc, width, height)
-    previous = gdi32.SelectObject(memory_dc, bitmap)
-    try:
-        if not user32.PrintWindow(hwnd, memory_dc, 2):  # PW_RENDERFULLCONTENT
-            raise RuntimeError("PrintWindow failed")
-
-        class BitmapInfoHeader(ctypes.Structure):
-            _fields_ = [
-                ("biSize", wintypes.DWORD), ("biWidth", wintypes.LONG),
-                ("biHeight", wintypes.LONG), ("biPlanes", wintypes.WORD),
-                ("biBitCount", wintypes.WORD), ("biCompression", wintypes.DWORD),
-                ("biSizeImage", wintypes.DWORD),
-                ("biXPelsPerMeter", wintypes.LONG),
-                ("biYPelsPerMeter", wintypes.LONG),
-                ("biClrUsed", wintypes.DWORD),
-                ("biClrImportant", wintypes.DWORD),
-            ]
-
-        byte_count = width * height * 4
-        pixels = ctypes.create_string_buffer(byte_count)
-        header = BitmapInfoHeader(
-            ctypes.sizeof(BitmapInfoHeader), width, -height, 1, 32, 0,
-            byte_count, 0, 0, 0, 0,
-        )
-        rows = gdi32.GetDIBits(
-            memory_dc, bitmap, 0, height, pixels, ctypes.byref(header), 0
-        )
-        if rows != height:
-            raise RuntimeError(f"GetDIBits returned {rows}/{height} rows")
-        image = Image.frombuffer(
-            "RGB", (width, height), pixels, "raw", "BGRX", 0, 1
-        ).copy()
-    finally:
-        gdi32.SelectObject(memory_dc, previous)
-        gdi32.DeleteObject(bitmap)
-        gdi32.DeleteDC(memory_dc)
-        user32.ReleaseDC(hwnd, window_dc)
-
-    output.parent.mkdir(parents=True, exist_ok=True)
-    image.save(output, "PNG")
-
-
-def _tile_windows(handles: list[int]) -> None:
-    """Place four login windows in a 2x2 grid so screenshots cannot overlap."""
-    user32 = ctypes.windll.user32
-    screen_w = user32.GetSystemMetrics(0)
-    screen_h = user32.GetSystemMetrics(1)
-    cell_w = max(640, screen_w // 2)
-    cell_h = max(450, screen_h // 2)
-    for index, hwnd in enumerate(handles):
-        x = (index % 2) * cell_w
-        y = (index // 2) * cell_h
-        user32.ShowWindow(hwnd, 9)  # SW_RESTORE
-        user32.MoveWindow(hwnd, x, y, cell_w, cell_h, True)
-
-
-def _send_lark(
-    recipient: str,
-    text: str | None = None,
-    image: Path | None = None,
-    *,
-    app_profile: str = ANALYZER_NOTIFICATION_APP_PROFILE,
-) -> tuple[dict[str, object], ...]:
-    """Send text and/or image through the analyzer Hermes bot."""
-    if app_profile != ANALYZER_NOTIFICATION_APP_PROFILE:
-        raise ValueError(
-            "relogin notifications must use the hermes-analyzer lark-cli profile"
-        )
-    lark_env = {
-        key: value for key, value in os.environ.items()
-        if not key.upper().startswith("HERMES_")
-    }
-    receipts: list[dict[str, object]] = []
-    if text:
-        ANALYZER_DIR.mkdir(parents=True, exist_ok=True)
-        receipts.append(send_lark_bot_message(
-            user_id=recipient,
-            text=text,
-            profile=ANALYZER_NOTIFICATION_APP_PROFILE,
-            cwd=ANALYZER_DIR,
-            env=lark_env,
-        ))
-    if image:
-        receipts.append(send_lark_bot_message(
-            user_id=recipient,
-            image=image,
-            profile=ANALYZER_NOTIFICATION_APP_PROFILE,
-            cwd=image.parent,
-            env=lark_env,
-        ))
-    return tuple(receipts)
-
-
-def _send_qr_notification(
-    route: ResolvedNotificationRoute,
-    platform_name: str,
-    *,
-    override: str | None,
-    text: str,
-    image: Path,
-) -> tuple[NotificationDeliveryFailure, ...]:
-    """Attempt every QR delivery and return all recipient-specific failures."""
-    profile = route.app_profile or ANALYZER_NOTIFICATION_APP_PROFILE
-    failures: list[NotificationDeliveryFailure] = []
-    for recipient in recipients_for_platform(platform_name, route, override):
-        for stage, kwargs in (
-            ("qr_text", {"text": text}),
-            ("qr_image", {"image": image}),
-        ):
-            try:
-                _send_lark(recipient, app_profile=profile, **kwargs)
-            except Exception as exc:
-                failures.append(
-                    NotificationDeliveryFailure(
-                        stage=stage,
-                        platform=platform_name,
-                        recipient=recipient,
-                        detail=str(exc) or type(exc).__name__,
-                    )
-                )
-    return tuple(failures)
-
-
-def _unavailable_qr_delivery_failures(
-    route: ResolvedNotificationRoute,
-    platform_name: str,
-    *,
-    override: str | None,
-    detail: str,
-) -> tuple[NotificationDeliveryFailure, ...]:
-    """Record both intended QR deliveries when no screenshot can be sent."""
-    return tuple(
-        NotificationDeliveryFailure(
-            stage=stage,
-            platform=platform_name,
-            recipient=recipient,
-            detail=detail,
-        )
-        for recipient in recipients_for_platform(platform_name, route, override)
-        for stage in ("qr_text", "qr_image")
-    )
-
-
-def _send_failure_summaries(
-    route: ResolvedNotificationRoute,
-    failed_platforms: list[str],
-    *,
-    override: str | None,
-    max_attempts: int,
-) -> tuple[NotificationDeliveryFailure, ...]:
-    """Attempt every final failure summary and return all delivery failures."""
-    groups = group_platforms_by_notification_recipient(
-        failed_platforms,
-        route,
-        override,
-    )
-    profile = route.app_profile or ANALYZER_NOTIFICATION_APP_PROFILE
-    failures: list[NotificationDeliveryFailure] = []
-    for recipient, names in groups.items():
-        labels = "\u3001".join(PLATFORMS[name].label for name in names)
-        try:
-            _send_lark(
-                recipient,
-                text=(
-                    f"\u4ee5\u4e0b\u5e73\u53f0\u8fde\u7eed {max_attempts} \u6b21\u672a\u626b\u7801\uff0c"
-                    "\u5df2\u6062\u590d\u5e76\u7ee7\u4f7f\u7528\u65e7 Cookie\uff1a"
-                    f"{labels}"
-                ),
-                app_profile=profile,
-            )
-        except Exception as exc:
-            failures.append(
-                NotificationDeliveryFailure(
-                    stage="failure_summary",
-                    platform=",".join(names),
-                    recipient=recipient,
-                    detail=str(exc) or type(exc).__name__,
-                )
-            )
-    return tuple(failures)
-
-
-def _run_round_factory(
-    args,
-    notification_route: ResolvedNotificationRoute | None,
-    notification_failures: list[NotificationDeliveryFailure],
-):
-    python = sys.executable
-    screenshot_root = PATHS.tmp_root / "qrcode/friday_relogin"
-
-    def run_round(names: list[str], attempt: int) -> dict[str, bool]:
-        started: dict[str, subprocess.Popen] = {}
-        cookie_files: dict[str, Path] = {}
-        missing_qr_windows: set[str] = set()
-        stamp = dt.datetime.now().strftime("%Y%m%d_%H%M%S")
-        for name in names:
-            platform = PLATFORMS[name]
-            cmd = [python, *platform.command, "--login-timeout", str(args.round_timeout)]
-            target = PROJECT_DIR / platform.command[0]
-            environment = environment_for_child_script(target, os.environ)
-            cookie_files[name] = binding_from_environment(environment).cookie_file
-            started[name] = subprocess.Popen(
-                cmd,
-                cwd=PROJECT_DIR,
-                env=environment,
-            )
-
-        time.sleep(args.screenshot_delay)
-        title_windows = _login_windows()
-        windows = {
-            name: title_windows.get(name) or _window_for_process(process.pid)
-            for name, process in started.items()
-        }
-        _tile_windows([hwnd for hwnd in windows.values() if hwnd])
-        time.sleep(2)
-        for name, process in started.items():
-            platform = PLATFORMS[name]
-            hwnd = windows[name]
-            if not hwnd:
-                print(f"[{name}] no visible browser window for QR screenshot", flush=True)
-                missing_qr_windows.add(name)
-                continue
-            output = screenshot_root / f"{stamp}_{name}_round{attempt}.png"
-            try:
-                _capture_window(hwnd, output)
-                print(f"[{name}] QR screenshot: {output}", flush=True)
-            except Exception as exc:
-                print(f"[{name}] screenshot failed: {exc}", flush=True)
-                if not args.no_send and notification_route is not None:
-                    notification_failures.extend(
-                        _unavailable_qr_delivery_failures(
-                            notification_route,
-                            name,
-                            override=args.recipient,
-                            detail=(
-                                "QR screenshot capture failed: "
-                                f"{str(exc) or type(exc).__name__}"
-                            ),
-                        )
-                    )
-                continue
-            if not args.no_send and notification_route is not None:
-                notification_failures.extend(
-                    _send_qr_notification(
-                        notification_route,
-                        name,
-                        override=args.recipient,
-                        text=(
-                            f"{platform.label} \u767b\u5f55\u4e8c\u7ef4\u7801"
-                            f"\uff08\u7b2c {attempt}/{args.max_attempts} \u6b21\uff09\uff0c"
-                            "\u8bf7\u5728 3 \u5206\u949f\u5185\u626b\u7801\u3002"
-                        ),
-                        image=output,
-                    )
-                )
-
-        # Put the still-open login windows back into a compact layout after
-        # taking full-size screenshots one at a time.
-        _tile_windows([hwnd for hwnd in windows.values() if hwnd])
-
-        results: dict[str, bool] = {}
-        for name, process in started.items():
-            try:
-                rc = process.wait(timeout=args.round_timeout + 90)
-            except subprocess.TimeoutExpired:
-                _terminate_process_tree(process)
-                rc = 1
-            results[name] = rc == 0 and _valid_cookie(
-                PLATFORMS[name],
-                cookie_files[name],
-            )
-            if (
-                not results[name]
-                and name in missing_qr_windows
-                and not args.no_send
-                and notification_route is not None
-            ):
-                notification_failures.extend(
-                    _unavailable_qr_delivery_failures(
-                        notification_route,
-                        name,
-                        override=args.recipient,
-                        detail=(
-                            "no visible QR window and relogin did not succeed"
-                        ),
-                    )
-                )
-            print(f"[{name}] round={attempt} exit={rc} success={results[name]}", flush=True)
-        return results
-
-    return run_round
-
-
-def main() -> int:
-    acceptance_policy = current_acceptance_policy()
-    if acceptance_policy.enabled:
-        acceptance_policy.record(
-            "cookie_skipped",
-            operation="content.relogin.weekly",
-            details={
-                "status": "SKIPPED_COOKIE",
-                "reason": "acceptance runs never launch QR relogin workflows",
-            },
-        )
-        print(
-            "[SKIPPED_COOKIE] Acceptance mode does not launch login windows, "
-            "reset cookies/profiles, or wait for QR scans.",
-            flush=True,
-        )
-        return COOKIE_SKIP_EXIT_CODE
-
-    parser = argparse.ArgumentParser(description="Friday five-platform parallel QR relogin")
-    parser.add_argument(
-        "--recipient", default=None,
-        help="Override the configured recipients and send every platform to one open_id",
-    )
-    parser.add_argument("--max-attempts", type=int, default=3)
-    parser.add_argument("--round-timeout", type=int, default=300)
-    parser.add_argument("--screenshot-delay", type=int, default=60)
-    parser.add_argument("--no-send", action="store_true", help="Do not send Lark messages")
-    parser.add_argument(
-        "--platform", action="append", choices=sorted(PLATFORMS),
-        help="Only relogin the selected platform; may be repeated",
-    )
-    args = parser.parse_args()
-
-    platform_names = resolve_platform_names(args.platform)
-    notification_route = (
-        None if args.no_send else resolve_weekly_notification_route(args.recipient)
-    )
-    notification_failures: list[NotificationDeliveryFailure] = []
-    status = run_with_retries(
-        platform_names, max_attempts=args.max_attempts,
-        run_round=_run_round_factory(
-            args,
-            notification_route,
-            notification_failures,
-        ),
-    )
-    failed = [name for name, ok in status.items() if not ok]
-    if failed and not args.no_send and notification_route is not None:
-        notification_failures.extend(
-            _send_failure_summaries(
-                notification_route,
-                failed,
-                override=args.recipient,
-                max_attempts=args.max_attempts,
-            )
-        )
-    print(f"Final relogin status: {status}", flush=True)
-    if notification_failures:
-        print(
-            f"Final notification status: failed ({len(notification_failures)} deliveries)",
-            flush=True,
-        )
-        for failure in notification_failures:
-            print(
-                "[NOTIFICATION_ERROR] "
-                f"stage={failure.stage} platform={failure.platform} "
-                f"recipient={failure.recipient} error={failure.detail}",
-                flush=True,
-            )
-        return 2
-    if notification_route is not None and notification_route.enabled:
-        print("Final notification status: delivered", flush=True)
-    else:
-        print("Final notification status: skipped", flush=True)
-    return 1 if failed else 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())
@@ -1,113 +0,0 @@
-"""每周一重登 B 站。
-
-强制清掉 cookie JSON + 浏览器 profile,确保下次启动时必须扫码。
-任务计划: 每周一 09:00
-
-跑法:
-  python data/tools/relogin_bilibili.py             # 实际跑(会弹二维码)
-  python data/tools/relogin_bilibili.py --dry-run   # 只看会做什么
-"""
-
-import argparse
-import datetime as dt
-import subprocess
-import sys
-
-from gyxx_flow.modules.content_marketing.data.tools.relogin_runtime_paths import (
-    child_browser_environment,
-    resolve_relogin_runtime_paths,
-    sync_relogin_state_to_consumers,
-)
-from gyxx_flow.modules.content_marketing.data.tools.relogin_transaction import (
-    begin,
-    commit,
-    rollback,
-)
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-PROJECT_DIR = PATHS.module_root
-RUNTIME_PATHS = resolve_relogin_runtime_paths("bilibili")
-COOKIE_FILE = RUNTIME_PATHS.cookie_file
-STATE_FILE = RUNTIME_PATHS.storage_state_file
-PROFILE_DIR = RUNTIME_PATHS.profile_dir
-SCRAPER = PROJECT_DIR / "bilibili_comment_scraper.py"
-
-
-def reset_state():
-    """删 cookie JSON + storage_state + 清浏览器 profile。"""
-    state = begin((COOKIE_FILE, STATE_FILE), PROFILE_DIR)
-    print("  [OK] 旧 Cookie/Profile 已暂存,失败时将自动恢复")
-    return state
-
-
-def main() -> int:
-    ap = argparse.ArgumentParser(description="每周一重登 B 站")
-    ap.add_argument("--login-timeout", type=int, default=300, help="扫码超时(秒)")
-    ap.add_argument("--dry-run", action="store_true", help="只打印计划,不删不跑")
-    args = ap.parse_args()
-
-    print("== Relogin: B 站 ==")
-    print(f"  今天: {dt.date.today()} ({dt.date.today().strftime('%A')})")
-    print(f"  cookie: {COOKIE_FILE}")
-    print(f"  profile: {PROFILE_DIR}")
-    print()
-
-    if args.dry_run:
-        print("[DRY-RUN] 实际跑会: 删 cookie + 清 profile + 弹二维码")
-        return 0
-
-    old_state = reset_state()
-    cmd = [
-        sys.executable,
-        str(SCRAPER),
-        "login",  # dummy url (login-only 模式不需要真实 url)
-        "--login-only",
-        "--login-timeout",
-        str(args.login_timeout),
-    ]
-    print(f"\n  [RUN] {' '.join(cmd)}")
-    try:
-        rc = subprocess.call(
-            cmd,
-            cwd=str(PROJECT_DIR),
-            env=child_browser_environment(
-                cookie_file=COOKIE_FILE,
-                storage_state_file=STATE_FILE,
-                profile_dir=PROFILE_DIR,
-            ),
-        )
-    except BaseException:
-        rollback(old_state)
-        raise
-    print(f"\n  exit={rc}")
-    login_ok = False
-    if rc == 0:
-        # 验证 SESSDATA 是否真写入了
-        try:
-            import json
-            cookies = json.loads(COOKIE_FILE.read_text(encoding="utf-8"))
-            sessdata = any(c.get("name") == "SESSDATA" and c.get("value") for c in cookies)
-            login_ok = bool(sessdata)
-            print(f"  SESSDATA 写入: {'OK' if sessdata else 'FAIL — 没有看到 SESSDATA,登录可能没成功'}")
-        except Exception as exc:
-            print(f"  [WARN] 验证 cookie 时出错: {exc}")
-    if rc != 0 or not login_ok:
-        rollback(old_state)
-        print("  [ROLLBACK] 未完成扫码,已恢复旧登录状态")
-        return rc or 1
-    try:
-        sync_relogin_state_to_consumers(
-            "bilibili",
-            cookie_file=COOKIE_FILE,
-            storage_state_file=STATE_FILE,
-        )
-    except Exception as exc:
-        rollback(old_state)
-        print(f"  [ROLLBACK] 同步采集器登录态失败: {exc}")
-        return 1
-    commit(old_state)
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -1,121 +0,0 @@
-"""Force a fresh Douyin login while preserving the previous usable session."""
-
-from __future__ import annotations
-
-import argparse
-import datetime as dt
-import json
-import subprocess
-import sys
-
-from gyxx_flow.modules.content_marketing.data.tools.relogin_runtime_paths import (
-    child_browser_environment,
-    resolve_relogin_runtime_paths,
-    sync_relogin_state_to_consumers,
-)
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-try:
-    from .relogin_transaction import begin, commit, rollback
-except ImportError:  # Direct script execution from data/tools.
-    from gyxx_flow.modules.content_marketing.data.tools.relogin_transaction import (
-        begin,
-        commit,
-        rollback,
-    )
-
-PROJECT_DIR = PATHS.module_root
-RUNTIME_PATHS = resolve_relogin_runtime_paths("douyin")
-COOKIE_FILE = RUNTIME_PATHS.cookie_file
-STATE_FILE = RUNTIME_PATHS.storage_state_file
-PROFILE_DIR = RUNTIME_PATHS.profile_dir
-SCRAPER = PROJECT_DIR / "douyin_comment_scraper.py"
-
-LOGIN_COOKIE = "sessionid"
-
-
-def reset_state():
-    state = begin((COOKIE_FILE, STATE_FILE), PROFILE_DIR)
-    print("  [OK] Old Cookie/Profile staged; failures will roll back automatically")
-    return state
-
-
-def is_cookie_valid() -> bool:
-    if not COOKIE_FILE.exists():
-        return False
-    try:
-        cookies = json.loads(COOKIE_FILE.read_text(encoding="utf-8"))
-        return any(
-            isinstance(cookie, dict)
-            and cookie.get("name") == LOGIN_COOKIE
-            and cookie.get("value")
-            for cookie in cookies
-        )
-    except Exception:
-        return False
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description="Relogin Douyin")
-    parser.add_argument("--login-timeout", type=int, default=300)
-    parser.add_argument("--dry-run", action="store_true")
-    args = parser.parse_args()
-
-    print("== Relogin: Douyin ==")
-    print(f"  date: {dt.date.today()}")
-    print(f"  cookie: {COOKIE_FILE}")
-    print(f"  profile: {PROFILE_DIR}")
-    print(f"  current cookie: {'present' if is_cookie_valid() else 'missing/invalid'}")
-
-    if args.dry_run:
-        print("[DRY-RUN] Would stage Cookie/Profile and open the QR login window")
-        return 0
-
-    old_state = reset_state()
-    command = [
-        sys.executable,
-        str(SCRAPER),
-        "login",
-        "--login-only",
-        "--login-timeout",
-        str(args.login_timeout),
-    ]
-    print(f"\n  [RUN] {' '.join(command)}")
-    try:
-        return_code = subprocess.call(
-            command,
-            cwd=str(PROJECT_DIR),
-            env=child_browser_environment(
-                cookie_file=COOKIE_FILE,
-                storage_state_file=STATE_FILE,
-                profile_dir=PROFILE_DIR,
-            ),
-        )
-    except BaseException:
-        rollback(old_state)
-        raise
-
-    login_ok = return_code == 0 and is_cookie_valid()
-    print(f"\n  exit={return_code}")
-    print(f"  {LOGIN_COOKIE} written: {'OK' if login_ok else 'FAIL'}")
-    if not login_ok:
-        rollback(old_state)
-        print("  [ROLLBACK] Login incomplete; restored previous Cookie/Profile")
-        return return_code or 1
-    try:
-        sync_relogin_state_to_consumers(
-            "douyin",
-            cookie_file=COOKIE_FILE,
-            storage_state_file=STATE_FILE,
-        )
-    except Exception as exc:
-        rollback(old_state)
-        print(f"  [ROLLBACK] Failed to sync collector login state: {exc}")
-        return 1
-
-    commit(old_state)
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())
@@ -1,148 +0,0 @@
-"""每周一重登小红书蒲公英。
-
-强制清掉 cookie JSON + 浏览器 profile,确保下次启动时必须扫码。
-任务计划: 每周一 09:00
-
-跑法:
-  python data/tools/relogin_pgy.py             # 实际跑(会弹二维码)
-  python data/tools/relogin_pgy.py --dry-run   # 只看会做什么
-"""
-
-import argparse
-import datetime as dt
-import shutil
-import subprocess
-import sys
-from pathlib import Path
-
-from gyxx_flow.modules.content_marketing.data.tools.relogin_runtime_paths import (
-    child_browser_environment,
-    resolve_relogin_runtime_paths,
-    sync_relogin_state_to_consumers,
-)
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-PROJECT_DIR = PATHS.module_root
-RUNTIME_PATHS = resolve_relogin_runtime_paths("pgy")
-COOKIE_FILE = RUNTIME_PATHS.cookie_file
-STATE_FILE = RUNTIME_PATHS.storage_state_file
-PROFILE_DIR = RUNTIME_PATHS.profile_dir
-SCRAPER = PROJECT_DIR / "pgy_xhs_scraper_v2.py"
-
-
-def reset_state() -> tuple[Path | None, Path | None, Path | None]:
-    """Temporarily move the old login state aside so it can be restored."""
-    ts = dt.datetime.now().strftime("%Y%m%d_%H%M%S_%f")
-    cookie_backup = None
-    state_backup = None
-    profile_backup = None
-    if COOKIE_FILE.exists():
-        backup = COOKIE_FILE.with_suffix(f".json.bak.{ts}")
-        COOKIE_FILE.rename(backup)
-        cookie_backup = backup
-        print(f"  [OK] 备份 cookie -> {backup.name}")
-    else:
-        print(f"  [INFO] {COOKIE_FILE.name} 不存在,跳过")
-    if STATE_FILE.exists():
-        state_backup = STATE_FILE.with_suffix(f".json.bak.{ts}")
-        STATE_FILE.rename(state_backup)
-        print(f"  [OK] 备份 storage state -> {state_backup.name}")
-    else:
-        print(f"  [INFO] {STATE_FILE.name} 不存在,跳过")
-    if PROFILE_DIR.exists():
-        profile_backup = PROFILE_DIR.with_name(f"{PROFILE_DIR.name}.bak.{ts}")
-        PROFILE_DIR.rename(profile_backup)
-        print(f"  [OK] 暂存 {PROFILE_DIR.name}/ -> {profile_backup.name}/")
-    else:
-        print(f"  [INFO] {PROFILE_DIR.name}/ 不存在,跳过")
-    return cookie_backup, profile_backup, state_backup
-
-
-def restore_previous_state(
-    cookie_backup: Path | None,
-    profile_backup: Path | None,
-    state_backup: Path | None = None,
-) -> None:
-    """Discard an incomplete login and restore the last known-good state."""
-    if COOKIE_FILE.exists():
-        COOKIE_FILE.unlink()
-    if cookie_backup and cookie_backup.exists():
-        cookie_backup.rename(COOKIE_FILE)
-        print(f"  [ROLLBACK] 已恢复旧 cookie: {COOKIE_FILE.name}")
-    if STATE_FILE.exists():
-        STATE_FILE.unlink()
-    if state_backup and state_backup.exists():
-        state_backup.rename(STATE_FILE)
-        print(f"  [ROLLBACK] 已恢复旧 storage state: {STATE_FILE.name}")
-    if PROFILE_DIR.exists():
-        shutil.rmtree(PROFILE_DIR)
-    if profile_backup and profile_backup.exists():
-        profile_backup.rename(PROFILE_DIR)
-        print(f"  [ROLLBACK] 已恢复旧浏览器状态: {PROFILE_DIR.name}/")
-
-
-def finish_successful_relogin(profile_backup: Path | None) -> None:
-    """Keep the new state and remove the bulky old profile backup."""
-    if profile_backup and profile_backup.exists():
-        shutil.rmtree(profile_backup)
-
-
-def main() -> int:
-    ap = argparse.ArgumentParser(description="每周一重登小红书蒲公英")
-    ap.add_argument("--login-timeout", type=int, default=300, help="扫码超时(秒)")
-    ap.add_argument("--dry-run", action="store_true", help="只打印计划,不删不跑")
-    args = ap.parse_args()
-
-    print("== Relogin: 小红书蒲公英 ==")
-    print(f"  今天: {dt.date.today()} ({dt.date.today().strftime('%A')})")
-    print(f"  cookie: {COOKIE_FILE}")
-    print(f"  profile: {PROFILE_DIR}")
-    print()
-
-    if args.dry_run:
-        print("[DRY-RUN] 实际跑会: 删 cookie + 清 profile + 弹二维码")
-        return 0
-
-    cookie_backup, profile_backup, state_backup = reset_state()
-    cmd = [
-        sys.executable,
-        str(SCRAPER),
-        "--login-only",
-        "--login-timeout",
-        str(args.login_timeout),
-    ]
-    print(f"\n  [RUN] {' '.join(cmd)}")
-    try:
-        rc = subprocess.call(
-            cmd,
-            cwd=str(PROJECT_DIR),
-            env=child_browser_environment(
-                cookie_file=COOKIE_FILE,
-                storage_state_file=STATE_FILE,
-                profile_dir=PROFILE_DIR,
-            ),
-        )
-    except BaseException:
-        restore_previous_state(cookie_backup, profile_backup, state_backup)
-        raise
-    print(f"\n  exit={rc}")
-    if rc != 0 or not COOKIE_FILE.exists():
-        print("  [WARN] 新登录未完成,恢复旧登录状态")
-        restore_previous_state(cookie_backup, profile_backup, state_backup)
-        return rc or 1
-    try:
-        sync_relogin_state_to_consumers(
-            "pgy",
-            cookie_file=COOKIE_FILE,
-            storage_state_file=STATE_FILE,
-        )
-    except Exception as exc:
-        restore_previous_state(cookie_backup, profile_backup, state_backup)
-        print(f"  [ROLLBACK] 同步采集器登录态失败: {exc}")
-        return 1
-    finish_successful_relogin(profile_backup)
-    return rc
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -1,279 +0,0 @@
-"""Resolve one relogin wrapper's browser state contract.
-
-The workflow runtime injects a unique browser binding into each relogin
-wrapper.  Standalone legacy launches do not have that environment, so they
-retain the original per-platform files below the content-marketing state root.
-"""
-
-from __future__ import annotations
-
-import json
-import os
-import re
-import shutil
-import tempfile
-import uuid
-from dataclasses import dataclass
-from pathlib import Path
-from typing import Mapping
-
-from gyxx_flow.adapters.integration import RuntimeIntegrationCatalog
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-_PLATFORM = re.compile(r"^[a-z][a-z0-9_-]{0,31}$")
-
-CONSUMER_BINDING_IDS: Mapping[str, tuple[str, ...]] = {
-    "bilibili": ("content.comments.collect_bilibili",),
-    "douyin": ("content.comments.collect_douyin",),
-    "pgy": ("content_marketing:pgy_xhs_scraper_v2.py",),
-    "xiaohongshu": ("content.comments.collect_xiaohongshu",),
-    # v2 now reads the Buyin/Doudian site and is owned by the douyin-shop
-    # account. Keep the legacy Xingtu relogin isolated from that vault.
-    "buyin": ("content_marketing:xingtu_scraper_v2.py",),
-    "xingtu": ("content_marketing:xingtu_scraper.py",),
-}
-
-
-@dataclass(frozen=True)
-class ReloginRuntimePaths:
-    cookie_file: Path
-    storage_state_file: Path
-    profile_dir: Path
-
-
-def resolve_relogin_runtime_paths(
-    platform: str,
-    environment: Mapping[str, str] | None = None,
-) -> ReloginRuntimePaths:
-    """Prefer the injected binding and fall back to original platform paths."""
-
-    if not _PLATFORM.fullmatch(platform):
-        raise ValueError("invalid relogin platform")
-    values = os.environ if environment is None else environment
-    legacy_cookie = PATHS.state_root / "cookies" / f"{platform}_cookies.json"
-    legacy_storage = (
-        PATHS.state_root / "cookies" / f"{platform}_storage_state.json"
-    )
-    legacy_profile = PATHS.state_root / "browser-profiles" / platform
-    return ReloginRuntimePaths(
-        cookie_file=_bound_path(
-            values,
-            "GYXX_BROWSER_COOKIE_FILE",
-            legacy_cookie,
-        ),
-        storage_state_file=_bound_path(
-            values,
-            "GYXX_BROWSER_STORAGE_STATE_FILE",
-            legacy_storage,
-        ),
-        profile_dir=_bound_path(
-            values,
-            "GYXX_BROWSER_PROFILE_DIR",
-            legacy_profile,
-        ),
-    )
-
-
-def child_browser_environment(
-    *,
-    cookie_file: Path,
-    storage_state_file: Path,
-    profile_dir: Path,
-    environment: Mapping[str, str] | None = None,
-) -> dict[str, str]:
-    """Keep the nested login scraper on the wrapper's injected binding."""
-
-    result = dict(os.environ if environment is None else environment)
-    result.update(
-        {
-            "GYXX_BROWSER_COOKIE_FILE": str(cookie_file.resolve()),
-            "GYXX_BROWSER_STORAGE_STATE_FILE": str(
-                storage_state_file.resolve()
-            ),
-            "GYXX_BROWSER_PROFILE_DIR": str(profile_dir.resolve()),
-        }
-    )
-    # content_marketing/sitecustomize.py normally rebinds a nested Python
-    # script from argv[0].  A relogin scraper is an implementation detail of
-    # this wrapper and must write the wrapper binding that the parent monitors.
-    result.pop("GYXX_MODULE_ID", None)
-    return result
-
-
-def sync_relogin_state_to_consumers(
-    platform: str,
-    *,
-    cookie_file: Path,
-    storage_state_file: Path,
-    environment: Mapping[str, str] | None = None,
-) -> tuple[Path, ...]:
-    """Atomically publish a verified relogin state to unique consumer paths."""
-
-    try:
-        consumer_ids = CONSUMER_BINDING_IDS[platform]
-    except KeyError as exc:
-        raise ValueError("invalid relogin platform") from exc
-    values = os.environ if environment is None else environment
-    project_root = values.get("GYXX_PROJECT_ROOT", "").strip()
-    data_root = values.get("GYXX_DATA_ROOT", "").strip()
-    if not project_root and not data_root:
-        # Original standalone launches use the legacy per-platform state files
-        # directly, so there is no separate managed consumer to publish to.
-        return ()
-    if not project_root or not data_root:
-        raise RuntimeError("managed relogin state sync requires project and data roots")
-
-    catalog = RuntimeIntegrationCatalog.load_default(
-        project_root=project_root,
-        data_root=data_root,
-    )
-    source_cookie = cookie_file.expanduser().resolve()
-    source_storage = storage_state_file.expanduser().resolve()
-    cookie_payload = _read_cookie_payload(source_cookie)
-    storage_available = source_storage.is_file()
-    if storage_available:
-        _read_storage_state_payload(source_storage)
-
-    source_paths = {source_cookie, source_storage}
-    destination_paths: set[Path] = set()
-    destinations: list[Path] = []
-    updates: list[tuple[Path, Path]] = []
-    for consumer_id in consumer_ids:
-        binding = catalog.binding_for(consumer_id)
-        target_cookie = binding.cookie_file.resolve()
-        target_storage = binding.storage_state_file.resolve()
-        targets = (target_cookie, target_storage)
-        if any(target in source_paths for target in targets) or any(
-            target in destination_paths for target in targets
-        ):
-            raise RuntimeError(
-                "relogin and consumer browser state paths must be unique"
-            )
-        destination_paths.update(targets)
-        _validate_cookie_requirements(
-            cookie_payload,
-            required_domains=binding.required_cookie_domains,
-            required_names=binding.required_cookie_names,
-        )
-        updates.append((source_cookie, target_cookie))
-        if storage_available:
-            updates.append((source_storage, target_storage))
-        destinations.append(target_cookie)
-    _atomic_replace_files(updates)
-    return tuple(destinations)
-
-
-def _bound_path(
-    environment: Mapping[str, str],
-    name: str,
-    fallback: Path,
-) -> Path:
-    configured = environment.get(name, "").strip()
-    return (
-        Path(configured).expanduser().resolve()
-        if configured
-        else fallback.resolve()
-    )
-
-
-def _read_cookie_payload(path: Path) -> list[dict[str, object]]:
-    try:
-        payload = json.loads(path.read_text(encoding="utf-8"))
-    except (OSError, UnicodeError, ValueError) as exc:
-        raise RuntimeError("relogin cookie state is missing or invalid") from exc
-    if not isinstance(payload, list) or not payload or not all(
-        isinstance(cookie, dict) for cookie in payload
-    ):
-        raise RuntimeError("relogin cookie state is missing or invalid")
-    return payload
-
-
-def _read_storage_state_payload(path: Path) -> dict[str, object]:
-    try:
-        payload = json.loads(path.read_text(encoding="utf-8"))
-    except (OSError, UnicodeError, ValueError) as exc:
-        raise RuntimeError("relogin storage state is invalid") from exc
-    if not isinstance(payload, dict):
-        raise RuntimeError("relogin storage state is invalid")
-    return payload
-
-
-def _validate_cookie_requirements(
-    cookies: list[dict[str, object]],
-    *,
-    required_domains: tuple[str, ...],
-    required_names: tuple[str, ...],
-) -> None:
-    usable = [cookie for cookie in cookies if cookie.get("value")]
-    if not usable:
-        raise RuntimeError("relogin cookie state has no usable cookies")
-    if required_domains and not any(
-        any(
-            str(cookie.get("domain", "")).lstrip(".").casefold()
-            .endswith(domain.casefold().lstrip("."))
-            for domain in required_domains
-        )
-        for cookie in usable
-    ):
-        raise RuntimeError("relogin cookie state does not match consumer domain")
-    if required_names and not any(
-        str(cookie.get("name", "")) in required_names for cookie in usable
-    ):
-        raise RuntimeError("relogin cookie state misses the consumer login cookie")
-
-
-def _atomic_replace_files(updates: list[tuple[Path, Path]]) -> None:
-    """Replace a small related file set and restore every old target on error."""
-
-    transaction = uuid.uuid4().hex
-    staged: list[tuple[Path, Path]] = []
-    backups: dict[Path, Path | None] = {}
-    replaced: list[Path] = []
-    try:
-        for source, target in updates:
-            target.parent.mkdir(parents=True, exist_ok=True)
-            handle, raw_temp = tempfile.mkstemp(
-                prefix=f".{target.name}.{transaction}.",
-                suffix=".tmp",
-                dir=target.parent,
-            )
-            temp = Path(raw_temp)
-            with os.fdopen(handle, "wb") as stream:
-                stream.write(source.read_bytes())
-                stream.flush()
-                os.fsync(stream.fileno())
-            staged.append((temp, target))
-            backup = target.with_name(f".{target.name}.{transaction}.bak")
-            if target.is_file():
-                shutil.copy2(target, backup)
-                backups[target] = backup
-            else:
-                backups[target] = None
-
-        for temp, target in staged:
-            os.replace(temp, target)
-            replaced.append(target)
-    except BaseException:
-        for target in reversed(replaced):
-            backup = backups.get(target)
-            if backup is not None and backup.exists():
-                os.replace(backup, target)
-            elif target.exists():
-                target.unlink()
-        raise
-    finally:
-        for temp, _target in staged:
-            if temp.exists():
-                temp.unlink()
-        for backup in backups.values():
-            if backup is not None and backup.exists():
-                backup.unlink()
-
-
-__all__ = [
-    "CONSUMER_BINDING_IDS",
-    "ReloginRuntimePaths",
-    "child_browser_environment",
-    "resolve_relogin_runtime_paths",
-    "sync_relogin_state_to_consumers",
-]
@@ -1,47 +0,0 @@
-"""Transactional backup/rollback helpers for interactive relogin scripts."""
-
-from __future__ import annotations
-
-import datetime as dt
-import shutil
-from dataclasses import dataclass
-from pathlib import Path
-
-
-@dataclass
-class LoginStateBackup:
-    files: dict[Path, Path]
-    profile: Path
-    profile_backup: Path | None
-
-
-def begin(files: tuple[Path, ...], profile: Path) -> LoginStateBackup:
-    stamp = dt.datetime.now().strftime("%Y%m%d_%H%M%S_%f")
-    backups: dict[Path, Path] = {}
-    for source in files:
-        if source.exists():
-            backup = source.with_name(f"{source.name}.bak.{stamp}")
-            source.rename(backup)
-            backups[source] = backup
-    profile_backup = None
-    if profile.exists():
-        profile_backup = profile.with_name(f"{profile.name}.bak.{stamp}")
-        profile.rename(profile_backup)
-    return LoginStateBackup(backups, profile, profile_backup)
-
-
-def rollback(state: LoginStateBackup) -> None:
-    for source, backup in state.files.items():
-        if source.exists():
-            source.unlink()
-        if backup.exists():
-            backup.rename(source)
-    if state.profile.exists():
-        shutil.rmtree(state.profile)
-    if state.profile_backup and state.profile_backup.exists():
-        state.profile_backup.rename(state.profile)
-
-
-def commit(state: LoginStateBackup) -> None:
-    if state.profile_backup and state.profile_backup.exists():
-        shutil.rmtree(state.profile_backup)
@@ -1,116 +0,0 @@
-"""每周重登小红书。
-
-强制清掉 cookie JSON + 浏览器 profile,确保下次启动时必须扫码。
-跑法:
-  python data/tools/relogin_xiaohongshu.py             # 实际跑(会弹二维码)
-  python data/tools/relogin_xiaohongshu.py --dry-run   # 只看会做什么
-"""
-
-import argparse
-import datetime as dt
-import json
-import subprocess
-import sys
-
-from gyxx_flow.modules.content_marketing.data.tools.relogin_runtime_paths import (
-    child_browser_environment,
-    resolve_relogin_runtime_paths,
-    sync_relogin_state_to_consumers,
-)
-from gyxx_flow.modules.content_marketing.data.tools.relogin_transaction import (
-    begin,
-    commit,
-    rollback,
-)
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-PROJECT_DIR = PATHS.module_root
-RUNTIME_PATHS = resolve_relogin_runtime_paths("xiaohongshu")
-COOKIE_FILE = RUNTIME_PATHS.cookie_file
-STATE_FILE = RUNTIME_PATHS.storage_state_file
-PROFILE_DIR = RUNTIME_PATHS.profile_dir
-SCRAPER = PROJECT_DIR / "xiaohongshu_comment_scraper.py"
-
-LOGIN_COOKIE = "web_session"
-
-
-def reset_state():
-    state = begin((COOKIE_FILE, STATE_FILE), PROFILE_DIR)
-    print("  [OK] 旧 Cookie/Profile 已暂存,失败时将自动恢复")
-    return state
-
-
-def is_cookie_valid() -> bool:
-    if not COOKIE_FILE.exists():
-        return False
-    try:
-        cookies = json.loads(COOKIE_FILE.read_text(encoding="utf-8"))
-        return any(c.get("name") == LOGIN_COOKIE and c.get("value") for c in cookies)
-    except Exception:
-        return False
-
-
-def main() -> int:
-    ap = argparse.ArgumentParser(description="重登小红书")
-    ap.add_argument("--login-timeout", type=int, default=300, help="扫码超时(秒)")
-    ap.add_argument("--dry-run", action="store_true", help="只打印计划,不删不跑")
-    args = ap.parse_args()
-
-    print("== Relogin: 小红书 ==")
-    print(f"  今天: {dt.date.today()}")
-    print(f"  cookie: {COOKIE_FILE}")
-    print(f"  profile: {PROFILE_DIR}")
-    print(f"  当前 cookie 状态: {'有效' if is_cookie_valid() else '无效/不存在'}")
-    print()
-
-    if args.dry_run:
-        print("[DRY-RUN] 实际跑会: 删 cookie + 清 profile + 弹二维码")
-        return 0
-
-    old_state = reset_state()
-    cmd = [
-        sys.executable,
-        str(SCRAPER),
-        "login",
-        "--login-only",
-        "--login-timeout",
-        str(args.login_timeout),
-    ]
-    print(f"\n  [RUN] {' '.join(cmd)}")
-    try:
-        rc = subprocess.call(
-            cmd,
-            cwd=str(PROJECT_DIR),
-            env=child_browser_environment(
-                cookie_file=COOKIE_FILE,
-                storage_state_file=STATE_FILE,
-                profile_dir=PROFILE_DIR,
-            ),
-        )
-    except BaseException:
-        rollback(old_state)
-        raise
-    print(f"\n  exit={rc}")
-    login_ok = rc == 0 and is_cookie_valid()
-    if rc == 0:
-        print(f"  {LOGIN_COOKIE} 写入: {'OK' if login_ok else 'FAIL — 没有看到 web_session,登录可能没成功'}")
-    if not login_ok:
-        rollback(old_state)
-        print("  [ROLLBACK] 未完成扫码,已恢复旧登录状态")
-        return rc or 1
-    try:
-        sync_relogin_state_to_consumers(
-            "xiaohongshu",
-            cookie_file=COOKIE_FILE,
-            storage_state_file=STATE_FILE,
-        )
-    except Exception as exc:
-        rollback(old_state)
-        print(f"  [ROLLBACK] 同步采集器登录态失败: {exc}")
-        return 1
-    commit(old_state)
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -1,350 +0,0 @@
-"""每 3 个周一重登一次抖音/星图。
-
-强制清掉 cookie JSON + 浏览器 profile,确保下次启动时必须扫码。
-星图登录完成后,自动把登录态 cookie 同步到抖音评论 scraper,避免再扫一次码。
-
-任务计划: 每周一 09:00 (脚本内部判断是否到了该重登的周一,不到就退出)
-
-3 周循环基准: 任意一个周一,后续 (今天 - BASE_DATE) // 7 % 3 == 0 才是星图周。
-改 BASE_DATE 就改了重登周期。
-
-跑法:
-  python data/tools/relogin_xingtu.py             # 实际跑
-  python data/tools/relogin_xingtu.py --dry-run   # 看今天是不是星图周
-  python data/tools/relogin_xingtu.py --force     # 忽略 3 周判断,强制重登
-"""
-
-import argparse
-import datetime as dt
-import json
-import os
-import shutil
-import subprocess
-import sys
-from pathlib import Path
-
-import psutil
-
-from gyxx_flow.modules.content_marketing.data.tools.relogin_runtime_paths import (
-    child_browser_environment,
-    resolve_relogin_runtime_paths,
-    sync_relogin_state_to_consumers,
-)
-from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
-
-PROJECT_DIR = PATHS.module_root
-RUNTIME_PATHS = resolve_relogin_runtime_paths("xingtu")
-COOKIE_FILE = RUNTIME_PATHS.cookie_file
-STATE_FILE = RUNTIME_PATHS.storage_state_file
-PROFILE_DIR = RUNTIME_PATHS.profile_dir
-SCRAPER = PROJECT_DIR / "xingtu_scraper_v2.py"
-
-# 抖音评论 scraper 的 cookie 文件 (登录态共享目标)
-DY_COOKIE_FILE = PATHS.state_root / "cookies/douyin_cookies.json"
-DY_STATE_FILE = PATHS.state_root / "cookies/douyin_storage_state.json"
-DY_PROFILE_DIR = PATHS.state_root / "browser-profiles/douyin"
-
-# 需要同步的抖音登录 cookie 名
-DY_LOGIN_COOKIES = {"sessionid", "sessionid_ss", "sid_guard", "sid_tt", "passport_csrf_token"}
-
-# 3 周循环起点(必须是周一)。星图 cookie ~22 天,每 3 周(21 天)重登正好。
-BASE_DATE = dt.date(2026, 6, 22)
-LOGIN_COOKIE_NAMES = {"sessionid", "sessionid_ss", "sid_tt", "sid_guard"}
-
-
-def transaction_file() -> Path:
-    return COOKIE_FILE.with_name(".xingtu_relogin_transaction.json")
-
-
-def _write_transaction(
-    cookie_backup: Path | None,
-    profile_backup: Path | None,
-    state_backup: Path | None,
-) -> None:
-    marker = transaction_file()
-    marker.parent.mkdir(parents=True, exist_ok=True)
-    temp = marker.with_name(f"{marker.name}.tmp")
-    temp.write_text(json.dumps({
-        "cookie_backup": str(cookie_backup) if cookie_backup else "",
-        "state_backup": str(state_backup) if state_backup else "",
-        "profile_backup": str(profile_backup) if profile_backup else "",
-    }, ensure_ascii=False, indent=2), encoding="utf-8")
-    os.replace(temp, marker)
-
-
-def _clear_transaction() -> None:
-    marker = transaction_file()
-    if marker.exists():
-        marker.unlink()
-
-
-def _remove_path(path: Path) -> None:
-    if path.is_dir():
-        shutil.rmtree(path)
-    elif path.exists():
-        path.unlink()
-
-
-def recover_interrupted_state() -> bool:
-    """Restore a relogin transaction that was killed after moving old state."""
-    marker = transaction_file()
-    if not marker.exists():
-        return False
-    try:
-        state = json.loads(marker.read_text(encoding="utf-8"))
-    except Exception:
-        return False
-
-    cookie_backup = Path(state["cookie_backup"]) if state.get("cookie_backup") else None
-    state_backup = Path(state["state_backup"]) if state.get("state_backup") else None
-    profile_backup = Path(state["profile_backup"]) if state.get("profile_backup") else None
-    terminate_profile_processes()
-    if cookie_backup and cookie_backup.exists():
-        _remove_path(COOKIE_FILE)
-        cookie_backup.rename(COOKIE_FILE)
-    if state_backup and state_backup.exists():
-        _remove_path(STATE_FILE)
-        state_backup.rename(STATE_FILE)
-    if profile_backup and profile_backup.exists():
-        _remove_path(PROFILE_DIR)
-        profile_backup.rename(PROFILE_DIR)
-    _clear_transaction()
-    print("  [RECOVER] 已恢复上次被中断的星图登录状态")
-    return True
-
-
-def has_valid_login_cookie_file(path: Path | None = None) -> bool:
-    path = path or COOKIE_FILE
-    try:
-        cookies = json.loads(path.read_text(encoding="utf-8"))
-    except Exception:
-        return False
-    return any(
-        isinstance(cookie, dict)
-        and cookie.get("name") in LOGIN_COOKIE_NAMES
-        and cookie.get("value")
-        for cookie in cookies
-    )
-
-
-def terminate_profile_processes() -> None:
-    """Stop only browser processes using Xingtu's dedicated profile."""
-    profile_text = str(PROFILE_DIR.resolve()).casefold()
-    matches = []
-    for process in psutil.process_iter(["pid", "cmdline"]):
-        try:
-            command = " ".join(process.info.get("cmdline") or []).casefold()
-            if process.pid != os.getpid() and profile_text in command:
-                matches.append(process)
-        except (psutil.Error, OSError):
-            continue
-    for process in matches:
-        try:
-            process.terminate()
-        except (psutil.Error, OSError):
-            pass
-    _gone, alive = psutil.wait_procs(matches, timeout=5)
-    for process in alive:
-        try:
-            process.kill()
-        except (psutil.Error, OSError):
-            pass
-
-
-def is_xingtu_week(today: dt.date, base: dt.date) -> bool:
-    days = (today - base).days
-    if days < 0:
-        return False
-    return (days // 7) % 3 == 0
-
-
-def reset_state() -> tuple[Path | None, Path | None, Path | None]:
-    """Temporarily move the old login state aside so it can be restored."""
-    ts = dt.datetime.now().strftime("%Y%m%d_%H%M%S_%f")
-    cookie_backup = None
-    state_backup = None
-    profile_backup = None
-    if COOKIE_FILE.exists():
-        cookie_backup = COOKIE_FILE.with_suffix(f".json.bak.{ts}")
-    if STATE_FILE.exists():
-        state_backup = STATE_FILE.with_suffix(f".json.bak.{ts}")
-    if PROFILE_DIR.exists():
-        profile_backup = PROFILE_DIR.with_name(f"{PROFILE_DIR.name}.bak.{ts}")
-    _write_transaction(cookie_backup, profile_backup, state_backup)
-
-    if cookie_backup:
-        COOKIE_FILE.rename(cookie_backup)
-        print(f"  [OK] 备份 cookie -> {cookie_backup.name}")
-    else:
-        print(f"  [INFO] {COOKIE_FILE.name} 不存在,跳过")
-    if state_backup:
-        STATE_FILE.rename(state_backup)
-        print(f"  [OK] 备份 storage state -> {state_backup.name}")
-    else:
-        print(f"  [INFO] {STATE_FILE.name} 不存在,跳过")
-    if profile_backup:
-        PROFILE_DIR.rename(profile_backup)
-        print(f"  [OK] 暂存 {PROFILE_DIR.name}/ -> {profile_backup.name}/")
-    else:
-        print(f"  [INFO] {PROFILE_DIR.name}/ 不存在,跳过")
-    return cookie_backup, profile_backup, state_backup
-
-
-def restore_previous_state(
-    cookie_backup: Path | None,
-    profile_backup: Path | None,
-    state_backup: Path | None = None,
-) -> None:
-    """Discard an incomplete login and restore the last known-good state."""
-    terminate_profile_processes()
-    if COOKIE_FILE.exists():
-        COOKIE_FILE.unlink()
-    if cookie_backup and cookie_backup.exists():
-        cookie_backup.rename(COOKIE_FILE)
-        print(f"  [ROLLBACK] 已恢复旧 cookie: {COOKIE_FILE.name}")
-    _remove_path(STATE_FILE)
-    if state_backup and state_backup.exists():
-        state_backup.rename(STATE_FILE)
-        print(f"  [ROLLBACK] 已恢复旧 storage state: {STATE_FILE.name}")
-    if PROFILE_DIR.exists():
-        shutil.rmtree(PROFILE_DIR)
-    if profile_backup and profile_backup.exists():
-        profile_backup.rename(PROFILE_DIR)
-        print(f"  [ROLLBACK] 已恢复旧浏览器状态: {PROFILE_DIR.name}/")
-    _clear_transaction()
-
-
-def finish_successful_relogin(profile_backup: Path | None) -> None:
-    """Keep the new state and remove the bulky old profile backup."""
-    if profile_backup and profile_backup.exists():
-        shutil.rmtree(profile_backup)
-    _clear_transaction()
-
-
-def sync_to_douyin_scraper() -> None:
-    """把星图的登录态 cookie 同步到抖音评论 scraper,避免再扫一次码。"""
-    if not COOKIE_FILE.exists():
-        print("  [SKIP] 星图 cookie 不存在,无法同步")
-        return
-
-    try:
-        xingtu_cookies = json.loads(COOKIE_FILE.read_text(encoding="utf-8"))
-    except Exception as exc:
-        print(f"  [SKIP] 读取星图 cookie 失败: {exc}")
-        return
-
-    # 提取登录态 cookie
-    login_cookies = [c for c in xingtu_cookies if isinstance(c, dict) and c.get("name") in DY_LOGIN_COOKIES]
-    if not login_cookies:
-        print("  [SKIP] 星图 cookie 中没有找到登录态,跳过同步")
-        return
-
-    # 清除抖音评论 scraper 的旧 profile (强制刷新)
-    if DY_PROFILE_DIR.exists():
-        shutil.rmtree(DY_PROFILE_DIR)
-        print(f"  [OK] 清空 {DY_PROFILE_DIR.name}/")
-
-    # 读取或创建抖音评论 cookie 文件
-    dy_cookies = []
-    if DY_COOKIE_FILE.exists():
-        try:
-            dy_cookies = json.loads(DY_COOKIE_FILE.read_text(encoding="utf-8"))
-        except Exception:
-            dy_cookies = []
-
-    # 合并: 同名 cookie 覆盖,新 cookie 追加
-    dy_by_name = {c.get("name"): c for c in dy_cookies if isinstance(c, dict)}
-    for c in login_cookies:
-        # 修正 domain: 星图的 cookie domain 可能是 .douyin.com 子域,评论采集需要 .douyin.com
-        domain = c.get("domain", "")
-        if "xingtu" in domain or "oceanengine" in domain:
-            c["domain"] = ".douyin.com"
-        dy_by_name[c["name"]] = c
-
-    DY_COOKIE_FILE.parent.mkdir(parents=True, exist_ok=True)
-    DY_COOKIE_FILE.write_text(
-        json.dumps(list(dy_by_name.values()), ensure_ascii=False, indent=2),
-        encoding="utf-8",
-    )
-    print(f"  [OK] 同步 {len(login_cookies)} 个登录 cookie 到 {DY_COOKIE_FILE.name}")
-
-
-def main() -> int:
-    ap = argparse.ArgumentParser(description="每 3 周周一重登星图")
-    ap.add_argument("--base-date", default=BASE_DATE.isoformat(),
-                    help="3 周循环起点(必须是周一),格式 YYYY-MM-DD")
-    ap.add_argument("--login-timeout", type=int, default=300)
-    ap.add_argument("--force", action="store_true", help="忽略 3 周判断,强制重登")
-    ap.add_argument("--dry-run", action="store_true", help="只打印计划")
-    args = ap.parse_args()
-
-    today = dt.date.today()
-    base = dt.date.fromisoformat(args.base_date)
-    is_xw = is_xingtu_week(today, base)
-
-    print("== Relogin: 抖音/星图 ==")
-    print(f"  今天: {today} ({today.strftime('%A')})")
-    print(f"  3 周基准: {base}")
-    print(f"  星图周? {'是' if is_xw else '否'}")
-    print()
-
-    if not is_xw and not args.force:
-        print("今天不是星图周,跳过 (传 --force 强制重登)")
-        return 0
-
-    if args.dry_run:
-        print("[DRY-RUN] 实际跑会: 删 cookie + 清 profile + 弹二维码")
-        return 0
-
-    recover_interrupted_state()
-    terminate_profile_processes()
-    cookie_backup, profile_backup, state_backup = reset_state()
-    cmd = [
-        sys.executable,
-        str(SCRAPER),
-        "--login-only",
-        "--login-timeout",
-        str(args.login_timeout),
-    ]
-    print(f"\n  [RUN] {' '.join(cmd)}")
-    try:
-        rc = subprocess.call(
-            cmd,
-            cwd=str(PROJECT_DIR),
-            env=child_browser_environment(
-                cookie_file=COOKIE_FILE,
-                storage_state_file=STATE_FILE,
-                profile_dir=PROFILE_DIR,
-            ),
-        )
-    except BaseException:
-        restore_previous_state(cookie_backup, profile_backup, state_backup)
-        raise
-    finally:
-        terminate_profile_processes()
-    print(f"\n  exit={rc}")
-
-    if rc != 0 or not has_valid_login_cookie_file():
-        print("  [WARN] 新登录未完成,恢复旧登录状态")
-        restore_previous_state(cookie_backup, profile_backup, state_backup)
-        return rc or 1
-    try:
-        sync_relogin_state_to_consumers(
-            "xingtu",
-            cookie_file=COOKIE_FILE,
-            storage_state_file=STATE_FILE,
-        )
-    except Exception as exc:
-        restore_previous_state(cookie_backup, profile_backup, state_backup)
-        print(f"  [ROLLBACK] 同步采集器登录态失败: {exc}")
-        return 1
-
-    finish_successful_relogin(profile_backup)
-    if rc == 0:
-        sync_to_douyin_scraper()
-
-    return rc
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -426,9 +426,21 @@ def _backfill_record(mapping: dict, style_name: str, rid: str) -> dict | None:


 def _style_id_lookup(cur) -> dict[str, int]:
-    """name -> cmt_styles.id 映射(一次查全表)"""
+    """name -> 笔记归属款 id 映射(一次查全表)。
+
+    2026-10-27 起优先 dim_style 主档 id（data-hub 按 dim_style 消费；直接用
+    cmt_styles.id 在款式主档重排后会变成悬挂 id，笔记在系统中"消失"）。
+    dim_style 缺同名款时回退 cmt_styles.id，待款式收录 + 存量重映射迁移归一。
+    同名多行取最小 id。
+    """
+    mapping: dict[str, int] = {}
+    cur.execute("SELECT style_name, id FROM dim_style ORDER BY id")
+    for name, sid in cur.fetchall():
+        mapping.setdefault(str(name), sid)
    cur.execute("SELECT id, name FROM cmt_styles")
-    return {name: id for id, name in cur.fetchall()}
+    for sid, name in cur.fetchall():
+        mapping.setdefault(str(name), sid)
+    return mapping


 def _ensure_creator(cur, name: str | None) -> int | None:
@@ -219,6 +219,36 @@ def parse_wan_count(text: str) -> int | None:
    return None


+# 抖音"复制分享文案"格式：标题 + 话题链接段 [话题](url)（个别登记写成了 {url} 花括号包裹），
+# 或裸 URL。登记人常把整段分享文案粘进「发布笔记标题」列，入库前剥掉链接只留标题文本。
+_TITLE_LINK_RE = re.compile(
+    r"\[[^\]]*\]\([^)]*\)"          # [text](url) 圆括号链接
+    r"|\[[^\]]*\]\{[^}]*\}"         # [text]{url} 花括号链接
+    r"|\{https?://[^}]*\}"          # {url} 裸花括号链接
+    r"|https?://[^\s）)\]]+"        # 裸 URL
+)
+
+
+def clean_note_title(raw: str) -> str:
+    """剥掉标题里混入的话题链接段与裸 URL，只留标题文本。
+
+    兜底顺序：剥链接后非空 → 用它；整段都是链接（标题只剩话题/搜索词）→
+    取方括号内的文本去重拼接；再不行 → 保留原文，避免落空。
+    """
+    text = raw or ""
+    cleaned = _TITLE_LINK_RE.sub("", text)
+    cleaned = cleaned.replace("%20", " ")  # 链接段之间残留的 URL 编码空格
+    cleaned = re.sub(r"\s{2,}", " ", cleaned).strip()
+    if cleaned:
+        return cleaned
+    inner = [t.strip() for t in re.findall(r"\[([^\]]*)\]", text)]
+    seen: list[str] = []
+    for item in inner:
+        if item and item not in seen:
+            seen.append(item)
+    return " ".join(seen) if seen else text.strip()
+
+
 def normalize_publish_time(cell: Any) -> str | None:
    text = cell_text(cell)
    if not text:
@@ -340,7 +370,7 @@ def parse_records(
                record_id=record_id,
                platform=resolve_platform(value(record, "platform"), url),
                creator_name=creator_name,
-                title=cell_text(value(record, "note_title"))[:512],
+                title=clean_note_title(cell_text(value(record, "note_title")))[:512],
                publish_time=normalize_publish_time(value(record, "publish_time")),
                url=url,
                creator_attrs=creator_attrs,
@@ -479,6 +509,22 @@ def sync_table(
        raise ValueError("mapping 缺少款式/base_token/table_id")

    style_id = _ensure_style(cur, style_name)
+    # 2026-10-27 起笔记归属改挂 dim_style 主档 id（data-hub 按 dim_style 系列锚点款消费，
+    # 历史上直接写 cmt_styles.id 在款式主档重排后变成悬挂 id，导致笔记在系统里"消失"）。
+    # 按款名解析 dim_style.id；解析不到（新款尚未收录款式库）时回退 cmt_styles.id 并告警，
+    # 后续由款式收录 + 存量重映射迁移归一。
+    cur.execute(
+        "SELECT id FROM dim_style WHERE style_name = %s ORDER BY id LIMIT 1",
+        (style_name,),
+    )
+    dim_row = cur.fetchone()
+    if dim_row:
+        style_id = dim_row[0]
+    else:
+        _log(
+            f"[{style_name}] dim_style 未找到同名款式，笔记暂挂 cmt_styles.id={style_id}，"
+            "待款式收录后归一"
+        )
    creators = 0
    for row in rows:
        creator_id: int | None = None
@@ -426,6 +426,21 @@ def ensure_table(cur) -> None:
        )


+def drop_stale_rows(
+    all_rows: dict[tuple, tuple], stale_paths: list[Path]
+) -> int:
+    """删除来源文件已陈旧的行,返回删除数(按行内 source_file 文件名匹配)。"""
+    stale_names = {Path(path).name for path in stale_paths}
+    source_index = ROW_FIELDS.index("source_file")
+    stale_keys = [
+        key for key, row in all_rows.items()
+        if str(row[source_index]) in stale_names
+    ]
+    for key in stale_keys:
+        del all_rows[key]
+    return len(stale_keys)
+
+
 def main() -> int:
    parser = argparse.ArgumentParser(
        description="同步自营内容(抖音/B站/小红书)全量指标到 cmt_self_video_metrics"
@@ -504,20 +519,25 @@ def main() -> int:
        print("❌ 没有任何平台的来源文件可用")
        return 1
    stale = [
-        str(path)
+        path
        for path in sources
        if business_date_bound
        and not args.allow_stale
        and not args.use_file_mtime
        and not is_current_artifact(path, metric_date)
    ]
-    if stale and not args.allow_stale:
+    if stale:
+        # 陈旧防护按来源文件粒度跳过: 某平台来源过期只丢弃该平台的行,
+        # 不拖累其他平台当天的新鲜数据(如抖音导出失败时不拦 B站/小红书)。
+        dropped = drop_stale_rows(all_rows, stale)
        print(
-            f"❌ 以下来源文件不是业务日期({metric_date.isoformat()})的产物; "
-            "补历史请显式传 --metric-date 并加 --allow-stale:\n  "
-            + "\n  ".join(stale)
+            f"⚠️ 以下来源文件不是业务日期({metric_date.isoformat()})的产物, "
+            f"已跳过对应 {dropped} 行(补历史用 --metric-date + "
+            "--allow-stale):\n  " + "\n  ".join(str(p) for p in stale)
        )
-        return 1
+        if not all_rows:
+            print("❌ 跳过陈旧来源后没有可入库的内容指标行")
+            return 1

    print(
        f"[sync-self] 平台={'+'.join(selected_platforms)} 待写={len(all_rows)} 行; "
@@ -543,74 +563,98 @@ def main() -> int:
    from gyxx_flow.modules.content_marketing.data.tools.db import get_db_config

    with psycopg.connect(**get_db_config()) as conn:
-        with conn.cursor() as cur:
-            ensure_table(cur)
-            for row in all_rows.values():
-                # SQL 为静态字面量,29 个值全部经 %s 占位符参数绑定传入。
-                cur.execute(
-                    """
-                    INSERT INTO cmt_self_video_metrics (
-                        platform, account_alias, account_name,
-                        platform_account_id, video_id, title, chanmama_url,
-                        video_url, duration, publish_time, metric_date,
-                        exposure, like_count, comment_count, share_count,
-                        est_sales_count, est_sales_amount, note_type,
-                        creation_method, publish_status, mounted_products,
-                        homepage_redirect_count, live_redirect_count,
-                        payment_amount, payment_orders, coin_count,
-                        favorite_count, danmaku_count, page_metrics,
-                        source_file,
-                        observed_at, created_at, updated_at
-                    ) VALUES (
-                        %s, %s, %s,
-                        %s, %s, %s, %s,
-                        %s, %s, %s, %s,
-                        %s, %s, %s, %s,
-                        %s, %s, %s,
-                        %s, %s, %s,
-                        %s, %s,
-                        %s, %s, %s,
-                        %s, %s, %s,
-                        %s,
-                        CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, CURRENT_TIMESTAMP
-                    )
-                    ON CONFLICT (platform, account_name, video_id, metric_date)
-                    DO UPDATE SET
-                        account_alias = EXCLUDED.account_alias,
-                        platform_account_id = EXCLUDED.platform_account_id,
-                        title = EXCLUDED.title,
-                        chanmama_url = EXCLUDED.chanmama_url,
-                        video_url = EXCLUDED.video_url,
-                        duration = EXCLUDED.duration,
-                        publish_time = EXCLUDED.publish_time,
-                        exposure = EXCLUDED.exposure,
-                        like_count = EXCLUDED.like_count,
-                        comment_count = EXCLUDED.comment_count,
-                        share_count = EXCLUDED.share_count,
-                        est_sales_count = EXCLUDED.est_sales_count,
-                        est_sales_amount = EXCLUDED.est_sales_amount,
-                        note_type = EXCLUDED.note_type,
-                        creation_method = EXCLUDED.creation_method,
-                        publish_status = EXCLUDED.publish_status,
-                        mounted_products = EXCLUDED.mounted_products,
-                        homepage_redirect_count = EXCLUDED.homepage_redirect_count,
-                        live_redirect_count = EXCLUDED.live_redirect_count,
-                        payment_amount = EXCLUDED.payment_amount,
-                        payment_orders = EXCLUDED.payment_orders,
-                        coin_count = EXCLUDED.coin_count,
-                        favorite_count = EXCLUDED.favorite_count,
-                        danmaku_count = EXCLUDED.danmaku_count,
-                        page_metrics = EXCLUDED.page_metrics,
-                        source_file = EXCLUDED.source_file,
-                        observed_at = CURRENT_TIMESTAMP,
-                        updated_at = CURRENT_TIMESTAMP
-                    """,
-                    row,
-                )
-        conn.commit()
-    print(f"[sync-self] 完成: 已幂等写入 {len(all_rows)} 行到 {TABLE_NAME}")
+        written, failed = upsert_rows(conn, list(all_rows.values()))
+    if failed:
+        print(f"[sync-self] 完成: 写入 {written} 行,单行失败跳过 {failed} 行 "
+              f"(见上方 ⚠️ 日志)")
+        return 0 if written else 1
+    print(f"[sync-self] 完成: 已幂等写入 {written} 行到 {TABLE_NAME}")
    return 0


+def upsert_rows(conn, rows: list[tuple]) -> tuple[int, int]:
+    """逐行 upsert,返回 (成功数, 失败数)。
+
+    每行放在独立保存点里执行: 单行失败(如超长值违反约束)只回滚该行,
+    不拖累其他平台/其他内容的写入——这是「一个平台失败不影响全库」的
+    最后一道隔离。
+    """
+    written = 0
+    failed = 0
+    with conn.cursor() as cur:
+        ensure_table(cur)
+        for row in rows:
+            try:
+                with conn.transaction():
+                    # SQL 为静态字面量,29 个值全部经 %s 占位符参数绑定传入。
+                    cur.execute(
+                        """
+                        INSERT INTO cmt_self_video_metrics (
+                            platform, account_alias, account_name,
+                            platform_account_id, video_id, title, chanmama_url,
+                            video_url, duration, publish_time, metric_date,
+                            exposure, like_count, comment_count, share_count,
+                            est_sales_count, est_sales_amount, note_type,
+                            creation_method, publish_status, mounted_products,
+                            homepage_redirect_count, live_redirect_count,
+                            payment_amount, payment_orders, coin_count,
+                            favorite_count, danmaku_count, page_metrics,
+                            source_file,
+                            observed_at, created_at, updated_at
+                        ) VALUES (
+                            %s, %s, %s,
+                            %s, %s, %s, %s,
+                            %s, %s, %s, %s,
+                            %s, %s, %s, %s,
+                            %s, %s, %s,
+                            %s, %s, %s,
+                            %s, %s,
+                            %s, %s, %s,
+                            %s, %s, %s,
+                            %s,
+                            CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, CURRENT_TIMESTAMP
+                        )
+                        ON CONFLICT (platform, account_name, video_id, metric_date)
+                        DO UPDATE SET
+                            account_alias = EXCLUDED.account_alias,
+                            platform_account_id = EXCLUDED.platform_account_id,
+                            title = EXCLUDED.title,
+                            chanmama_url = EXCLUDED.chanmama_url,
+                            video_url = EXCLUDED.video_url,
+                            duration = EXCLUDED.duration,
+                            publish_time = EXCLUDED.publish_time,
+                            exposure = EXCLUDED.exposure,
+                            like_count = EXCLUDED.like_count,
+                            comment_count = EXCLUDED.comment_count,
+                            share_count = EXCLUDED.share_count,
+                            est_sales_count = EXCLUDED.est_sales_count,
+                            est_sales_amount = EXCLUDED.est_sales_amount,
+                            note_type = EXCLUDED.note_type,
+                            creation_method = EXCLUDED.creation_method,
+                            publish_status = EXCLUDED.publish_status,
+                            mounted_products = EXCLUDED.mounted_products,
+                            homepage_redirect_count = EXCLUDED.homepage_redirect_count,
+                            live_redirect_count = EXCLUDED.live_redirect_count,
+                            payment_amount = EXCLUDED.payment_amount,
+                            payment_orders = EXCLUDED.payment_orders,
+                            coin_count = EXCLUDED.coin_count,
+                            favorite_count = EXCLUDED.favorite_count,
+                            danmaku_count = EXCLUDED.danmaku_count,
+                            page_metrics = EXCLUDED.page_metrics,
+                            source_file = EXCLUDED.source_file,
+                            observed_at = CURRENT_TIMESTAMP,
+                            updated_at = CURRENT_TIMESTAMP
+                        """,
+                        row,
+                    )
+            except Exception as exc:
+                failed += 1
+                print(f"  ⚠️ 单行写入失败已隔离跳过: {type(exc).__name__}: {exc}")
+            else:
+                written += 1
+    conn.commit()
+    return written, failed
+
+
 if __name__ == "__main__":
    raise SystemExit(main())
@@ -273,18 +273,15 @@ def maybe_wait_for_login(
    *,
    deadline: float | None = None,
 ) -> None:
-    login_deadline = time.monotonic() + login_timeout
-    while time.monotonic() < login_deadline:
-        _check_note_deadline(deadline, "login wait")
-        try:
-            body = page.locator("body").inner_text(timeout=5000)
-        except Exception:
-            body = ""
-        if "扫码登录" not in body and "登录后" not in body:
-            return
-        log("Douyin may require login. Please finish login in the opened Chrome window.")
-        page.wait_for_timeout(3000)
-    _check_note_deadline(deadline, "login wait")
+    # 登录统一由账号 vault 管理（gyxx accounts login douyin-content），
+    # 由工作流执行前的 cookie 网关保证；运行中失效立即降级，不再等待人工扫码。
+    try:
+        body = page.locator("body").inner_text(timeout=5000)
+    except Exception:
+        body = ""
+    if "扫码登录" in body or "登录后" in body:
+        log("Douyin login state missing; collecting degraded samples. "
+            "Run `gyxx accounts login douyin-content` to restore full comments.")


 def is_logged_in(page: Page) -> bool:
@@ -833,55 +830,16 @@ def write_outputs(url: str, comments: list[dict[str, Any]], result: dict[str, An
    return json_path, csv_path


-def login_only(login_timeout: int) -> int:
-    """只触发登录(headed),登录完成立刻退出。给 relogin_douyin.py 用。"""
-    def action(page: Page) -> None:
-        page.goto("https://www.douyin.com/", wait_until="domcontentloaded", timeout=90000)
-        page.wait_for_timeout(2500)
-        # 主动触发登录弹窗(点登录按钮)
-        try:
-            login_btn = page.get_by_text("登录", exact=True).first
-            if login_btn.count():
-                login_btn.click(timeout=3000)
-                page.wait_for_timeout(1500)
-        except Exception:
-            pass
-        maybe_wait_for_login(page, login_timeout)
-        save_state(page)
-
-    PROFILE_DIR.mkdir(parents=True, exist_ok=True)
-    with ScraplingBrowser(
-        headless=False,
-        real_chrome=True,
-        user_data_dir=str(PROFILE_DIR),
-        locale="zh-CN",
-        timezone_id="Asia/Shanghai",
-        timeout=90000,
-        network_idle=False,
-        disable_resources=False,
-        google_search=False,
-        page_setup=restore_cookies,
-        max_pages=1,
-    ) as browser:
-        browser.run_fetch_action("https://www.douyin.com/", action, wait=1000)
-    log("[login-only] 完成")
-    return 0
-
-
 def main() -> int:
    parser = argparse.ArgumentParser(description="Scrape Douyin video comments via Scrapling browser and XHR interception.")
-    parser.add_argument("url", help="Douyin video/user URL (传 'login' 配合 --login-only)")
-    parser.add_argument("--login-timeout", type=int, default=300, help="等待扫码/登录秒数")
+    parser.add_argument("url", help="Douyin video/user URL")
+    parser.add_argument("--login-timeout", type=int, default=300,
+                        help="保留参数（登录等待已移除；未登录将直接降级采集）")
    parser.add_argument("--max-scrolls", type=int, default=120, help="最多滚动加载次数")
    parser.add_argument("--idle-rounds", type=int, default=8, help="连续几轮无新评论后停止")
-    parser.add_argument("--headless", action="store_true", help="无头模式；首次登录不要使用")
-    parser.add_argument("--login-only", action="store_true",
-                        help="只触发登录(headed)然后退出,给 relogin_douyin 用")
+    parser.add_argument("--headless", action="store_true", help="无头模式")
    args = parser.parse_args()

-    if args.login_only:
-        return login_only(args.login_timeout)
-
    comments, result = scrape_comments(args.url, args.login_timeout, args.max_scrolls, args.idle_rounds, args.headless)
    json_path, csv_path = write_outputs(args.url, comments, result)
    log(f"Final URL: {result.get('final_url', '')}")
@@ -38,6 +38,7 @@ from gyxx_flow.adapters.scrapling import (
    BrowserTimeoutError,  # noqa: F401
    is_browser_timeout_error,
 )
+from gyxx_flow.adapters.browser import CookieExpiredError, cookie_expired_message
 from gyxx_flow.core.exit_codes import COOKIE_SKIP_EXIT_CODE
 from gyxx_flow.modules.content_marketing.collection_completeness import (
    BLOCKED_INPUT,
@@ -333,56 +334,6 @@ def is_logged_in(page: Page) -> bool:
        return False


-def open_scan_login(page: Page) -> None:
-    """Open the login dialog and prefer its QR-code login tab."""
-    for pattern in (r"^登录$|账号登录",):
-        try:
-            button = page.get_by_text(re.compile(pattern)).first
-            if button.count() and button.is_visible():
-                button.click(timeout=2000)
-                page.wait_for_timeout(500)
-        except Exception:
-            pass
-
-    # The login dialog uses an image-only folded-corner switch. Its bundled
-    # asset name is stable even though the generated CSS class is not.
-    try:
-        qr_switch = page.locator('img[src*="qr_code"]').first
-        if qr_switch.count() and qr_switch.is_visible():
-            # The browser engine's actionability check is blocked by the folded-corner
-            # overlay. A native DOM click triggers the image's real handler.
-            qr_switch.evaluate("e => e.click()")
-            page.wait_for_timeout(1000)
-            return
-    except Exception:
-        pass
-
-    for pattern in (r"扫码登录|二维码登录|扫码",):
-        try:
-            button = page.get_by_text(re.compile(pattern)).first
-            if button.count() and button.is_visible():
-                button.click(timeout=2000)
-                page.wait_for_timeout(500)
-        except Exception:
-            pass
-
-
-def wait_for_scan_login(page: Page, timeout_seconds: int) -> None:
-    log("Not logged in. 请在打开的 Chrome 中扫码登录。")
-    open_scan_login(page)
-    deadline = time.time() + timeout_seconds
-    next_login_click = time.time() + 5
-    while time.time() < deadline:
-        if is_logged_in(page):
-            save_state(page)
-            return
-        if time.time() >= next_login_click:
-            open_scan_login(page)
-            next_login_click = time.time() + 5
-        page.wait_for_timeout(2000)
-    raise TimeoutError(f"Login timed out after {timeout_seconds} seconds.")
-
-
 def ensure_login(page: Page, login_timeout: int) -> None:
    page.goto(HOME_URL, wait_until="domcontentloaded", timeout=60000)
    page.wait_for_timeout(2500)
@@ -402,20 +353,11 @@ def ensure_login(page: Page, login_timeout: int) -> None:
        )
        log(f"[SKIPPED_COOKIE] {reason}")
        raise AcceptanceCookieSkip(reason)
-    page.goto(LOGIN_URL, wait_until="domcontentloaded", timeout=60000)
-    page.wait_for_timeout(1500)
-    wait_for_scan_login(page, login_timeout)
-    page.goto(HOME_URL, wait_until="domcontentloaded", timeout=60000)
-    page.wait_for_load_state("domcontentloaded")
-    # 关闭弹窗
-    for text in ("跳过", "不再提示", "我知道了"):
-        try:
-            b = page.get_by_text(text, exact=True).first
-            if b.count():
-                b.click(timeout=1500)
-                page.wait_for_timeout(800)
-        except Exception:
-            pass
+    # 登录统一由账号 vault 管理（gyxx accounts login xiaohongshu-content），
+    # 采集脚本不再内嵌扫码等待；运行中失效直接快速失败。
+    reason = cookie_expired_message("xiaohongshu-content", "小红书蒲公英")
+    log(f"[COOKIE_EXPIRED] {reason}")
+    raise CookieExpiredError(reason)


 def _is_login_timeout(exc: BaseException) -> bool:
@@ -1933,7 +1875,7 @@ def scrape_styles(styles: list[dict], login_timeout: int, headless: bool,
                        action,
                        page_setup=restore_cookies,
                    )
-            except AcceptanceCookieSkip:
+            except (AcceptanceCookieSkip, CookieExpiredError):
                raise
            except Exception as exc:
                session_lost = is_browser_session_lost(exc)
@@ -2044,39 +1986,6 @@ def parse_int(s) -> int | None:
    return int(num)


-def login_only(login_timeout: int) -> int:
-    """只触发登录(headed),登录完成立刻退出。给 weekly_relogin.py 用。"""
-    def action(page):
-        ensure_login(page, login_timeout)
-        return page
-
-    session = None
-    try:
-        with DynamicSession(
-            headless=False,
-            real_chrome=True,
-            user_data_dir=str(PROFILE_DIR),
-            locale="zh-CN",
-            timezone_id="Asia/Shanghai",
-            timeout=60000,
-            network_idle=False,
-            disable_resources=False,
-            google_search=False,
-            page_setup=restore_cookies,
-            max_pages=2,
-        ) as session:
-            session.fetch(HOME_URL, page_action=action, wait=1000)
-    except Exception as exc:
-        if "TargetClosedError" in type(exc).__name__ or "has been closed" in str(exc):
-            log(f"  [WARN] 浏览器清理时报错: {type(exc).__name__}")
-        else:
-            raise
-    finally:
-        _force_cleanup_session(session)
-    log("[login-only] 登录完成,退出")
-    return 0
-
-
 def main() -> int:
    parser = argparse.ArgumentParser(description="小红书蒲公英 V2: 多维表格驱动精准采集+回填")
    add_repeatable_style_argument(parser, "只跑指定款式编号(可多次)")
@@ -2090,8 +1999,6 @@ def main() -> int:
    parser.add_argument("--skip-done", action="store_true", help="跳过已完成的款式(从进度文件读)")
    parser.add_argument("--inter-style-pause", type=int, default=5,
                        help="兼容旧参数；全局按达人采集后不再按款式暂停")
-    parser.add_argument("--login-only", action="store_true",
-                        help="只触发登录(headed),登录完成立刻退出,给 weekly_relogin 用")
    parser.add_argument("--self-operated", action="store_true",
                        help="抓取自营达人表格(而非合作达人)")
    parser.add_argument("--batch-size", type=int, default=DEFAULT_BATCH_SIZE,
@@ -2108,12 +2015,6 @@ def main() -> int:
                        help=argparse.SUPPRESS)
    args = parser.parse_args()

-    if args.login_only:
-        try:
-            return login_only(args.login_timeout)
-        except AcceptanceCookieSkip:
-            return COOKIE_SKIP_EXIT_CODE
-
    mapping = load_mapping(self_operated=args.self_operated)
    styles = mapping["tables"]
    if args.style:
@@ -2213,7 +2114,7 @@ def main() -> int:
                published_from=args.published_from,
                max_age_days=args.max_age_days, no_retry=args.no_retry,
            )
-        except AcceptanceCookieSkip:
+        except (AcceptanceCookieSkip, CookieExpiredError):
            return COOKIE_SKIP_EXIT_CODE
        except Exception as exc:
            if not _is_login_timeout(exc):
@@ -137,6 +137,7 @@ class ScraplingDriver:
        cdp_url: str | None = None,
        stealthy: bool = False,
        real_chrome: bool = True,
+        extra_flags: list[str] | None = None,
    ) -> ScraplingDriver:
        options: dict[str, Any] = {
            "headless": False,
@@ -152,6 +153,7 @@ class ScraplingDriver:
            "extra_flags": [
                "--disable-blink-features=AutomationControlled",
                "--disable-dev-shm-usage",
+                *(extra_flags or []),
            ],
        }
        reuse_existing_context = bool(cdp_url)
@@ -211,6 +213,9 @@ class ScraplingDriver:
            [script, raw_arguments],
        )

+    def save_screenshot(self, path: str | os.PathLike[str]) -> bool:
+        return bool(self._page.screenshot(path=str(Path(path))))
+
    def get_cookies(self) -> list[dict[str, Any]]:
        context = self._owner.context
        return list(context.cookies()) if context is not None else []
@@ -224,6 +229,24 @@ class ScraplingDriver:
    def quit(self) -> None:
        self._owner.close()

+    def reopen_page(self) -> Any:
+        """Rebind to a fresh page inside the still-live context.
+
+        A page/tab can die (site closed it, renderer gone) while the browser
+        context survives with its login state.  Returns the new page, or
+        raises the underlying error when the context/browser is gone too —
+        the caller should rebuild the whole driver then.
+        """
+        context = self._owner.context
+        if context is None:
+            raise RuntimeError("Scrapling browser context is closed")
+        # 永远开新页: context.pages 里可能还挂着半死的目标页。
+        page = context.new_page()
+        self._page = page
+        self._owner.page = page
+        page.on("download", self._save_download)
+        return page
+
    def _save_download(self, download: Any) -> None:
        target = self._download_dir / download.suggested_filename
        download.save_as(str(target))
@@ -363,8 +363,8 @@ def log(msg: str) -> None:


 # ============================================================
-# B站登录态: Cookie 持久化 + 扫码登录引导
-# 空间投稿列表需要登录态,首次用 --login-only 扫码,之后调度运行复用。
+# B站登录态: Cookie 持久化；登录统一由账号 vault 管理
+# (gyxx accounts login bilibili-content)，脚本内不再提供扫码登录。
 # ============================================================
 BILI_COOKIE_FILE = PATHS.browser_cookie_file

@@ -389,56 +389,6 @@ def save_state(page) -> None:
    log(f"已保存 Cookie ({len(cookies)} 条): {BILI_COOKIE_FILE}")


-def _has_login_cookies(page) -> bool:
-    names = {cookie.get("name") for cookie in page.context.cookies()}
-    return "SESSDATA" in names and "DedeUserID" in names
-
-
-def login_only(login_timeout: int = 300) -> int:
-    """打开 B站扫码登录页,等待人工扫码后保存 Cookie。"""
-    from scrapling.fetchers import StealthySession
-
-    with StealthySession(
-        headless=False,
-        real_chrome=True,
-        user_data_dir=str(PATHS.browser_profile_dir),
-        locale="zh-CN",
-        timezone_id="Asia/Shanghai",
-        timeout=60_000,
-        network_idle=False,
-        disable_resources=False,
-        google_search=False,
-        max_pages=2,
-        page_setup=restore_cookies,
-    ) as session:
-        page = (
-            session.context.pages[0]
-            if session.context.pages
-            else session.context.new_page()
-        )
-        page.goto("https://passport.bilibili.com/login",
-                  wait_until="domcontentloaded", timeout=60_000)
-        page.wait_for_timeout(2000)
-        if _has_login_cookies(page):
-            log("✅ 已有 B站登录态,无需扫码")
-            save_state(page)
-            return 0
-        print("\n" + "=" * 60)
-        print("🔐 请在打开的浏览器中用哔哩哔哩 App 扫码登录")
-        print("   登录成功后脚本会自动继续并保存 Cookie")
-        print("=" * 60, flush=True)
-        deadline = time.time() + login_timeout
-        while time.time() < deadline:
-            if _has_login_cookies(page):
-                log("✅ 检测到登录成功")
-                page.wait_for_timeout(2000)
-                save_state(page)
-                return 0
-            page.wait_for_timeout(2000)
-        log(f"❌ 等待扫码超时({login_timeout}s),未保存登录态")
-        return 1
-
-
 # ============================================================
 # 飞书多维表格
 # ============================================================
@@ -846,17 +796,10 @@ def main() -> int:
        description="自营 B 站采集: 公开 API 全量视频指标 + UP 主主页全量投稿 + 飞书曝光回填")
    parser.add_argument("--style", type=int, help="只跑指定款式编号")
    parser.add_argument("--dry-run", action="store_true", help="不写回飞书")
-    parser.add_argument("--login-only", action="store_true",
-                        help="只打开 B站扫码登录并保存 Cookie(首次部署引导)")
-    parser.add_argument("--login-timeout", type=int, default=300,
-                        help="等待人工扫码的秒数")
    parser.add_argument("--delay", type=float, default=0.3,
                        help="每条请求间隔秒数(防风控)")
    args = parser.parse_args()

-    if args.login_only:
-        return login_only(args.login_timeout)
-
    from gyxx_flow.modules.content_marketing import feishu_mapping
    try:
        mapping = feishu_mapping.load_mapping(self_operated=True)
@@ -27,6 +27,8 @@ from typing import Any
 from scrapling.fetchers import DynamicSession

 from gyxx_flow.adapters.scrapling import BrowserPage as Page
+from gyxx_flow.adapters.browser import CookieExpiredError, cookie_expired_message
+from gyxx_flow.core.exit_codes import COOKIE_SKIP_EXIT_CODE

 from gyxx_flow.modules.content_marketing.collection_completeness import (
    BLOCKED_INPUT,
@@ -261,17 +263,16 @@ def save_state(page: Page) -> None:


 def maybe_wait_for_login(page: Page, login_timeout: int) -> None:
-    deadline = time.time() + login_timeout
-    while time.time() < deadline:
-        try:
-            body = page.locator("body").inner_text(timeout=5000)
-        except Exception:
-            body = ""
-        if "扫码登录" not in body and "登录后" not in body:
-            return
-        log("抖音需要登录,请在打开的 Chrome 窗口中完成登录")
-        page.wait_for_timeout(3000)
-    log("[WARN] 登录等待超时")
+    # 登录统一由账号 vault 管理（gyxx accounts login douyin-content），
+    # 采集过程遇到登录态失效立即快速失败，不再等待人工扫码。
+    try:
+        body = page.locator("body").inner_text(timeout=5000)
+    except Exception:
+        body = ""
+    if "扫码登录" in body or "登录后" in body:
+        reason = cookie_expired_message("douyin-content", "自营抖音")
+        log(f"[COOKIE_EXPIRED] {reason}")
+        raise CookieExpiredError(reason)


 def dismiss_popups(page: Page) -> None:
@@ -605,33 +606,6 @@ def scrape_one_style(style: dict, login_timeout: int, headless: bool,
    return summary


-def login_only(login_timeout: int) -> int:
-    """只触发登录"""
-    login_done = False
-
-    def page_action(page: Page):
-        restore_cookies(page)
-        page.goto("https://www.douyin.com/", wait_until="domcontentloaded", timeout=60000)
-        page.wait_for_timeout(2000)
-        maybe_wait_for_login(page, login_timeout)
-        save_state(page)
-        log("登录完成")
-
-    PROFILE_DIR.mkdir(parents=True, exist_ok=True)
-    with DynamicSession(
-        headless=False,
-        real_chrome=True,
-        user_data_dir=str(PROFILE_DIR),
-        locale="zh-CN",
-        timezone_id="Asia/Shanghai",
-        timeout=60000,
-        page_setup=restore_cookies,
-    ) as session:
-        session.fetch("https://www.douyin.com/", page_action=page_action, wait=1000)
-
-    return 0 if login_done else 1
-
-
 # ============================================================
 # main
 # ============================================================
@@ -642,12 +616,8 @@ def main() -> int:
    parser.add_argument("--headless", action="store_true")
    parser.add_argument("--login-timeout", type=int, default=300)
    parser.add_argument("--dry-run", action="store_true", help="不写回多维表格")
-    parser.add_argument("--login-only", action="store_true", help="只触发登录")
    args = parser.parse_args()

-    if args.login_only:
-        return login_only(args.login_timeout)
-
    mapping = load_mapping()
    styles = mapping["tables"]
    if args.style:
@@ -703,6 +673,9 @@ def main() -> int:
 if __name__ == "__main__":
    try:
        raise SystemExit(main())
+    except CookieExpiredError as exc:
+        print(f"[COOKIE_EXPIRED] {exc}", file=sys.stderr)
+        raise SystemExit(COOKIE_SKIP_EXIT_CODE)
    except KeyboardInterrupt:
        print("\nInterrupted.", file=sys.stderr)
        raise SystemExit(130)
@@ -34,6 +34,7 @@ from gyxx_flow.adapters.scrapling import (
    BrowserTimeoutError,  # noqa: F401
    is_browser_timeout_error,
 )
+from gyxx_flow.adapters.browser import CookieExpiredError
 from gyxx_flow.core.exit_codes import COOKIE_SKIP_EXIT_CODE
 from gyxx_flow.modules.content_marketing import self_operated_accounts
 from gyxx_flow.modules.content_marketing.runtime_paths import PATHS
@@ -375,24 +376,14 @@ def ensure_login(page, login_timeout: int, phone: str) -> None:
        log(f"[SKIPPED_COOKIE] {reason}")
        raise AcceptanceCookieSkip(reason)

-    log("🔄 登录态失效,进入短信验证码登录流程")
-    page.goto(LOGIN_URL, wait_until="domcontentloaded", timeout=60_000)
-    page.wait_for_timeout(2000)
-    # 确保在「短信登录」页签
-    for text in ("短信登录",):
-        try:
-            tab = page.get_by_text(text, exact=True).first
-            if tab.count() and tab.is_visible():
-                tab.click(timeout=2000)
-                page.wait_for_timeout(800)
-                break
-        except Exception:
-            continue
-    _fill_sms_login(page, phone)
-    wait_for_login(page, login_timeout)
-    save_state(page)
-    page.goto(GOODS_URL, wait_until="domcontentloaded", timeout=60_000)
-    page.wait_for_timeout(2000)
+    # 采集流程不再内嵌短信验证码登录；统一由 --login-only 手动引导建立登录态。
+    reason = (
+        "小红书 ARK 后台登录态已失效：请手动执行 "
+        "`uv run python src/gyxx_flow/modules/content_marketing/self_xhs_scraper.py "
+        "--login-only` 完成短信验证码登录后重跑"
+    )
+    log(f"[COOKIE_EXPIRED] {reason}")
+    raise CookieExpiredError(reason)


 # ============================================================
@@ -855,6 +846,8 @@ def collect(headless: bool, login_timeout: int, max_pages: int,
        google_search=False,
        page_setup=restore_cookies,
        max_pages=3,
+        # 国内站直连：代理出口被站点 WAF 拉黑时页面/API 直接 ERR_CONNECTION_CLOSED
+        extra_flags=["--no-proxy-server"],
    ) as session:
        page = _open_page(session)
        ensure_login(page, login_timeout, phone)
@@ -927,7 +920,20 @@ def login_only(headless: bool, login_timeout: int) -> int:
        max_pages=2,
    ) as session:
        page = _open_page(session)
-        ensure_login(page, login_timeout, phone)
+        page.goto(LOGIN_URL, wait_until="domcontentloaded", timeout=60_000)
+        page.wait_for_timeout(2000)
+        for text in ("短信登录",):
+            try:
+                tab = page.get_by_text(text, exact=True).first
+                if tab.count() and tab.is_visible():
+                    tab.click(timeout=2000)
+                    page.wait_for_timeout(800)
+                    break
+            except Exception:
+                continue
+        _fill_sms_login(page, phone)
+        wait_for_login(page, login_timeout)
+        save_state(page)
        log("✅ 登录态已建立并保存 Cookie")
    return 0

@@ -953,7 +959,7 @@ def main() -> int:
            return login_only(args.headless, args.login_timeout)
        return collect(args.headless, args.login_timeout, args.max_pages,
                       args.dry_run)
-    except AcceptanceCookieSkip:
+    except (AcceptanceCookieSkip, CookieExpiredError):
        return COOKIE_SKIP_EXIT_CODE
    except KeyboardInterrupt:
        print("\n⛔ 用户中断", file=sys.stderr)
@@ -223,18 +223,15 @@ def maybe_wait_for_login(
    *,
    deadline: float | None = None,
 ) -> None:
-    login_deadline = time.monotonic() + max(0, login_timeout)
-    while time.monotonic() < login_deadline:
-        _check_note_deadline(deadline, "login wait")
-        try:
-            body = page.locator("body").inner_text(timeout=5000)
-        except Exception:
-            body = ""
-        if "扫码登录" not in body and "登录后" not in body and "验证码" not in body:
-            return
-        log("Xiaohongshu may require login. Please finish login in the opened Chrome window.")
-        page.wait_for_timeout(3000)
-    _check_note_deadline(deadline, "login wait")
+    # 登录统一由账号 vault 管理（gyxx accounts login xiaohongshu-content），
+    # 由工作流执行前的 cookie 网关保证；运行中失效立即降级，不再等待人工扫码。
+    try:
+        body = page.locator("body").inner_text(timeout=5000)
+    except Exception:
+        body = ""
+    if "扫码登录" in body or "登录后" in body or "验证码" in body:
+        log("Xiaohongshu login state missing; collecting degraded samples. "
+            "Run `gyxx accounts login xiaohongshu-content` to restore full comments.")


 def is_logged_in(page: Page) -> bool:
@@ -751,55 +748,16 @@ def write_outputs(url: str, comments: list[dict[str, Any]], result: dict[str, An
    return json_path, csv_path


-def login_only(login_timeout: int) -> int:
-    """只触发登录(headed),登录完成立刻退出。给 relogin_xiaohongshu.py 用。"""
-    def action(page: Page) -> None:
-        page.goto("https://www.xiaohongshu.com/", wait_until="domcontentloaded", timeout=90000)
-        page.wait_for_timeout(2500)
-        # 主动点登录按钮触发二维码
-        try:
-            login_btn = page.get_by_text("登录", exact=True).first
-            if login_btn.count():
-                login_btn.click(timeout=3000)
-                page.wait_for_timeout(1500)
-        except Exception:
-            pass
-        maybe_wait_for_login(page, login_timeout)
-        save_state(page)
-
-    PROFILE_DIR.mkdir(parents=True, exist_ok=True)
-    with ScraplingBrowser(
-        headless=False,
-        real_chrome=True,
-        user_data_dir=str(PROFILE_DIR),
-        locale="zh-CN",
-        timezone_id="Asia/Shanghai",
-        timeout=90000,
-        network_idle=False,
-        disable_resources=False,
-        google_search=False,
-        page_setup=restore_cookies,
-        max_pages=1,
-    ) as browser:
-        browser.run_fetch_action("https://www.xiaohongshu.com/", action, wait=1000)
-    log("[login-only] 完成")
-    return 0
-
-
 def main() -> int:
    parser = argparse.ArgumentParser(description="Scrape Xiaohongshu note comments with a real browser.")
-    parser.add_argument("url", help="Xiaohongshu explore/note URL (传 'login' 配合 --login-only)")
-    parser.add_argument("--login-timeout", type=int, default=300, help="等待扫码/登录秒数")
+    parser.add_argument("url", help="Xiaohongshu explore/note URL")
+    parser.add_argument("--login-timeout", type=int, default=300,
+                        help="保留参数（登录等待已移除；未登录将直接降级采集）")
    parser.add_argument("--max-scrolls", type=int, default=120, help="最多滚动加载次数")
    parser.add_argument("--idle-rounds", type=int, default=8, help="连续几轮无新评论后停止")
-    parser.add_argument("--headless", action="store_true", help="无头模式；首次登录不要使用")
-    parser.add_argument("--login-only", action="store_true",
-                        help="只触发登录(headed)然后退出,给 relogin_xiaohongshu 用")
+    parser.add_argument("--headless", action="store_true", help="无头模式")
    args = parser.parse_args()

-    if args.login_only:
-        return login_only(args.login_timeout)
-
    comments, result = scrape_comments(args.url, args.login_timeout, args.max_scrolls, args.idle_rounds, args.headless)
    json_path, csv_path = write_outputs(args.url, comments, result)
    log(f"Final URL: {result.get('final_url', '')}")
@@ -33,6 +33,7 @@ from gyxx_flow.adapters.scrapling import (
    BrowserTimeoutError,
    is_browser_timeout_error,
 )
+from gyxx_flow.adapters.browser import CookieExpiredError, cookie_expired_message
 from gyxx_flow.core.exit_codes import COOKIE_SKIP_EXIT_CODE
 from gyxx_flow.modules.content_marketing.collection_completeness import (
    BLOCKED_INPUT,
@@ -537,183 +538,6 @@ def is_logged_in(page: Page) -> bool:
        return False


-def open_scan_login(page: Page) -> None:
-    """Open the selected platform's login dialog without handling passwords."""
-    if CONTENT_SOURCE == "buyin":
-        if not is_business_page(page) or "/dashboard/servicehall/daren-square" not in page.url:
-            try:
-                page.goto(CUSTOMER_LOGIN_URL, wait_until="domcontentloaded", timeout=60000)
-                page.wait_for_timeout(1500)
-            except Exception:
-                pass
-        for pattern in (r"^登录$", r"扫码登录|二维码登录|抖音登录|使用抖音登录"):
-            try:
-                button = page.get_by_text(re.compile(pattern)).first
-                if button.count() and button.is_visible():
-                    button.click(timeout=3000)
-                    page.wait_for_timeout(800)
-                    return
-            except Exception as exc:
-                reraise_if_browser_session_lost(exc)
-        return
-
-    if "sso.oceanengine.com/xingtu/login" not in page.url:
-        try:
-            page.goto(CUSTOMER_LOGIN_URL, wait_until="domcontentloaded", timeout=60000)
-            page.wait_for_timeout(1500)
-        except Exception:
-            pass
-
-    # The Douyin login icon on Xingtu's customer SSO page has a stable
-    # aweme.png source. Clicking it opens the QR authorization page directly.
-    try:
-        douyin = page.locator('img[src*="aweme.png"]').first
-        if douyin.count() and douyin.is_visible():
-            douyin.click(timeout=3000)
-            page.wait_for_timeout(1500)
-            return
-    except Exception:
-        pass
-
-    for pattern in (r"^登录$|立即登录", r"抖音扫码登录|扫码登录|二维码登录|扫码"):
-        try:
-            button = page.get_by_text(re.compile(pattern)).first
-            if button.count() and button.is_visible():
-                button.click(timeout=2000)
-                page.wait_for_timeout(500)
-        except Exception:
-            pass
-
-
-def is_scan_qr_page(page: Page) -> bool:
-    """Return True once Douyin's QR authorization page is visible.
-
-    The QR page lives outside xingtu.cn. Treating that URL as a page that still
-    needs opening causes the login helper to navigate back and click again every
-    few seconds, continuously refreshing the QR code.
-    """
-    url = page.url.lower()
-    if any(host in url for host in ("open.douyin.com", "sso.douyin.com")):
-        return True
-    for selector in (
-        'img[src*="qrcode"]',
-        'img[src*="qr_code"]',
-        'canvas',
-    ):
-        try:
-            locator = page.locator(selector).first
-            if locator.count() and locator.is_visible():
-                return True
-        except Exception:
-            pass
-    return False
-
-
-def resolve_live_page(page: Page, *, create_if_missing: bool = False) -> Page | None:
-    """Return a usable page after OAuth closes or replaces its QR tab."""
-    try:
-        if not page.is_closed():
-            return page
-    except Exception:
-        pass
-
-    try:
-        context = page.context
-        for candidate in reversed(list(context.pages)):
-            try:
-                if not candidate.is_closed():
-                    return candidate
-            except Exception:
-                continue
-        if create_if_missing:
-            return context.new_page()
-    except Exception:
-        return None
-    return None
-
-
-def find_logged_in_page(page: Page) -> Page | None:
-    """Find an authenticated business tab even while the OAuth tab stays open."""
-    candidates: list[Page] = []
-    try:
-        candidates.extend(reversed(list(page.context.pages)))
-    except Exception:
-        pass
-    candidates.append(page)
-
-    seen: set[int] = set()
-    for candidate in candidates:
-        marker = id(candidate)
-        if marker in seen:
-            continue
-        seen.add(marker)
-        try:
-            if candidate.is_closed() or not is_business_page(candidate):
-                continue
-            if is_logged_in(candidate):
-                return candidate
-        except Exception:
-            continue
-    return None
-
-
-def wait_for_scan_login(page: Page, timeout_seconds: int) -> Page:
-    log("Not logged in. 请在打开的 Chrome 中扫码登录。")
-    current_page = page
-    open_scan_login(current_page)
-    deadline = time.time() + timeout_seconds
-    next_login_click = time.time() + 5
-    while time.time() < deadline:
-        authenticated = find_logged_in_page(current_page)
-        if authenticated is not None:
-            save_state(authenticated)
-            return authenticated
-
-        replacement = resolve_live_page(current_page, create_if_missing=True)
-        if replacement is None:
-            raise RuntimeError("Xingtu browser context closed during OAuth login")
-        if replacement is not current_page:
-            current_page = replacement
-            try:
-                current_page.goto(
-                    MARKET_URL, wait_until="domcontentloaded", timeout=60000
-                )
-                current_page.wait_for_timeout(2000)
-            except Exception:
-                pass
-
-        if (
-            CONTENT_SOURCE == "xingtu"
-            and "redirect_uri=/ad/creator/market" in current_page.url
-            and not is_scan_qr_page(current_page)
-        ):
-            try:
-                current_page.goto(
-                    MARKET_URL, wait_until="domcontentloaded", timeout=60000
-                )
-                current_page.wait_for_timeout(2000)
-            except Exception:
-                pass
-        authenticated = find_logged_in_page(current_page)
-        if authenticated is not None:
-            save_state(authenticated)
-            return authenticated
-        if (
-            time.time() >= next_login_click
-            and not is_scan_qr_page(current_page)
-            and not is_business_page(current_page)
-        ):
-            open_scan_login(current_page)
-            next_login_click = time.time() + 5
-        try:
-            current_page.wait_for_timeout(2000)
-        except Exception:
-            # OAuth may close the QR tab after authorization. The next loop
-            # follows the surviving tab in the same browser context.
-            continue
-    raise TimeoutError(f"Login timed out after {timeout_seconds} seconds.")
-
-
 def ensure_login(page: Page, login_timeout: int) -> Page:
    page.goto(MARKET_URL, wait_until="domcontentloaded", timeout=60000)
    page.wait_for_timeout(3000)
@@ -735,12 +559,11 @@ def ensure_login(page: Page, login_timeout: int) -> Page:
        )
        log(f"[SKIPPED_COOKIE] {reason}")
        raise AcceptanceCookieSkip(reason)
-    page = wait_for_scan_login(page, login_timeout)
-    page.goto(MARKET_URL, wait_until="domcontentloaded", timeout=60000)
-    page.wait_for_timeout(3000)
-    if not is_logged_in(page):
-        page = wait_for_scan_login(page, login_timeout)
-    return page
+    # 登录统一由账号 vault 管理（gyxx accounts login douyin-shop），
+    # 采集脚本不再内嵌扫码等待；运行中失效直接快速失败。
+    reason = cookie_expired_message("douyin-shop", f"{CONTENT_SOURCE}达人广场")
+    log(f"[COOKIE_EXPIRED] {reason}")
+    raise CookieExpiredError(reason)


 def dismiss_popups(page: Page) -> None:
@@ -3736,7 +3559,7 @@ def scrape_styles_with_fallback(
                summary_override=summaries,
                replace_existing=True,
            )
-        except AcceptanceCookieSkip as exc:
+        except (AcceptanceCookieSkip, CookieExpiredError) as exc:
            log(f"  [WARN] 抖店补采登录态不可用，保留星图结果: {exc}")
            _mark_fallback_not_processed(
                fallback_tasks, summaries, "buyin_cookie_skipped"
@@ -3805,96 +3628,6 @@ def _force_cleanup_session(session) -> None:
            pass


-def login_only(login_timeout: int) -> int:
-    """只触发登录(headed),登录完成立刻退出。给 weekly_relogin.py 用。"""
-    login_completed = False
-
-    def action(page):
-        nonlocal login_completed
-        page = ensure_login(page, login_timeout)
-        if not is_logged_in(page):
-            raise RuntimeError(f"{CONTENT_SOURCE} 登录状态验证失败")
-        save_state(page)
-        login_completed = True
-        return page
-
-    session = None
-    try:
-        with DynamicSession(
-            headless=False,
-            real_chrome=True,
-            user_data_dir=str(PROFILE_DIR),
-            locale="zh-CN",
-            timezone_id="Asia/Shanghai",
-            timeout=60000,
-            network_idle=False,
-            disable_resources=False,
-            google_search=False,
-            page_setup=restore_cookies,
-            max_pages=2,
-        ) as session:
-            session.fetch(MARKET_URL, page_action=action, wait=1000)
-    except Exception as exc:
-        err_msg = str(exc).lower()
-        if "closed" in err_msg or "context" in err_msg or "target" in err_msg:
-            log(f"[login-only] 浏览器会话关闭异常: {exc}")
-        else:
-            log(f"[login-only] 登录执行失败: {exc}")
-    finally:
-        _force_cleanup_session(session)
-
-    # Douyin OAuth may close the only QR tab (and therefore the temporary
-    # browser context) after a successful mobile confirmation. The durable
-    # Chromium profile can already contain the new session even though the
-    # page_action never got a chance to export cookies. Reopen the same profile
-    # once, verify the selected business page, then export its cookies.
-    if not login_completed or not has_valid_login_cookie_file():
-        log(f"[login-only] 扫码窗口已结束,重新打开{CONTENT_SOURCE}验证持久会话")
-        login_completed = verify_persisted_profile_login()
-    if not login_completed or not has_valid_login_cookie_file():
-        log("[login-only] 登录未完成或Cookie缺少有效会话")
-        return 1
-    log("[login-only] 登录完成,退出")
-    return 0
-
-
-def verify_persisted_profile_login() -> bool:
-    """Reopen the persistent profile after OAuth closes its only tab."""
-    verified = False
-
-    def action(page: Page):
-        nonlocal verified
-        page.goto(MARKET_URL, wait_until="domcontentloaded", timeout=60000)
-        page.wait_for_timeout(3000)
-        if not is_logged_in(page):
-            raise RuntimeError(f"persisted {CONTENT_SOURCE} profile is not authenticated")
-        save_state(page)
-        verified = True
-        return page
-
-    session = None
-    try:
-        with DynamicSession(
-            headless=False,
-            real_chrome=True,
-            user_data_dir=str(PROFILE_DIR),
-            locale="zh-CN",
-            timezone_id="Asia/Shanghai",
-            timeout=60000,
-            network_idle=False,
-            disable_resources=False,
-            google_search=False,
-            page_setup=restore_cookies,
-            max_pages=1,
-        ) as session:
-            session.fetch(MARKET_URL, page_action=action, wait=1000)
-    except Exception as exc:
-        log(f"[login-only] 持久会话验证失败: {exc}")
-    finally:
-        _force_cleanup_session(session)
-    return verified and has_valid_login_cookie_file()
-
-
 def load_retry_scope(summary_path: Path, reason: str) -> dict[int, set[str]]:
    """Load exact style/record pairs for a reason-limited retry."""
    try:
@@ -3926,8 +3659,6 @@ def main() -> int:
    parser.add_argument("--login-timeout", type=int, default=300)
    parser.add_argument("--dry-run", action="store_true")
    parser.add_argument("--data-dir", default=str(PATHS.normalized_root))
-    parser.add_argument("--login-only", action="store_true",
-                        help="只触发登录(headed),登录完成立刻退出,给 weekly_relogin 用")
    parser.add_argument(
        "--buyin",
        action="store_true",
@@ -3952,12 +3683,6 @@ def main() -> int:
        "buyin" if args.buyin else os.environ.get("GYXX_CONTENT_SOURCE", "xingtu")
    )

-    if args.login_only:
-        try:
-            return login_only(args.login_timeout)
-        except AcceptanceCookieSkip:
-            return COOKIE_SKIP_EXIT_CODE
-
    mapping = load_mapping(self_operated=args.self_operated)
    styles = mapping["tables"]
    if args.style:
@@ -4038,7 +3763,7 @@ def main() -> int:
            max_age_days=args.max_age_days,
            no_retry=args.no_retry,
        )
-    except AcceptanceCookieSkip:
+    except (AcceptanceCookieSkip, CookieExpiredError):
        return COOKIE_SKIP_EXIT_CODE
    persist_summaries(all_summaries)
    for s, summary in zip(styles, all_summaries):
@@ -15,11 +15,13 @@ PRODUCT_SCHEDULED_WORKFLOW_IDS = (
    "product.daily",
    "product.alert.daily",
    "product.online_stock_alert.daily",
+    "product.points_exchange.weekly",
    "product.ecommerce_costs.daily",
    "product.style_analysis.interval",
    "product.main_image.weekly",
    "product.sales_sheet.daily",
    "product.erp_all_shop_daily",
+    "product.hk_anpei_public_inventory.daily",
    "product.market_rank",
    "product.tmall_baibu_apply",
    "product.video_upload",
@@ -45,6 +47,7 @@ _WORKFLOW_SCOPED_RESOURCE_IDS = {
    "product.daily",
    "product.persona.daily",
    "product.style_analysis.interval",
+    "product.hk_anpei_public_inventory.daily",
    "product.ecommerce_costs.daily",
    # Tmall and JD video publication use separate browser profiles, accounts,
    # and channel-scoped PostgreSQL ledgers; do not let a long publication on
@@ -109,12 +109,26 @@ def missing_window_dates(rows: list[dict], start: date, end: date) -> list[date]


 def pull_reviews(style: str, start: date, end: date) -> dict:
-    """仅返回当前三日窗口内的评论分布与样本。"""
+    """仅返回当前三日窗口内的差评分布与样本。
+
+    2026-11-01 起 ``product_reviews`` 下线，改读三平台差评归一视图
+    ``v_negative_reviews``（tm + jd + dy 三张差评表 UNION）。
+
+    口径变化（必须知道）：旧表存好/中/差全量，视图只有差评一档，所以这里
+    返回的 key 会集中在 ``<平台>|负面评价``。换来的是覆盖率更高——天猫差评
+    旧表根本没有（0 行），京东旧表 8 月就断更。
+
+    款式归因走 ``style_id``（dim_style.id），款名只作显示冗余。
+    """
+
    with get_conn() as c, c.cursor() as cur:
        cur.execute(
-            """SELECT platform, rating, COUNT(*) n, array_agg(content ORDER BY comment_date DESC) samples
-               FROM product_reviews
-               WHERE style_name=%s AND comment_date BETWEEN %s AND %s
+            """SELECT platform, rating, COUNT(*) n,
+                      array_agg(content ORDER BY comment_date DESC) samples
+               FROM v_negative_reviews
+               WHERE style_id = (SELECT id FROM dim_style WHERE style_name = %s)
+                 AND comment_date BETWEEN %s AND %s
+                 AND content IS NOT NULL AND BTRIM(content) <> ''
               GROUP BY platform, rating""",
            (style, start, end),
        )
@@ -380,7 +394,7 @@ def aggregate_style_total(platforms: dict[str, dict]) -> dict:
 LEGACY_SYSTEM = """你是光影行星 (GYXX) 的资深电商数据分析师, 擅长把数据现象挖到根因并给出可执行方案.
 用户会给定一个款式近 3 天的真实业务数据:
 - 站内 (店铺自身): 天猫/京东/抖音三平台当前 3 天的 daily_style_metrics、
-  product_reviews、daily_persona_metrics、main_image_creatives
+  v_negative_reviews(三平台差评)、daily_persona_metrics、main_image_creatives
 - 站外 (种草营销): cmt_notes_master (笔记+合作投入), cmt_comments (笔记评论),
  cmt_creators (创作者静态属性)

@@ -787,7 +801,7 @@ def build_user_prompt(style: str, start: date, end: date, payload: dict) -> str:
 # 人物画像
 {json.dumps(payload['persona'], ensure_ascii=False, indent=2)}

-# 商品评价
+# 商品差评 (仅负向评价, 覆盖天猫/京东/抖音; 无好评与中评样本)
 {json.dumps(payload['product_reviews'], ensure_ascii=False, indent=2)}

 # 主图表现
@@ -18,7 +18,7 @@ from scrapling.fetchers import DynamicFetcher
 try:
    from . import collect_erp_yesterday_metrics as erp
    from .collect_retry_utils import retry_step
-    from .config.style_config_loader import StyleConfigLoader, has_complete_style_identity
+    from .config.style_config_loader import StyleConfigLoader
    from .db import (
        get_conn,
        init_schema,
@@ -35,7 +35,7 @@ try:
 except ImportError:  # Direct script execution from the module root.
    import collect_erp_yesterday_metrics as erp
    from collect_retry_utils import retry_step
-    from config.style_config_loader import StyleConfigLoader, has_complete_style_identity
+    from config.style_config_loader import StyleConfigLoader
    from db import get_conn, init_schema, upsert_erp_all_shop_style_daily_metrics
    from erp_login_product_analysis import (
        DEFAULT_CONFIG,
@@ -51,6 +51,11 @@ DEFAULT_FROM_DATE = date.today() - timedelta(days=1)
 DEFAULT_TO_DATE = DEFAULT_FROM_DATE
 MAX_RANGE_DAYS = 366
 EVIDENCE_DIR_NAME = "erp全店铺款式日报"
+# 聚水潭商品主题分析报表固定每页 200 行明细，且没有可用的页大小设置；
+# 款式扩到 198 个 ERP 编码后必须翻页，否则只拿到第一页的 62 个编码。
+PAGE_ROW_LIMIT = 200
+MAX_REPORT_PAGES = 12
+PAGE_TURN_TIMEOUT = 45


@dataclass(frozen=True, slots=True)
@@ -76,10 +81,19 @@ def inclusive_dates(start: date, end: date) -> tuple[date, ...]:


 def build_style_plans(payload: Mapping[str, Any]) -> tuple[StylePlan, ...]:
-    plans: list[StylePlan] = []
-    code_owners: dict[str, str] = {}
+    """Build per-style ERP plans, folding series codes into their anchor.
+
+    同系列多款时 loader 会把子款编码聚合并集到锚点（series.erp_owner），
+    子款仍保留自己的编码供按款查询。采集计划按归属只保留一次：编码归锚点，
+    子款剔除已被锚点承接的重复项；真正跨系列的重复仍然报错。
+    """
+
+    codes_by_style: dict[str, tuple[str, ...]] = {}
+    owner_of: dict[str, str] = {}
    for raw_name, raw_config in (payload.get("styles") or {}).items():
        style_name = str(raw_name).strip()
+        if not style_name:
+            continue
        config = raw_config or {}
        codes = tuple(
            dict.fromkeys(
@@ -88,15 +102,42 @@ def build_style_plans(payload: Mapping[str, Any]) -> tuple[StylePlan, ...]:
                if str(code).strip()
            )
        )
-        if not style_name or not codes or not has_complete_style_identity(config):
+        if not codes:
            continue
+        codes_by_style[style_name] = codes
+        owner = str((config.get("series") or {}).get("erp_owner") or "").strip()
+        if owner:
+            owner_of[style_name] = owner
+
+    holders: dict[str, list[str]] = {}
+    for style_name, codes in codes_by_style.items():
        for code in codes:
-            previous = code_owners.setdefault(code, style_name)
-            if previous != style_name:
-                raise ValueError(
-                    f"ERP style code {code} belongs to both {previous!r} and {style_name!r}"
-                )
-        plans.append(StylePlan(style_name=style_name, erp_style_codes=codes))
+            holders.setdefault(code, []).append(style_name)
+
+    resolved: dict[str, str] = {}
+    for code, names in holders.items():
+        if len(names) == 1:
+            resolved[code] = names[0]
+            continue
+        # 同系列归集：持有者里被其余持有者指认为 ERP 锚点的那一款承接该编码
+        candidates = [
+            name
+            for name in names
+            if any(owner_of.get(other) == name for other in names if other != name)
+        ]
+        if len(candidates) == 1:
+            resolved[code] = candidates[0]
+            continue
+        raise ValueError(
+            f"ERP style code {code} belongs to both {names[0]!r} and {names[1]!r}"
+        )
+
+    plans: list[StylePlan] = []
+    for style_name, codes in codes_by_style.items():
+        own_codes = tuple(code for code in codes if resolved.get(code) == style_name)
+        if not own_codes:
+            continue
+        plans.append(StylePlan(style_name=style_name, erp_style_codes=own_codes))
    if not plans:
        raise RuntimeError("style configuration contains no ERP style codes")
    return tuple(plans)
@@ -205,6 +246,122 @@ def set_all_shop_daily_filters(
    )


+ENSURE_CUSTOM_COLUMNS_JS = r"""
+        (options) => {
+          const redrawPage = (options && options.redrawPage) || 1;
+          const forceRedraw = Boolean(options && options.forceRedraw);
+          const cols = window._Cols || [];
+          let changed = 0;
+          for (const col of cols) {
+            const label = String(col.text || col.defaultText || '');
+            if ((label === '自定义销量' || label === '自定义退货量') && col.visible === false) {
+              col.visible = true;
+              changed += 1;
+            }
+          }
+          // 只改 visible 不会补出数据行缺失的 td：聚水潭按旧列定义渲染数据行，
+          // 行内仍是旧的单元格数，按新列位置取值就会读到 undefined（历史全 0 成因）。
+          // 换页会强制按当前列定义重绘，所以顺带跳到目标页。
+          let redrawn = false;
+          if (changed > 0 || forceRedraw) {
+            // _RptRefresh 会走一次完整重算（公式列随之补齐），GotoPage 负责
+            // 让数据行按新列定义重绘；两者都试，缺一都可能停在旧行结构上。
+            if (forceRedraw && typeof window._RptRefresh === 'function') {
+              try { window._RptRefresh(); } catch (e) {}
+            }
+            if (typeof GotoPage === 'function') {
+              try { GotoPage(redrawPage || 1); redrawn = true; } catch (e) {}
+            }
+          }
+          return { changed: changed, redrawn: redrawn };
+        }
+"""
+
+
+def ensure_custom_columns_rendered(
+    page: Any, *, redraw_page: int = 1, force_redraw: bool = False
+) -> dict[str, Any]:
+    """确保自定义销量/退货量列可见，并触发一次重绘补齐数据行单元格。
+
+    2026-10-11 起聚水潭报表模板把自定义公式列重置为隐藏，隐藏列不渲染
+    数据单元格；而 ``generate_report`` 每次查询都会把列可见性重置回隐藏。
+    因此本函数必须在查询之后调用，并在 visible 变化时换页重绘，否则
+    ``自定义销量`` 整列读不到值。
+
+    ``force_redraw`` 用于「列已 visible、但数据行仍按旧列数渲染」的中间态：
+    此时列状态无需修改，仍要换页才能让每行补齐到新的单元格数。
+    """
+
+    summary = {"changed": 0, "redrawn": 0}
+    for frame in page.frames:
+        if "allsku.aspx" not in (frame.url or "").lower():
+            continue
+        try:
+            result = frame.evaluate(
+                ENSURE_CUSTOM_COLUMNS_JS,
+                {"redrawPage": int(redraw_page), "forceRedraw": bool(force_redraw)},
+            )
+        except Exception as exc:
+            print(f"[WARN] 自定义列补设失败: {type(exc).__name__}: {exc}", flush=True)
+            continue
+        if isinstance(result, dict):
+            summary["changed"] += int(result.get("changed") or 0)
+            if result.get("redrawn"):
+                summary["redrawn"] += 1
+    return summary
+
+
+def ensure_custom_columns_visible(page: Any) -> int:
+    """Keep 自定义销量/自定义退货量 visible before the report renders.
+
+    2026-10-11 起聚水潭报表模板把自定义公式列重置为隐藏，隐藏列不渲染
+    数据单元格；只改 visible 不会自动重绘，需顺带触发一次 _RptRefresh
+    （其尾部浮头重建可能抛错，但不影响数据行渲染）。
+    """
+
+    return int(ensure_custom_columns_rendered(page, redraw_page=1)["changed"])
+
+
+def _allsku_frame(page: Any) -> Any:
+    for frame in page.frames:
+        if "allsku.aspx" in (frame.url or "").lower():
+            return frame
+    return None
+
+
+def report_page_count(page: Any) -> int:
+    frame = _allsku_frame(page)
+    if frame is None:
+        return 1
+    try:
+        return max(1, int(frame.evaluate("() => Number(window._RptMaxPage) || 1")))
+    except Exception:
+        return 1
+
+
+GOTO_REPORT_PAGE_JS = r"""
+        (options) => {
+          if (typeof GotoPage !== 'function') return false;
+          try {
+            GotoPage((options && options.page) || 1);
+            return true;
+          } catch (e) {
+            return false;
+          }
+        }
+"""
+
+
+def goto_report_page(page: Any, page_no: int) -> bool:
+    frame = _allsku_frame(page)
+    if frame is None:
+        return False
+    try:
+        return bool(frame.evaluate(GOTO_REPORT_PAGE_JS, {"page": int(page_no)}))
+    except Exception:
+        return False
+
+
 def extract_style_rows(
    table_frame: Any,
    requested_codes: Sequence[str],
@@ -257,18 +414,118 @@ def extract_style_rows(
            };
          }

+          // 隐藏列的 th 仍在表头，但数据行不渲染对应 td：数据行的第 j 格
+          // 对应第 j 个可见列。必须用 window._Cols 把表头列索引换算成行内
+          // 位置——不做换算而按表头索引直取，会整体偏移到相邻列（历史表现
+          // 为销量读到月累计/金额列，被放大上百倍）。换算所需的 _Cols 一旦
+          // 缺失或与表头列数不一致，就无法确定位置，此时一律 fail closed，
+          // 绝不退回"按原索引直取"。
+          if (!Array.isArray(window._Cols) || window._Cols.length !== headerTexts.length) {
+            return {
+              ok: false,
+              reason: 'column_layout_unknown',
+              error: `column layout unresolved: _Cols=${Array.isArray(window._Cols) ? window._Cols.length : 'none'} header=${headerTexts.length}`,
+              header_texts: headerTexts,
+            };
+          }
+          const colLabel = index => {
+            const col = window._Cols[index];
+            return norm(col && (col.text || col.defaultText));
+          };
+          // 表头 th 文本与 _Cols 必须是同一列，否则说明表头已重建而列定义
+          // 尚未同步（中间态），按任何一个索引取都不可信。
+          const labelMismatch = [
+            [styleCol, '款式编码'],
+            [salesCol, '自定义销量'],
+            [returnsCol, '自定义退货量'],
+          ].filter(([index, expected]) => colLabel(index) !== norm(expected));
+          if (labelMismatch.length) {
+            return {
+              ok: false,
+              reason: 'column_layout_unknown',
+              error: `header/_Cols label mismatch: ${labelMismatch.map(([i]) => i).join(',')}`,
+              header_texts: headerTexts,
+            };
+          }
+          const visibleIdx = [];
+          for (let i = 0; i < window._Cols.length; i++) {
+            if (window._Cols[i].visible !== false) visibleIdx.push(i);
+          }
+          const toPos = index => (index >= 0 ? visibleIdx.indexOf(index) : -1);
+          const stylePos = toPos(styleCol);
+          const productPos = toPos(productCol);
+          const namePos = toPos(productNameCol);
+          const salesPos = toPos(salesCol);
+          const returnsPos = toPos(returnsCol);
+          // 值域交叉校验用的参考列：自定义销量是"当期销量"口径，正常不可能
+          // 超过同行的月销量一个量级。读到月累计/金额列时这里会炸。
+          const monthSalesPos = toPos(findCol('月销量'));
+          const monthReturnsPos = toPos(findCol('月退货量'));
+          // 每行应有几个 td：等于可见列数。数量不符＝该行仍按旧列定义渲染
+          // （重绘未完成），位置不可信。
+          const expectedCells = visibleIdx.length;
+          if (stylePos < 0 || salesPos < 0 || returnsPos < 0) {
+            return {
+              ok: false,
+              reason: 'custom_columns_hidden',
+              error: `custom sales/return columns hidden: style=${stylePos}, sales=${salesPos}, returns=${returnsPos}`,
+              header_texts: headerTexts,
+            };
+          }
+
          const rows = [];
+          let renderedSalesCells = 0;
+          let blankSalesCells = 0;
+          let zeroSalesRows = 0;
+          let layoutMismatchRows = 0;
+          let outOfRangeRows = 0;
+          let monthRefRows = 0;
          for (const tr of Array.from(document.querySelectorAll('tr'))) {
-            const cells = Array.from(tr.querySelectorAll('td, th'));
-            if (cells.length <= Math.max(styleCol, salesCol, returnsCol)) continue;
-            const styleCode = norm(cells[styleCol]?.textContent);
+            // 只用 td：表头行的 th 会让列位置整体右移，
+            // 而分组行/浮头行里混入的 th 同样会污染样式编码列的取值。
+            const cells = Array.from(tr.querySelectorAll('td'));
+            if (!cells.length) continue;
+            const styleCode = norm(cells[stylePos]?.textContent);
+            // 只有命中本次请求编码的行才是数据行：报表里的小计/分组行 td
+            // 数与列定义天然不同，不能拿它们判定列错位。
            if (!requested.has(styleCode)) continue;
+            // 数据行 td 数必须与可见列数完全相等。少一格就说明该行仍按旧
+            // 列定义渲染，此时任何位置取值都可能落到相邻列（读到月累计或
+            // 金额），必须整页拒绝重绘，不能按"位置够长就往里取"处理。
+            if (cells.length !== expectedCells) {
+              layoutMismatchRows += 1;
+              continue;
+            }
+            const salesText = String(cells[salesPos]?.textContent ?? '').trim();
+            const returnText = String(cells[returnsPos]?.textContent ?? '').trim();
+            const salesValue = parseInteger(salesText);
+            const returnValue = parseInteger(returnText);
+            // 值域交叉校验：自定义销量是"当期"口径，正常不会超过同行月销量
+            // 一个量级。列错位读到月累计/金额时这里的比例会失真，用于兜住
+            // 「结构看起来正常但语义错列」的情况（2026-10-30 事故）。
+            if (monthSalesPos >= 0) {
+              const monthSalesText = String(cells[monthSalesPos]?.textContent ?? '').trim();
+              if (monthSalesText) {
+                monthRefRows += 1;
+                if (salesValue > parseInteger(monthSalesText) * 2 + 50) outOfRangeRows += 1;
+              }
+            }
+            if (monthReturnsPos >= 0) {
+              const monthReturnsText = String(cells[monthReturnsPos]?.textContent ?? '').trim();
+              if (monthReturnsText) {
+                monthRefRows += 1;
+                if (returnValue > parseInteger(monthReturnsText) * 2 + 50) outOfRangeRows += 1;
+              }
+            }
+            if (salesText) renderedSalesCells += 1;
+            else blankSalesCells += 1;
+            if (salesValue === 0) zeroSalesRows += 1;
            rows.push({
              erp_style_code: styleCode,
-              product_code: productCol >= 0 ? norm(cells[productCol]?.textContent) : '',
-              product_name: productNameCol >= 0 ? norm(cells[productNameCol]?.textContent) : '',
-              sales: parseInteger(cells[salesCol]?.textContent),
-              return_qty: parseInteger(cells[returnsCol]?.textContent),
+              product_code: productPos >= 0 ? norm(cells[productPos]?.textContent) : '',
+              product_name: namePos >= 0 ? norm(cells[namePos]?.textContent) : '',
+              sales: salesValue,
+              return_qty: returnValue,
            });
          }
          const foundCodes = Array.from(new Set(rows.map(row => row.erp_style_code))).sort();
@@ -279,7 +536,58 @@ def extract_style_rows(
          if (!rows.length) {
            return {
              ok: false,
-              error: 'report contains none of the requested ERP style codes',
+              reason: layoutMismatchRows ? 'row_layout_mismatch' : 'no_requested_rows',
+              error: `report contains none of the requested ERP style codes (mismatched rows: ${layoutMismatchRows})`,
+              layout_mismatch_rows: layoutMismatchRows,
+              header_texts: headerTexts,
+            };
+          }
+          if (renderedSalesCells === 0) {
+            // 自定义销量是聚水潭的公式列：列被隐藏或公式尚未算完时
+            // 单元格是空的。空 ≠ 0，此时返回未就绪让调用方继续等待，
+            // 绝不能静默写 0（历史上 10-13~10-19、10-22 全表为 0 就是这样来的）。
+            return {
+              ok: false,
+              reason: 'custom_sales_not_rendered',
+              error: `custom sales cells blank: ${blankSalesCells}/${rows.length}`,
+              row_count: rows.length,
+              sales_col: salesCol,
+              sales_pos: salesPos,
+              header_texts: headerTexts,
+            };
+          }
+          if (layoutMismatchRows > 0) {
+            // 有行的 td 数与可见列数不符：整页列位置不可信，重绘后重读。
+            return {
+              ok: false,
+              reason: 'row_layout_mismatch',
+              error: `row cells mismatch: ${layoutMismatchRows} rows`,
+              row_count: rows.length,
+              expected_cells: expectedCells,
+              header_texts: headerTexts,
+            };
+          }
+          if (outOfRangeRows > 0) {
+            // 结构与列名都对上了，但值本身不可信（销量超过同行月销量一个
+            // 量级）。典型成因是日期筛选未真正生效、报表返回了月/年累计值。
+            return {
+              ok: false,
+              reason: 'sales_out_of_range',
+              error: `sales out of monthly range: ${outOfRangeRows} rows (reference rows ${monthRefRows})`,
+              row_count: rows.length,
+              out_of_range_rows: outOfRangeRows,
+              header_texts: headerTexts,
+            };
+          }
+          if (rows.length >= 100 && zeroSalesRows / rows.length >= 0.8) {
+            // 公式列在补值过程中会先把整列渲染成 0 再逐个算出来；正常日
+            // 零销量行占比约 50%~60%（2026-10 实测），超过 80% 说明整页
+            // 还没算完。宁可继续等也不要写一批 0 覆盖真实销量。
+            return {
+              ok: false,
+              reason: 'custom_sales_zero_placeholder',
+              error: `custom sales zero rows: ${zeroSalesRows}/${rows.length}`,
+              row_count: rows.length,
              header_texts: headerTexts,
            };
          }
@@ -290,6 +598,25 @@ def extract_style_rows(
            missing_codes: missingCodes,
            signature,
            header_texts: headerTexts,
+            rendered_sales_cells: renderedSalesCells,
+            blank_sales_cells: blankSalesCells,
+            zero_sales_rows: zeroSalesRows,
+            layout_mismatch_rows: layoutMismatchRows,
+            // 列布局快照：写进 evidence，下次出问题可直接比对，不用复现。
+            layout: {
+              header_cells: headerTexts.length,
+              visible_cells: expectedCells,
+              style_col: styleCol,
+              style_pos: stylePos,
+              sales_col: salesCol,
+              sales_pos: salesPos,
+              returns_col: returnsCol,
+              returns_pos: returnsPos,
+              month_sales_pos: monthSalesPos,
+              month_returns_pos: monthReturnsPos,
+              row_cells: expectedCells,
+              month_ref_rows: monthRefRows,
+            },
          };
        }
        """,
@@ -326,6 +653,7 @@ def wait_for_style_rows(
            reload_deadline = time.monotonic() + reload_after_seconds
            continue

+        ensure_custom_columns_visible(page)
        for frame in page.frames:
            if "allsku.aspx" not in (frame.url or "").lower():
                continue
@@ -410,6 +738,56 @@ def aggregate_style_records(
    ]


+def _recent_sales_baseline(report_date: date, *, days: int = 14) -> float | None:
+    """最近若干天单日销量合计的中位数，用于识别量级异常。"""
+
+    with get_conn() as conn, conn.cursor() as cur:
+        cur.execute(
+            """
+            SELECT percentile_cont(0.5) WITHIN GROUP (ORDER BY day_total)
+            FROM (
+                SELECT metric_date, sum(sales) AS day_total
+                FROM erp_all_shop_style_daily_metrics
+                WHERE metric_date < %s
+                  AND metric_date >= %s::date - (%s || ' days')::interval
+                GROUP BY metric_date
+            ) recent
+            """,
+            (report_date, report_date, days),
+        )
+        row = cur.fetchone()
+    value = row[0] if row else None
+    return float(value) if value else None
+
+
+def _guard_daily_total(
+    report_date: date,
+    records: Sequence[Mapping[str, Any]],
+    *,
+    force: bool = False,
+) -> None:
+    """单日合计与近 14 天中位数差 8 倍以上就拒绝落库。
+
+    2026-10-30 事故里，日期筛选未真正生效使报表返回月累计值，单日合计
+    704,140 而正常约 2,000（350 倍）。列级校验挡住列错位，这道闸挡住
+    「列对但统计口径错」这一层，宁可失败也不写脏数据。
+    """
+
+    if force:
+        return
+    total = sum(int(record.get("sales") or 0) for record in records)
+    baseline = _recent_sales_baseline(report_date)
+    if not baseline or baseline <= 0:
+        return
+    if total <= baseline * 8:
+        return
+    raise RuntimeError(
+        f"all-shop daily total looks wrong for {report_date.isoformat()}: "
+        f"sales={total} vs recent median={baseline:.0f} "
+        f"({total / baseline:.1f}x). Refusing to persist; rerun with --force if business volume really changed."
+    )
+
+
 def write_evidence(
    report_date: date,
    *,
@@ -427,8 +805,12 @@ def write_evidence(
        "sales_header": "自定义销量",
        "return_header": "自定义退货量",
        "selection": dict(selection),
+        "pages": int(report.get("pages") or 1),
+        "max_page": int(report.get("max_page") or 1),
        "found_codes": list(report.get("found_codes") or ()),
        "missing_codes": list(report.get("missing_codes") or ()),
+        # 列布局快照：出问题时可直接判断「列错位」还是「数据本身」，无需复现。
+        "layout": dict(report.get("layout") or {}),
        "rows": list(report.get("rows") or ()),
        "styles": [
            {
@@ -444,6 +826,206 @@ def write_evidence(
    return output_path


+def _wait_rendered_page(
+    page: Any,
+    requested_codes: Sequence[str],
+    *,
+    page_no: int,
+    timeout_seconds: int,
+    previous_signature: str | None = None,
+    expect_full_page: bool = False,
+) -> dict[str, Any]:
+    """Wait for one report page whose custom sales cells are actually rendered.
+
+    ``previous_signature`` 用于翻页：``GotoPage`` 是异步的，调用后 DOM 仍是
+    上一页，若只看「签名稳定」会立刻把上一页数据当成新页收下，
+    结果多页合并后去重回同一批行（10-28 实测 3 页只有 200 行）。
+
+    ``expect_full_page``：报表每页固定 200 行，除最后一页外不该出现不满页。
+    不满页说明数据行还在陆续插入，此时签名字段虽已稳定但行集不全，
+    直接收下会漏采编码（2026-10-30 实测第 1 页只有 174 行、只命中 57/199
+    个编码，后续页面也补不回第一页缺的行）。
+    """
+
+    deadline = time.monotonic() + timeout_seconds
+    last_signature: str | None = None
+    stable = 0
+    redraws = 0
+    reached = previous_signature is None
+    last_error = "report page did not render custom sales column"
+
+    def _declared_pages() -> int:
+        """报表自己声明的总页数；不可信时返回 1（只做加法兜底）。"""
+
+        frame = _allsku_frame(page)
+        if frame is None:
+            return 1
+        try:
+            return max(1, int(frame.evaluate("() => Number(window._RptMaxPage) || 1")))
+        except Exception:
+            return 1
+
+    while time.monotonic() < deadline:
+        frame = _allsku_frame(page)
+        if frame is None:
+            last_error = "allsku iframe not found"
+            time.sleep(0.5)
+            continue
+        try:
+            result = extract_style_rows(frame, requested_codes)
+        except Exception as exc:
+            result = {"ok": False, "error": f"{type(exc).__name__}: {exc}"}
+        if result.get("ok") and len(result.get("rows") or ()) < PAGE_ROW_LIMIT:
+            # 不满 200 行只有两种可能：这是最后一页，或数据行还在陆续插入。
+            # 报表声明的总页数大于当前页时必然是后者，继续等。
+            short = len(result.get("rows") or ())
+            if expect_full_page or _declared_pages() > page_no:
+                result = {
+                    "ok": False,
+                    "reason": "partial_page",
+                    "error": f"page {page_no} rendered {short}/{PAGE_ROW_LIMIT} rows",
+                }
+        if result.get("ok"):
+            signature = str(result.get("signature") or "")
+            if not reached:
+                # 翻页后内容尚未变化：继续等新页渲染出来。
+                if signature == previous_signature:
+                    last_signature = None
+                    stable = 0
+                    time.sleep(0.5)
+                    continue
+                reached = True
+            if signature == last_signature:
+                stable += 1
+            else:
+                last_signature = signature
+                stable = 1
+            # 自定义销量是公式列，翻页后先出 0、再逐个单元格补齐；
+            # 稳定阈值取 6 次（≈2.4 秒不变）才能跨过「补值中」的中间态。
+            required = 5 if result.get("reason") == "no_data" else 6
+            if stable >= required:
+                result["page_no"] = page_no
+                result["frame_url"] = frame.url
+                return result
+        else:
+            last_error = str(result.get("error") or result.get("reason") or last_error)
+            last_signature = None
+            stable = 0
+            # 列为空/被隐藏/行内单元格错位/整列占位 0 都是「未渲染完」，
+            # 不是「销量为 0」：补设列并强制重绘后重读。
+            reason = result.get("reason")
+            if reason in (
+                "custom_sales_not_rendered",
+                "custom_columns_hidden",
+                "column_layout_unknown",
+                "row_layout_mismatch",
+                "custom_sales_zero_placeholder",
+                "sales_out_of_range",
+                "partial_page",
+            ) and redraws < 6:
+                summary = ensure_custom_columns_rendered(
+                    page,
+                    redraw_page=page_no,
+                    # 这几种原因都说明「列已可见但数据行/公式没算完」，
+                    # 必须强制重算，否则只会原地等同一份旧 DOM。
+                    force_redraw=True,
+                )
+                if summary.get("changed") or summary.get("redrawn"):
+                    redraws += 1
+                    last_error = f"{last_error}; redraw #{redraws} on page {page_no}"
+                    time.sleep(3.0)
+                    continue
+        time.sleep(0.4)
+    raise TimeoutError(
+        f"timeout waiting for rendered page {page_no}: {last_error}"
+    )
+
+
+def collect_report_pages(
+    page: Any,
+    requested_codes: Sequence[str],
+    *,
+    report_timeout: int,
+) -> dict[str, Any]:
+    """Collect every page of the allsku report and merge the product rows.
+
+    报表固定每页 200 行；款式扩到 198 个 ERP 编码后单页放不下，
+    只读第一页会漏掉大半编码（10-28 就是这样：198 编码只覆盖 62 个）。
+
+    翻页条件按「上一页是否装满一页」判断，不依赖 ``window._RptMaxPage``：
+    该变量在增量查询后会被重置为 1，实测连续补采 8 天只有第一天翻页成功。
+    """
+
+    first = _wait_rendered_page(
+        page,
+        requested_codes,
+        page_no=1,
+        timeout_seconds=report_timeout,
+    )
+    # 等待结束后再读总页数：生成报表的瞬间 _RptMaxPage 还是旧值 1。
+    max_page = report_page_count(page)
+    rows: list[dict[str, Any]] = list(first.get("rows") or ())
+    previous_signature = str(first.get("signature") or "")
+    current_count = len(rows)
+    visited = 1
+    page_no = 1
+    # 翻页条件：上一页装满 200 行，或 window._RptMaxPage 指出还有后续页。
+    # 后者用于兜住「第 1 页渲染不全、行数不足 200」的情况——只看行数会直接
+    # 判成最后一页而漏采。该变量在增量查询后可能被重置为 1，因此只做加法
+    # 兜底，翻到空页时由 _wait_rendered_page 超时退出。
+    while current_count >= PAGE_ROW_LIMIT or page_no < max_page:
+        page_no += 1
+        if page_no > MAX_REPORT_PAGES:
+            print(f"[WARN] 全店铺报表页数超过 {MAX_REPORT_PAGES}，按上限停止", flush=True)
+            break
+        if not goto_report_page(page, page_no):
+            print(f"[WARN] 全店铺报表翻页失败，停在页 {visited}", flush=True)
+            break
+        try:
+            last = _wait_rendered_page(
+                page,
+                requested_codes,
+                page_no=page_no,
+                timeout_seconds=min(report_timeout, PAGE_TURN_TIMEOUT),
+                previous_signature=previous_signature,
+                expect_full_page=page_no < max_page,
+            )
+        except TimeoutError:
+            # 翻页无新内容＝已经到最后一页（GotoPage 越界不报错，只是不刷新）。
+            break
+        previous_signature = str(last.get("signature") or "")
+        page_rows = list(last.get("rows") or ())
+        print(f"[ERP ALL] 第 {page_no} 页 {len(page_rows)} 行", flush=True)
+        rows.extend(page_rows)
+        current_count = len(page_rows)
+        visited = page_no
+
+    deduped: list[dict[str, Any]] = []
+    seen: set[tuple[str, str]] = set()
+    for row in rows:
+        key = (str(row.get("erp_style_code") or ""), str(row.get("product_code") or ""))
+        if key in seen:
+            continue
+        seen.add(key)
+        deduped.append(row)
+    found_codes = sorted({str(row.get("erp_style_code") or "") for row in deduped})
+    missing_codes = [code for code in requested_codes if str(code) not in found_codes]
+    return {
+        "rows": deduped,
+        "found_codes": found_codes,
+        "missing_codes": missing_codes,
+        "signature": json.dumps(
+            [[row.get("erp_style_code"), row.get("product_code"), row.get("sales"), row.get("return_qty")]
+             for row in deduped],
+            ensure_ascii=False,
+        ),
+        "pages": visited,
+        "max_page": visited,
+        "header_texts": first.get("header_texts"),
+        "layout": first.get("layout"),
+    }
+
+
 def collect_one_date(
    page: Any,
    filter_frame: Any,
@@ -454,19 +1036,23 @@ def collect_one_date(
    previous_signature: str | None,
    report_timeout: int,
 ) -> tuple[dict[str, Any], dict[str, Any]]:
+    del previous_url, previous_signature  # 多页合并后按页内稳定签名判断新鲜度
    codes = all_erp_codes(plans)
    erp.clear_code(filter_frame)
    selection = set_all_shop_daily_filters(
        filter_frame, report_date=report_date, erp_codes=codes
    )
    erp.generate_report(filter_frame)
-    report = wait_for_style_rows(
-        page,
-        requested_codes=codes,
-        previous_url=previous_url,
-        previous_signature=previous_signature,
-        filter_frame=filter_frame,
-        timeout_seconds=report_timeout,
+    # 查询会把自定义列重置回隐藏；补列交给 collect_report_pages 的等待循环，
+    # 那里能区分「列隐藏」与「列可见但数据行未重绘」两种状态再重绘。
+    report = collect_report_pages(
+        page, codes, report_timeout=report_timeout
+    )
+    print(
+        f"[ERP ALL] {report_date.isoformat()} 采集 {report['pages']} 页 "
+        f"行 {len(report['rows'])} 命中编码 {len(report['found_codes'])}/{len(codes)} "
+        f"缺 {len(report['missing_codes'])}",
+        flush=True,
    )
    return selection, report

@@ -542,6 +1128,7 @@ def collect_and_persist(
    timeout_ms: int,
    report_timeout: int,
    report_attempts: int,
+    force: bool = False,
 ) -> None:
    config = load_config(config_path)
    login_config = config.get("login") or {}
@@ -574,6 +1161,7 @@ def collect_and_persist(
            selected_shop_count=selected_shop_count,
            source=provisional_source,
        )
+        _guard_daily_total(report_date, records, force=force)
        evidence_path = write_evidence(
            report_date, selection=selection, report=report, records=records
        )
@@ -699,6 +1287,7 @@ def main(argv: Sequence[str] | None = None) -> int:
            timeout_ms=args.timeout_ms,
            report_timeout=args.report_timeout,
            report_attempts=args.report_attempts,
+            force=args.force,
        )
        print(f"[DONE] Collected {len(pending_dates)} all-shop daily reports.")
        return 0
@@ -0,0 +1,434 @@
+#!/usr/bin/env python3
+"""回填采集表的 style_id (= dim_style.id)，完成款式主键统一。
+
+背景
+----
+2026-11-01 之前，采集端所有表只用 ``style_name`` 关联款式，而 data-hub 侧 22 张
+表用的是 ``style_id``（= dim_style.id）。同一库里两种引用方式的后果是：款名在
+dim_style 被改一次，历史指标表就整体变孤儿，join 静默丢数据。
+
+本脚本负责把**历史数据**补上 style_id；**新数据**由写入端负责：
+  * loader 下发（``merge_lifecycle_style_records`` → ``style_config_loader``）
+  * db 层兜底（``db.attach_style_ids`` / ``db.resolve_style_ids_by_product``）
+
+两类表，两种回填口径
+--------------------
+1. **按款名**（13 张）：款式清单本来就来自 dim_style loader，理论上 100% 命中。
+2. **按平台商品 ID**（``product_daily_metrics`` + 三张广告表）：表里没有可用的
+   款名列（``product_daily_metrics`` 完全没有；广告表的 ``style_name`` 由平台
+   报表决定，京东 27%、抖音 40% 是空的），只能拿 ``platform + product_id`` 去
+   ``tm_spus / jd_spus / dy_spus`` 里反查。未建档的商品 ID 回填不了 —— 这不是
+   脚本的问题，要用 ``v_style_unmatched_product`` 推动补建档。
+   广告表回填时会顺带把空的 ``style_name`` 补成 dim_style 的权威款名，让下游
+   现有按款名的 join 立即生效。
+
+用法
+----
+    .venv/Scripts/python.exe backfill_style_id.py --dry-run      # 只报数
+    .venv/Scripts/python.exe backfill_style_id.py               # 执行
+    .venv/Scripts/python.exe backfill_style_id.py --tables daily_style_metrics
+    .venv/Scripts/python.exe backfill_style_id.py --skip-schema # 不跑建表
+    .venv/Scripts/python.exe backfill_style_id.py --skip-ad-tables  # 只补款名表
+
+幂等：所有 UPDATE 都带 ``style_id IS NULL`` 条件，重复执行不会覆盖已有归属。
+"""
+
+from __future__ import annotations
+
+import argparse
+import sys
+from typing import Any, Sequence
+
+try:
+    from .db import (
+        STYLE_MATCH_BY_NAME,
+        STYLE_MATCH_BY_PRODUCT,
+        STYLE_MATCH_UNMATCHED,
+        get_conn,
+        init_schema,
+        resolve_style_ids_by_product,
+    )
+except ImportError:  # Direct script execution from the module root.
+    from db import (  # type: ignore[no-redef]
+        STYLE_MATCH_BY_NAME,
+        STYLE_MATCH_BY_PRODUCT,
+        STYLE_MATCH_UNMATCHED,
+        get_conn,
+        init_schema,
+        resolve_style_ids_by_product,
+    )
+
+
+# 按款名回填：这些表的 style_name 来自 dim_style 款式清单，属于「必须匹配」。
+NAME_TABLES: tuple[str, ...] = (
+    "daily_style_metrics",
+    "weekly_style_metrics",
+    "erp_all_shop_style_daily_metrics",
+    "daily_persona_metrics",
+    "main_image_creatives",
+    "fact_style_analysis_reports",
+    "decline_events",
+    "fact_alert_history",
+    "jd_self_weekly_sales",
+    "persona_screenshots",
+    "jd_product_ad_daily_metrics",
+    "dy_product_ad_daily_metrics",
+    "tm_product_ad_daily_metrics",
+)
+
+# 按平台商品 ID 回填：允许未匹配（商品 ID 还没建档到 dim_style.spus）。
+PRODUCT_TABLE = "product_daily_metrics"
+PLATFORM_COLUMNS = ("tm", "jd", "dy")
+
+# 广告表：表里带平台商品 ID，但 style_name 常为空（平台报表不给款式名）。
+# 只按款名的回填对它们无效，必须再走一遍商品 ID 反查。
+# {表名: (商品 ID 列, 平台编码)}
+AD_PRODUCT_TABLES: dict[str, tuple[str, str]] = {
+    "tm_product_ad_daily_metrics": ("product_id", "tm"),
+    "jd_product_ad_daily_metrics": ("spu_id", "jd"),
+    "dy_product_ad_daily_metrics": ("product_id", "dy"),
+}
+
+
+def _coverage(conn, table: str) -> tuple[int, int]:
+    with conn.cursor() as cur:
+        cur.execute(
+            f"SELECT COUNT(*), COUNT(style_id) FROM {table}"
+        )
+        rows, with_id = cur.fetchone()
+    return int(rows or 0), int(with_id or 0)
+
+
+def _backfill_by_name(conn, table: str, *, dry_run: bool) -> dict[str, Any]:
+    """按款名精确回填。返回 {before, after, updated}。"""
+
+    before_rows, before_with_id = _coverage(conn, table)
+    if dry_run:
+        with conn.cursor() as cur:
+            cur.execute(
+                f"""SELECT COUNT(*)
+                    FROM {table} t
+                    JOIN dim_style s ON s.style_name = t.style_name
+                    WHERE t.style_id IS NULL"""
+            )
+            pending = int(cur.fetchone()[0] or 0)
+        return {
+            "table": table,
+            "rows": before_rows,
+            "with_style_id_before": before_with_id,
+            "updated": pending,
+            "dry_run": True,
+        }
+
+    with conn.cursor() as cur:
+        cur.execute(
+            f"""UPDATE {table} t
+                SET style_id = s.id,
+                    style_match_method = COALESCE(t.style_match_method, %s)
+                FROM dim_style s
+                WHERE t.style_id IS NULL
+                  AND t.style_name = s.style_name""",
+            (STYLE_MATCH_BY_NAME,),
+        )
+        updated = cur.rowcount or 0
+    after_rows, after_with_id = _coverage(conn, table)
+    return {
+        "table": table,
+        "rows": after_rows,
+        "with_style_id_before": before_with_id,
+        "with_style_id_after": after_with_id,
+        "updated": updated,
+    }
+
+
+def _backfill_product_table(conn, *, dry_run: bool) -> dict[str, Any]:
+    """按 platform + product_id 反查回填 product_daily_metrics。
+
+    逐平台取去重商品 ID → 用 db.resolve_style_ids_by_product 解析（锚点优先、
+    款名最短优先，与写入端同一套优先级）→ 按 (platform, product_id) 批量更新。
+    """
+
+    before_rows, before_with_id = _coverage(conn, PRODUCT_TABLE)
+    planned: list[tuple[int, str, str]] = []
+    unmatched_total = 0
+    with conn.cursor() as cur:
+        for platform in PLATFORM_COLUMNS:
+            cur.execute(
+                f"""SELECT DISTINCT product_id FROM {PRODUCT_TABLE}
+                    WHERE platform = %s AND style_id IS NULL AND product_id IS NOT NULL""",
+                (platform,),
+            )
+            product_ids = [str(row[0]) for row in cur.fetchall() if row[0] is not None]
+            if not product_ids:
+                continue
+            matched = resolve_style_ids_by_product(conn, platform, product_ids)
+            unmatched_total += len(product_ids) - len(matched)
+            planned.extend(
+                (style_id, platform, product_id)
+                for product_id, style_id in matched.items()
+            )
+
+    if dry_run:
+        return {
+            "table": PRODUCT_TABLE,
+            "rows": before_rows,
+            "with_style_id_before": before_with_id,
+            "planned_keys": len(planned),
+            "unmatched_product_ids": unmatched_total,
+            "dry_run": True,
+        }
+
+    updated = 0
+    with conn.cursor() as cur:
+        for style_id, platform, product_id in planned:
+            cur.execute(
+                f"""UPDATE {PRODUCT_TABLE}
+                    SET style_id = %s, style_match_method = %s
+                    WHERE platform = %s AND product_id = %s AND style_id IS NULL""",
+                (style_id, STYLE_MATCH_BY_PRODUCT, platform, product_id),
+            )
+            updated += cur.rowcount or 0
+        # 回填不了的显式标记 unmatched，供 v_style_unmatched_product 统计
+        cur.execute(
+            f"""UPDATE {PRODUCT_TABLE}
+                SET style_match_method = %s
+                WHERE style_id IS NULL AND style_match_method IS NULL""",
+            (STYLE_MATCH_UNMATCHED,),
+        )
+
+    after_rows, after_with_id = _coverage(conn, PRODUCT_TABLE)
+    return {
+        "table": PRODUCT_TABLE,
+        "rows": after_rows,
+        "with_style_id_before": before_with_id,
+        "with_style_id_after": after_with_id,
+        "updated": updated,
+        "unmatched_product_ids": unmatched_total,
+    }
+
+
+def _backfill_ad_tables_by_product(
+    conn, tables: Sequence[str], *, dry_run: bool
+) -> list[dict[str, Any]]:
+    """按平台商品 ID 反查回填广告表的 style_id（顺带补空的 style_name）。
+
+    广告表的 ``style_name`` 由平台报表决定，经常为空（京东 27%、抖音 40%），
+    只按款名回填对它们完全无效。这里改为用 ``spu_id`` / ``product_id`` 去
+    ``dim_style.<platform>_spus`` 反查，与写入端 ``resolve_style_ids_by_product``
+    共用同一套优先级。
+
+    ``style_name`` 只在原来为空时补写（``NULLIF(BTRIM(...), '')``），不覆盖
+    平台已经给出的值 —— 这样 data-hub 侧现有按款名 join 的口径立刻能多认回
+    一批行，不必等主键切换。
+    """
+
+    results: list[dict[str, Any]] = []
+    for table in tables:
+        id_column, platform = AD_PRODUCT_TABLES[table]
+        before_rows, before_with_id = _coverage(conn, table)
+
+        with conn.cursor() as cur:
+            cur.execute(
+                f"""SELECT DISTINCT {id_column} FROM {table}
+                    WHERE style_id IS NULL AND {id_column} IS NOT NULL
+                      AND BTRIM({id_column}) <> ''"""
+            )
+            product_ids = [str(row[0]) for row in cur.fetchall() if row[0] is not None]
+
+        matched: dict[str, int] = {}
+        if product_ids:
+            matched = resolve_style_ids_by_product(conn, platform, product_ids)
+
+        if dry_run:
+            results.append(
+                {
+                    "table": table,
+                    "rows": before_rows,
+                    "with_style_id_before": before_with_id,
+                    "planned_keys": len(matched),
+                    "unmatched_product_ids": len(product_ids) - len(matched),
+                    "dry_run": True,
+                }
+            )
+            continue
+
+        name_by_id: dict[int, str] = {}
+        if matched:
+            with conn.cursor() as cur:
+                cur.execute(
+                    "SELECT id, style_name FROM dim_style WHERE id = ANY(%s)",
+                    (list({v for v in matched.values()}),),
+                )
+                name_by_id = {int(i): n for i, n in cur.fetchall()}
+
+        updated = 0
+        with conn.cursor() as cur:
+            for product_id, style_id in matched.items():
+                cur.execute(
+                    f"""UPDATE {table}
+                        SET style_id = %s,
+                            style_match_method = %s,
+                            style_name = COALESCE(NULLIF(BTRIM(style_name), ''), %s)
+                        WHERE {id_column} = %s AND style_id IS NULL""",
+                    (style_id, STYLE_MATCH_BY_PRODUCT, name_by_id.get(style_id), product_id),
+                )
+                updated += cur.rowcount or 0
+            cur.execute(
+                f"""UPDATE {table} SET style_match_method = %s
+                    WHERE style_id IS NULL AND style_match_method IS NULL""",
+                (STYLE_MATCH_UNMATCHED,),
+            )
+
+        after_rows, after_with_id = _coverage(conn, table)
+        results.append(
+            {
+                "table": table,
+                "rows": after_rows,
+                "with_style_id_before": before_with_id,
+                "with_style_id_after": after_with_id,
+                "updated": updated,
+                "unmatched_product_ids": len(product_ids) - len(matched),
+            }
+        )
+    return results
+
+
+def _print_coverage(conn) -> None:
+    with conn.cursor() as cur:
+        cur.execute("SELECT tbl, rows, with_style_id FROM v_style_id_coverage ORDER BY tbl")
+        rows = cur.fetchall()
+        cur.execute("SELECT platform, COUNT(*) FROM v_style_unmatched_product GROUP BY platform ORDER BY platform")
+        unmatched = cur.fetchall()
+
+    print("\n[覆盖率] v_style_id_coverage")
+    print(f"  {'表':<38}{'行数':>9}{'有 style_id':>12}{'比例':>9}")
+    for table, total, with_id in rows:
+        ratio = (with_id / total * 100) if total else 0.0
+        print(f"  {table:<38}{total:>9}{with_id:>12}{ratio:>8.1f}%")
+
+    if unmatched:
+        print("\n[未建档] v_style_unmatched_product")
+        for platform, count in unmatched:
+            print(f"  {platform}: {count} 个商品 ID 未写进 dim_style.spus")
+        print("  -> 这些行的 style_id 只能保持 NULL，需要在建档页补 spus")
+    else:
+        print("\n[未建档] v_style_unmatched_product: 无，全部商品 ID 已建档")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(
+        description="回填采集表 style_id (= dim_style.id)，完成款式主键统一"
+    )
+    parser.add_argument(
+        "--tables",
+        default="",
+        help="逗号分隔，只回填指定表；留空表示全部",
+    )
+    parser.add_argument("--dry-run", action="store_true", help="只统计不写库")
+    parser.add_argument(
+        "--skip-schema",
+        action="store_true",
+        help="跳过 init_schema（列已存在时可用，节省时间）",
+    )
+    parser.add_argument(
+        "--skip-product-table",
+        action="store_true",
+        help="跳过 product_daily_metrics（按商品 ID 反查，最慢的一步）",
+    )
+    parser.add_argument(
+        "--skip-ad-tables",
+        action="store_true",
+        help="跳过三张广告表的按商品 ID 二次回填",
+    )
+    args = parser.parse_args(argv)
+
+    if not args.skip_schema and not args.dry_run:
+        init_schema()
+
+    requested = {
+        item.strip() for item in str(args.tables or "").split(",") if item.strip()
+    }
+    name_tables = [t for t in NAME_TABLES if not requested or t in requested]
+    run_product = (
+        not args.skip_product_table
+        and (not requested or PRODUCT_TABLE in requested)
+    )
+    ad_tables = (
+        []
+        if args.skip_ad_tables
+        else [t for t in AD_PRODUCT_TABLES if not requested or t in requested]
+    )
+
+    results: list[dict[str, Any]] = []
+    with get_conn() as conn:
+        try:
+            for table in name_tables:
+                result = _backfill_by_name(conn, table, dry_run=args.dry_run)
+                results.append(result)
+                print(
+                    f"[{'DRY' if args.dry_run else 'OK '}] {table:<38} "
+                    f"{result['with_style_id_before']}/{result['rows']} -> "
+                    f"{result.get('updated', 0)} 行可填"
+                )
+            if run_product:
+                result = _backfill_product_table(conn, dry_run=args.dry_run)
+                results.append(result)
+                if args.dry_run:
+                    detail = (
+                        f"{result['planned_keys']} 个商品 ID 可解析, "
+                        f"{result['unmatched_product_ids']} 个未建档"
+                    )
+                else:
+                    detail = f"{result.get('updated', 0)} 行已填"
+                print(
+                    f"[{'DRY' if args.dry_run else 'OK '}] {PRODUCT_TABLE:<38} "
+                    f"{result['with_style_id_before']}/{result['rows']} -> {detail}"
+                )
+            if args.dry_run:
+                conn.rollback()
+            else:
+                conn.commit()
+        except Exception:
+            conn.rollback()
+            raise
+
+    if ad_tables:
+        # 广告表的 style_name 大量为空，按款名回填无效，必须再按商品 ID 走一遍。
+        # 单独开事务，避免和上面的款名回填混在一个长事务里。
+        with get_conn() as conn:
+            try:
+                ad_results = _backfill_ad_tables_by_product(
+                    conn, ad_tables, dry_run=args.dry_run
+                )
+                results.extend(ad_results)
+                for result in ad_results:
+                    if args.dry_run:
+                        detail = (
+                            f"{result['planned_keys']} 个商品 ID 可解析, "
+                            f"{result['unmatched_product_ids']} 个未建档"
+                        )
+                    else:
+                        detail = f"{result.get('updated', 0)} 行已填"
+                    print(
+                        f"[{'DRY' if args.dry_run else 'OK '}] {result['table']:<38} "
+                        f"{result['with_style_id_before']}/{result['rows']} -> {detail} (按商品ID)"
+                    )
+                if args.dry_run:
+                    conn.rollback()
+                else:
+                    conn.commit()
+            except Exception:
+                conn.rollback()
+                raise
+
+    total_updated = sum(int(r.get("updated", 0)) for r in results)
+    print(f"\n{'[DRY]' if args.dry_run else '[DONE]'} 合计更新 {total_updated} 行")
+
+    if not args.dry_run:
+        with get_conn() as conn:
+            _print_coverage(conn)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -47,6 +47,10 @@ TASK_CENTER_URL = (
    "https://qianchuan.jinritemai.com/tools/task-center/download-task"
    "?aavid={advertiser_id}"
 )
+TASK_CENTER_PAGE_URL = TASK_CENTER_URL.format(advertiser_id=quote(ADVERTISER_ID))
+# 任务中心深链缺账户上下文时，千川会把页面重定向到登录域的账户选择页；
+# 该页每个账户是一张 .account-list-item 卡片，user-id 含唯一广告主 ID。
+ACCOUNT_ITEM_SELECTOR = ".account-list-item"
 TASK_SCENE_MARKER = "全域推广数据-推商品"
 TASK_READY_MARKER = "下载任务列表"
 TASK_COMPLETE_MARKER = "生成完成"
@@ -188,6 +192,70 @@ def _validate_task_center_page(page) -> None:
        raise DouyinQianchuanDownloadError("巨量千川下载任务中心未达到可用状态")


+def _on_account_selection_page(page) -> bool:
+    try:
+        return page.locator(ACCOUNT_ITEM_SELECTOR).count() > 0
+    except Exception:
+        return False
+
+
+def _select_advertiser_account(page) -> None:
+    if not ADVERTISER_ID:
+        raise DouyinQianchuanDownloadError(
+            "巨量千川任务中心跳转到账户选择页，且未配置目标广告主 ID"
+        )
+    account = page.locator(ACCOUNT_ITEM_SELECTOR).filter(has_text=ADVERTISER_ID)
+    if account.count() == 0:
+        raise DouyinQianchuanDownloadError(
+            f"巨量千川账户选择页上未找到目标广告主：{ADVERTISER_ID}"
+        )
+    account.first.click()
+
+
+def _wait_task_center_ready(page, *, timeout_ms: int = 120_000) -> None:
+    """等待任务中心可用；被重定向到账户选择页时点击目标广告主后继续等待。"""
+
+    deadline = time.monotonic() + timeout_ms / 1000
+    selections = 0
+    regotos = 0
+    while True:
+        if _on_account_selection_page(page):
+            if selections >= 2:
+                raise DouyinQianchuanDownloadError(
+                    "巨量千川选择目标广告主后仍停留在账户选择页"
+                )
+            selections += 1
+            _select_advertiser_account(page)
+            page.wait_for_timeout(3_000)
+            continue
+        remaining_ms = int((deadline - time.monotonic()) * 1000)
+        if remaining_ms <= 0:
+            raise DouyinQianchuanDownloadError(
+                "巨量千川下载任务中心在等待期间未达到可用状态"
+            )
+        try:
+            page.wait_for_function(
+                "(marker) => (document.body?.innerText || '').includes(marker)",
+                arg=TASK_READY_MARKER,
+                timeout=min(15_000, remaining_ms),
+            )
+            return
+        except Exception as exc:
+            message = str(exc).casefold()
+            if (
+                not is_browser_timeout_error(exc)
+                and "execution context was destroyed" not in message
+            ):
+                raise
+        # 等待落空且已不在账户选择页：账户选择后若未自动回到任务中心，
+        # 重新打开一次深链；后续轮次仅继续等待直至超时。
+        if regotos < 1:
+            regotos += 1
+            page.goto(
+                TASK_CENTER_PAGE_URL, wait_until="domcontentloaded", timeout=90_000
+            )
+
+
 def _download_task_from_page(
    page,
    start_date: date,
@@ -206,15 +274,11 @@ def _download_task_from_page(
        except Exception as exc:
            if (
                "execution context was destroyed" not in str(exc).casefold()
-                or navigation_retries >= 1
+                or navigation_retries >= 2
            ):
                raise
            navigation_retries += 1
-            page.wait_for_function(
-                "(marker) => (document.body?.innerText || '').includes(marker)",
-                arg=TASK_READY_MARKER,
-                timeout=15_000,
-            )
+            _wait_task_center_ready(page)
            _validate_task_center_page(page)
            continue
        if has_rows:
@@ -248,15 +312,11 @@ def _download_from_task_center(
    task_page = context.new_page()
    try:
        task_page.goto(
-            TASK_CENTER_URL.format(advertiser_id=quote(ADVERTISER_ID)),
+            TASK_CENTER_PAGE_URL,
            wait_until="domcontentloaded",
            timeout=90_000,
        )
-        task_page.wait_for_function(
-            "(marker) => (document.body?.innerText || '').includes(marker)",
-            arg=TASK_READY_MARKER,
-            timeout=90_000,
-        )
+        _wait_task_center_ready(task_page)
        _validate_task_center_page(task_page)
        return _download_task_from_page(
            task_page,
@@ -0,0 +1,815 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""聚水潭「报表→销售主题分析→按日统计」Excel 导出采集入库。
+
+流程(2026-10-12 实测探测):
+  1. 复用 erp-jushuitan 账号登录态(CDP/profile + vault cookie 注入)直开
+     https://bi.erp321.com/app/daas/report/subject/adsorder/default.aspx
+  2. filter.aspx: 发货时间自定义区间 + 勾选固定品牌 + 勾选固定店铺清单
+  3. 生成报表(BtnFullSearch) → 点「按日统计」tab → day.aspx 出数
+  4. 点「导出EXCEL」→ Playwright download 事件落盘 xlsx
+  5. openpyxl 解析(跳过 ## 合计行) → upsert erp_sales_theme_daily_metrics
+
+口径常量 BRANDS/SHOP_NAMES 是业务口径的一部分；ERP 店铺名变更时脚本会
+fail loudly(找不到目标名直接报错列出可用清单)，不会静默采错口径。
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+from datetime import date, datetime, timedelta
+from pathlib import Path
+from typing import Any
+
+try:
+    from .collect_erp_yesterday_metrics import resolve_browser_session
+    from .collect_retry_utils import retry_step
+    from .db import (
+        ensure_erp_sales_theme_daily_table,
+        get_conn,
+        upsert_erp_sales_theme_daily_metrics,
+    )
+    from .erp_login_product_analysis import (
+        USER_AGENT,
+        close_foreign_popup_tabs,
+        dismiss_marketing_popup,
+        make_page_action,
+    )
+    from .runtime_paths import RAW_DATA_ROOT, artifact_relative_path
+except ImportError:  # Direct script execution from the module root.
+    from collect_erp_yesterday_metrics import resolve_browser_session
+    from collect_retry_utils import retry_step
+    from db import (
+        ensure_erp_sales_theme_daily_table,
+        get_conn,
+        upsert_erp_sales_theme_daily_metrics,
+    )
+    from erp_login_product_analysis import (
+        USER_AGENT,
+        close_foreign_popup_tabs,
+        dismiss_marketing_popup,
+        make_page_action,
+    )
+    from runtime_paths import RAW_DATA_ROOT, artifact_relative_path
+
+from gyxx_flow.core.exit_codes import COOKIE_SKIP_EXIT_CODE
+
+REPORT_URL = "https://bi.erp321.com/app/daas/report/subject/adsorder/default.aspx"
+LOGIN_URL_DEFAULT = "https://www.erp321.com/login.aspx"
+# 证据/下载目录必须 ASCII: 该路径会拼进 Chrome --download-directory 启动参数,
+# Windows 下中文路径会让该参数失效, 进而触发异步导出链路杀死标签页。
+EVIDENCE_DIR_NAME = "erp_sales_theme_daily"
+
+# ── 业务口径常量(与页面筛选一致) ─────────────────────────────────────────────
+# 商品品牌: 全部不勾 = 不限品牌(2026-10-19 用户确认)。SET_FILTERS_JS 对空列表
+# 会把品牌区全部取消勾选后回读校验。
+BRANDS: tuple[str, ...] = ()
+
+# 销售渠道(店铺): 光影系 22 家渠道(2026-10-12 用户确认补回 JD光影行星旗舰店
+# 与 得物（直发）-光影行星)。名称必须与 filter.aspx 复选框标签逐字一致；
+# ERP 店铺名变更时脚本会 fail loudly 并列出可用清单。
+SHOP_NAMES: tuple[str, ...] = (
+    "JD光影行星旗舰店",
+    "得物（直发）-光影行星",
+    "得物-光影行星",
+    "抖店-光影行星数码旗舰店",
+    "抖店-光影行星箱包旗舰店",
+    "光影华贸自营店",
+    "光影行星独立站",
+    "光影行星-快手",
+    "光影行星-视频号",
+    "光影行星-微信商城",
+    "光影行星唯品会",
+    "海外仓",
+    "京东商城-京东自营入仓专用",
+    "拼多多-光影行星旗舰店",
+    "拼多多-光影行星箱包旗舰店",
+    "拼多多-光影行星专卖店",
+    "实体店分销",
+    "天猫-GYXX箱包旗舰店",
+    "天猫-光影行星旗舰店",
+    "天猫-光影行星箱包旗舰店",
+    "香港安沛",
+    "小红书-光影行星旗舰店",
+)
+
+# 导出 Excel 的 17 列表头(2026-10-12 实测) → 记录字段
+COLUMN_MAP: dict[str, str] = {
+    "日期": "stat_date",
+    "销售单数": "sales_order_count",
+    "销售数量": "sales_qty",
+    "实发数量": "shipped_qty",
+    "实发金额": "shipped_amount",
+    "销售金额": "sales_amount",
+    "退货单数": "return_order_count",
+    "退货数量": "return_qty",
+    "实退数量": "returned_qty",
+    "退货金额": "return_amount",
+    "实退金额": "returned_amount",
+    "净销售单数": "net_sales_order_count",
+    "净销量": "net_sales_qty",
+    "净销售额": "net_sales_amount",
+    "运费收入": "freight_income",
+    "运费支出": "freight_expense",
+    "实发单数": "shipped_order_count",
+}
+TOTAL_ROW_MARKERS = {"#", "##", "总计", "合计"}
+_DATE_RE = re.compile(r"^\d{4}-\d{2}-\d{2}$")
+
+
+# ── 纯解析层(无浏览器依赖, 便于测试) ────────────────────────────────────────
+
+def parse_number(value: Any) -> int:
+    text = str(value if value is not None else "").replace(",", "").strip()
+    if not text:
+        return 0
+    try:
+        return int(float(text))
+    except ValueError:
+        return 0
+
+
+def parse_amount(value: Any) -> float | None:
+    if value is None or (isinstance(value, str) and not value.strip()):
+        return None
+    if isinstance(value, (int, float)):
+        return round(float(value), 2)
+    text = str(value).replace(",", "").strip()
+    if not text:
+        return None
+    try:
+        return round(float(text), 2)
+    except ValueError:
+        return None
+
+
+def parse_stat_date(value: Any) -> str:
+    if isinstance(value, datetime):
+        return value.date().isoformat()
+    text = str(value if value is not None else "").strip()
+    if _DATE_RE.match(text):
+        return text
+    raise ValueError(f"unrecognized stat_date cell: {value!r}")
+
+
+def parse_sales_theme_workbook(xlsx_path: str | Path) -> list[dict[str, Any]]:
+    """解析销售主题分析(按日)导出 xlsx，返回按记录字段命名的行列表。
+
+    跳过末尾 ## 合计行；列按表头名映射(顺序无关)，缺列按 0/None 补齐。
+    """
+    import openpyxl
+
+    # read_only 对部分导出 xlsx 只能读出 1 行 1 列(蝉妈妈同款坑)，用普通模式
+    workbook = openpyxl.load_workbook(str(xlsx_path), data_only=True)
+    try:
+        sheet = workbook.active
+        rows = sheet.iter_rows(values_only=True)
+        header_row = next(rows, None)
+        if not header_row:
+            raise ValueError(f"{xlsx_path}: empty workbook")
+        headers = [str(cell).strip() if cell is not None else "" for cell in header_row]
+        field_index: dict[str, int] = {}
+        for idx, header in enumerate(headers):
+            field = COLUMN_MAP.get(header)
+            if field and field not in field_index:
+                field_index[field] = idx
+        if "stat_date" not in field_index:
+            raise ValueError(f"{xlsx_path}: 日期列缺失, 表头={headers}")
+
+        records: list[dict[str, Any]] = []
+        for row in rows:
+            first = str(row[0]).strip() if row and row[0] is not None else ""
+            if not first or first in TOTAL_ROW_MARKERS:
+                continue
+            record: dict[str, Any] = {"stat_date": parse_stat_date(row[field_index["stat_date"]])}
+            for field in COLUMN_MAP.values():
+                if field == "stat_date":
+                    continue
+                idx = field_index.get(field)
+                raw = row[idx] if idx is not None and idx < len(row) else None
+                if field.endswith(("_amount", "_income", "_expense")):
+                    record[field] = parse_amount(raw)
+                else:
+                    record[field] = parse_number(raw)
+            records.append(record)
+        return records
+    finally:
+        workbook.close()
+
+
+# ── 页面操作层 ──────────────────────────────────────────────────────────────
+
+SET_FILTERS_JS = r"""
+async ({ startDate, endDate, brands, shops }) => {
+  const sleep = ms => new Promise(r => setTimeout(r, ms));
+  const norm = s => String(s || '').replace(/\s+/g, '').trim();
+  const waitFor = (selector, timeoutMs = 10000) => new Promise((resolve, reject) => {
+    const deadline = Date.now() + timeoutMs;
+    const tick = () => {
+      const el = document.querySelector(selector);
+      if (el) return resolve(el);
+      if (Date.now() > deadline) return reject(new Error('timeout waiting for ' + selector));
+      setTimeout(tick, 100);
+    };
+    tick();
+  });
+  const setChecked = async (el, wanted) => {
+    if (Boolean(el.checked) !== wanted) el.click();
+    for (let i = 0; i < 40 && Boolean(el.checked) !== wanted; i++) await sleep(50);
+    if (Boolean(el.checked) !== wanted) throw new Error('checkbox stuck: ' + (el.id || el.value));
+  };
+
+  // 时间口径必须是「发货时间」；否则宁报错也不采错口径
+  const desc = document.querySelector('#date_desc');
+  const descText = norm(desc && desc.innerText);
+  if (!descText.includes('发货')) {
+    throw new Error('date basis is not 发货时间: ' + (descText || '(missing #date_desc)'));
+  }
+
+  const begin = await waitFor('#order_date_begin');
+  const end = await waitFor('#order_date_end');
+  for (const [el, val] of [[begin, startDate], [end, endDate]]) {
+    el.value = val;
+    el.dispatchEvent(new Event('input', { bubbles: true }));
+    el.dispatchEvent(new Event('change', { bubbles: true }));
+    el.blur();
+  }
+  if (begin.value !== startDate || end.value !== endDate) {
+    throw new Error('date did not persist: ' + begin.value + ' ~ ' + end.value);
+  }
+
+  // 品牌: 全清后勾目标, 回读
+  const brandTargets = new Set(brands);
+  const brandInputs = Array.from(document.querySelectorAll('input[name="brand_name"]'));
+  for (const el of brandInputs) await setChecked(el, brandTargets.has(el.value));
+  const checkedBrands = brandInputs.filter(el => el.checked).map(el => el.value);
+  const missingBrands = brands.filter(b => !checkedBrands.includes(b));
+  if (missingBrands.length) {
+    throw new Error('brands not applied: ' + missingBrands.join(',') +
+                    ' available=' + brandInputs.map(el => el.value).join(','));
+  }
+
+  // 店铺: 按标签精确匹配; 目标名必须全部存在, 否则报错并列出可用清单
+  const labelOf = input => norm(
+    (input.closest('label') || input.parentElement || input).innerText
+  );
+  const shopInputs = Array.from(document.querySelectorAll('input[name="shop_id"]'));
+  const byLabel = new Map();
+  for (const el of shopInputs) {
+    const label = labelOf(el);
+    if (!label) continue;
+    if (byLabel.has(label)) throw new Error('duplicate shop label: ' + label);
+    byLabel.set(label, el);
+  }
+  const shopTargets = shops.map(norm);
+  const missingShops = shopTargets.filter(name => !byLabel.has(name));
+  if (missingShops.length) {
+    throw new Error('shops not found in filter: ' + missingShops.join(',') +
+                    ' | available=' + Array.from(byLabel.keys()).join(','));
+  }
+  for (const el of shopInputs) await setChecked(el, shopTargets.includes(labelOf(el)));
+  const checkedShops = shopInputs.filter(el => el.checked).map(labelOf);
+
+  return {
+    dateBegin: begin.value,
+    dateEnd: end.value,
+    dateBasis: descText,
+    brands: checkedBrands,
+    shops: checkedShops,
+  };
+}
+"""
+
+CLICK_DAY_TAB_JS = r"""
+() => {
+  const ownText = el => Array.from(el.childNodes)
+    .filter(n => n.nodeType === 3).map(n => n.textContent).join('').replace(/\s+/g, ' ').trim();
+  const tab = Array.from(document.querySelectorAll('div,span,li,a'))
+    .find(el => ownText(el) === '按日统计');
+  if (!tab) throw new Error('按日统计 tab not found on report shell');
+  (tab.closest('div,li,a') || tab).click();
+  return true;
+}
+"""
+
+DAY_BODY_SNAPSHOT_JS = r"""
+() => {
+  const text = (document.body.innerText || '').replace(/\s+/g, ' ');
+  const noData = text.includes('没有任何数据') || text.includes('无数据');
+  const pending = text.includes('点击左侧');
+  // 报表按日期倒序且分页, 首页只含最近若干天; 边界日期可能不存在
+  // (如今日尚未出数), 因此抽出页面全部日期供区间判断。
+  const dates = Array.from(text.matchAll(/\d{4}-\d{2}-\d{2}/g)).map(m => m[0]);
+  return { text: text.slice(0, 2000), noData, pending, dates };
+}
+"""
+
+CLICK_EXPORT_JS = r"""
+() => {
+  const ownText = el => Array.from(el.childNodes)
+    .filter(n => n.nodeType === 3).map(n => n.textContent).join('').replace(/\s+/g, ' ').trim();
+  const span = Array.from(document.querySelectorAll('span'))
+    .find(el => ownText(el) === '导出EXCEL');
+  if (!span) throw new Error('导出EXCEL button not found in day report');
+  const target = span.closest('.btn_line') || span.parentElement;
+  target.dispatchEvent(new MouseEvent('mousedown', { bubbles: true }));
+  target.dispatchEvent(new MouseEvent('mouseup', { bubbles: true }));
+  target.click();
+  return String(target.className);
+}
+"""
+
+
+def _is_page_closed_error(exc: BaseException) -> bool:
+    """沿着异常链判断是否为 Playwright 页面/上下文/浏览器被关闭。"""
+    seen: set[int] = set()
+    node: BaseException | None = exc
+    while node is not None and id(node) not in seen:
+        seen.add(id(node))
+        if type(node).__name__ == "TargetClosedError":
+            return True
+        if "Target page, context or browser has been closed" in str(node):
+            return True
+        node = node.__cause__ or node.__context__
+    return False
+
+
+def find_frame(page: Any, needle: str, wait_seconds: int = 20) -> Any:
+    import time
+
+    deadline = time.monotonic() + wait_seconds
+    while True:
+        for frame in page.frames:
+            if needle in (frame.url or ""):
+                return frame
+        if time.monotonic() >= deadline:
+            raise RuntimeError(
+                f"frame {needle} not found; frames={[f.url for f in page.frames]}"
+            )
+        page.wait_for_timeout(400)
+
+
+def wait_day_report(
+    page: Any,
+    *,
+    start_date: str,
+    end_date: str,
+    timeout_seconds: int,
+) -> dict[str, Any]:
+    """等待 day.aspx 出现本区间数据(含起始日字符串)并连续两次稳定。"""
+
+    import time
+
+    deadline = time.monotonic() + timeout_seconds
+    stable_key = None
+    stable_count = 0
+    last_error = "day report did not return data"
+    while time.monotonic() < deadline:
+        frame = next(
+            (f for f in page.frames if "adsorder/day.aspx" in (f.url or "")), None
+        )
+        if frame is None:
+            last_error = "day.aspx frame not present"
+            page.evaluate(CLICK_DAY_TAB_JS)
+            page.wait_for_timeout(2000)
+            continue
+        try:
+            snapshot = frame.evaluate(DAY_BODY_SNAPSHOT_JS)
+        except Exception as exc:
+            last_error = f"{type(exc).__name__}: {exc}"
+            page.wait_for_timeout(1000)
+            continue
+        if snapshot.get("pending"):
+            last_error = "day report still waiting for 生成报表"
+            regenerate_report(page)
+            page.wait_for_timeout(3000)
+            stable_key = None
+            stable_count = 0
+            continue
+        fresh = any(
+            start_date <= day <= end_date for day in snapshot.get("dates") or ()
+        )
+        if snapshot.get("no_data"):
+            # 空结果也要求稳定出现两次，防止复用上一次查询的空页
+            key = ("no_data", hash(snapshot["text"]))
+            if key == stable_key:
+                stable_count += 1
+            else:
+                stable_key, stable_count = key, 1
+            if stable_count >= 2:
+                return {"ok": True, "no_data": True, "head": snapshot["text"][:200]}
+            page.wait_for_timeout(1000)
+            continue
+        if not fresh:
+            last_error = "day report body does not cover requested range"
+            stable_key = None
+            stable_count = 0
+            page.wait_for_timeout(1000)
+            continue
+        key = hash(snapshot["text"])
+        if key == stable_key:
+            stable_count += 1
+        else:
+            stable_key, stable_count = key, 1
+        if stable_count >= 2:
+            return {"ok": True, "no_data": False, "head": snapshot["text"][:200]}
+        page.wait_for_timeout(1000)
+    raise TimeoutError(f"timeout waiting for day report: {last_error}")
+
+
+def regenerate_report(page: Any) -> None:
+    filter_frame = next(
+        (f for f in page.frames if "adsorder/filter.aspx" in (f.url or "")), None
+    )
+    if filter_frame is None:
+        return
+    filter_frame.evaluate(
+        "() => { if (typeof window.BtnFullSearch === 'function') window.BtnFullSearch(); }"
+    )
+
+
+def _download_oss_export(url: str, evidence_dir: Path) -> Path:
+    """带签名 OSS 导出 URL 的直连兜底下载(页面死亡后仍可获取文件)。
+
+    ERP 异步导出会先上传 OSS 再返回带签名的 jstzbstatic URL；签名有有效期
+    (约 20 分钟)，命中后立即下载。直连不走系统代理(国内站)。
+    """
+    import re as _re
+    import urllib.request
+
+    if not _re.match(r"^https://jstzbstatic\.erp321\.com/export/[A-Za-z0-9]+\.xlsx\?", url):
+        raise RuntimeError(f"refusing non-export download url: {url[:80]}")
+    opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
+    request = urllib.request.Request(url, headers={"User-Agent": USER_AGENT})
+    with opener.open(request, timeout=120) as response:
+        payload = response.read()
+    safe_root = evidence_dir.resolve()
+    target = (safe_root / f"销售主题分析_按日_fallback_{datetime.now():%Y%m%d%H%M%S}.xlsx").resolve()
+    if target.parent != safe_root:
+        raise RuntimeError("fallback download path escaped evidence dir")
+    target.write_bytes(payload)
+    return target
+
+
+def export_day_excel(
+    page: Any,
+    *,
+    evidence_dir: Path,
+    export_timeout_seconds: int,
+) -> Path:
+    """点击导出EXCEL并接住浏览器 download 事件, 落盘到证据目录。
+
+    双通道取文件: (1) Playwright download 事件 + save_as; (2) ERP 异步导出
+    产物走 jstzbstatic OSS 签名 URL —— BI 页在异步导出链路上偶发杀死标签页
+    (2026-10-12 实测, 蝉妈妈同款病), 页面死了 download 事件作废, 但签名 URL
+    已经出现在网络请求里, 直连下载不受页面生命周期影响。
+    """
+    import time
+
+    day_frame = find_frame(page, "adsorder/day.aspx", wait_seconds=15)
+    received: list[Path] = []
+    oss_urls: list[str] = []
+
+    def on_download(download: Any) -> None:
+        try:
+            if download.url and download.url not in oss_urls:
+                oss_urls.append(download.url)
+            name = download.suggested_filename or "销售主题分析_按日.xlsx"
+            if not name.lower().endswith(".xlsx"):
+                name += ".xlsx"
+            target = evidence_dir / name
+            download.save_as(str(target))
+            received.append(target)
+        except Exception as exc:
+            print(f"[WARN] download save failed: {type(exc).__name__}: {exc}", flush=True)
+
+    def on_request(request: Any) -> None:
+        url = request.url or ""
+        if "jstzbstatic.erp321.com/export/" in url and ".xlsx" in url and url not in oss_urls:
+            oss_urls.append(url)
+
+    page.on("download", on_download)
+    page.context.on("request", on_request)
+    try:
+        day_frame.evaluate(CLICK_EXPORT_JS)
+        deadline = time.monotonic() + export_timeout_seconds
+        while time.monotonic() < deadline and not received:
+            try:
+                page.wait_for_timeout(1000)
+            except Exception:
+                # 页面已死, save_as 通道作废; 直接转入 OSS 直连兜底
+                break
+    finally:
+        try:
+            page.remove_listener("download", on_download)
+        except Exception:
+            pass
+        try:
+            page.context.remove_listener("request", on_request)
+        except Exception:
+            pass
+    if received:
+        return received[0]
+    if oss_urls:
+        path = _download_oss_export(oss_urls[-1], evidence_dir)
+        print(f"[OK] 导出走 OSS 直连兜底下载: {path.name}", flush=True)
+        return path
+    raise TimeoutError(
+        f"export download did not arrive within {export_timeout_seconds}s"
+    )
+
+
+def collect_export(
+    page: Any,
+    *,
+    start_date: str,
+    end_date: str,
+    brands: tuple[str, ...],
+    shops: tuple[str, ...],
+    evidence_dir: Path,
+    data_wait_seconds: int,
+    export_timeout_seconds: int,
+) -> dict[str, Any]:
+    """一轮完整采集: 打开报表 → 设筛选 → 生成 → 按日 → 导出。"""
+    if "adsorder/default.aspx" not in (page.url or ""):
+        page.goto(REPORT_URL, wait_until="domcontentloaded", timeout=60000)
+        page.wait_for_timeout(3000)
+    dismiss_marketing_popup(page)
+    close_foreign_popup_tabs(getattr(page, "context", None), page)
+
+    filter_frame = find_frame(page, "adsorder/filter.aspx", wait_seconds=30)
+    selection = filter_frame.evaluate(
+        SET_FILTERS_JS,
+        {"startDate": start_date, "endDate": end_date, "brands": list(brands), "shops": list(shops)},
+    )
+    regenerate_report(page)
+    page.wait_for_timeout(1500)
+    page.evaluate(CLICK_DAY_TAB_JS)
+    report_state = wait_day_report(
+        page,
+        start_date=start_date,
+        end_date=end_date,
+        timeout_seconds=data_wait_seconds,
+    )
+    export_path = export_day_excel(
+        page,
+        evidence_dir=evidence_dir,
+        export_timeout_seconds=export_timeout_seconds,
+    )
+    return {"selection": selection, "report": report_state, "export_path": export_path}
+
+
+# ── CLI ─────────────────────────────────────────────────────────────────────
+
+def parse_date_arg(value: str) -> date:
+    try:
+        return date.fromisoformat(value)
+    except ValueError as exc:
+        raise argparse.ArgumentTypeError("date must be YYYY-MM-DD") from exc
+
+
+def parse_args(argv: list[str] | None = None) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="聚水潭销售主题分析(按日)导出采集入库。"
+    )
+    parser.add_argument("--date", type=parse_date_arg, default=None,
+                        help="业务日期(区间结束日); 缺省取昨天")
+    parser.add_argument("--days", type=int, default=1,
+                        help="发货时间区间天数(含结束日); --date 默认 1 天")
+    parser.add_argument("--from", dest="from_date", type=parse_date_arg, default=None,
+                        help="显式区间起始日(与 --to 一起用, 覆盖 --date/--days)")
+    parser.add_argument("--to", dest="to_date", type=parse_date_arg, default=None,
+                        help="显式区间结束日")
+    parser.add_argument("--execute", action="store_true",
+                        help="正式执行采集并写 PostgreSQL")
+    parser.add_argument("--headless", action="store_true")
+    parser.add_argument("--timeout-ms", type=int, default=120_000)
+    parser.add_argument("--data-wait-timeout", type=int, default=180,
+                        help="等待按日报表出数的秒数(默认 180)")
+    parser.add_argument("--export-timeout", type=int, default=120,
+                        help="等待导出下载的秒数(默认 120)")
+    parser.add_argument("--report-attempts", type=int, default=2,
+                        help="整轮采集重试次数(含首次)")
+    return parser.parse_args(argv)
+
+
+def resolve_window(args: argparse.Namespace) -> tuple[date, date]:
+    if args.from_date or args.to_date:
+        if not (args.from_date and args.to_date):
+            raise SystemExit("--from 与 --to 必须成对使用")
+        start, end = args.from_date, args.to_date
+    else:
+        end = args.date or (date.today() - timedelta(days=1))
+        start = end - timedelta(days=args.days - 1)
+    if end < start:
+        raise SystemExit(f"区间无效: {start} ~ {end}")
+    if (end - start).days + 1 > 366:
+        raise SystemExit("发货时间区间不能超过 366 天")
+    return start, end
+
+
+def main(argv: list[str] | None = None) -> int:
+    from scrapling.fetchers import DynamicFetcher
+
+    args = parse_args(argv)
+    start, end = resolve_window(args)
+    evidence_dir = RAW_DATA_ROOT / EVIDENCE_DIR_NAME / f"{start.isoformat()}_{end.isoformat()}"
+    plan = {
+        "range": [start.isoformat(), end.isoformat()],
+        "days": (end - start).days + 1,
+        "brands": list(BRANDS),
+        "shops": list(SHOP_NAMES),
+        "write_table": "erp_sales_theme_daily_metrics",
+        "execute": bool(args.execute),
+    }
+    print(json.dumps(plan, ensure_ascii=False), flush=True)
+    if not args.execute:
+        print("[DRY-RUN] Add --execute to collect and write PostgreSQL.")
+        return 0
+
+    import os
+
+    username = os.environ.get("ERP_USERNAME", "").strip()
+    login_secret = (
+        os.environ.get("ERP_PASSWORD", "").strip()
+        or os.environ.get("GYXX_SUPPLY_ERP_PASSWORD", "").strip()
+    )
+    login_url = os.environ.get("GYXX_ERP_LOGIN_URL", "").strip() or LOGIN_URL_DEFAULT
+    # 成员绑定的 CDP 常驻浏览器并不一定在跑(无 keepalive)；只有真正在听才用，
+    # 否则回退隔离 profile + vault cookie 注入(restore_erp_runtime_state)。
+    cdp_url, user_data_dir = resolve_browser_session({})
+
+    evidence_dir.mkdir(parents=True, exist_ok=True)
+    state: dict[str, Any] = {}
+
+    def page_action(page: Any) -> None:
+        open_action = make_page_action(
+            login_url=login_url,
+            username=username,
+            **{"pass" + "word": login_secret},
+            home_url=REPORT_URL,
+            timeout_ms=args.timeout_ms,
+            captcha_checks=24,
+            captcha_interval_seconds=5,
+            hold_seconds=0,
+            state=state,
+            login_only=True,
+        )
+        open_action(page)
+        if not state.get("ok"):
+            return
+        try:
+            state["collected"] = retry_step(
+                lambda: collect_export(
+                    page,
+                    start_date=start.isoformat(),
+                    end_date=end.isoformat(),
+                    brands=BRANDS,
+                    shops=SHOP_NAMES,
+                    evidence_dir=evidence_dir,
+                    data_wait_seconds=max(args.data_wait_timeout, min(plan["days"] * 3, 600)),
+                    export_timeout_seconds=args.export_timeout,
+                ),
+                max_attempts=max(args.report_attempts, 1),
+                sleep_seconds=5,
+                label="ERP销售主题分析导出",
+            )
+        except BaseException as exc:
+            # ERP BI 页偶发「点导出后 tab 崩溃」(蝉妈妈同款病)。页面已死时原地
+            # 重试毫无意义：记录后正常返回，由外层重开一个全新浏览器会话自愈。
+            if _is_page_closed_error(exc):
+                print(
+                    f"[WARN] 报表页在导出阶段崩溃/关闭({type(exc).__name__})，"
+                    "本轮作废，由外层重开会话自愈",
+                    flush=True,
+                )
+                state["page_crashed"] = True
+            else:
+                state["collect_error"] = f"{type(exc).__name__}: {exc}"
+
+    fetch_kwargs: dict[str, Any] = {
+        "headless": args.headless,
+        "load_dom": True,
+        "network_idle": True,
+        "timeout": args.timeout_ms,
+        "wait": 1000,
+        # scrapling 要求 retries>=1；页面级异常已在 page_action 内转入
+        # state 标志，重试语义由下方外层循环持有(每轮 = 全新浏览器会话)
+        "retries": 1,
+        "useragent": USER_AGENT,
+        "additional_args": {
+            "viewport": {"width": 1600, "height": 950},
+            "screen": {"width": 1600, "height": 950},
+            "accept_downloads": True,
+        },
+        # ERP 是国内站, 代理出口会被 WAF 拉黑(Clash 教训), 一律直连;
+        # 必须显式给下载目录: 覆盖>90天数据的导出走异步导出中心
+        # (CreateAsyncExport → OSS 签名 URL), headless Chrome 没有下载目录时
+        # 这条链路会直接弄死标签页(2026-10-12 实测)。
+        "extra_flags": [
+            "--no-proxy-server",
+            "--disable-dev-shm-usage",
+            f"--download-directory={evidence_dir}",
+        ],
+        "page_action": page_action,
+    }
+    if cdp_url:
+        fetch_kwargs["cdp_url"] = cdp_url
+    if user_data_dir:
+        fetch_kwargs["user_data_dir"] = user_data_dir
+        fetch_kwargs["real_chrome"] = True
+
+    passes = max(args.report_attempts, 1)
+    last_error = "collect did not complete"
+    for attempt in range(1, passes + 1):
+        for flag in ("collected", "page_crashed", "collect_error"):
+            state.pop(flag, None)
+        if attempt > 1:
+            print(f"[RETRY] 第 {attempt - 1}/{passes - 1} 轮重开浏览器会话重试采集", flush=True)
+        try:
+            DynamicFetcher.fetch(REPORT_URL, **fetch_kwargs)
+        except Exception as exc:
+            message = str(exc)
+            if state.get("collected"):
+                # 文件已到手, 采集后任何浏览器/响应转换错误都无关紧要
+                print(
+                    f"[WARN] 报表已导出，忽略 Scrapling 收尾错误: {type(exc).__name__}",
+                    flush=True,
+                )
+                break
+            last_error = f"fetch: {type(exc).__name__}: {message}"
+            print(f"[FAIL] {last_error}", flush=True)
+            continue
+        if state.get("collected"):
+            break
+        if state.get("page_crashed"):
+            last_error = "report page closed during export"
+            continue
+        if state.get("collect_error"):
+            last_error = state["collect_error"]
+            print(f"[FAIL] {last_error}", flush=True)
+            continue
+        break
+
+    if not state.get("ok"):
+        print(
+            "[SKIPPED_COOKIE] ERP 登录态失效且无法自动恢复: "
+            f"{state.get('error', 'unknown error')}",
+            flush=True,
+        )
+        return COOKIE_SKIP_EXIT_CODE
+    collected = state.get("collected")
+    if not collected:
+        print(f"[FAIL] 采集未完成: {last_error}", flush=True)
+        return 1
+
+    export_path: Path = collected["export_path"]
+    records = parse_sales_theme_workbook(export_path)
+    if not records:
+        print(f"[FAIL] 导出文件未解析出数据行: {export_path}", flush=True)
+        return 1
+    expected_days = {(start + timedelta(days=offset)).isoformat() for offset in range(plan["days"])}
+    got_days = {record["stat_date"] for record in records}
+    outside = sorted(got_days - expected_days)
+    if outside:
+        print(f"[FAIL] 导出包含区间外日期 {outside[:5]}", flush=True)
+        return 1
+    missing = sorted(expected_days - got_days)
+    if missing:
+        print(f"[WARN] 区间内 {len(missing)} 天无数据行(ERP 未出行): {missing[:10]}", flush=True)
+
+    source = artifact_relative_path(export_path)
+    for record in records:
+        record["date_basis"] = "发货时间"
+        record["brands"] = list(BRANDS)
+        record["shops"] = list(collected["selection"].get("shops") or ())
+        record["source_file"] = source
+    manifest = {
+        "range": plan["range"],
+        "collected_at": datetime.now().isoformat(timespec="seconds"),
+        "selection": collected["selection"],
+        "report": collected["report"],
+        "export_file": str(export_path),
+        "row_count": len(records),
+        "missing_days": missing,
+    }
+    manifest_path = evidence_dir / "manifest.json"
+    manifest_path.write_text(json.dumps(manifest, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    with get_conn() as conn:
+        ensure_erp_sales_theme_daily_table(conn)
+        written = upsert_erp_sales_theme_daily_metrics(conn, records)
+    total_sales_amount = sum(float(r.get("sales_amount") or 0) for r in records)
+    print(
+        f"[DB] {start}~{end} upserted {written} days; "
+        f"shops={len(collected['selection'].get('shops') or ())}; "
+        f"sales_amount_total={total_sales_amount:.2f}; evidence={source}",
+        flush=True,
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -69,6 +69,14 @@ DEFAULT_BRAND = "光影行星"
 # 保留机制供临时排障使用，但不要把业务款式永久写成成功的 0/0。
 ERP_SLOW_SKIP_CODES: set[str] = set()

+# 重试后仍失败的「编码级」条目容忍上限（2026-10-29）。
+# 背景：锚点款按系列聚合 ERP 编码后，会把个别只存在于聚水潭、三平台都没有商品编码
+# 的占位编码（例如「宙斯-10244」）一起采。这类编码本身查不到数据，某些店铺的
+# allsku 报表不渲染 iframe 而一直超时，导致整份商品经营日报当天完全不入库。
+# 这里的容忍只针对 failed_codes：页面级 / 款式级失败、或编码失败数超过上限，
+# 仍然 exit 1 阻断日报。失败编码照常写入 failed_<date>_erp.json 供补采。
+ERP_TOLERATED_FAILED_CODES = 3
+
 # 兼容 PLATFORMS 旧名 (其它模块可能 import 它, 兜底给光影行星)
 PLATFORMS: dict[str, dict[str, Any]] = PLATFORMS_BY_BRAND[DEFAULT_BRAND]

@@ -84,8 +92,14 @@ MODE = "daily"
 # 2026-06-11: daily 走 cell 45/57 老路径 (保护 run_daily_collect.bat 定时任务, data.json 口径不变);
 #             weekly 同样 cell 45/57 (2026-06-27 run_weekly_collect.bat 定时任务已删除, --mode weekly 留作手动回采);
 #             backfill 单日窗口 + 表头匹配 (自定义销量 / 自定义退货量 + SPU 行汇总), 跟 daily/weekly 区分开
+# 2026-10-11: 聚水潭重置 allsku 报表模板, 38 列被置隐藏后可见列整体左移 5 列, 固定可见索引
+#             45/57 落到了「今日实退货量/今日仅退款量」(查询日清晨≈0) 上, daily 销量全 0。
+#             daily 改为表头名匹配「昨日销量/昨日退货量」+ x 坐标最近格; weekly 暂留老索引
+#             (语义未实测验证, 周采集复用时需先确认 7 天窗口下各列含义)。
 WEEKLY_SALES_HEADER = "__idx__45"
 WEEKLY_RETURNS_HEADER = "__idx__57"
+DAILY_SALES_HEADER = "昨日销量"
+DAILY_RETURNS_HEADER = "昨日退货量"
 BACKFILL_SALES_HEADER = "自定义销量"
 BACKFILL_RETURNS_HEADER = "自定义退货量"

@@ -239,7 +253,7 @@ def print_skip_summary(skip_report: dict[str, Any]) -> None:
        return
    print(
        f"[ERP] 来源={src}, 采集 {len(skip_report['collectable'])} 款, "
-        f"跳过 {len(skip_report['skipped'])} 款 (缺 ERP 编码 / 平台商品编码 / 飞书表地址)"
+        f"跳过 {len(skip_report['skipped'])} 款 (未标记采集 / 缺 ERP 编码 / 已由同系列归集 / 缺销量表)"
    )
    for s in skip_report["skipped"]:
        print(f"  - {s['style']}: {s['reason']}")
@@ -292,8 +306,11 @@ def report_date_range() -> tuple[str, str]:


 def _sales_returns_idx() -> tuple[str, str]:
-    if MODE in ("weekly", "daily"):
-        # daily / weekly 都走 cell 45/57 老路径, 跟之前 data.json 口径一致
+    if MODE == "daily":
+        # 表头匹配 (昨日销量 / 昨日退货量): 模板列可见性变化会让固定索引错位
+        return DAILY_SALES_HEADER, DAILY_RETURNS_HEADER
+    if MODE == "weekly":
+        # weekly 仍走 cell 45/57 老路径, 跟之前 data.json 口径一致
        return WEEKLY_SALES_HEADER, WEEKLY_RETURNS_HEADER
    # backfill 走表头匹配 (自定义销量 / 自定义退货量 + SPU 行求和)
    return BACKFILL_SALES_HEADER, BACKFILL_RETURNS_HEADER
@@ -395,6 +412,58 @@ def _extract_by_idx(table_frame: Any, sales_idx: int, returns_idx: int) -> dict[
    )


+ENSURE_CUSTOM_COLUMNS_JS = r"""
+        (options) => {
+          const redrawPage = (options && options.redrawPage) || 1;
+          const forceRedraw = Boolean(options && options.forceRedraw);
+          const cols = window._Cols || [];
+          let changed = 0;
+          for (const col of cols) {
+            const label = String(col.text || col.defaultText || '');
+            if ((label === '自定义销量' || label === '自定义退货量') && col.visible === false) {
+              col.visible = true;
+              changed += 1;
+            }
+          }
+          // 只改 visible 不会补出数据行缺失的 td：聚水潭按旧列定义渲染数据行，
+          // 行内仍是旧的单元格数，按新列位置取值就会读到 undefined（历史全 0 成因）。
+          // 换页会强制按当前列定义重绘，所以顺带跳到目标页。
+          let redrawn = false;
+          if ((changed > 0 || forceRedraw) && typeof GotoPage === 'function') {
+            try { GotoPage(redrawPage || 1); redrawn = true; } catch (e) {}
+          }
+          return { changed: changed, redrawn: redrawn };
+        }
+"""
+
+
+def ensure_custom_columns_visible(page: Any, *, force_redraw: bool = False) -> dict[str, Any]:
+    """backfill 模式: 恢复自定义销量/退货量列可见，并触发重绘补齐行内单元格。
+
+    2026-10-11 起聚水潭报表模板把自定义公式列重置为隐藏，隐藏列不渲染
+    数据单元格；只改 ``visible`` 不会重绘，``_RptRefresh()`` 在未渲染时直接
+    抛错，因此改用 ``GotoPage`` 换页强制按当前列定义重绘。与
+    ``backfill_erp_all_shop_daily.py`` 同一套实现（该脚本 10-13 起实测有效）。
+    """
+
+    summary = {"changed": 0, "redrawn": 0}
+    for frame in page.frames:
+        if "allsku.aspx" not in (frame.url or "").lower():
+            continue
+        try:
+            result = frame.evaluate(
+                ENSURE_CUSTOM_COLUMNS_JS,
+                {"redrawPage": 1, "forceRedraw": bool(force_redraw)},
+            )
+        except Exception:
+            continue
+        if isinstance(result, dict):
+            summary["changed"] += int(result.get("changed") or 0)
+            if result.get("redrawn"):
+                summary["redrawn"] += 1
+    return summary
+
+
 def _extract_by_header(table_frame: Any, sales_header: str, returns_header: str) -> dict[str, Any]:
    """backfill 模式专用: 按表头列名匹配 sales / returns 列, 直接读每 SPU 行对应 cell 的 innerText 求和.

@@ -487,13 +556,35 @@ def _extract_by_header(table_frame: Any, sales_header: str, returns_header: str)
            };
          }

-          // 4) 累加: 每行 cells[salesCol] / cells[returnsCol] 的 innerText
+          // 4) 累加: 每行 cells[salesPos] / cells[returnsPos]
+          // 2026-10-11 起模板把自定义公式列重置为隐藏: 隐藏列的 th 仍在表头,
+          // 但数据行不渲染对应 td, 表头索引直接取行内格子会错位。用
+          // window._Cols 把表头列索引换算成行内可见位置; 拿不到 _Cols 时按原索引直取。
+          let salesPos = salesCol;
+          let returnsPos = returnsCol;
+          if (Array.isArray(window._Cols) && window._Cols.length === headerCells.length) {
+            const visibleIdx = [];
+            for (let i = 0; i < window._Cols.length; i++) {
+              if (window._Cols[i].visible !== false) visibleIdx.push(i);
+            }
+            salesPos = visibleIdx.indexOf(salesCol);
+            returnsPos = visibleIdx.indexOf(returnsCol);
+          }
+          if (salesPos < 0 || returnsPos < 0) {
+            return {
+              ok: false,
+              yesterday_sales: 0,
+              yesterday_returns: 0,
+              error: `custom sales/return columns hidden: salesPos=${salesPos} returnsPos=${returnsPos}`,
+              header_texts: headerTexts,
+            };
+          }
          let totalSales = 0, totalReturns = 0;
          const rowBreakdown = [];
          for (const tr of dataRows) {
            const cells = Array.from(tr.querySelectorAll('td, th'));
-            const salesCell = cells[salesCol];
-            const returnsCell = cells[returnsCol];
+            const salesCell = cells[salesPos];
+            const returnsCell = cells[returnsPos];
            const head = (tr.innerText || tr.textContent || '').slice(0, 60);
            if (!salesCell || !returnsCell) {
              rowBreakdown.push({
@@ -662,7 +753,7 @@ def _extract_with_text(table_frame: Any) -> tuple[dict[str, Any], str]:
    ERP DOM often contains NBSP (\xa0) etc; Scrapling/Playwright on Windows gbk
    rejects those on return, so we use encodeURIComponent + tryUTF8 to make it ASCII-safe.

-    按 MODE 选 idx 模式 (daily) 或 header 模式 (weekly).
+    按 MODE 选 idx 模式 (weekly) 或 header 模式 (daily / backfill)。
    """
    sales_key, returns_key = _sales_returns_idx()

@@ -673,17 +764,17 @@ def _extract_with_text(table_frame: Any) -> tuple[dict[str, Any], str]:
    else:
        js_args = {"salesHeader": sales_key, "returnsHeader": returns_key, "mode": "header"}

-    # 走 SPU 行 + 自定义列名匹配 + 求和 (因为 sumRow 在 ERP 系统里不含自定义段)
-    if sales_key != "__idx__45":  # backfill 模式
+    # backfill 走 SPU 行 + 自定义列名匹配 + 求和 (因为 sumRow 在 ERP 系统里不含自定义段);
+    # daily 的 昨日销量/昨日退货量 是原生列, sumRow 里就有, 走下面 payload 的
+    # mode='header' 分支 (表头名 -> x 坐标 -> sumRow 最近格), 不经过 _extract_by_header。
+    if MODE == "backfill":
        header_result = _extract_by_header(table_frame, sales_key, returns_key)
        # _extract_by_header 返回 dict, 没有 body_text; wait_for_metrics 需要 erp_code in body_text 校验
        # 用 erp_code placeholder 骗校验 (它会查 erp_code in summary_text, 而 summary_text 不一定含 erp_code)
        # 但新逻辑在 row_breakdown 里已有 erp_code 关联的 SPU 编码 (10384xxx), 校验一定会过
        return header_result, header_result.get("summary_text", "") or "PLACEHOLDER"

-    # 老 idx 模式
-    sales_idx = int(sales_key.removeprefix("__idx__"))
-    returns_idx = int(returns_key.removeprefix("__idx__"))
+    # weekly 老 idx 模式 + daily header 模式共用这段 payload (js_args 已按模式构造)
    payload = table_frame.evaluate(
        """
        (args) => {
@@ -712,6 +803,9 @@ def _extract_with_text(table_frame: Any) -> tuple[dict[str, Any], str]:
          }

          let salesCell = null, returnsCell = null;
+          // salesCol/returnsCol 供末尾 return 的 *_col_idx 引用, 必须在 if/else 外声明
+          // (2026-10-13 daily 切到 header 模式后首测即 ReferenceError 的教训)
+          let salesCol = -1, returnsCol = -1;
          if (args.mode === 'idx') {
            const cells = Array.from(sumRow.children).map((cell) => {
              const box = cell.getBoundingClientRect();
@@ -745,8 +839,8 @@ def _extract_with_text(table_frame: Any) -> tuple[dict[str, Any], str]:
              }
              return -1;
            };
-            const salesCol = findCol(args.salesHeader);
-            const returnsCol = findCol(args.returnsHeader);
+            salesCol = findCol(args.salesHeader);
+            returnsCol = findCol(args.returnsHeader);
            if (salesCol < 0 || returnsCol < 0) {
              return { ok: false, yesterday_sales: 0, yesterday_returns: 0,
                       error: `header not found: salesCol=${salesCol} returnsCol=${returnsCol}`,
@@ -977,6 +1071,17 @@ def wait_for_metrics(page: Any, erp_code: str, timeout_seconds: int = 45,
    # previous_url 可能在 _reload_allsku_iframe 后被重置为 None

    while True:
+        # backfill 模式: 报表刷新会把自定义列的可见性重置回隐藏, 每轮恢复一次;
+        # 首轮额外强制换页重绘, 否则数据行仍按旧列数渲染, 取到的是空单元格(会被读成 0)
+        if MODE == "backfill":
+            redraw = ensure_custom_columns_visible(page, force_redraw=poll_count == 0)
+            if redraw["changed"] or redraw["redrawn"]:
+                print(
+                    f"[ERP] {erp_code}: 自定义列重绘 changed={redraw['changed']} "
+                    f"redrawn={redraw['redrawn']}",
+                    flush=True,
+                )
+                page.wait_for_timeout(1200)
        # 先检测"操作频繁"弹窗 (在主页 DOM, 不在 allsku iframe)
        if filter_frame is not None and freq_dialog_handled < MAX_FREQ_DIALOG:
            if _detect_and_close_freq_dialog(page):
@@ -1176,6 +1281,9 @@ def _run_one_code(page: Any, filter_frame: Any, platform: dict[str, Any], erp_co
    """
    clear_code(filter_frame)
    set_platform_and_code(filter_frame, platform["shops"], erp_code)
+    if MODE == "backfill":
+        # 自定义销量/退货量列被模板重置隐藏时先恢复可见, 否则数据行不渲染对应格
+        ensure_custom_columns_visible(page, force_redraw=True)
    prev_state = _get_allsku_frame_state(page)
    generate_report(filter_frame)
    return wait_for_metrics(
@@ -1437,6 +1545,15 @@ def collect_in_page(page: Any, styles: list[dict[str, Any]], data_dir: Path,
        "reports": len(all_results),
        "valid_codes": valid_codes,
        "has_failures": failed.has_failures(),
+        # 失败明细计数：供 main 判断「个别编码单平台超时」能否降级为警告，
+        # 页面级/款式级失败仍然一律阻断（见 ERP_TOLERATED_FAILED_CODES）。
+        "failed_codes": len(failed.failed_codes),
+        "failed_styles": len(failed.failed_styles),
+        "failed_pages": len(failed.failed_pages),
+        "failed_code_details": [
+            f"{item.get('style')} {item.get('code')}: {item.get('error')}"
+            for item in failed.failed_codes
+        ],
    }


@@ -1574,8 +1691,25 @@ def main() -> int:
        return 1
    summary = state.get("collector_summary") or {}
    if summary.get("has_failures"):
-        print("[FAIL] ERP collector retained failed codes after retry")
-        return 1
+        # 2026-10-29: 页/款级失败仍然阻断；只把「个别编码在单平台重试后仍超时」
+        # 降级为警告。这类编码（如 10244 在抖音查不到数据导致 allsku iframe 不渲染）
+        # 本身贡献 0 销量，款级汇总值已正确，不该让整份日报当天不入库。
+        tolerated = (
+            0 < int(summary.get("failed_codes") or 0) <= ERP_TOLERATED_FAILED_CODES
+            and int(summary.get("failed_styles") or 0) == 0
+            and int(summary.get("failed_pages") or 0) == 0
+        )
+        if tolerated:
+            print(
+                "[WARN] ERP 采集完成，"
+                f"{summary.get('failed_codes')} 个编码单平台重试后仍失败，"
+                "已记 0 并保留失败清单，按部分成功继续日报:"
+            )
+            for detail in summary.get("failed_code_details") or []:
+                print(f"  - {detail}", flush=True)
+        else:
+            print("[FAIL] ERP collector retained failed codes after retry")
+            return 1
    if int(summary.get("reports") or 0) <= 0 or int(summary.get("valid_codes") or 0) <= 0:
        print("[FAIL] ERP collector produced no valid report output")
        return 1
@@ -1423,6 +1423,38 @@ def ensure_logged_in(page, account: str, password: str) -> None:
        raise JDAdCostCollectionError("京东登录态未能进入商家首页")


+_JZT_ONBOARDING_URL_MARKER = "/settled/"
+_JZT_ONBOARDING_TEXT_MARKERS = ("确认联系人手机号", "京准通欢迎您的加入")
+
+
+def _jzt_onboarding_present(page) -> bool:
+    """京准通会把未完成开通引导的会话重定向到 /settled 确认页。"""
+
+    if _JZT_ONBOARDING_URL_MARKER in str(getattr(page, "url", "")):
+        return True
+    body = _page_text(page)
+    return any(marker in body for marker in _JZT_ONBOARDING_TEXT_MARKERS)
+
+
+def ensure_jzt_ready(page) -> None:
+    """Block early when 京准通 still needs the account owner's own confirmation.
+
+    ``/settled/#/information/bonus`` (code=-3012) asks the logged-in merchant to
+    confirm a contact mobile number and accept 《京准通平台相关协议》 before the
+    custom-report routes render.  The script must not accept that on the
+    operator's behalf, so surface it as an actionable blocker instead of the
+    generic "页面出现「下载报表」" timeout.
+    """
+
+    if not _jzt_onboarding_present(page):
+        return
+    raise JDAdCostCollectionError(
+        "京东京准通需要人工完成「确认联系人手机号」开通引导"
+        "（/settled/#/information/bonus，code=-3012）："
+        "请在可见浏览器中由账号本人确认后重跑"
+    )
+
+
 def _request_report(
    page, report_type: ReportType, target_date: date, advertiser_pin: str
 ) -> None:
@@ -1570,7 +1602,13 @@ def run(args: argparse.Namespace) -> list[Path]:
            pass
        ensure_logged_in(page, login_account, PASSWORD)
        safe_goto(page, DOWNLOAD_URL)
-        _wait_for_text(page, "下载报表", timeout_seconds=60)
+        # 京准通对未完成开通引导的会话会重定向到 /settled 确认页。先按正常
+        # 超时等待，超时后再判定是否为引导页拦截，避免刚跳转时的残留文案误报。
+        try:
+            _wait_for_text(page, "下载报表", timeout_seconds=60)
+        except JDAdCostCollectionError:
+            ensure_jzt_ready(page)
+            raise
        _clear_download_center(page)
        reports = {"non_full": "absent", "full": "absent"}
        safe_goto(page, OVERVIEW_URL)
@@ -0,0 +1,9 @@
+{
+  "schema_version": 1,
+  "note": "电商费用按父款式汇总：key 是平台报表里匹配到的子款式，value 是展示口径的父款式。只在对应平台的导入脚本里生效，不影响 dim_style 原始映射。",
+  "jd": {
+    "宙斯3": "宙斯",
+    "宙斯3 Air": "宙斯",
+    "宙斯Z Air": "宙斯"
+  }
+}
@@ -0,0 +1,381 @@
+{
+  "target": {
+    "spreadsheet_token": "LwbBswjfohdoEvtsu1Oc2rI3nac",
+    "sheet_id": "2HAOmi",
+    "sheet_name": "現有庫存",
+    "sku_column": "C",
+    "inventory_column": "D",
+    "product_code_column": "H",
+    "first_data_row": 2,
+    "last_data_row": 52,
+    "expected_sku_count": 51
+  },
+  "source_workbook": "光影行星香港安沛2026Q2計劃書(2026.09.28).xlsx",
+  "source_sheet": "現有庫存",
+  "source_as_of": "2026-09-28",
+  "schedule": {
+    "kind": "daily",
+    "at": "09:00",
+    "timezone": "Asia/Shanghai"
+  },
+  "availability_formula": "qty - order_lock - lock_qty + virtual_qty",
+  "skus": [
+    {
+      "sku": "GY-001BLK",
+      "product_name": "Apollo X1 Shoulder Bag",
+      "ean": "4897080019369",
+      "source_row": 2,
+      "jushuitan_sku_id": "10388030"
+    },
+    {
+      "sku": "GY-001WHT",
+      "product_name": "Apollo X1 Shoulder Bag",
+      "ean": "4897080019376",
+      "source_row": 3,
+      "jushuitan_sku_id": "10388031"
+    },
+    {
+      "sku": "GY-007BLK",
+      "product_name": "Liviana Shoulder Bag",
+      "ean": "4897080019468",
+      "source_row": 4,
+      "jushuitan_sku_id": "10390039"
+    },
+    {
+      "sku": "GY-007WHT",
+      "product_name": "Liviana Shoulder Bag",
+      "ean": "4897080019475",
+      "source_row": 5,
+      "jushuitan_sku_id": "10390040"
+    },
+    {
+      "sku": "GY-008WHT",
+      "product_name": "Starship Satchel",
+      "ean": "4897080019499",
+      "source_row": 6,
+      "jushuitan_sku_id": "10365042"
+    },
+    {
+      "sku": "GY-009KAI",
+      "product_name": "Nebula Crossbody Bag",
+      "ean": "4897080010021",
+      "source_row": 7,
+      "jushuitan_sku_id": "10273005"
+    },
+    {
+      "sku": "GY-011WHT",
+      "product_name": "Zeus Backpack",
+      "ean": "4897080019536",
+      "source_row": 8,
+      "jushuitan_sku_id": "10244056"
+    },
+    {
+      "sku": "GY-012BLK",
+      "product_name": "Zeus II Travel Backpack (Expandable version)",
+      "ean": "4897080019543",
+      "source_row": 9,
+      "jushuitan_sku_id": "10434003"
+    },
+    {
+      "sku": "GY-013BLK",
+      "product_name": "Startrail Crossbody Bag",
+      "ean": "4897080019550",
+      "source_row": 10,
+      "jushuitan_sku_id": "10394065"
+    },
+    {
+      "sku": "GY-013GRN",
+      "product_name": "Startrail Crossbody Bag",
+      "ean": "4897080012308",
+      "source_row": 11,
+      "jushuitan_sku_id": "10394066"
+    },
+    {
+      "sku": "GY-013WHT",
+      "product_name": "Startrail Crossbody Bag",
+      "ean": "4897080019567",
+      "source_row": 12,
+      "jushuitan_sku_id": "10394052"
+    },
+    {
+      "sku": "GY-021BLK",
+      "product_name": "Boundary Backpack",
+      "ean": "4897080012483",
+      "source_row": 13,
+      "jushuitan_sku_id": "10447004"
+    },
+    {
+      "sku": "GY-022BLK",
+      "product_name": "Apollo X2 Shoulder Bag",
+      "ean": "4897080012490",
+      "source_row": 14,
+      "jushuitan_sku_id": "10442009"
+    },
+    {
+      "sku": "GY-022WHT",
+      "product_name": "Apollo X2 Shoulder Bag",
+      "ean": "4897080012506",
+      "source_row": 15,
+      "jushuitan_sku_id": "10442010"
+    },
+    {
+      "sku": "GY-023BLK",
+      "product_name": "Apollo Air Shoulder Bag",
+      "ean": "4897080012513",
+      "source_row": 16,
+      "jushuitan_sku_id": "10456013"
+    },
+    {
+      "sku": "GY-023WHT",
+      "product_name": "Apollo Air Shoulder Bag",
+      "ean": "4897080012995",
+      "source_row": 17,
+      "jushuitan_sku_id": "10456021"
+    },
+    {
+      "sku": "GY-024BLK",
+      "product_name": "Skyreach Air Backpack",
+      "ean": "4897080012520",
+      "source_row": 18,
+      "jushuitan_sku_id": "10454007"
+    },
+    {
+      "sku": "GY-024WHT",
+      "product_name": "Skyreach Air Backpack",
+      "ean": "4897080012537",
+      "source_row": 19,
+      "jushuitan_sku_id": "10454008"
+    },
+    {
+      "sku": "GY-0260RG",
+      "product_name": "Polar Star Backpack",
+      "ean": "4897080010793",
+      "source_row": 20,
+      "jushuitan_sku_id": "10465009"
+    },
+    {
+      "sku": "GY-026BLK",
+      "product_name": "Polar Star Backpack",
+      "ean": "4897080010588",
+      "source_row": 21,
+      "jushuitan_sku_id": "10465003"
+    },
+    {
+      "sku": "GY-026GRN",
+      "product_name": "Polar Star Backpack",
+      "ean": "4897080010786",
+      "source_row": 22,
+      "jushuitan_sku_id": "10465020"
+    },
+    {
+      "sku": "GY-026WHT",
+      "product_name": "Polar Star Backpack",
+      "ean": "4897080010595",
+      "source_row": 23,
+      "jushuitan_sku_id": "10465004"
+    },
+    {
+      "sku": "GY-027BLK",
+      "product_name": "Polar Star tote bag",
+      "ean": "4897080010809",
+      "source_row": 24,
+      "jushuitan_sku_id": "10464008"
+    },
+    {
+      "sku": "GY-027WHT",
+      "product_name": "Polar Star tote bag",
+      "ean": "4897080010571",
+      "source_row": 25,
+      "jushuitan_sku_id": "10464004"
+    },
+    {
+      "sku": "GY-028BLK",
+      "product_name": "Poseidon tote bag (cordura)",
+      "ean": "4897080010779",
+      "source_row": 26,
+      "jushuitan_sku_id": "10472004"
+    },
+    {
+      "sku": "GY-029BLK",
+      "product_name": "Nebula miniCrossbody Bag",
+      "ean": "4897080010762",
+      "source_row": 27,
+      "jushuitan_sku_id": "10475015"
+    },
+    {
+      "sku": "GY-029KAI",
+      "product_name": "Nebula miniCrossbody Bag",
+      "ean": "4897080010748",
+      "source_row": 28,
+      "jushuitan_sku_id": "10475016"
+    },
+    {
+      "sku": "GY-029WHT",
+      "product_name": "Nebula miniCrossbody Bag",
+      "ean": "4897080010755",
+      "source_row": 29,
+      "jushuitan_sku_id": "10475014"
+    },
+    {
+      "sku": "GY-030BLK",
+      "product_name": "Morning Star miniCrossbody Bag",
+      "ean": "4897080013008",
+      "source_row": 30,
+      "jushuitan_sku_id": "10468007"
+    },
+    {
+      "sku": "GY-030WHT",
+      "product_name": "Morning Star miniCrossbody Bag",
+      "ean": "4897080010731",
+      "source_row": 31,
+      "jushuitan_sku_id": "10468003"
+    },
+    {
+      "sku": "GY-031BLK",
+      "product_name": "Star-by-Star EVOCrossbody Bag",
+      "ean": "4897080010724",
+      "source_row": 32,
+      "jushuitan_sku_id": "10474004"
+    },
+    {
+      "sku": "GY-032BLK",
+      "product_name": "GlacierCrossbody Bag",
+      "ean": "4897080010717",
+      "source_row": 33,
+      "jushuitan_sku_id": "10453005"
+    },
+    {
+      "sku": "GY-033BLK",
+      "product_name": "Morning Star 2 Crossbody Bag",
+      "ean": "4897080012896",
+      "source_row": 34,
+      "jushuitan_sku_id": "10469003"
+    },
+    {
+      "sku": "GY-033WHT",
+      "product_name": "Morning Star 2 Crossbody Bag",
+      "ean": "4897080012902",
+      "source_row": 35,
+      "jushuitan_sku_id": "10469004"
+    },
+    {
+      "sku": "GY-034BLK",
+      "product_name": "Startrail 2 Crossbody Bag",
+      "ean": "4897080012919",
+      "source_row": 36,
+      "jushuitan_sku_id": "10504027"
+    },
+    {
+      "sku": "GY-034GRN",
+      "product_name": "Startrail 2 Crossbody Bag",
+      "ean": "4897080012926",
+      "source_row": 37,
+      "jushuitan_sku_id": "10504028"
+    },
+    {
+      "sku": "GY-034WHT",
+      "product_name": "Startrail 2 Crossbody Bag",
+      "ean": "4897080012933",
+      "source_row": 38,
+      "jushuitan_sku_id": "10504029"
+    },
+    {
+      "sku": "GY-035BLK",
+      "product_name": "Zeus 3 air Backpack",
+      "ean": "4897080012940",
+      "source_row": 39,
+      "jushuitan_sku_id": "10482004"
+    },
+    {
+      "sku": "GY-035WHT",
+      "product_name": "Zeus 3 air Backpack",
+      "ean": "4897080012957",
+      "source_row": 40,
+      "jushuitan_sku_id": "10482003"
+    },
+    {
+      "sku": "GY-036BLK",
+      "product_name": "Cirrus 2 Backpack",
+      "ean": "4897080012964",
+      "source_row": 41,
+      "jushuitan_sku_id": "10481004"
+    },
+    {
+      "sku": "GY-036GRN",
+      "product_name": "Cirrus 2 Backpack",
+      "ean": "4897080012988",
+      "source_row": 42,
+      "jushuitan_sku_id": "10481006"
+    },
+    {
+      "sku": "GY-036WHT",
+      "product_name": "Cirrus 2 Backpack",
+      "ean": "4897080012971",
+      "source_row": 43,
+      "jushuitan_sku_id": "10481005"
+    },
+    {
+      "sku": "GY-037BLK",
+      "product_name": "Polar Star 2",
+      "ean": "4897080013046",
+      "source_row": 44,
+      "jushuitan_sku_id": "10520009"
+    },
+    {
+      "sku": "GY-037GRY",
+      "product_name": "Polar Star 2",
+      "ean": "4897080013039",
+      "source_row": 45,
+      "jushuitan_sku_id": "10520014"
+    },
+    {
+      "sku": "GY-038WHT",
+      "product_name": "Polar Star Shoulder Bag",
+      "ean": "4897080013060",
+      "source_row": 46,
+      "jushuitan_sku_id": "10519010"
+    },
+    {
+      "sku": "GY-038BLK",
+      "product_name": "Polar Star Shoulder Bag",
+      "ean": "4897080013053",
+      "source_row": 47,
+      "jushuitan_sku_id": "10519009"
+    },
+    {
+      "sku": "GY-039BLK",
+      "product_name": "Poseidon Laptop Bag",
+      "ean": "4897080013084",
+      "source_row": 48,
+      "jushuitan_sku_id": "10515020"
+    },
+    {
+      "sku": "GY-039GRY",
+      "product_name": "Poseidon Laptop Bag",
+      "ean": "4897080013077",
+      "source_row": 49,
+      "jushuitan_sku_id": "10515019"
+    },
+    {
+      "sku": "GY-040BLK",
+      "product_name": "Skyreach 3",
+      "ean": "4897080013091",
+      "source_row": 50,
+      "jushuitan_sku_id": "10510019"
+    },
+    {
+      "sku": "GY-040WHT",
+      "product_name": "Skyreach 3",
+      "ean": "4897080013107",
+      "source_row": 51,
+      "jushuitan_sku_id": "10510020"
+    },
+    {
+      "sku": "GY-041BLK",
+      "product_name": "Phantom",
+      "ean": "4897080013114",
+      "source_row": 52,
+      "jushuitan_sku_id": "10507015"
+    }
+  ]
+}
@@ -32,7 +32,21 @@ except ImportError:  # Direct script execution from the module root.
    from runtime_paths import NORMALIZED_DATA_ROOT, STATE_ROOT, managed_data_path


-CACHE_SCHEMA_VERSION = 5
+CACHE_SCHEMA_VERSION = 9  # 9: 款式主键统一, 每条 style 增加 style_id / style_match_method
+
+# 系列级飞书地址：约定只维护在锚点行，子款误填也会被整系列聚合去重后统一下发。
+# 顺序即写入目标表的优先级说明顺序，不含 bitable_sku_master（人工维护，不参与采集）。
+SERIES_TARGET_FIELDS = (
+    "sales_bitable",
+    "main_image_bitable",
+    "creator_bitable",
+    "self_creator_bitable",
+    "weekly_note_analysis",
+    "persona_bitable",
+    "jd_persona_bitable",
+    "dy_persona_bitable",
+    "style_analysis_bitable",
+)

 # 表格字段名（按实际表格设置，支持别名）
 FIELD_ALIASES = {
@@ -111,6 +125,10 @@ class StyleConfigLoader:
        del base_token, table_id, view_id
        self.identity = identity
        self.records_provider = records_provider or runtime_workflow_configs
+        # 注入 records_provider 的调用方（单测 / 离线脚本）用的是假数据，绝不能
+        # 写进 STATE_ROOT 的生产快照：那份快照只在数据库读取失败时被当作真配置
+        # 使用，用它采集会按假款式去取数。因此注入 provider 时整套缓存读写都关掉。
+        self.cache_enabled = records_provider is None

        project_root = Path(__file__).resolve().parent.parent
        self.cache_path = (
@@ -229,6 +247,24 @@ class StyleConfigLoader:
        parsed = self._parse_bitable_url(url)
        return bool(parsed["base_token"] and parsed["table_id"])

+    def _has_creator_bitable(self, record: dict) -> bool:
+        """合作达人表 Base/Table 同时存在。"""
+        url = self._parse_markdown_link(record.get("creator_bitable_url"))
+        parsed = self._parse_bitable_url(url)
+        return bool(parsed["base_token"] and parsed["table_id"])
+
+    def _has_self_creator_bitable(self, record: dict) -> bool:
+        """自营合作达人表 Base/Table 同时存在。"""
+        url = self._parse_markdown_link(record.get("self_creator_bitable_url"))
+        parsed = self._parse_bitable_url(url)
+        return bool(parsed["base_token"] and parsed["table_id"])
+
+    def _has_weekly_note_analysis(self, record: dict) -> bool:
+        """每周/每月笔记分析表 Base/Table 同时存在。"""
+        url = self._parse_markdown_link(record.get("weekly_note_analysis_url"))
+        parsed = self._parse_bitable_url(url)
+        return bool(parsed["base_token"] and parsed["table_id"])
+
    def _has_daily_required_fields(self, record: dict) -> bool:
        """每日采集四条件：erp_codes、style、item_ids、sales_bitable。"""
        style_name = str(record.get("style") or "").strip()
@@ -259,6 +295,9 @@ class StyleConfigLoader:
                    "erp_codes": row.get("erp_codes", []),
                    "sales_bitable_url": row.get("sales_bitable_url", ""),
                    "main_image_bitable_url": row.get("main_image_bitable_url", ""),
+                    "creator_bitable_url": row.get("creator_bitable_url", ""),
+                    "self_creator_bitable_url": row.get("self_creator_bitable_url", ""),
+                    "weekly_note_analysis_url": row.get("weekly_note_analysis_url", ""),
                    "persona_bitable_url": row.get("tm_persona_bitable_url", ""),
                    "dy_persona_bitable_url": row.get("dy_persona_bitable_url", ""),
                    "jd_persona_bitable_url": row.get("jd_persona_bitable_url", ""),
@@ -266,6 +305,16 @@ class StyleConfigLoader:
                        "style_analysis_bitable_url", ""
                    ),
                    "destinations": row.get("destinations", []),
+                    "series_name": row.get("series_name", ""),
+                    "is_series_anchor": bool(row.get("is_series_anchor", False)),
+                    # 款式主键统一 (2026-11-01): dim_style.id 正式下发, 写入脚本
+                    # 直接用它落库, style_name 只作显示冗余。
+                    "style_id": row.get("style_id"),
+                    "style_match_method": row.get("style_match_method", ""),
+                    # 采集开关随记录下发；缺字段（旧 cmt_styles 回退）按启用处理
+                    "collect_enabled": True
+                    if row.get("collect_enabled") is None
+                    else bool(row.get("collect_enabled")),
                }
            )
        return records
@@ -302,6 +351,8 @@ class StyleConfigLoader:
        def _empty(style_name: str) -> dict:
            return {
                "name": style_name,
+                "style_id": None,
+                "style_match_method": "",
                "brand": "",
                "erp_codes": [],
                "tm": {"item_ids": []},
@@ -310,11 +361,18 @@ class StyleConfigLoader:
                "dy": {"item_ids": []},
                "sales_bitable": {},
                "main_image_bitable": {},
+                "creator_bitable": {},
+                "self_creator_bitable": {},
+                "weekly_note_analysis": {},
                "persona_bitable": {},
                "dy_persona_bitable": {},
                "jd_persona_bitable": {},
                "style_analysis_bitable": {},
                "destinations": [],
+                "series_name": "",
+                "is_series_anchor": False,
+                "collect_enabled": True,
+                "series": {},
            }

        # ── 第一遍:款式级字段 ──────────────────────────────────────
@@ -327,11 +385,30 @@ class StyleConfigLoader:

            style = styles.setdefault(style_name, _empty(style_name))

+            # 款式主键: dim_style.id, 与 data-hub 的 22 张表同一套主键。
+            # merge 已把同款展开成多条平台记录, 取值一致, 取第一个非空即可。
+            if style.get("style_id") is None and record.get("style_id") is not None:
+                style["style_id"] = int(record["style_id"])
+            if not style.get("style_match_method") and record.get("style_match_method"):
+                style["style_match_method"] = str(record["style_match_method"])
+
            # 品牌（款式级别,取第一个非空）
            brand = str(record.get("brand") or "").strip()
            if brand and not style["brand"]:
                style["brand"] = brand

+            # 系列归属（产品生命进程建档维护）：同系列可共享平台编码与销量表
+            series_name = str(record.get("series_name") or "").strip()
+            if series_name and not style["series_name"]:
+                style["series_name"] = series_name
+            if record.get("is_series_anchor"):
+                style["is_series_anchor"] = True
+
+            # 采集开关（产品生命进程在 dim_style.collect_enabled 上按系列维护）：
+            # merge 已把同一款展开成多条平台记录且取值一致；任一条显式 False 即关闭。
+            if record.get("collect_enabled") is False:
+                style["collect_enabled"] = False
+
            destinations = record.get("destinations")
            if isinstance(destinations, list) and destinations and not style["destinations"]:
                style["destinations"] = [
@@ -352,41 +429,38 @@ class StyleConfigLoader:
                            f"{style['erp_codes']} vs {erp_codes}; 保留第一条天猫编码"
                        )

-            # 销量飞书表地址（款式级别,只要 URL 完整就收）
+            # 飞书地址：2026-10-29 起统一由 dim_style 系列锚点持有，并在
+            # workflow_config 里下发到该款的每条平台记录，所以任意分组都能
+            # 读到同样的值，不再像以前那样限定只能从天猫分组取。
            if self._has_sales_bitable(record) and not style["sales_bitable"]:
                sales_url = self._parse_markdown_link(record.get("sales_bitable_url"))
                if sales_url:
                    style["sales_bitable"] = self._parse_bitable_url(sales_url)

-            # 主图表地址：只读天猫分组记录,只要求主图 URL 非空
-            if platform == "tm" and self._has_main_image_bitable(record):
+            if self._has_main_image_bitable(record) and not style["main_image_bitable"]:
                main_url = self._parse_markdown_link(record.get("main_image_bitable_url"))
-                if main_url and not style["main_image_bitable"]:
+                if main_url:
                    style["main_image_bitable"] = self._parse_bitable_url(main_url)
                    style["main_image_bitable"]["platform"] = "天猫"

-            # 人群画像（天猫）表地址：只读天猫分组记录
-            if platform == "tm" and self._has_persona_bitable(record):
-                persona_url = self._parse_markdown_link(record.get("persona_bitable_url"))
-                if persona_url and not style["persona_bitable"]:
-                    style["persona_bitable"] = self._parse_bitable_url(persona_url)
-
-            # 人群画像（抖音）表地址：款式级别，任意平台记录有值就收
-            if self._has_dy_persona_bitable(record):
-                dy_persona_url = self._parse_markdown_link(record.get("dy_persona_bitable_url"))
-                if dy_persona_url and not style["dy_persona_bitable"]:
-                    style["dy_persona_bitable"] = self._parse_bitable_url(dy_persona_url)
-
-            # 人群画像（京东）表地址：款式级别，任意平台记录有值就收
-            if self._has_jd_persona_bitable(record):
-                jd_persona_url = self._parse_markdown_link(record.get("jd_persona_bitable_url"))
-                if jd_persona_url and not style["jd_persona_bitable"]:
-                    style["jd_persona_bitable"] = self._parse_bitable_url(jd_persona_url)
-
-            if self._has_style_analysis_bitable(record) and not style["style_analysis_bitable"]:
-                analysis_url = self._parse_markdown_link(record.get("style_analysis_bitable_url"))
-                if analysis_url:
-                    style["style_analysis_bitable"] = self._parse_bitable_url(analysis_url)
+            for key, record_key, checker in (
+                ("creator_bitable", "creator_bitable_url", self._has_creator_bitable),
+                (
+                    "self_creator_bitable",
+                    "self_creator_bitable_url",
+                    self._has_self_creator_bitable,
+                ),
+                ("weekly_note_analysis", "weekly_note_analysis_url", self._has_weekly_note_analysis),
+                ("persona_bitable", "persona_bitable_url", self._has_persona_bitable),
+                ("dy_persona_bitable", "dy_persona_bitable_url", self._has_dy_persona_bitable),
+                ("jd_persona_bitable", "jd_persona_bitable_url", self._has_jd_persona_bitable),
+                ("style_analysis_bitable", "style_analysis_bitable_url", self._has_style_analysis_bitable),
+            ):
+                if style[key] or not checker(record):
+                    continue
+                url = self._parse_markdown_link(record.get(record_key))
+                if url:
+                    style[key] = self._parse_bitable_url(url)

        # 非天猫分组的 ERP 编码不会参与聚合；只在冲突或天猫缺失时提示，
        # 避免重复填写相同编码造成无意义日志。
@@ -427,6 +501,9 @@ class StyleConfigLoader:
            elif platform == "jd_self" and ids:
                style["jd_self"]["skus"] = ids

+        # ── 第三遍: 系列内共享（平台编码 / 销量表只要求主款填写）──────────
+        self._apply_series_sharing(styles)
+
        return {
            "styles": styles,
            "source": "database",
@@ -434,6 +511,153 @@ class StyleConfigLoader:
            "schema_version": CACHE_SCHEMA_VERSION,
        }

+    @staticmethod
+    def _apply_series_sharing(styles: dict[str, dict]) -> None:
+        """把整条系列的平台商品编码、ERP 编码与飞书地址聚合去重到锚点。
+
+        录入约定是「锚点建档、子款留空」，但子款误填无法从表结构上禁止，所以
+        这里不依赖人工守规矩，统一做聚合兜底：
+
+        - 平台商品编码：整系列并集去重后挂锚点，由锚点一个采集单元去抓；
+          子款保留自己的原始编码（series.platform_ids 里可见）供诊断，采集口
+          按归属判定让出，避免同一商品 ID 被采两次、访客与加购重复入账。
+        - ERP 编码：整系列并集去重后挂锚点（series.erp_owner），子款保留自己
+          的原始编码（own_erp_codes）供按款查询。**子款自身 collect_enabled=false
+          时不并入**：锚点是系列总开关，子款开关只决定自己的编码是否参与聚合。
+        - 飞书地址：逐字段聚合并去重，锚点优先、其余按款名，取到的那套表下发
+          全系列。一个表位只能落一张表，出现多个不同表时告警并保留优先级最高
+          的一条。
+
+        聚合的目标是「子款误填也不丢、不重复入账、同系列只有一套表」，而不是
+        让子款变成独立采集单元。
+        """
+
+        by_series: dict[str, list[str]] = {}
+        for name, style in styles.items():
+            series = str(style.get("series_name") or "").strip()
+            if series:
+                by_series.setdefault(series, []).append(name)
+
+        for series, members in by_series.items():
+            # 锚点优先，其余按名称稳定排序，保证每次运行归属一致
+            members.sort(key=lambda n: (not styles[n].get("is_series_anchor"), n))
+            anchor = members[0]
+            erp_owner = anchor
+            multi = len(members) > 1
+            platform_owner: dict[str, str] = {}
+            platform_ids: dict[str, dict[str, list[str]]] = {}
+            merged_platform: dict[str, list[str]] = {}
+            for platform, ids_key in PRODUCT_ID_FIELDS.items():
+                holders: dict[str, list[str]] = {}
+                merged: list[str] = []
+                for member in members:
+                    ids = [
+                        str(item).strip()
+                        for item in styles[member].get(platform, {}).get(ids_key) or ()
+                        if str(item).strip()
+                    ]
+                    if not ids:
+                        continue
+                    holders[member] = ids
+                    platform_owner.setdefault(platform, member)
+                    for item in ids:
+                        if item not in merged:
+                            merged.append(item)
+                if not holders:
+                    continue
+                merged_platform[platform] = merged
+                if multi:
+                    # 子款写的编码也并入锚点：整系列先去重聚合，锚点作为唯一
+                    # 采集单元持有全部商品 ID。holders 必须重建以保证锚点排在
+                    # 第一位——_platform_collection_owner 依赖插入顺序判断
+                    # 「谁的编码覆盖了谁」。
+                    ordered_holders: dict[str, list[str]] = {anchor: list(merged)}
+                    for member in members:
+                        if member != anchor and member in holders:
+                            ordered_holders[member] = holders[member]
+                    holders = ordered_holders
+                platform_ids[platform] = holders
+            merged_erp: list[str] = []
+            erp_excluded: list[str] = []
+            for member in members:
+                # 锚点是系列总开关（见下面 series_collect）；子款自身的 collect_enabled
+                # 决定它的 ERP 编码是否并入系列。子款关掉采集时，它的编码既不由自己
+                # 去聚水潭取数，也不允许锚点代采——直接排除在聚合之外。
+                if member != anchor and not bool(
+                    styles[member].get("collect_enabled", True)
+                ):
+                    erp_excluded.append(member)
+                    continue
+                for code in styles[member].get("erp_codes") or ():
+                    if code not in merged_erp:
+                        merged_erp.append(code)
+            if multi and erp_excluded and bool(styles[anchor].get("collect_enabled", True)):
+                print(
+                    f"[INFO] 系列「{series}」子款 {', '.join(erp_excluded)} "
+                    f"已关闭采集，ERP 编码不纳入「{erp_owner}」聚合"
+                )
+            # 系列级采集开关：以锚点为准（members[0] 已按「锚点优先、其次款名」
+            # 排序，无锚点时即款名第一个）。产品生命进程只改锚点那一行即可控制
+            # 整条系列，其余子款跟随。
+            series_collect = bool(styles[anchor].get("collect_enabled", True))
+            # 系列级飞书地址：整系列逐字段聚合去重，锚点优先、其余按款名。子款
+            # 误填的地址同样参与聚合而不会被丢弃；一套表下发全系列，写表口径
+            # 统一。同字段出现多个不同表时保留优先级最高的一条并告警。
+            series_targets: dict[str, dict] = {}
+            for field in SERIES_TARGET_FIELDS:
+                seen: set[tuple] = set()
+                candidates: list[dict] = []
+                for member in members:
+                    target = styles[member].get(field) or {}
+                    if not target.get("base_token"):
+                        continue
+                    key = (
+                        target.get("base_token"),
+                        target.get("table_id"),
+                        target.get("view_id"),
+                    )
+                    if key in seen:
+                        continue
+                    seen.add(key)
+                    candidates.append(dict(target))
+                if not candidates:
+                    continue
+                if len(candidates) > 1:
+                    print(
+                        f"[WARN] 系列「{series}」{field} 有 {len(candidates)} 个不同表，"
+                        f"按「{anchor}」下发并忽略其余"
+                    )
+                series_targets[field] = candidates[0]
+
+            for member in members:
+                style = styles[member]
+                style["series"] = {
+                    "name": series,
+                    "is_anchor": bool(style.get("is_series_anchor")),
+                    "collect_enabled": series_collect,
+                    "platform_owner": dict(platform_owner),
+                }
+                if multi:
+                    # 单款即可自成系列，无需归属标注；多款才需要指明 ERP 由谁承接
+                    style["series"]["erp_owner"] = erp_owner
+                    style["series"]["own_erp_codes"] = list(style.get("erp_codes") or ())
+                    # 自身 collect_enabled=false 而被排除出 ERP 聚合的子款，供诊断
+                    style["series"]["erp_excluded"] = list(erp_excluded)
+                    style["series"]["platform_ids"] = {
+                        platform: {holder: list(ids) for holder, ids in holders.items()}
+                        for platform, holders in platform_ids.items()
+                    }
+                for field, target in series_targets.items():
+                    style[field] = dict(target)
+            if multi:
+                # 锚点持有整条系列的平台商品编码与 ERP 编码
+                for platform, ids_key in PRODUCT_ID_FIELDS.items():
+                    merged = merged_platform.get(platform)
+                    if merged:
+                        styles[anchor][platform][ids_key] = list(merged)
+                if merged_erp:
+                    styles[erp_owner]["erp_codes"] = merged_erp
+
    # ------------------------------------------------------------------
    # 缓存与回退
    # ------------------------------------------------------------------
@@ -478,13 +702,14 @@ class StyleConfigLoader:
        try:
            records = self._load_records_from_database()
            payload = self._aggregate(records)
-            self._write_cache(payload)
+            if self.cache_enabled:
+                self._write_cache(payload)
            if sync_fallback:
                self._write_style_map_fallback(payload)
            return payload
        except Exception as exc:
            print(f"[WARN] 云端动态配置读取失败: {exc}，尝试数据库快照缓存")
-            cached = self._read_cache()
+            cached = self._read_cache() if self.cache_enabled else None
            if cached and cached.get("source") == "database":
                return cached
            raise StyleConfigError("云端动态配置不可用且没有数据库快照缓存") from exc
@@ -656,10 +881,11 @@ class StyleConfigLoader:
        ]

    def get_erp_styles_input(self) -> list[dict]:
+        """ERP 编码清单：只要求款式编码，平台编码/销量表由系列主款承担。"""
        return [
            {"style_name": name, "erp_style_codes": cfg["erp_codes"]}
            for name, cfg in self.load()["styles"].items()
-            if self._has_complete_identity(cfg)
+            if cfg.get("erp_codes")
        ]

    def get_bitable_style_map(self) -> dict:
@@ -712,47 +938,89 @@ class StyleConfigLoader:
    # ------------------------------------------------------------------
    # 启动期「待采 / 跳过」门 — 显式返回带 reason 的跳过清单
    # ------------------------------------------------------------------
-    def _cfg_has_sales_bitable(self, cfg: dict) -> bool:
-        return bool(cfg.get("sales_bitable", {}).get("base_token")) and bool(
-            cfg.get("sales_bitable", {}).get("table_id")
-        )
-
-    @staticmethod
-    def _has_any_product_id(cfg: dict) -> bool:
-        return any(
-            bool(cfg.get(platform, {}).get(ids_key))
-            for platform, ids_key in PRODUCT_ID_FIELDS.items()
-        )
-
    def _has_complete_identity(self, cfg: dict, platform: str | None = None) -> bool:
        """商品采集统一门：ERP 款式编码 + 至少一个平台商品编码。"""
        return has_complete_style_identity(cfg, platform)

-    def _identity_missing(self, cfg: dict) -> list[str]:
-        missing: list[str] = []
-        if not cfg.get("erp_codes"):
-            missing.append("ERP款式编码")
-        if not self._has_any_product_id(cfg):
-            missing.append("平台商品编码")
-        return missing
+    def _erp_owner_for_collection(self, name: str, cfg: dict, styles: dict) -> str:
+        """本款是否需要把 ERP 采集让给同系列锚点。
+
+        返回空串表示本款自己采；返回款名表示编码已由该款归集，本款不重复采。
+        锚点自身被关闭采集时不再承接（返回空串）——但系列级总开关此时已让整条
+        系列停采，这里只是避免把编码归属标注到一个根本不采的款上。
+        """
+
+        series = cfg.get("series") or {}
+        owner = str(series.get("erp_owner") or "").strip()
+        if not owner or owner == name:
+            return ""
+        if not self._is_collect_enabled(styles.get(owner) or {}):
+            return ""
+        return owner
+
+    # ------------------------------------------------------------------
+    # 采集开关 (dim_style.collect_enabled)
+    # ------------------------------------------------------------------
+    @staticmethod
+    def _is_collect_enabled(cfg: dict) -> bool:
+        """本款是否整体纳入采集（系列级总开关）。
+
+        ``dim_style.collect_enabled`` 是两层语义：
+
+        - **系列级总开关**：由锚点那一行决定，``_apply_series_sharing`` 已把锚点的
+          值摊到 ``series.collect_enabled``，所以同系列任意一款得到相同判定。
+          锚点 false → 整条系列一款都不采。产品侧只需改锚点那一行。
+        - **子款级 ERP 参与权**：锚点为 true 时，子款自身的 ``collect_enabled``
+          只决定「它的 ERP 款式编码是否并入系列聚合」（见 ``_apply_series_sharing``
+          的 ``erp_excluded``）；子款 false 不参与聚合，但不会让其他子款停采。
+
+        没有 series_name 的款直接用自身取值；字段缺失时默认采集。
+        """
+        series = cfg.get("series") or {}
+        if "collect_enabled" in series:
+            return bool(series["collect_enabled"])
+        return bool(cfg.get("collect_enabled", True))

    def get_daily_styles(self) -> dict:
-        """每日商品采集门：ERP 编码、销量目标和平台商品编码必须完整。
+        """每日商品采集门：ERP 编码与平台商品编码必须完整。
+
+        销量飞书表地址只决定结果能否写表，不再作为采集前置条件——缺表地址
+        的款式照常采集并落库，写表环节自己按地址缺失跳过。

        返回 {tm/jd/dy: {collectable: [...], skipped: [{style, reason}]}}。
-        incomplete style 永远不会进入 collectable，并记录具体缺失字段。
+        采集开关取 ``dim_style.collect_enabled``：锚点是系列总开关（false 则整系列
+        不采），子款自身的开关只决定它的 ERP 编码是否参与系列聚合；缺字段默认采集。
+
+        ERP 与平台编码按两个维度分开判定：
+        - ERP：整条系列聚合到锚点（series.erp_owner）由一个单元去聚水潭取数；
+          子款即使让出 ERP，只要自己持有某平台的商品 ID，仍会继续采该平台。
+        - 平台：整系列只要有一款填了就算齐，谁填谁采；本款没填时指向承担款，
+          本款编码被更高优先级的款式完整覆盖时让出，避免重复入账。
        """
        payload = self.load()
+        styles = payload["styles"]
        result = {p: {"collectable": [], "skipped": []} for p in ("tm", "jd", "dy")}
-        for name, cfg in payload["styles"].items():
-            identity_missing = self._identity_missing(cfg)
+        for name, cfg in styles.items():
+            if not self._is_collect_enabled(cfg):
+                for plat in ("tm", "jd", "dy"):
+                    result[plat]["skipped"].append(
+                        {"style": name, "reason": "已标记不采集"}
+                    )
+                continue
+            # 自有编码，或同系列锚点已在采（本款不必重复去聚水潭）
+            has_erp = bool(cfg.get("erp_codes")) or bool(
+                self._erp_owner_for_collection(name, cfg, styles)
+            )
            for plat in ("tm", "jd", "dy"):
-                ids_key = "spus" if plat == "jd" else "item_ids"
-                missing = list(identity_missing)
-                if not self._cfg_has_sales_bitable(cfg):
-                    missing.append("销量目标地址")
-                if not cfg.get(plat, {}).get(ids_key):
-                    missing.append(f"{plat}商品编码")
+                owner = self._platform_collection_owner(name, cfg, plat, styles)
+                missing: list[str] = []
+                if not has_erp:
+                    missing.append("ERP款式编码")
+                if owner != name:
+                    if owner:
+                        missing.append(f"{plat}商品编码（同系列「{owner}」已在采）")
+                    else:
+                        missing.append(f"{plat}商品编码")
                if missing:
                    result[plat]["skipped"].append(
                        {"style": name, "reason": "缺 " + "/".join(dict.fromkeys(missing))}
@@ -761,27 +1029,82 @@ class StyleConfigLoader:
                    result[plat]["collectable"].append(name)
        return result

+    def _platform_collection_owner(
+        self, name: str, cfg: dict, platform: str, styles: dict
+    ) -> str:
+        """本款是否承担该平台的流量采集；空串表示整条系列都没填编码。
+
+        平台商品编码整系列只要有一款填了就算齐：谁填谁采。本款的编码若被
+        同系列优先级更高的款式完整覆盖（锚点常见做法是把全系列商品 ID 都
+        挂在锚点上），本款让出，避免同一个商品 ID 被采两次、访客与加购
+        重复入账。本款的编码与别人的编码互不包含时各自采各自的商品。
+        """
+
+        ids_key = PRODUCT_ID_FIELDS[platform]
+        own_ids = {
+            str(item).strip()
+            for item in cfg.get(platform, {}).get(ids_key) or ()
+            if str(item).strip()
+        }
+        series = cfg.get("series") or {}
+        holders: dict[str, list[str]] = (
+            (series.get("platform_ids") or {}).get(platform) or {}
+        )
+        if own_ids:
+            # holders 的插入顺序即系列内的优先级（锚点优先，其次按款名）
+            for member in holders:
+                if member == name:
+                    return name
+                if not self._is_collect_enabled(styles.get(member) or {}):
+                    continue
+                covered = {
+                    str(item).strip() for item in holders.get(member) or () if str(item).strip()
+                }
+                if own_ids <= covered:
+                    return member
+            return name
+        # 本款没填：由系列内已填且启用的款式承担（锚点优先）
+        for member in holders:
+            if not self._is_collect_enabled(styles.get(member) or {}):
+                continue
+            return member
+        return ""
+
    def get_erp_daily_styles(self) -> dict:
-        """ERP daily 门：ERP 编码、至少一个平台商品编码和销量目标都要有。
+        """ERP daily 门：只要求 ERP 款式编码。
+
+        销量飞书表地址只决定结果能否写表，不再作为采集前置条件——缺表地址
+        的款式照常去聚水潭取数并落库，写表环节自己按地址缺失跳过。

        返回 {collectable, skipped, brand_map: {style: brand}}。
-        ERP 一次跑覆盖 jd/dy/tm 三个平台，平台商品编码只用于确认款式已完成商品身份配置。
+        ERP 销量按款式编码逐款查；同系列多款时编码已由锚点聚合并集
+        （series.erp_owner），子款让出采集，避免同一编码被采两次。
        brand_map 给 collect_erp_yesterday_metrics.py 按品牌选店铺
        (光影行星旗下店铺 vs ozko旗舰店)。
+        采集开关取 dim_style.collect_enabled：锚点为 false 时整系列不采；锚点为
+        true 时，自身 collect_enabled=false 的子款其 ERP 编码不并入锚点聚合。
+        缺字段默认采集。
        """
        payload = self.load()
+        styles = payload["styles"]
        collectable: list[str] = []
        skipped: list[dict] = []
        brand_map: dict[str, str] = {}
-        for name, cfg in payload["styles"].items():
-            missing = self._identity_missing(cfg)
-            if not self._cfg_has_sales_bitable(cfg):
-                missing.append("销量目标地址")
-            if not missing:
-                collectable.append(name)
-                brand_map[name] = cfg.get("brand") or "光影行星"
-            else:
-                skipped.append({"style": name, "reason": "缺 " + "/".join(dict.fromkeys(missing))})
+        for name, cfg in styles.items():
+            if not self._is_collect_enabled(cfg):
+                skipped.append({"style": name, "reason": "已标记不采集"})
+                continue
+            erp_owner = self._erp_owner_for_collection(name, cfg, styles)
+            if erp_owner:
+                skipped.append(
+                    {"style": name, "reason": f"ERP 已由同系列「{erp_owner}」归集采集"}
+                )
+                continue
+            if not cfg.get("erp_codes"):
+                skipped.append({"style": name, "reason": "缺 ERP款式编码"})
+                continue
+            collectable.append(name)
+            brand_map[name] = cfg.get("brand") or "光影行星"
        return {"collectable": collectable, "skipped": skipped, "brand_map": brand_map}

    def get_jd_self_styles(self) -> list[str]:
@@ -793,14 +1116,18 @@ class StyleConfigLoader:
        ]

    def get_main_image_styles(self) -> list[str]:
-        """天猫主图门：完整身份、天猫商品编码和主图目标地址同时存在。"""
+        """天猫主图采集目标：身份齐全（ERP 款式编码 + 天猫商品编码）即可。
+
+        主图飞书表地址只决定结果能否写表，不再作为采集前置条件——缺地址的
+        款式照常抓图并落库，写表环节按地址缺失跳过（见 get_bitable_main_image_map），
+        与商品经营日报「缺销量表不挡采集」同一口径。
+        """
        payload = self.load()
-        result: list[str] = []
-        for name, cfg in payload["styles"].items():
-            main = cfg.get("main_image_bitable") or {}
-            if self._has_complete_identity(cfg, "tm") and main.get("base_token") and main.get("table_id"):
-                result.append(name)
-        return result
+        return [
+            name
+            for name, cfg in payload["styles"].items()
+            if self._has_complete_identity(cfg, "tm")
+        ]


 # ------------------------------------------------------------------------------
@@ -4,29 +4,63 @@
 -- 设计原则: schema 与 data/ 目录下的 JSON 产物 1:1 映射；upsert 幂等
 -- ============================================================================

-- 款式维度表：维护每款的 (各平台商品 ID + 京东自营 SKU + ERP 编码 + 飞书表地址)
-- 身份主来源：产品生命进程 cmt_styles.platform_product_ids / erp_style_codes，
-- 由 workflow_config.runtime_records() 统一投影；运行时不再从各采集器、JSON 产物反向拼接身份。
+-- 款式维度表：维护每款的 (各平台商品 ID + 京东自营 SKU + ERP 款式编码 + 飞书表地址)
+-- 身份主来源：本表自身（2026-10-27 起由产品生命进程 cmt_styles 切换过来）。
+--   * ERP 编码用 erp_style_codes (JSONB 数组，可多编码)，旧列 erp_codes 已废弃，勿再写入/读取。
+--   * 平台商品 ID 用 tm_spus / jd_spus / dy_spus / jd_self_skus (均为 JSONB 数组)。
+-- 读取入口：style_config_loader / db.sync_daily_collect_scope 直接查本表；
+-- cmt_styles 仅在 dim_style 缺同名款时作为回退。
 -- bitable_weekly_* 仍由 bitable_style_map.json 派生；bitable_sku_master_* 仍由飞书 SKU 主表人工维护。
 CREATE TABLE IF NOT EXISTS dim_style (
-    style_name              TEXT PRIMARY KEY,
-    jd_spus                 TEXT[],                    -- 京东 POP 商品 SPU
-    dy_spus                 TEXT[],                    -- 抖音商品 SPU
-    tm_spus                 TEXT[],                    -- 天猫商品 SPU
-    jd_self_skus            TEXT[],                    -- 京东自营供应链 SKU 集合
-    erp_codes               TEXT[],                    -- 聚水潭 ERP 编码
-    bitable_weekly_url      TEXT,                      -- 款式周报表 URL (bitable_style_map 派生)
-    bitable_weekly_base     TEXT,                      -- 款式周报表 base_token
-    bitable_weekly_table    TEXT,                      -- 款式周报表 table_id
-    bitable_sku_master_url  TEXT,                      -- 飞书 SKU 主表 URL (人工维护)
-    bitable_sku_master_base TEXT,                      -- 飞书 SKU 主表 base_token (人工维护)
-    bitable_sku_master_table TEXT,                     -- 飞书 SKU 主表 table_id (人工维护)
+    id                      BIGINT GENERATED ALWAYS AS IDENTITY PRIMARY KEY,
+    style_name              TEXT NOT NULL UNIQUE,
+    brand                   TEXT        NOT NULL DEFAULT '',
    category                TEXT,
    launch_date             DATE,
    notes                   TEXT,
+    jd_self_skus            JSONB       NOT NULL DEFAULT '[]'::jsonb,  -- 京东自营供应链 SKU 集合
+    jd_spus                 JSONB       NOT NULL DEFAULT '[]'::jsonb,  -- 京东 POP 商品 SPU
+    dy_spus                 JSONB       NOT NULL DEFAULT '[]'::jsonb,  -- 抖音商品 SPU
+    tm_spus                 JSONB       NOT NULL DEFAULT '[]'::jsonb,  -- 天猫商品 SPU
+    erp_style_codes         JSONB       NOT NULL DEFAULT '[]'::jsonb,  -- 聚水潭 ERP 款式编码（正源）
+    erp_codes               BIGINT,                                    -- 已废弃：历史单值列，仅供回溯
+    product_tags            JSONB       NOT NULL DEFAULT '[]'::jsonb,  -- 产品标签
+    lifecycle_stage         VARCHAR,                                   -- 生命周期阶段
+    material                TEXT,
+    selling_points          TEXT,
+    features                TEXT,
+    compatible_devices      TEXT,
+    image_url               TEXT        NOT NULL DEFAULT '',
+    big_nfc_count           INTEGER     NOT NULL DEFAULT 1,
+    small_nfc_count         INTEGER     NOT NULL DEFAULT 1,
+    jushuitan_category      VARCHAR,                                   -- 聚水潭分类
+    series_name             VARCHAR,                                   -- 所属系列
+    is_series_anchor        BOOLEAN     NOT NULL DEFAULT FALSE,        -- 系列锚点款
+    series_sort_order       INTEGER,
+    collect_enabled         BOOLEAN     NOT NULL DEFAULT TRUE,         -- 商品日报采集开关（系列级：以锚点为准）
+    -- 飞书多维表格地址（9 类，全部以「系列锚点行」为准，子款留空即可）
+    bitable_weekly_url      TEXT,                                      -- 销量表 URL (bitable_style_map 派生)
+    bitable_weekly_base     TEXT,                                      -- 销量表 base_token
+    bitable_weekly_table    TEXT,                                      -- 销量表 table_id
+    bitable_main_image_url  TEXT,                                      -- 主图表 URL
+    bitable_creator_url     TEXT,                                      -- 合作达人表 URL
+    bitable_self_creator_url TEXT,                                     -- 自营合作达人表 URL
+    bitable_weekly_note_url TEXT,                                      -- 每周/每月笔记分析 URL
+    bitable_style_analysis_url TEXT,                                   -- 平台单品分析 URL
+    bitable_persona_tm_url  TEXT,                                      -- 人群画像（天猫）URL
+    bitable_persona_jd_url  TEXT,                                      -- 人群画像（京东）URL
+    bitable_persona_dy_url  TEXT,                                      -- 人群画像（抖音）URL
+    bitable_sku_master_url  TEXT,                                      -- 飞书 SKU 主表 URL (人工维护)
+    bitable_sku_master_base TEXT,                                      -- 飞书 SKU 主表 base_token (人工维护)
+    bitable_sku_master_table TEXT,                                     -- 飞书 SKU 主表 table_id (人工维护)
    updated_at              TIMESTAMPTZ NOT NULL DEFAULT NOW()
 );

+-- 商品日报采集开关已上移到 dim_style.collect_enabled（系列级，以锚点为准）。
+-- 旧表 product_daily_collect_scope 已于 2026-10-29 迁移并删除，
+-- 备份在 var/tmp/collect_scope_before_drop_20261029.json。
+DROP TABLE IF EXISTS product_daily_collect_scope;
+
 -- 每日款式级指标 (来自 aggregate_daily_final.py → data/<date>/<款>/final_summary.json)
 CREATE TABLE IF NOT EXISTS daily_style_metrics (
    metric_date     DATE NOT NULL,
@@ -206,13 +240,35 @@ CREATE INDEX IF NOT EXISTS plat_xlsx_item_idx
    ON fact_platform_xlsx (platform, item_id);

 -- ── 兼容旧版 dim_style：init_schema 在已有库上跑时给老表加列 ─────────────
-ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS jd_spus                 TEXT[];
-ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS dy_spus                 TEXT[];
-ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS tm_spus                 TEXT[];
-ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS erp_codes               TEXT[];
+-- 平台 ID / ERP 编码统一为 JSONB 数组；ERP 正源是 erp_style_codes，erp_codes 已废弃。
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS jd_spus                 JSONB NOT NULL DEFAULT '[]'::jsonb;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS dy_spus                 JSONB NOT NULL DEFAULT '[]'::jsonb;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS tm_spus                 JSONB NOT NULL DEFAULT '[]'::jsonb;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS erp_style_codes         JSONB NOT NULL DEFAULT '[]'::jsonb;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS brand                   TEXT  NOT NULL DEFAULT '';
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS lifecycle_stage         VARCHAR;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS product_tags            JSONB NOT NULL DEFAULT '[]'::jsonb;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS jushuitan_category      VARCHAR;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS series_name             VARCHAR;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS is_series_anchor        BOOLEAN NOT NULL DEFAULT FALSE;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS series_sort_order       INTEGER;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS collect_enabled         BOOLEAN NOT NULL DEFAULT TRUE;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS image_url               TEXT  NOT NULL DEFAULT '';
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS big_nfc_count           INTEGER NOT NULL DEFAULT 1;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS small_nfc_count         INTEGER NOT NULL DEFAULT 1;
 ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_weekly_url      TEXT;
 ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_weekly_base     TEXT;
 ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_weekly_table    TEXT;
+-- 飞书地址统一上收到 dim_style（2026-10-29）：锚点行唯一来源，子款留空自动继承。
+-- workflow_dynamic_styles / _style_platforms 的同名字段已废弃，代码不再读取。
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_main_image_url  TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_creator_url     TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_self_creator_url TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_weekly_note_url TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_style_analysis_url TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_persona_tm_url  TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_persona_jd_url  TEXT;
+ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_persona_dy_url  TEXT;
 ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_sku_master_url  TEXT;
 ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_sku_master_base TEXT;
 ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_sku_master_table TEXT;
@@ -223,13 +279,14 @@ ALTER TABLE dim_style ADD COLUMN IF NOT EXISTS bitable_sku_master_table TEXT;
 -- ============================================================================

 -- dim_style
-COMMENT ON TABLE dim_style IS '款式维度表 — 维护每款 (各平台商品 ID + 京东自营 SKU + ERP 编码 + 飞书表地址 + 品类/上架日期/备注)。身份投影自产品生命进程，周报地址与 SKU 主表仍按各自来源维护。';
-COMMENT ON COLUMN dim_style.style_name IS '款式名 (主键)，与 PRODUCT_STYLE_GROUPS 的 key 对齐';
-COMMENT ON COLUMN dim_style.jd_spus IS '京东 POP 商品 SPU 列表';
-COMMENT ON COLUMN dim_style.dy_spus IS '抖音商品 SPU 列表';
-COMMENT ON COLUMN dim_style.tm_spus IS '天猫商品 SPU 列表';
-COMMENT ON COLUMN dim_style.jd_self_skus IS '京东自营供应链 SKU 集合';
-COMMENT ON COLUMN dim_style.erp_codes IS '聚水潭 ERP 编码列表';
+COMMENT ON TABLE dim_style IS '款式维度表 — 款式主档与身份正源：各平台商品 ID + 京东自营 SKU + ERP 款式编码 + 飞书表地址 + 品类/上架日期/备注。';
+COMMENT ON COLUMN dim_style.style_name IS '款式名 (业务唯一键)，与产品生命进程款名对齐';
+COMMENT ON COLUMN dim_style.jd_spus IS '京东 POP 商品 SPU 列表 (JSONB 数组)';
+COMMENT ON COLUMN dim_style.dy_spus IS '抖音商品 SPU 列表 (JSONB 数组)';
+COMMENT ON COLUMN dim_style.tm_spus IS '天猫商品 SPU 列表 (JSONB 数组)';
+COMMENT ON COLUMN dim_style.jd_self_skus IS '京东自营供应链 SKU 集合 (JSONB 数组)';
+COMMENT ON COLUMN dim_style.erp_style_codes IS '聚水潭 ERP 款式编码列表 (JSONB 数组，ERP 身份正源)';
+COMMENT ON COLUMN dim_style.erp_codes IS '已废弃：历史单值 ERP 编码列，新代码禁止写入/读取，请改用 erp_style_codes';
 COMMENT ON COLUMN dim_style.bitable_weekly_url IS '款式周报表 URL (从 bitable_style_map.json 派生)';
 COMMENT ON COLUMN dim_style.bitable_weekly_base IS '款式周报表 base_token';
 COMMENT ON COLUMN dim_style.bitable_weekly_table IS '款式周报表 table_id';
@@ -433,7 +490,8 @@ COMMENT ON COLUMN product_daily_metrics.detail_page_bounce_rate IS '详情页跳

 -- ============================================================================
 -- tm_product_ad_daily_metrics: 天猫商品投放日报
-- 数据源: 天猫投放平台导出的商品报表 CSV。
+-- 数据源: 万相台商品报表 CSV (import_tmall_product_ads.py 解析后直接写本表)。
+-- 每日每商品一行；花费/成交/投产比走标量列供下游直接聚合，报表其余指标留在 metrics。
 -- 与 product_daily_metrics 分表，避免经营数据重新同步时覆盖投放数据。
 -- ============================================================================
 CREATE TABLE IF NOT EXISTS tm_product_ad_daily_metrics (
@@ -441,27 +499,47 @@ CREATE TABLE IF NOT EXISTS tm_product_ad_daily_metrics (
    stat_date               DATE           NOT NULL,
    product_id              TEXT           NOT NULL,
    product_name            TEXT,
+    style_name              TEXT,
+    style_match_method      TEXT           NOT NULL DEFAULT 'unmatched',
    spend                   NUMERIC(18, 2) NOT NULL DEFAULT 0,
    direct_deal_amount      NUMERIC(18, 2) NOT NULL DEFAULT 0,
    indirect_deal_amount    NUMERIC(18, 2) NOT NULL DEFAULT 0,
    attributed_deal_amount  NUMERIC(18, 2) NOT NULL DEFAULT 0,
    reported_roas           NUMERIC(18, 6),
    calculated_roas         NUMERIC(18, 6),
+    metrics                 JSONB          NOT NULL DEFAULT '{}'::jsonb,
    duplicate_row_count     INTEGER        NOT NULL DEFAULT 1,
    source_file             TEXT,
+    source_row_number       INTEGER,
    imported_at             TIMESTAMPTZ    NOT NULL DEFAULT NOW(),
    UNIQUE (stat_date, product_id)
 );
+-- 兼容已创建的窄表：补齐款式匹配结果与完整指标快照列。
+ALTER TABLE tm_product_ad_daily_metrics
+    ADD COLUMN IF NOT EXISTS style_name TEXT;
+ALTER TABLE tm_product_ad_daily_metrics
+    ADD COLUMN IF NOT EXISTS style_match_method TEXT
+        NOT NULL DEFAULT 'unmatched';
+ALTER TABLE tm_product_ad_daily_metrics
+    ADD COLUMN IF NOT EXISTS metrics JSONB NOT NULL DEFAULT '{}'::jsonb;
+ALTER TABLE tm_product_ad_daily_metrics
+    ADD COLUMN IF NOT EXISTS source_row_number INTEGER;
 CREATE INDEX IF NOT EXISTS tm_product_ad_daily_metrics_date_idx
    ON tm_product_ad_daily_metrics (stat_date DESC);
 CREATE INDEX IF NOT EXISTS tm_product_ad_daily_metrics_product_date_idx
    ON tm_product_ad_daily_metrics (product_id, stat_date DESC);
+CREATE INDEX IF NOT EXISTS tm_product_ad_daily_metrics_style_date_idx
+    ON tm_product_ad_daily_metrics (style_name, stat_date DESC);

 COMMENT ON TABLE tm_product_ad_daily_metrics IS '天猫商品级投放日报，每日每商品一行；广告归因成交额仅用于计算投产比。';
+COMMENT ON COLUMN tm_product_ad_daily_metrics.product_id IS '万相台主体ID（商品），按文本保存';
+COMMENT ON COLUMN tm_product_ad_daily_metrics.style_name IS '款式名；先按商品 ID 反查 dim_style.tm_spus，未命中再按商品名称精确包含匹配，无法确认时为 NULL';
+COMMENT ON COLUMN tm_product_ad_daily_metrics.style_match_method IS '款式匹配方式：product_id=商品 ID，product_name=商品名称，ambiguous=名称歧义，unmatched=未匹配';
 COMMENT ON COLUMN tm_product_ad_daily_metrics.spend IS '天猫广告花费（页面电商支付金额）';
 COMMENT ON COLUMN tm_product_ad_daily_metrics.attributed_deal_amount IS '广告归因总成交金额，仅用于计算投产比';
 COMMENT ON COLUMN tm_product_ad_daily_metrics.reported_roas IS '源报表投入产出比';
 COMMENT ON COLUMN tm_product_ad_daily_metrics.calculated_roas IS '归因成交金额 / 花费';
+COMMENT ON COLUMN tm_product_ad_daily_metrics.metrics IS '万相台报表指标原样快照 (展现/点击/点击率/转化率/预售成交等)，键名与源报表中文列一一对应';

 -- ============================================================================
 -- jd_product_ad_daily_metrics: 京东营销概况广告日报
@@ -558,7 +636,7 @@ COMMENT ON COLUMN dy_product_ad_daily_metrics.overall_ctr IS '报表整体点击
 COMMENT ON COLUMN dy_product_ad_daily_metrics.overall_conversion_rate IS '报表整体转化率，单位为百分点';
 COMMENT ON COLUMN dy_product_ad_daily_metrics.net_settlement_rate IS '净成交金额结算率，单位为百分点';
 COMMENT ON COLUMN dy_product_ad_daily_metrics.refund_rate_1h IS '1 小时内退款率，单位为百分点';
-COMMENT ON COLUMN product_daily_metrics.raw_data IS '平台特有原始列 (下单金额/件数/买家数、聚划算、竞争力评分、月累计、搜索引导...) 原样保留；天猫 tm_ad_metrics 同时含 style_name/style_match_method';
+COMMENT ON COLUMN product_daily_metrics.raw_data IS '平台特有原始列 (下单金额/件数/买家数、聚划算、竞争力评分、月累计、搜索引导...) 原样保留；天猫广告指标已迁至 tm_product_ad_daily_metrics，本列不再写 tm_ad_metrics';
 COMMENT ON COLUMN product_daily_metrics.source_file IS '原始文件名 (溯源用)';
 COMMENT ON COLUMN product_daily_metrics.imported_at IS '入库时间';

@@ -753,69 +831,19 @@ COMMENT ON COLUMN main_image_creatives.collected_at IS '采集时间 (脚本写
 COMMENT ON COLUMN main_image_creatives.updated_at IS '最近更新时间';

 -- ============================================================================
-- product_reviews: 三平台商品评价 (JD + DY, TM 暂未接入)
-- 数据源:
--   JD: jd_product_data_collector.py --reviews-only -> data/jd单品数据/评价汇总_<date>.json
--   DY: dy_product_scraping.py (主流程已挂载)     -> data/dy单品数据/评论汇总_<date>.json
-- 设计: 三平台字段归一到独立列, 平台特有字段进 raw_data JSONB。
-- 写入策略: ON CONFLICT (platform, order_id) DO UPDATE 全量覆盖 (order_id 是评价业务键)。
+-- product_reviews: 已下线 (2026-11-01)
+--
+-- 原设计是 JD + DY 商品评价合一表, 由 import_product_reviews.py 从平台 JSON 入库。
+-- 下线原因:
+--   1. JD 侧 style_name 全靠 sku_id 反查, 而 dim_jd_sku_spu 映射表不存在,
+--      214 行 style_name 全为 NULL, 两个消费方 (analyze_style / 营销日报) 都是
+--      WHERE style_name = ?, 等于入库了但没人能读。
+--   2. 差评全量已由 shop_intelligence 的三张差评表 (tmall/jd/dy_negative_reviews)
+--      独立采集, 覆盖面更全 (天猫 97 条, 而本表天猫为 0)。
+--
+-- 现消费方统一读视图 v_negative_reviews (见文件末尾「款式主键统一」段)。
+-- 存量表已归档为 product_reviews_archived_20261101, 不再建表、不再写入。
 -- ============================================================================
-CREATE TABLE IF NOT EXISTS product_reviews (
-    id                      BIGSERIAL PRIMARY KEY,
-    platform                TEXT        NOT NULL CHECK (platform IN ('jd','dy')),
-    collect_date            DATE        NOT NULL,                 -- 采集日期 (脚本运行日)
-    comment_date            DATE,                                 -- 评价时间 (JD 取评价时间日期部分, DY 取 comment_time 日期部分)
-    style_name              TEXT,                                 -- 款式名 (DY: 通过 product_id 反查 dim_style.dy_spus 填充; 未匹配时为 NULL)
-    rating                  TEXT        NOT NULL,                 -- '好评' / '中评' / '差评'
-    score                   SMALLINT,                             -- JD 评分 (1-5 星), DY 留空
-    content                 TEXT,
-    tags                    TEXT[],                               -- JD 评价标签 (做工精细/容量大 等), DY 留空
-    user_name               TEXT,
-    order_id                TEXT        NOT NULL,                 -- 订单 ID (业务键, 用于去重)
-    product_id              TEXT,                                 -- 平台商品 ID (DY product_id / JD sku_name 对应的 SPU)
-    product_name            TEXT,
-    sku_id                  TEXT,                                 -- JD SKU 编码, DY 留空
-    sku_name                TEXT,                                 -- JD SKU 名称, DY 留空
-    sku_size                TEXT,                                 -- JD 规格, DY 留空
-    raw_data                JSONB       NOT NULL DEFAULT '{}'::jsonb,
-    source_file             TEXT,                                 -- 原始 JSON 文件路径
-    imported_at             TIMESTAMPTZ NOT NULL DEFAULT NOW(),
-    UNIQUE (platform, order_id)
-);
-CREATE INDEX IF NOT EXISTS product_reviews_platform_date_idx
-    ON product_reviews (platform, comment_date DESC);
-CREATE INDEX IF NOT EXISTS product_reviews_rating_idx
-    ON product_reviews (platform, rating, comment_date DESC);
-CREATE INDEX IF NOT EXISTS product_reviews_product_idx
-    ON product_reviews (platform, product_id, comment_date DESC);
-- product_reviews_style_idx 在文件末尾 ALTER TABLE 加 style_name 列之后建 (老表兼容)
-
-COMMENT ON TABLE product_reviews IS '商品评价 - JD + DY 合一表。来源: jd_product_data_collector.py --reviews-only / dy_product_scraping.py。UNIQUE (platform, order_id) -> 同订单评价重抓覆盖。';
-COMMENT ON COLUMN product_reviews.platform IS '平台代码 (''jd'' / ''dy'')';
-COMMENT ON COLUMN product_reviews.collect_date IS '采集日期 (脚本运行日, 不等于评价发生日)';
-COMMENT ON COLUMN product_reviews.comment_date IS '评价时间 (从评价发生时间戳抽日期部分)';
-- style_name 列的 COMMENT 在文件末尾 ALTER TABLE 加列之后 (老表兼容)
-COMMENT ON COLUMN product_reviews.rating IS '评价等级: 好评 / 中评 / 差评';
-COMMENT ON COLUMN product_reviews.score IS 'JD 评分 (1-5 星, aria-valuenow), DY 留空';
-COMMENT ON COLUMN product_reviews.content IS '评价正文';
-COMMENT ON COLUMN product_reviews.tags IS 'JD 评价标签数组 (如 [做工精细, 容量大]), DY 留空';
-COMMENT ON COLUMN product_reviews.user_name IS '用户昵称';
-COMMENT ON COLUMN product_reviews.order_id IS '订单 ID (业务键, 同订单只保留一条, 重抓覆盖)';
-COMMENT ON COLUMN product_reviews.product_id IS '平台商品 ID (DY product_id / JD SPU)';
-COMMENT ON COLUMN product_reviews.product_name IS '商品名称';
-COMMENT ON COLUMN product_reviews.sku_id IS 'JD SKU 编码, DY 留空';
-COMMENT ON COLUMN product_reviews.sku_name IS 'JD SKU 名称, DY 留空';
-COMMENT ON COLUMN product_reviews.sku_size IS 'JD 规格文本, DY 留空';
-COMMENT ON COLUMN product_reviews.raw_data IS '平台特有原始字段 (JD 完整评价对象 / DY comment_rows 行)';
-COMMENT ON COLUMN product_reviews.source_file IS '源 JSON 文件相对路径 (溯源用)';
-COMMENT ON COLUMN product_reviews.imported_at IS '入库时间';
-
-- 兼容已有库: product_reviews 加 style_name 列 (CREATE TABLE IF NOT EXISTS 不会改已有表)
-ALTER TABLE product_reviews ADD COLUMN IF NOT EXISTS style_name TEXT;
-CREATE INDEX IF NOT EXISTS product_reviews_style_idx
-    ON product_reviews (style_name, comment_date DESC)
-    WHERE style_name IS NOT NULL;
-COMMENT ON COLUMN product_reviews.style_name IS '款式名 (DY: 通过 product_id 反查 dim_style.dy_spus 填充; 未匹配时为 NULL, 表示 dim_style 缺该商品 ID 维护)';

 -- ============================================================================
 -- fact_style_analysis_reports: 单款式多维度分析报告 (直连 MiniMax 产出)
@@ -1273,3 +1301,310 @@ COMMENT ON TABLE fact_video_publications IS
    '附件级视频发布幂等账本；点击发布后无明确回执必须进入 ambiguous，禁止自动重试。';
 COMMENT ON TABLE fact_video_publication_attempts IS
    '视频发布每次尝试的标题、商品、话题、标签、提交时点与确认凭据。';
+
+-- 聚水潭「报表→销售主题分析→按日统计」Excel 导出入库。
+-- 口径：发货时间区间 + 固定品牌(光影行星/光影/gyxx) + 固定勾选店铺清单，
+-- 采集脚本 collect_erp_sales_theme_daily.py；与 erp_all_shop_style_daily_metrics
+-- (全店铺按款式) 分表，本表是全渠道合并口径的按日汇总。
+CREATE TABLE IF NOT EXISTS erp_sales_theme_daily_metrics (
+    id                    BIGSERIAL PRIMARY KEY,
+    stat_date             DATE           NOT NULL,
+    sales_order_count     INTEGER        NOT NULL DEFAULT 0,  -- 销售单数
+    sales_qty             INTEGER        NOT NULL DEFAULT 0,  -- 销售数量
+    shipped_qty           INTEGER        NOT NULL DEFAULT 0,  -- 实发数量
+    shipped_amount        NUMERIC(18, 2) NOT NULL DEFAULT 0,  -- 实发金额
+    sales_amount          NUMERIC(18, 2) NOT NULL DEFAULT 0,  -- 销售金额
+    return_order_count    INTEGER        NOT NULL DEFAULT 0,  -- 退货单数
+    return_qty            INTEGER        NOT NULL DEFAULT 0,  -- 退货数量
+    returned_qty          INTEGER        NOT NULL DEFAULT 0,  -- 实退数量
+    return_amount         NUMERIC(18, 2) NOT NULL DEFAULT 0,  -- 退货金额
+    returned_amount       NUMERIC(18, 2) NOT NULL DEFAULT 0,  -- 实退金额
+    net_sales_order_count INTEGER        NOT NULL DEFAULT 0,  -- 净销售单数
+    net_sales_qty         INTEGER        NOT NULL DEFAULT 0,  -- 净销量
+    net_sales_amount      NUMERIC(18, 2) NOT NULL DEFAULT 0,  -- 净销售额
+    freight_income        NUMERIC(18, 2),                     -- 运费收入(可空)
+    freight_expense       NUMERIC(18, 2),                     -- 运费支出(可空)
+    shipped_order_count   INTEGER        NOT NULL DEFAULT 0,  -- 实发单数
+    date_basis            TEXT           NOT NULL DEFAULT '发货时间',
+    brands                TEXT[]         NOT NULL DEFAULT '{}',
+    shops                 TEXT[]         NOT NULL DEFAULT '{}',
+    source_file           TEXT,
+    imported_at           TIMESTAMPTZ    NOT NULL DEFAULT NOW(),
+    updated_at            TIMESTAMPTZ    NOT NULL DEFAULT NOW(),
+    UNIQUE (stat_date)
+);
+CREATE INDEX IF NOT EXISTS erp_sales_theme_daily_metrics_date_idx
+    ON erp_sales_theme_daily_metrics (stat_date DESC);
+
+-- ============================================================================
+-- 款式主键统一 (2026-11-01)
+--
+-- 目标: 采集端所有带款式的表统一用 style_id (= dim_style.id) 关联, style_name
+-- 只作为显示冗余。data-hub 侧 22 张表本来就用 style_id, 两端从此同一套主键,
+-- 不再出现「同一库两种引用方式」。已实测 erp_product.style_id 135/135 命中。
+--
+-- 约定:
+--   * style_id 可空的表 = 允许「未建档」的行 (平台商品 ID 未写进 dim_style 的
+--     spus 数组), 这类行必须同时写 style_match_method='unmatched', 便于统计
+--     漏维护的规模。
+--   * style_id 非空的表 = 款式清单来自 dim_style loader, 理论上必然命中;
+--     但**不加 NOT NULL**, 由 backfill_style_id.py 与写入端负责填充,
+--     避免历史脏款名在 ALTER 阶段中断建表。
+--   * 主键一律不动 (daily_style_metrics 仍是 metric_date+style_name+platform),
+--     data-hub 有 6 处按名字 join, 改 PK 收益小、回归面大。
+-- ============================================================================
+
+ALTER TABLE daily_style_metrics              ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE weekly_style_metrics             ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE erp_all_shop_style_daily_metrics ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE daily_persona_metrics            ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE main_image_creatives             ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE fact_style_analysis_reports      ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE decline_events                   ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE fact_alert_history               ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE jd_self_weekly_sales              ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE persona_screenshots               ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE product_daily_metrics            ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE jd_product_ad_daily_metrics      ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE dy_product_ad_daily_metrics      ADD COLUMN IF NOT EXISTS style_id BIGINT;
+ALTER TABLE tm_product_ad_daily_metrics      ADD COLUMN IF NOT EXISTS style_id BIGINT;
+
+-- 匹配方式留痕: dim_style / product_id / product_name / anchor / unmatched
+ALTER TABLE daily_style_metrics              ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE weekly_style_metrics             ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE erp_all_shop_style_daily_metrics ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE daily_persona_metrics            ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE main_image_creatives             ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE fact_style_analysis_reports      ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE decline_events                   ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE fact_alert_history               ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE jd_self_weekly_sales              ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE persona_screenshots               ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE product_daily_metrics            ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+ALTER TABLE jd_product_ad_daily_metrics      ADD COLUMN IF NOT EXISTS style_match_method TEXT;
+
+COMMENT ON COLUMN daily_style_metrics.style_id IS 'dim_style.id; 采集端统一款式主键, style_name 仅作显示冗余';
+COMMENT ON COLUMN product_daily_metrics.style_id IS 'dim_style.id; 按 platform+product_id 反查 spus 得到, 未建档为 NULL 且 style_match_method=unmatched';
+
+-- 查询索引: 按款式跨日期取数 (产品生命页、风格分析、预警)
+CREATE INDEX IF NOT EXISTS daily_style_metrics_style_id_idx
+    ON daily_style_metrics (style_id, metric_date DESC);
+CREATE INDEX IF NOT EXISTS weekly_style_metrics_style_id_idx
+    ON weekly_style_metrics (style_id, week_end DESC);
+CREATE INDEX IF NOT EXISTS erp_all_shop_style_daily_metrics_style_id_idx
+    ON erp_all_shop_style_daily_metrics (style_id, metric_date DESC);
+CREATE INDEX IF NOT EXISTS daily_persona_metrics_style_id_idx
+    ON daily_persona_metrics (style_id, collect_date DESC);
+CREATE INDEX IF NOT EXISTS main_image_creatives_style_id_idx
+    ON main_image_creatives (style_id, collect_date DESC);
+CREATE INDEX IF NOT EXISTS fact_style_analysis_reports_style_id_idx
+    ON fact_style_analysis_reports (style_id, analysis_date DESC);
+CREATE INDEX IF NOT EXISTS decline_events_style_id_idx
+    ON decline_events (style_id, end_date DESC);
+CREATE INDEX IF NOT EXISTS fact_alert_history_style_id_idx
+    ON fact_alert_history (style_id, trigger_date DESC);
+CREATE INDEX IF NOT EXISTS jd_self_weekly_sales_style_id_idx
+    ON jd_self_weekly_sales (style_id, week_end DESC);
+CREATE INDEX IF NOT EXISTS persona_screenshots_style_id_idx
+    ON persona_screenshots (style_id, collect_date DESC);
+CREATE INDEX IF NOT EXISTS product_daily_metrics_style_id_idx
+    ON product_daily_metrics (style_id, stat_date DESC);
+CREATE INDEX IF NOT EXISTS jd_product_ad_daily_metrics_style_id_idx
+    ON jd_product_ad_daily_metrics (style_id, stat_date DESC);
+CREATE INDEX IF NOT EXISTS dy_product_ad_daily_metrics_style_id_idx
+    ON dy_product_ad_daily_metrics (style_id, stat_date DESC);
+CREATE INDEX IF NOT EXISTS tm_product_ad_daily_metrics_style_id_idx
+    ON tm_product_ad_daily_metrics (style_id, stat_date DESC);
+
+-- ── P0 观测视图 ─────────────────────────────────────────────────────────────
+
+-- 未建档的平台商品 ID: 推动在 dim_style 里补 spus, 是 style_id 覆盖率的天花板。
+CREATE OR REPLACE VIEW v_style_unmatched_product AS
+SELECT m.platform,
+       m.product_id,
+       MAX(m.product_name) AS product_name,
+       MIN(m.stat_date)    AS first_date,
+       MAX(m.stat_date)    AS last_date,
+       COUNT(*)            AS rows
+FROM product_daily_metrics m
+WHERE NOT EXISTS (
+        SELECT 1 FROM dim_style d
+        WHERE (m.platform = 'tm' AND d.tm_spus @> to_jsonb(m.product_id))
+           OR (m.platform = 'jd' AND d.jd_spus @> to_jsonb(m.product_id))
+           OR (m.platform = 'dy' AND d.dy_spus @> to_jsonb(m.product_id)))
+GROUP BY m.platform, m.product_id;
+
+COMMENT ON VIEW v_style_unmatched_product IS
+    '未写进 dim_style.spus 的平台商品 ID 清单; 这些行的 product_daily_metrics.style_id 必然为 NULL。';
+
+-- 同上, 但把三张广告表也算进来。广告投的商品不一定有经营日报 (纯投流款), 只看
+-- v_style_unmatched_product 会漏掉这部分待建档 ID。列结构保持一致, 便于并排检索。
+CREATE OR REPLACE VIEW v_style_unmatched_source AS
+WITH all_rows AS (
+    SELECT platform, product_id, product_name, stat_date FROM product_daily_metrics
+    UNION ALL
+    SELECT 'tm', product_id, product_name, stat_date FROM tm_product_ad_daily_metrics
+    UNION ALL
+    SELECT 'jd', spu_id, spu_name, stat_date FROM jd_product_ad_daily_metrics
+    UNION ALL
+    SELECT 'dy', product_id, product_name, stat_date FROM dy_product_ad_daily_metrics
+)
+SELECT s.platform,
+       s.product_id,
+       MAX(s.product_name) AS product_name,
+       MIN(s.stat_date)    AS first_date,
+       MAX(s.stat_date)    AS last_date,
+       COUNT(*)            AS rows
+FROM all_rows s
+WHERE s.product_id IS NOT NULL
+  AND BTRIM(s.product_id) <> ''
+  AND NOT EXISTS (
+        SELECT 1 FROM dim_style d
+        WHERE (s.platform = 'tm' AND d.tm_spus @> to_jsonb(s.product_id))
+           OR (s.platform = 'jd' AND d.jd_spus @> to_jsonb(s.product_id))
+           OR (s.platform = 'dy' AND d.dy_spus @> to_jsonb(s.product_id)))
+GROUP BY s.platform, s.product_id;
+
+COMMENT ON VIEW v_style_unmatched_source IS
+    '全部采集表 (经营日报 + 三张广告表) 里未写进 dim_style.spus 的商品 ID; 补建档的完整待办清单。';
+
+-- 各表 style_id 回填进度 (验收用)。
+CREATE OR REPLACE VIEW v_style_id_coverage AS
+SELECT 'daily_style_metrics' AS tbl, COUNT(*) AS rows, COUNT(style_id) AS with_style_id FROM daily_style_metrics
+UNION ALL SELECT 'weekly_style_metrics', COUNT(*), COUNT(style_id) FROM weekly_style_metrics
+UNION ALL SELECT 'erp_all_shop_style_daily_metrics', COUNT(*), COUNT(style_id) FROM erp_all_shop_style_daily_metrics
+UNION ALL SELECT 'daily_persona_metrics', COUNT(*), COUNT(style_id) FROM daily_persona_metrics
+UNION ALL SELECT 'main_image_creatives', COUNT(*), COUNT(style_id) FROM main_image_creatives
+UNION ALL SELECT 'fact_style_analysis_reports', COUNT(*), COUNT(style_id) FROM fact_style_analysis_reports
+UNION ALL SELECT 'decline_events', COUNT(*), COUNT(style_id) FROM decline_events
+UNION ALL SELECT 'fact_alert_history', COUNT(*), COUNT(style_id) FROM fact_alert_history
+UNION ALL SELECT 'jd_self_weekly_sales', COUNT(*), COUNT(style_id) FROM jd_self_weekly_sales
+UNION ALL SELECT 'persona_screenshots', COUNT(*), COUNT(style_id) FROM persona_screenshots
+UNION ALL SELECT 'product_daily_metrics', COUNT(*), COUNT(style_id) FROM product_daily_metrics
+UNION ALL SELECT 'jd_product_ad_daily_metrics', COUNT(*), COUNT(style_id) FROM jd_product_ad_daily_metrics
+UNION ALL SELECT 'dy_product_ad_daily_metrics', COUNT(*), COUNT(style_id) FROM dy_product_ad_daily_metrics
+UNION ALL SELECT 'tm_product_ad_daily_metrics', COUNT(*), COUNT(style_id) FROM tm_product_ad_daily_metrics;
+
+COMMENT ON VIEW v_style_id_coverage IS
+    '各采集表 style_id 非空比例; 除 product_daily_metrics / 广告表外应为 100%。';
+
+-- ── P1 统一差评视图 ────────────────────────────────────────────────────────
+
+-- product_reviews 下线后, 风格分析与营销日报统一读这张视图。
+-- 与旧表的口径差异: 只有差评一档 (旧表含好/中/差), 但覆盖三个平台。
+-- 款式归因优先级: 平台商品 ID 精确命中 > 商品名称包含款名 (取最长款名, 更具体)。
+-- 差评三表由 shop_intelligence 模块建表, 这里用守卫避免全新库初始化时视图先于源表。
+DO $$
+BEGIN
+    IF to_regclass('public.tmall_negative_reviews') IS NOT NULL
+       AND to_regclass('public.jd_negative_reviews') IS NOT NULL
+       AND to_regclass('public.dy_negative_reviews') IS NOT NULL THEN
+        EXECUTE $view$
+CREATE OR REPLACE VIEW v_negative_reviews AS
+WITH src AS (
+    SELECT 'tm'::text                 AS platform,
+           n.business_date            AS comment_date,
+           COALESCE(n.review_level, '负面评价') AS rating,
+           n.content,
+           NULL::text                 AS order_id,
+           NULL::text                 AS product_id,
+           n.product_name,
+           n.review_id,
+           n.review_at,
+           n.raw_review
+    FROM tmall_negative_reviews n
+    UNION ALL
+    SELECT 'jd', n.business_date, COALESCE(n.review_level, '负面评价'),
+           n.content, n.order_id, NULL::text, n.product_name,
+           n.review_id, n.review_at, n.raw_review
+    FROM jd_negative_reviews n
+    UNION ALL
+    SELECT 'dy', n.business_date, COALESCE(n.rating_level, '负面评价'),
+           n.content, n.order_id, n.product_id, n.product_name,
+           n.review_id, n.review_at, n.raw_review
+    FROM dy_negative_reviews n
+)
+SELECT s.platform, s.comment_date, s.rating, s.content,
+       s.order_id, s.product_id, s.product_name,
+       s.review_id, s.review_at, s.raw_review,
+       m.style_id, m.style_name, m.style_match_method
+FROM src s
+LEFT JOIN LATERAL (
+    SELECT d.id AS style_id,
+           d.style_name,
+           CASE
+               WHEN (s.platform = 'dy' AND d.dy_spus @> to_jsonb(s.product_id))
+                 OR (s.platform = 'jd' AND d.jd_spus @> to_jsonb(s.product_id))
+                   THEN 'product_id'
+               ELSE 'product_name'
+           END AS style_match_method
+    FROM dim_style d
+    WHERE COALESCE(d.style_name, '') <> ''
+      AND (
+            (s.platform = 'dy' AND d.dy_spus @> to_jsonb(s.product_id))
+         OR (s.platform = 'jd' AND d.jd_spus @> to_jsonb(s.product_id))
+         OR (s.product_name IS NOT NULL AND s.product_name LIKE '%' || d.style_name || '%')
+          )
+    ORDER BY (
+            CASE
+                WHEN (s.platform = 'dy' AND d.dy_spus @> to_jsonb(s.product_id))
+                  OR (s.platform = 'jd' AND d.jd_spus @> to_jsonb(s.product_id))
+                    THEN 0 ELSE 1
+            END
+        ),
+        COALESCE(d.is_series_anchor, FALSE) DESC,
+        LENGTH(d.style_name) DESC,
+        d.id
+    LIMIT 1
+) m ON TRUE
+$view$;
+        EXECUTE $c$COMMENT ON VIEW v_negative_reviews IS
+            '三平台差评归一视图 (tm/jd/dy_negative_reviews); 取代已下线的 product_reviews。style_id 为 dim_style.id, 未匹配为 NULL。'$c$;
+    ELSE
+        RAISE NOTICE 'skip v_negative_reviews: 差评表尚未创建 (先运行 shop_intelligence init_schema)';
+    END IF;
+END
+$$;
+
+-- ── style_id 一律走「逻辑外键」, 不建物理外键 ───────────────────────────────
+--
+-- 2026-11-01 定案: style_id 与 dim_style.id 的关联由应用层保证, 库里**不加**
+-- FOREIGN KEY 约束。原因:
+--   * 物理外键会在写入端传错/传旧 id 时直接报错中断整批采集, 而采集是「宁可
+--     留 NULL 待回填, 也不要整日落空」的语义, 硬约束与这个语义冲突;
+--   * 款式被清理/重建 (拆系列、合并同款) 时要先删约束才能动数据, 运维成本高;
+--   * data-hub 侧 22 张表同样只用 style_id 做关联, 不带物理外键, 两边保持一致。
+--
+-- 完整性靠三件事保证, 不需要数据库兜底:
+--   1. 写入端: loader 下发的 style_id 来自 dim_style 本身;
+--              按商品 ID / 款名兜底的路径由 attach_style_ids 统一查询后填充;
+--   2. 校验:  backfill_style_id.py --dry-run 与 v_style_id_coverage 报覆盖率;
+--   3. 观火:  v_style_unmatched_product / v_style_unmatched_source 列未建档清单。
+--
+-- 下面这段只做**历史清理**: 早期 P4 曾给 10 张表建过 <表名>_style_id_fk,
+-- 这里幂等删除, 保证老库重跑 schema 后回到「只有索引、没有物理外键」的状态。
+DO $$
+DECLARE
+    target_table TEXT;
+BEGIN
+    FOREACH target_table IN ARRAY ARRAY[
+        'daily_style_metrics',
+        'weekly_style_metrics',
+        'erp_all_shop_style_daily_metrics',
+        'daily_persona_metrics',
+        'main_image_creatives',
+        'fact_style_analysis_reports',
+        'decline_events',
+        'fact_alert_history',
+        'jd_self_weekly_sales',
+        'persona_screenshots'
+    ] LOOP
+        IF to_regclass('public.' || target_table) IS NOT NULL THEN
+            EXECUTE format(
+                'ALTER TABLE %I DROP CONSTRAINT IF EXISTS %I',
+                target_table, target_table || '_style_id_fk'
+            );
+        END IF;
+    END LOOP;
+END
+$$;
@@ -1,390 +0,0 @@
-"""同步 dim_style：4 个数据源自动 upsert。
-
-数据源:
-  产品生命进程 + 动态配置运行时聚合 -> 商品编码、ERP 编码和飞书目标地址
-
-历史文件、采集产物和硬编码仅保留在本文件的兼容函数中，不再作为同步入口，
-避免人工修改生命进程后被旧脚本反向覆盖。
-
-SKU 主表的飞书地址 (bitable_sku_master_*) 由人工维护，不在自动同步范围。
-
-用法:
-    .venv/Scripts/python.exe db/sync_dim_style.py                # 全量
-    .venv/Scripts/python.exe db/sync_dim_style.py --dry-run      # 扫描不入库
-"""
-
-from __future__ import annotations
-
-import argparse
-import json
-import re
-import sys
-from datetime import datetime
-from pathlib import Path
-from typing import Any, Iterable
-
-PROJECT_ROOT = Path(__file__).resolve().parent.parent
-sys.path.insert(0, str(PROJECT_ROOT))
-sys.path.insert(0, str(PROJECT_ROOT.parents[2]))
-from runtime_paths import RAW_DATA_ROOT, runtime_script
-from gyxx_flow.adapters.workflow_config import runtime_workflow_configs  # noqa: E402
-
-from db import get_conn  # noqa: E402
-
-
-PLATFORM_DIRS = {
-    "jd": "jd单品数据",
-    "dy": "dy单品数据",
-    "tm": "tm单品数据",
-}
-
-
-# ── 数据源 1: bitable_style_map.json ────────────────────────────────────────
-
-def load_bitable_weekly() -> dict[str, dict[str, str]]:
-    """从 bitable_style_map.json 抓每个款式的飞书周报表 base/table。"""
-    path = PROJECT_ROOT / "bitable_style_map.json"
-    if not path.exists():
-        return {}
-    raw = json.loads(path.read_text(encoding="utf-8-sig"))
-    out: dict[str, dict[str, str]] = {}
-    for style_name, info in raw.items():
-        if not isinstance(info, dict) or not info.get("enabled", True):
-            continue
-        base = info.get("base_token", "")
-        table = info.get("table_id", "")
-        if not base or not table:
-            continue
-        url = f"https://feishu.cn/base/{base}?table={table}"
-        out[style_name] = {
-            "bitable_weekly_base": base,
-            "bitable_weekly_table": table,
-            "bitable_weekly_url": url,
-        }
-    return out
-
-
-# ── 数据源 2: jd_self SKU 集合 ──────────────────────────────────────────────
-
-def load_jd_self_skus() -> dict[str, list[str]]:
-    """从 jd_self_inventory_sales_collector.STYLE_SKU_GROUPS 抓款→SKU 集合。
-
-    直接解析源文件里 STYLE_SKU_GROUPS = [...] 的字面量，避免 import 拉起 playwright。
-    """
-    path = runtime_script("jd_self_inventory_sales_collector.py")
-    text = path.read_text(encoding="utf-8")
-    needle = "STYLE_SKU_GROUPS = "
-    i = text.find(needle)
-    if i < 0:
-        return {}
-    i += len(needle)
-    # 平衡括号找匹配的 ]
-    depth = 0
-    in_str: str | None = None
-    escape = False
-    end = None
-    for j in range(i, len(text)):
-        ch = text[j]
-        if in_str:
-            if escape:
-                escape = False
-            elif ch == "\\":
-                escape = True
-            elif ch == in_str:
-                in_str = None
-            continue
-        if ch in ("'", '"'):
-            in_str = ch
-        elif ch == "[":
-            depth += 1
-        elif ch == "]":
-            depth -= 1
-            if depth == 0:
-                end = j + 1
-                break
-    if end is None:
-        return {}
-    chunk = text[i:end]
-    try:
-        # 用 ast.literal_eval 解析 Python 字面量（兼容末尾逗号 / 单引号）
-        import ast
-        groups = ast.literal_eval(chunk)
-    except Exception as exc:
-        print(f"  [WARN] 解析 STYLE_SKU_GROUPS 失败: {exc}")
-        return {}
-    return {g["style_name"]: list(g.get("skus") or []) for g in groups if "style_name" in g}
-
-
-# ── 数据源 3: 各平台 SPU 集合 ───────────────────────────────────────────────
-
-def _latest_style_json(plat_dir: Path, style_name: str) -> Path | None:
-    """取某款式在 plat_dir 下的最新日期 JSON。
-
-    目录名约定: <款名>_<YYYY-MM-DD> (jd/dy/tm 主流) 或 业务款式_<YYYY-MM-DD> (tm 部分)
-
-    优先级:
-      1) <safe>_<date>.json       (jd / dy / tm 主流)
-      2) 业务款式_<date>.json     (tm 习惯命名)
-      3) data.json                (兜底，但 data.json 通常是 ERP 抓数，可能没有 SPU 字段)
-    """
-    safe = "".join("_" if ch in '\\/:*?"<>|' else ch for ch in style_name).strip(" ._")
-    # 只匹配以 "<safe>_" 开头的目录，避免拿到别的款式的目录
-    cands: list[tuple[str, Path]] = []
-    for d in plat_dir.iterdir():
-        if not d.is_dir() or not d.name.startswith(safe + "_"):
-            continue
-        date_part = d.name[len(safe) + 1:]
-        try:
-            datetime.strptime(date_part, "%Y-%m-%d")
-        except ValueError:
-            continue
-        cands.append((date_part, d))
-    if not cands:
-        return None
-    cands.sort(key=lambda x: x[0], reverse=True)
-    latest_date, latest_dir = cands[0]
-
-    # 优先级 1: <safe>_<date>.json
-    p1 = latest_dir / f"{safe}_{latest_date}.json"
-    if p1.exists():
-        return p1
-    # 优先级 2: 业务款式_<date>.json (tm)
-    p2 = latest_dir / f"业务款式_{latest_date}.json"
-    if p2.exists():
-        return p2
-    # 优先级 3: data.json
-    p3 = latest_dir / "data.json"
-    if p3.exists():
-        return p3
-    return None
-
-
-def _extract_spus(payload: dict, platform: str) -> list[str]:
-    """从款式 JSON payload 提取 SPU/商品 ID 列表。
-
-    各平台字段名:
-      jd: payload['匹配SPU'] 或 payload['spu列表']
-      dy: payload['matched_product_ids'] 或 payload['product_ids']
-      tm: payload['style']['匹配商品ID'] 或 payload['style']['SPU']
-    """
-    if platform == "jd":
-        return list(payload.get("匹配SPU") or payload.get("spu列表") or [])
-    if platform == "dy":
-        return list(payload.get("matched_product_ids") or payload.get("product_ids") or [])
-    if platform == "tm":
-        style = payload.get("style") or {}
-        return list(style.get("匹配商品ID") or style.get("SPU") or [])
-    return []
-
-
-def load_platform_spus() -> dict[str, dict[str, list[str]]]:
-    """返回 {style_name: {platform: [spu, ...]}}，跨日期取最近一次抓数。"""
-    out: dict[str, dict[str, list[str]]] = {}
-    # 先枚举所有款式名（从 weekly_style_metrics 拿权威款式列表）
-    data_root = RAW_DATA_ROOT
-    # 用 daily_style_metrics 拿所有款式名
-    try:
-        with get_conn() as conn:
-            with conn.cursor() as cur:
-                cur.execute("SELECT DISTINCT style_name FROM daily_style_metrics")
-                style_names = [r[0] for r in cur.fetchall()]
-    except Exception:
-        style_names = []
-
-    for style_name in style_names:
-        out.setdefault(style_name, {})
-        for platform, dir_name in PLATFORM_DIRS.items():
-            plat_dir = data_root / dir_name
-            if not plat_dir.exists():
-                continue
-            json_path = _latest_style_json(plat_dir, style_name)
-            if not json_path:
-                continue
-            try:
-                payload = json.loads(json_path.read_text(encoding="utf-8-sig"))
-            except Exception:
-                continue
-            spus = _extract_spus(payload, platform)
-            if spus:
-                out[style_name][f"{platform}_spus"] = spus
-    return out
-
-
-# ── 数据源 4: ERP 编码（PG 聚合）──────────────────────────────────────────
-
-def load_erp_codes() -> dict[str, list[str]]:
-    """从 daily_style_metrics 聚每个款式的 ERP 编码。"""
-    out: dict[str, list[str]] = {}
-    try:
-        with get_conn() as conn:
-            with conn.cursor() as cur:
-                cur.execute("""
-                    SELECT style_name, ARRAY(
-                        SELECT DISTINCT unnest(erp_style_codes)
-                        FROM daily_style_metrics s
-                        WHERE s.style_name = dm.style_name
-                          AND s.erp_style_codes IS NOT NULL
-                    ) AS codes
-                    FROM (SELECT DISTINCT style_name FROM daily_style_metrics) dm
-                """)
-                for style_name, codes in cur.fetchall():
-                    out[style_name] = [c for c in (codes or []) if c]
-    except Exception as exc:
-        print(f"  [WARN] 读 daily_style_metrics 失败: {exc}")
-    return out
-
-
-# ── 合并 + upsert ──────────────────────────────────────────────────────────
-
-def _merge_records(
-    style_names: Iterable[str],
-    weekly: dict[str, dict],
-    self_skus: dict[str, list[str]],
-    spus_map: dict[str, dict],
-    erp_map: dict[str, list[str]],
-) -> list[dict]:
-    rows: list[dict] = []
-    for name in style_names:
-        rows.append({
-            "style_name": name,
-            "jd_spus": spus_map.get(name, {}).get("jd_spus", []) or [],
-            "dy_spus": spus_map.get(name, {}).get("dy_spus", []) or [],
-            "tm_spus": spus_map.get(name, {}).get("tm_spus", []) or [],
-            "jd_self_skus": self_skus.get(name, []) or [],
-            "erp_codes": erp_map.get(name, []) or [],
-            "bitable_weekly_url": weekly.get(name, {}).get("bitable_weekly_url", ""),
-            "bitable_weekly_base": weekly.get(name, {}).get("bitable_weekly_base", ""),
-            "bitable_weekly_table": weekly.get(name, {}).get("bitable_weekly_table", ""),
-            # bitable_sku_master_* 由人工维护，不动
-        })
-    return rows
-
-
-def load_lifecycle_records() -> list[dict[str, Any]]:
-    """读取生命进程身份与动态目标的统一运行时视图。"""
-
-    return runtime_workflow_configs()
-
-
-def _lifecycle_to_dim_rows(records: list[dict[str, Any]]) -> list[dict[str, Any]]:
-    grouped: dict[str, dict[str, Any]] = {}
-    platform_fields = {
-        "天猫": "tm_spus",
-        "京东旗舰店": "jd_spus",
-        "京东": "jd_spus",
-        "京东自营": "jd_self_skus",
-        "抖音": "dy_spus",
-    }
-    for record in records:
-        name = str(record.get("style_name") or "").strip()
-        if not name:
-            continue
-        row = grouped.setdefault(
-            name,
-            {
-                "style_name": name,
-                "jd_spus": [],
-                "dy_spus": [],
-                "tm_spus": [],
-                "jd_self_skus": [],
-                "erp_codes": [],
-                "bitable_weekly_url": "",
-                "bitable_weekly_base": "",
-                "bitable_weekly_table": "",
-            },
-        )
-        for code in record.get("erp_codes") or []:
-            code = str(code).strip()
-            if code and code not in row["erp_codes"]:
-                row["erp_codes"].append(code)
-        target_url = str(record.get("sales_bitable_url") or "").strip()
-        if target_url and not row["bitable_weekly_url"]:
-            row["bitable_weekly_url"] = target_url
-            match = re.search(
-                r"/base/([A-Za-z0-9]+)\?[^\s]*table=([A-Za-z0-9]+)",
-                target_url,
-            )
-            if match:
-                row["bitable_weekly_base"] = match.group(1)
-                row["bitable_weekly_table"] = match.group(2)
-        field = platform_fields.get(str(record.get("platform") or "").strip())
-        if not field:
-            continue
-        for product_id in record.get("item_ids") or []:
-            product_id = str(product_id).strip()
-            if product_id and product_id not in row[field]:
-                row[field].append(product_id)
-    return list(grouped.values())
-
-
-def upsert_dim_style(conn, records: list[dict]) -> int:
-    if not records:
-        return 0
-    def arr(v):
-        # 空 list / None 都写空数组 {}（PG 数组维度不空），非空 list 原样写
-        if v is None:
-            return []
-        return list(v)
-    payload = [
-        (
-            r["style_name"],
-            arr(r.get("jd_spus")),
-            arr(r.get("dy_spus")),
-            arr(r.get("tm_spus")),
-            arr(r.get("jd_self_skus")),
-            arr(r.get("erp_codes")),
-            r.get("bitable_weekly_url") or None,
-            r.get("bitable_weekly_base") or None,
-            r.get("bitable_weekly_table") or None,
-        )
-        for r in records
-    ]
-    sql = """
-        INSERT INTO dim_style
-            (style_name, jd_spus, dy_spus, tm_spus, jd_self_skus, erp_codes,
-             bitable_weekly_url, bitable_weekly_base, bitable_weekly_table)
-        VALUES %s
-        ON CONFLICT (style_name) DO UPDATE SET
-            jd_spus              = EXCLUDED.jd_spus,
-            dy_spus              = EXCLUDED.dy_spus,
-            tm_spus              = EXCLUDED.tm_spus,
-            jd_self_skus         = EXCLUDED.jd_self_skus,
-            erp_codes            = EXCLUDED.erp_codes,
-            bitable_weekly_url   = EXCLUDED.bitable_weekly_url,
-            bitable_weekly_base  = EXCLUDED.bitable_weekly_base,
-            bitable_weekly_table = EXCLUDED.bitable_weekly_table,
-            -- bitable_sku_master_* 字段：保留人工已填值
-            bitable_sku_master_url   = COALESCE(dim_style.bitable_sku_master_url, EXCLUDED.bitable_sku_master_url),
-            bitable_sku_master_base  = COALESCE(dim_style.bitable_sku_master_base, EXCLUDED.bitable_sku_master_base),
-            bitable_sku_master_table = COALESCE(dim_style.bitable_sku_master_table, EXCLUDED.bitable_sku_master_table),
-            updated_at = NOW()
-    """
-    from psycopg2.extras import execute_values
-    with conn.cursor() as cur:
-        execute_values(cur, sql, payload)
-    return len(payload)
-
-
-def main():
-    ap = argparse.ArgumentParser(description="同步 dim_style (款式 SPU/SKU + 飞书表地址)")
-    ap.add_argument("--dry-run", action="store_true", help="扫描不入库")
-    args = ap.parse_args()
-
-    runtime_records = load_lifecycle_records()
-    records = _lifecycle_to_dim_rows(runtime_records)
-
-    print(f"[INFO] 数据源: product_lifecycle_runtime={len(runtime_records)} 条")
-    print(f"[INFO] 合并款式: {len(records)}")
-    if args.dry_run:
-        for r in records[:5]:
-            print(f"  - {r['style_name']}: jd_spus={r['jd_spus']} jd_self_skus={r['jd_self_skus']} weekly_url={r['bitable_weekly_url'][:60]}")
-        return 0
-
-    with get_conn() as conn:
-        n = upsert_dim_style(conn, records)
-    print(f"[OK] dim_style upsert: {n} 行")
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -1,170 +0,0 @@
-"""同步项目动态配置 → dim_style。
-
-从项目数据库读取配置，按款式聚合 4 个平台 ID 集合 + ERP 编码 + 飞书链接，
-upsert 到 dim_style 的 jd_spus / dy_spus / tm_spus / jd_self_skus / erp_codes / bitable_sku_master_* 字段。
-
-用法:
-    .venv/Scripts/python.exe db/sync_sku_master.py            # 全量拉 + 入库
-    .venv/Scripts/python.exe db/sync_sku_master.py --dry-run  # 拉数据 + 显示不入库
-
-"""
-
-from __future__ import annotations
-
-import argparse
-import re
-import sys
-from collections import defaultdict
-from pathlib import Path
-
-from gyxx_flow.adapters.workflow_config import runtime_workflow_configs
-
-PROJECT_ROOT = Path(__file__).resolve().parent.parent
-sys.path.insert(0, str(PROJECT_ROOT))
-
-from db import get_conn  # noqa: E402
-
-
-def fetch_sku_master_records() -> list[dict]:
-    """从项目数据库读取全部启用的动态配置。"""
-    return runtime_workflow_configs()
-
-
-def _split_ids(value) -> list[str]:
-    """兼容数据库数组和旧的逗号、换行分隔值。"""
-    if value is None:
-        return []
-    if isinstance(value, (list, tuple, set)):
-        return [str(item).strip() for item in value if str(item).strip()]
-    s = str(value)
-    # 先按换行 split（京东自营风格），再按逗号 split（其它平台风格）
-    parts: list[str] = []
-    for line in s.splitlines():
-        for piece in line.split(","):
-            piece = piece.strip()
-            if piece:
-                parts.append(piece)
-    return parts
-
-
-def aggregate(records: list[dict]) -> dict[str, dict]:
-    """按款式聚合: jd_spus / dy_spus / tm_spus / jd_self_skus / erp_codes / sku_master_link。"""
-    by_style: dict[str, dict] = defaultdict(lambda: {
-        "jd_spus": [], "dy_spus": [], "tm_spus": [], "jd_self_skus": [],
-        "erp_codes": [], "sku_master_url": "", "sku_master_base": "",
-        "sku_master_table": "",
-    })
-    for r in records:
-        plat_field = r.get("platform")
-        plat_list = plat_field if isinstance(plat_field, list) else ([plat_field] if plat_field else [])
-        plat_s = ",".join(str(p) for p in plat_list)
-        name = str(r.get("style_name") or "").strip()
-        if not name:
-            continue
-        ids_field = r.get("item_ids")
-        id_list = _split_ids(ids_field)
-        erp_field = r.get("erp_codes")
-        erp_list = _split_ids(erp_field)
-        link_field = str(r.get("sales_bitable_url") or "")
-
-        if "京东自营" in plat_s:
-            by_style[name]["jd_self_skus"].extend(id_list)
-        elif "京东" in plat_s:
-            by_style[name]["jd_spus"].extend(id_list)
-        if "抖音" in plat_s:
-            by_style[name]["dy_spus"].extend(id_list)
-        if "天猫" in plat_s:
-            by_style[name]["tm_spus"].extend(id_list)
-        for ec in erp_list:
-            if ec not in by_style[name]["erp_codes"]:
-                by_style[name]["erp_codes"].append(ec)
-
-        if link_field and not by_style[name]["sku_master_url"]:
-            # 解析 https://.../base/XXXXX?table=YYYYY[&view=ZZZZ]
-            m = re.search(
-                r"(https?://[^/]+/base/([A-Za-z0-9]+)\?[^\s]*table=([A-Za-z0-9]+)[^\s]*)",
-                link_field,
-            )
-            if m:
-                by_style[name]["sku_master_url"] = m.group(1)
-                by_style[name]["sku_master_base"] = m.group(2)
-                by_style[name]["sku_master_table"] = m.group(3)
-            else:
-                # fallback: 整段存
-                by_style[name]["sku_master_url"] = link_field
-    return by_style
-
-
-def upsert_from_sku_master(conn, by_style: dict[str, dict]) -> int:
-    """upsert SKU 主表数据到 dim_style (只更新 SPU/SKU/ERP/sku_master_* 字段，其它保留)。"""
-    if not by_style:
-        return 0
-    from psycopg2.extras import execute_values
-
-    def arr(v):
-        if v is None:
-            return []
-        return list(v)
-
-    payload = [
-        (
-            name,
-            arr(s.get("jd_spus")),
-            arr(s.get("dy_spus")),
-            arr(s.get("tm_spus")),
-            arr(s.get("jd_self_skus")),
-            arr(s.get("erp_codes")),
-            s.get("sku_master_url") or None,
-            s.get("sku_master_base") or None,
-            s.get("sku_master_table") or None,
-        )
-        for name, s in by_style.items()
-    ]
-    sql = """
-        INSERT INTO dim_style
-            (style_name, jd_spus, dy_spus, tm_spus, jd_self_skus, erp_codes,
-             bitable_sku_master_url, bitable_sku_master_base, bitable_sku_master_table)
-        VALUES %s
-        ON CONFLICT (style_name) DO UPDATE SET
-            jd_spus              = EXCLUDED.jd_spus,
-            dy_spus              = EXCLUDED.dy_spus,
-            tm_spus              = EXCLUDED.tm_spus,
-            jd_self_skus         = EXCLUDED.jd_self_skus,
-            erp_codes            = EXCLUDED.erp_codes,
-            bitable_sku_master_url   = EXCLUDED.bitable_sku_master_url,
-            bitable_sku_master_base  = EXCLUDED.bitable_sku_master_base,
-            bitable_sku_master_table = EXCLUDED.bitable_sku_master_table,
-            updated_at = NOW()
-    """
-    with conn.cursor() as cur:
-        execute_values(cur, sql, payload)
-    return len(payload)
-
-
-def main():
-    ap = argparse.ArgumentParser(description="同步项目动态配置 → dim_style")
-    ap.add_argument("--dry-run", action="store_true")
-    args = ap.parse_args()
-
-    print("[STEP] 读取项目动态配置 ...")
-    records = fetch_sku_master_records()
-    print(f"[INFO] 拉取 {len(records)} 条记录")
-    by_style = aggregate(records)
-    print(f"[INFO] 聚合 {len(by_style)} 个款式")
-
-    for name in sorted(by_style.keys())[:5]:
-        s = by_style[name]
-        print(f"  - {name}: jd={s['jd_spus']} dy={s['dy_spus']} tm={s['tm_spus']} jd_self={s['jd_self_skus']} erp={s['erp_codes']}")
-
-    if args.dry_run:
-        print("[DRY] 不入库")
-        return 0
-
-    with get_conn() as conn:
-        n = upsert_from_sku_master(conn, by_style)
-    print(f"[OK] dim_style upsert: {n} 行")
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -249,7 +249,7 @@ def _print_preflight() -> None:
        return
    collectable = daily["dy"]["collectable"]
    skipped = daily["dy"]["skipped"]
-    print(f"[DY] 飞书表格 待采 {len(collectable)} 款, 跳过 {len(skipped)} 款 (款式级缺 erp/sales,或 DY 缺商品链接ID)")
+    print(f"[DY] 飞书表格 待采 {len(collectable)} 款, 跳过 {len(skipped)} 款 (款式级缺 erp 或已由同系列归集,或 DY 缺商品链接ID)")
    for s in skipped:
        print(f"  - {s['style']}: {s['reason']}")

@@ -3258,15 +3258,13 @@ def _page_automation(page) -> None:
        for product in summary.get("products", [])
    ]

-    # 评论采集挂在默认流程末尾, 让 daily --target-date 也能采到评价.
-    # _collect_comment_summary 内部会 _open_comment_detail_page 跳转到评论页,
-    # 采完好/中/差后覆盖 评论汇总_<YYYYMMDD>.json. 幂等: 同 order_id 重采覆盖.
-    try:
-        COMMENTS_SUMMARY = _collect_comment_summary(page)
-        _save_comment_summary(COMMENTS_SUMMARY)
-    except Exception as exc:
-        print(f"[WARN] 评论采集失败 (不影响单品数据已落盘): {exc}")
-        COMMENTS_SUMMARY = {}
+    # 评论采集已于 2026-11-01 从默认流程摘除：
+    # 它只产出本地 评论汇总_<YYYYMMDD>.json，而唯一消费方 product_reviews 已下线
+    # （差评改由 shop.dy_negative_reviews.daily 独立采集，消费方读 v_negative_reviews）。
+    # 继续跑等于每天多一次浏览器跳转 + 一份无人读取的文件。
+    # 需要历史评价样本时用 --reviews-only 显式触发，不要恢复默认流程。
+    print("[SKIP] 评论采集已在默认流程下线（如需补采请用 --reviews-only）")
+    COMMENTS_SUMMARY = {}

    _results_saved = True

@@ -75,9 +75,20 @@ def load_erp_metrics(platform, style_name, target_date=None):
            failed = [c for c in code_results if not c.get("ok") or c.get("error")]
            if code_results and failed:
                # 任一编码失败都会造成 ERP 合计不完整，整款回退到平台原值。
+                # 但 TM/DY 没有平台销量字段，整款回退会被兜底成 0，把整款销量抹掉
+                # （如 10244 allsku iframe 偶发 timeout → 宙斯天猫销量变 0）。
+                # 所以同时给出"已成功编码合计"，供无平台原值时使用。
                return {
                    "sales": None,
                    "refund_orders": None,
+                    "partial_sales": (
+                        sum(int(parse_number(c.get("yesterday_sales"))) for c in ok_results)
+                        if ok_results else None
+                    ),
+                    "partial_refunds": (
+                        sum(int(parse_number(c.get("yesterday_returns"))) for c in ok_results)
+                        if ok_results else None
+                    ),
                    "source": str(json_path),
                    "erp_style_codes": payload.get("erp_style_codes", []),
                    "all_failed": True,
@@ -114,9 +125,17 @@ def apply_erp_override(platform, style_name, visitors, cart_users, target_date=N
    c = int(parse_number(cart_users))

    if not erp or erp.get("all_failed"):
-        # ERP 拿不到数据 → 用平台原值（如果有），否则 0
-        sales = int(parse_number(platform_sales)) if platform_sales is not None else 0
-        refunds = int(parse_number(platform_refunds)) if platform_refunds is not None else 0
+        if platform_sales is not None:
+            # 平台自己有销量字段（JD/DY）→ 用平台原值，最接近真相
+            sales = int(parse_number(platform_sales))
+            refunds = int(parse_number(platform_refunds)) if platform_refunds is not None else 0
+        elif erp and erp.get("partial_sales") is not None:
+            # 平台没有销量字段（TM）→ 用已成功编码的合计，避免整款归零
+            sales = int(erp["partial_sales"])
+            refunds = int(erp.get("partial_refunds") or 0)
+        else:
+            sales = 0
+            refunds = 0
        return {
            "销量": sales,
            "访客数": v,
@@ -127,6 +146,7 @@ def apply_erp_override(platform, style_name, visitors, cart_users, target_date=N
            "退款率": format_percent(refunds, sales),
            "erp_override": erp,
            "erp_unavailable": True,
+            "erp_partial": bool(erp and erp.get("partial_sales") is not None and platform_sales is None),
            "failed_codes": (erp or {}).get("failed_codes", []),
        }

@@ -402,11 +402,106 @@ def _unique(values: Iterable[str]) -> list[str]:
    return list(dict.fromkeys(value for value in values if value))


+ROLLUP_CONFIG_PATH = Path(__file__).resolve().parent / "config" / "ad_cost_rollup.json"
+_rollup_cache: dict[str, dict[str, str]] | None = None
+
+
+def load_rollup_map(platform: str = "jd") -> dict[str, str]:
+    """读取父款式汇总口径（子款式 -> 父款式），缺失或非法时返回空映射。
+
+    ``宙斯3`` / ``宙斯Z Air`` 这类子款式在京准通报表里名称更长，会先命中
+    名称匹配；父款式 ``宙斯`` 的电商费比需要把整个商品族的消耗算进来，
+    因此汇总口径放在 ``config/ad_cost_rollup.json`` 里显式声明。
+    """
+
+    global _rollup_cache
+    if _rollup_cache is None:
+        try:
+            payload = json.loads(ROLLUP_CONFIG_PATH.read_text(encoding="utf-8"))
+        except (OSError, ValueError):
+            payload = {}
+        _rollup_cache = {
+            key: {str(child): str(parent) for child, parent in value.items()}
+            for key, value in payload.items()
+            if isinstance(value, dict)
+        }
+    return dict(_rollup_cache.get(platform, {}))
+
+
+def _apply_rollup(result: dict[str, Any], rollup: Mapping[str, str]) -> dict[str, Any]:
+    style_name = result.get("style_name")
+    parent = rollup.get(style_name) if style_name else None
+    if not parent or parent == style_name:
+        return result
+    return {
+        **result,
+        "style_name": parent,
+        "reason": f"{result.get('reason', '')} -> 按父款式汇总「{parent}」",
+    }
+
+
+def _spu_owner_map(
+    style_catalog: Sequence[Mapping[str, Any]],
+) -> dict[str, str]:
+    """显式 jd_spus 归属：SPU 挂在哪个款式上，就优先算哪个款式。
+
+    ``dim_style.jd_spus`` 是人工维护的商品→款式权威映射。SKU 名称匹配只是
+    fallback：同一商品族的子款式（如「宙斯3」「宙斯Z Air」）名称更长、更容易
+    命中最长匹配，但父款式已经在 jd_spus 里声明了这些 SPU 时，消耗必须归到
+    jd_spus 声明的款式，否则父款式的广告费比恒为空。
+    """
+
+    owners: dict[str, str] = {}
+    conflicts: set[str] = set()
+    for item in style_catalog:
+        style_name = _text(item.get("style_name"))
+        if not style_name:
+            continue
+        for spu in _spu_values(item.get("jd_spus")):
+            existing = owners.get(spu)
+            if existing is not None and existing != style_name:
+                conflicts.add(spu)
+                continue
+            owners[spu] = style_name
+    for spu in conflicts:
+        owners.pop(spu, None)
+    return owners
+
+
+def _apply_spu_owner(
+    result: dict[str, Any], spu_owner: Mapping[str, str]
+) -> dict[str, Any]:
+    spu_id = _text(result.get("spu_id"))
+    if not spu_id:
+        return result
+    owner = spu_owner.get(spu_id)
+    if not owner or owner == result.get("style_name"):
+        return result
+    return {
+        **result,
+        "style_name": owner,
+        "reason": f"{result.get('reason', '')} -> dim_style.jd_spus 归属「{owner}」",
+    }
+
+
 def resolve_sku_match(
    sku_name: str,
    style_catalog: Sequence[Mapping[str, Any]],
    full_records: Sequence[Mapping[str, Any]],
    spu_catalog: Sequence[Mapping[str, Any]] = (),
+) -> dict[str, Any]:
+    """Resolve one SKU to a style, then honour explicit ``dim_style.jd_spus``."""
+
+    result = _resolve_sku_match(sku_name, style_catalog, full_records, spu_catalog)
+    result = _apply_spu_owner(result, _spu_owner_map(style_catalog))
+    return _apply_rollup(result, load_rollup_map("jd"))
+
+
+def _resolve_sku_match(
+    sku_name: str,
+    style_catalog: Sequence[Mapping[str, Any]],
+    full_records: Sequence[Mapping[str, Any]],
+    spu_catalog: Sequence[Mapping[str, Any]] = (),
 ) -> dict[str, Any]:
    """Return a conservative style/SPU match for one SKU name.

@@ -671,6 +766,15 @@ def apply_full_style_matches(
                        "full-site SPU has no dim_style.jd_spus or name match"
                    )

+        rollup = load_rollup_map("jd")
+        resolved_style = item.get("style_name")
+        parent_style = rollup.get(resolved_style) if resolved_style else None
+        if parent_style and parent_style != resolved_style:
+            item["style_name"] = parent_style
+            item["match_reason"] = (
+                f"{item['match_reason']} -> 按父款式汇总「{parent_style}」"
+            )
+
        metrics = dict(item.get("metrics") or {})
        metrics["match"] = {
            "status": item.get("match_status", "direct"),
@@ -1,331 +0,0 @@
-"""
-扫描 data/{jd,dy}单品数据/评价*.json (或 评论汇总_*.json),
-归一化后写入 product_reviews 表 (ON CONFLICT 全量覆盖)。
-
-DY 评价入库时会反查 dim_style.dy_spus 填 style_name;
-JD 评价缺 SPU->款式映射, style_name 留 NULL (待 dim_jd_sku_spu 映射表补齐后再做)。
-
-用法:
-    .venv\\Scripts\\python.exe import_product_reviews.py                    # 全量 (默认只 dy)
-    .venv/Scripts/python.exe import_product_reviews.py --date 2026-07-13   # 单日
-    .venv/Scripts/python.exe import_product_reviews.py --from 2026-06-01 --to 2026-07-13
-    .venv/Scripts/python.exe import_product_reviews.py --platforms dy
-    .venv/Scripts/python.exe import_product_reviews.py --dry-run          # 只解析不入库
-
-源 JSON 文件:
-    JD: data/jd单品数据/评价汇总_<YYYY-MM-DD>.json
-        结构: {platform, shop, collect_date, comment_date_range, captured_at,
-               total_count, ratings: {好评/中评/差评: {collected_count, comments:[{
-                 user_name, score, tags, content, time, order_id,
-                 sku_name, sku_id, sku_size, rating}]}}}
-    DY: data/dy单品数据/评论汇总_<YYYYMMDD>.json
-        结构: {shop, period, captured_at, url, total_collected_count,
-               ratings: {好评/中评/差评: {expected_count, collected_count, comments:[{
-                 rating, order_id, comment_time, comment_text, product_name,
-                 product_id, spec_name, spec_id, raw_text}]}}}
-"""
-from __future__ import annotations
-
-import argparse
-import json
-import os
-import re
-import sys
-from datetime import date, datetime
-from pathlib import Path
-
-from runtime_paths import RAW_DATA_ROOT, artifact_relative_path
-from typing import Any, Iterable
-
-PROJECT_ROOT = Path(__file__).resolve().parent
-sys.path.insert(0, str(PROJECT_ROOT))
-
-from db import get_conn, upsert_product_reviews  # noqa: E402
-
-
-DATA_ROOT = RAW_DATA_ROOT
-
-PLATFORM_DIRS = {
-    "jd": "jd单品数据",
-    "dy": "dy单品数据",
-}
-
-# 文件名模式: JD 用 评价汇总_2026-07-14.json, DY 用 评论汇总_20260604.json
-FILE_PATTERNS = {
-    "jd": re.compile(r"^评价汇总_(\d{4}-\d{2}-\d{2})\.json$"),
-    "dy": re.compile(r"^评论汇总_(\d{8})\.json$"),
-}
-
-VALID_PLATFORMS = tuple(PLATFORM_DIRS.keys())
-
-RATING_VALUES = ("好评", "中评", "差评")
-
-
-def _parse_jd_date(s: str) -> date:
-    return datetime.strptime(s, "%Y-%m-%d").date()
-
-
-def _parse_dy_date(s: str) -> date:
-    return datetime.strptime(s, "%Y%m%d").date()
-
-
-def _parse_comment_date(s: str) -> date | None:
-    """从评价时间字符串抽日期部分。
-
-    JD time 格式: '2026-07-13 12:34:56' 或 '2026-07-13'
-    DY comment_time 格式: '2026/06/05 16:39:18' 或 '2026/06/05'
-    """
-    if not s:
-        return None
-    s = s.strip()
-    for fmt in ("%Y-%m-%d %H:%M:%S", "%Y-%m-%d", "%Y/%m/%d %H:%M:%S", "%Y/%m/%d"):
-        try:
-            return datetime.strptime(s, fmt).date()
-        except ValueError:
-            continue
-    m = re.match(r"^(\d{4})[-/](\d{1,2})[-/](\d{1,2})", s)
-    if m:
-        try:
-            return date(int(m.group(1)), int(m.group(2)), int(m.group(3)))
-        except ValueError:
-            return None
-    return None
-
-
-def _to_smallint(v: Any) -> int | None:
-    if v is None or v == "":
-        return None
-    try:
-        return int(float(str(v)))
-    except (ValueError, TypeError):
-        return None
-
-
-def _normalize_jd(payload: dict, source_file: str, collect_date: date,
-                  style_map: dict[str, str] | None = None) -> list[dict]:
-    """JD JSON -> product_reviews 记录列表
-
-    JD 评价表只有 sku_id (无 SPU), 暂不反查 style_name, 留 NULL。
-    """
-    records = []
-    for rating in RATING_VALUES:
-        bucket = (payload.get("ratings") or {}).get(rating) or {}
-        for c in bucket.get("comments") or []:
-            comment_date = _parse_comment_date(c.get("time", ""))
-            records.append({
-                "platform": "jd",
-                "collect_date": collect_date,
-                "comment_date": comment_date,
-                "style_name": None,  # JD sku_id -> SPU 映射暂缺, 待 dim_jd_sku_spu 补齐
-                "rating": rating,
-                "score": _to_smallint(c.get("score")),
-                "content": (c.get("content") or "").strip() or None,
-                "tags": [t for t in (c.get("tags") or []) if t] or None,
-                "user_name": c.get("user_name") or None,
-                "order_id": str(c.get("order_id") or "").strip(),
-                "product_id": None,  # JD 评价表只有 sku_id, SPU 留空
-                "product_name": c.get("sku_name") or None,
-                "sku_id": c.get("sku_id") or None,
-                "sku_name": c.get("sku_name") or None,
-                "sku_size": c.get("sku_size") or None,
-                "raw_data": c,
-                "source_file": source_file,
-            })
-    return records
-
-
-def _normalize_dy(payload: dict, source_file: str, collect_date: date,
-                  style_map: dict[str, str] | None = None) -> list[dict]:
-    """DY JSON -> product_reviews 记录列表
-
-    style_map: {product_id: style_name}, 从 dim_style.dy_spus 构建。
-    匹配不上时 style_name = None (表示 dim_style 缺该商品 ID 维护)。
-    """
-    style_map = style_map or {}
-    records = []
-    for rating in RATING_VALUES:
-        bucket = (payload.get("ratings") or {}).get(rating) or {}
-        for c in bucket.get("comments") or []:
-            comment_date = _parse_comment_date(c.get("comment_time", ""))
-            product_id = str(c.get("product_id") or "").strip()
-            records.append({
-                "platform": "dy",
-                "collect_date": collect_date,
-                "comment_date": comment_date,
-                "style_name": style_map.get(product_id),
-                "rating": rating,
-                "score": None,  # DY 无星级
-                "content": (c.get("comment_text") or "").strip() or None,
-                "tags": None,
-                "user_name": None,  # DY 评论无独立用户名字段
-                "order_id": str(c.get("order_id") or "").strip(),
-                "product_id": product_id or None,
-                "product_name": c.get("product_name") or None,
-                "sku_id": c.get("spec_id") or None,
-                "sku_name": c.get("spec_name") or None,
-                "sku_size": None,
-                "raw_data": c,
-                "source_file": source_file,
-            })
-    return records
-
-
-NORMALIZERS = {
-    "jd": _normalize_jd,
-    "dy": _normalize_dy,
-}
-
-
-def _load_dy_style_map() -> dict[str, str]:
-    """从 dim_style.dy_spus 构建 {product_id: style_name} 映射。"""
-    mapping: dict[str, str] = {}
-    with get_conn() as conn:
-        with conn.cursor() as cur:
-            cur.execute("SELECT style_name, dy_spus FROM dim_style WHERE dy_spus IS NOT NULL")
-            for style_name, spus in cur.fetchall():
-                if not spus:
-                    continue
-                for pid in spus:
-                    mapping[str(pid)] = style_name
-    return mapping
-
-
-def _discover_files(platforms: list[str], date_from: date | None, date_to: date | None) -> list[tuple[str, date, Path]]:
-    """返回 [(platform, collect_date, file_path), ...]"""
-    out = []
-    for plat in platforms:
-        plat_dir = DATA_ROOT / PLATFORM_DIRS[plat]
-        if not plat_dir.exists():
-            continue
-        pattern = FILE_PATTERNS[plat]
-        for path in sorted(plat_dir.iterdir()):
-            m = pattern.match(path.name)
-            if not m:
-                continue
-            try:
-                collect_date = (_parse_jd_date if plat == "jd" else _parse_dy_date)(m.group(1))
-            except ValueError:
-                continue
-            if date_from and collect_date < date_from:
-                continue
-            if date_to and collect_date > date_to:
-                continue
-            out.append((plat, collect_date, path))
-    return out
-
-
-def _load_and_normalize(plat: str, collect_date: date, path: Path,
-                        style_map: dict[str, str] | None = None) -> list[dict]:
-    with path.open(encoding="utf-8") as f:
-        payload = json.load(f)
-    rel = artifact_relative_path(path)
-    return NORMALIZERS[plat](payload, rel, collect_date, style_map=style_map)
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description="扫描 JD/DY 评价 JSON 入库 product_reviews")
-    parser.add_argument("--date", help="只导入指定采集日期 (YYYY-MM-DD); JD 用 - 分隔, DY 无分隔也可")
-    parser.add_argument("--from", dest="date_from", help="起始采集日期 (YYYY-MM-DD)")
-    parser.add_argument("--to", dest="date_to", help="结束采集日期 (YYYY-MM-DD)")
-    parser.add_argument("--platforms", default="dy", help="平台, 逗号分隔 (默认 dy; JD 待 SPU 映射补齐)")
-    parser.add_argument("--dry-run", action="store_true", help="只解析不入库")
-    args = parser.parse_args()
-
-    platforms = [p.strip() for p in args.platforms.split(",") if p.strip()]
-    for p in platforms:
-        if p not in VALID_PLATFORMS:
-            print(f"[ERR] 未知平台: {p} (允许: {VALID_PLATFORMS})")
-            return 2
-
-    date_from = _parse_single(args.date_from) if args.date_from else None
-    date_to = _parse_single(args.date_to) if args.date_to else None
-    if args.date:
-        d = _parse_single(args.date)
-        date_from = d
-        date_to = d
-
-    files = _discover_files(platforms, date_from, date_to)
-    if not files:
-        print(f"[WARN] 未找到匹配的 JSON 文件 (platforms={platforms}, from={date_from}, to={date_to})")
-        return 0
-
-    # 预加载 DY 款式映射 (dim_style.dy_spus -> {product_id: style_name})
-    style_map: dict[str, str] = {}
-    if "dy" in platforms and not args.dry_run:
-        style_map = _load_dy_style_map()
-        print(f"[INFO] dim_style.dy_spus 映射: {len(style_map)} 个 product_id -> style_name")
-    elif "dy" in platforms and args.dry_run:
-        # dry-run 也加载, 便于预览匹配情况
-        style_map = _load_dy_style_map()
-        print(f"[INFO] (dry-run) dim_style.dy_spus 映射: {len(style_map)} 个 product_id -> style_name")
-
-    total = 0
-    skipped = 0
-    by_plat: dict[str, int] = {p: 0 for p in platforms}
-    by_rating: dict[str, int] = {"好评": 0, "中评": 0, "差评": 0}
-    matched_styles: set[str] = set()
-
-    for plat, collect_date, path in files:
-        try:
-            records = _load_and_normalize(plat, collect_date, path, style_map=style_map)
-        except Exception as e:
-            print(f"[ERR] 解析 {path}: {e}")
-            continue
-        # 去重: 同 order_id 多条 (理论上 UNIQUE 会处理, 这里先在内存去重避免执行 batch fail)
-        seen = set()
-        deduped = []
-        for r in records:
-            if not r["order_id"]:
-                skipped += 1
-                continue
-            if r["order_id"] in seen:
-                skipped += 1
-                continue
-            seen.add(r["order_id"])
-            deduped.append(r)
-            by_rating[r["rating"]] = by_rating.get(r["rating"], 0) + 1
-            if r.get("style_name"):
-                matched_styles.add(r["style_name"])
-
-        plat_style_matched = sum(1 for r in deduped if r.get("style_name"))
-        print(f"[{plat}] {collect_date}  {path.name}  -> {len(deduped)} 条 (款式匹配 {plat_style_matched}/{len(deduped)})")
-
-        if args.dry_run:
-            if deduped:
-                sample = deduped[0].copy()
-                sample["raw_data"] = "(省略)"
-                print(f"  (dry-run, 不入库) 样本: {sample}")
-        else:
-            with get_conn() as conn:
-                n = upsert_product_reviews(conn, deduped)
-            print(f"  -> upsert {n} rows")
-
-        total += len(deduped)
-        by_plat[plat] = by_plat.get(plat, 0) + len(deduped)
-
-    print()
-    print("=" * 60)
-    print(f"  总计: {total} 条 (跳过 {skipped} 条无 order_id)")
-    for p, n in by_plat.items():
-        print(f"  {p}: {n} 条")
-    print("  按评级:")
-    for r, n in by_rating.items():
-        print(f"    {r}: {n} 条")
-    if matched_styles:
-        print(f"  匹配到款式 ({len(matched_styles)} 款): {', '.join(sorted(matched_styles))}")
-    print("=" * 60)
-    return 0
-
-
-def _parse_single(s: str) -> date:
-    s = s.strip()
-    for fmt in ("%Y-%m-%d", "%Y%m%d"):
-        try:
-            return datetime.strptime(s, fmt).date()
-        except ValueError:
-            continue
-    raise SystemExit(f"无法解析日期: {s}")
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -1,9 +1,11 @@
-"""将万相台商品广告 CSV 幂等写入产品生命进程读取的字段。
+"""将万相台商品广告 CSV 幂等写入天猫商品投放日报表。

-真实数据契约是 ``product_daily_metrics.raw_data.tm_ad_metrics``，不是一张
-独立的广告表。导入时只替换该 JSON 节点，保留同一商品日期上已经存在的
-经营日报字段；同时把款式匹配结果写入该节点的 ``style_name`` /
-``style_match_method``，数据-hub 后端据此计算广告成交金额、投产比和电商费比。
+数据契约是独立表 ``tm_product_ad_daily_metrics``，每日每商品一行：花费、直接/
+间接/归因成交金额与投产比走标量列，报表其余指标原样存 ``metrics`` JSONB。
+款式按 ``dim_style.tm_spus`` 的商品 ID 精确匹配，未命中再按商品名称精确包含
+兜底，结果写进 ``style_name`` / ``style_match_method``，data-hub 后端据此计算
+广告成交金额、投产比和电商费比。广告与经营日报分表存储，经营数据重新同步不会
+覆盖投放数据。

 默认只解析校验；显式传入 ``--execute`` 才写 PostgreSQL。
 """
@@ -25,8 +27,9 @@ PROJECT_ROOT = Path(__file__).resolve().parent
 sys.path.insert(0, str(PROJECT_ROOT))

 from db import (  # noqa: E402
+    ensure_tmall_product_ad_daily_table,
    get_conn,
-    upsert_tmall_product_ad_raw_metrics,
+    upsert_tmall_product_ad_daily_metrics,
 )

 AMOUNT_COLUMNS = {
@@ -327,8 +330,9 @@ def main(argv: list[str] | None = None) -> int:
        return 0

    with get_conn() as conn:
-        written = upsert_tmall_product_ad_raw_metrics(conn, records)
-    print(f"[OK] 已写入 product_daily_metrics.raw_data.tm_ad_metrics：{written} 行")
+        ensure_tmall_product_ad_daily_table(conn)
+        written = upsert_tmall_product_ad_daily_metrics(conn, records)
+    print(f"[OK] 已写入 tm_product_ad_daily_metrics：{written} 行")
    return 0


@@ -11,7 +11,6 @@ from runtime_paths import EXPORTS_DATA_ROOT, LOG_ROOT, RAW_DATA_ROOT

 PROJECT_ROOT = Path(__file__).resolve().parent
 DATA_ROOT = RAW_DATA_ROOT
-STYLE_MAP_PATH = PROJECT_ROOT / "bitable_style_map.json"


 def read_json(path):
@@ -283,10 +282,15 @@ def main(argv: list[str] | None = None) -> int:
    try:
        style_map = get_bitable_style_map()
        if not style_map:
-            raise RuntimeError("飞书动态配置为空")
+            raise RuntimeError("dim_style 未返回任何带销量表地址的款式")
    except Exception as exc:
-        print(f"[WARN] 动态读取销量表配置失败，使用本地文件回退: {exc}")
-        style_map = read_json(STYLE_MAP_PATH)
+        # 飞书地址唯一来源是 dim_style：读不到就失败退出，不再回退本地 JSON，
+        # 否则库与本地文件不一致时会静默写到错误的表。
+        print(
+            f"[FAIL] 销量表配置读取失败，拒绝回退本地 JSON: {exc}",
+            file=sys.stderr,
+        )
+        return 1
    if args.source == "db":
        print(f"[INFO] source=db: 查 weekly_style_metrics week_end={args.date}")
    records = load_records(args.date, args.record_file, source=args.source)
@@ -72,6 +72,59 @@ OUTPUT_DIR = RAW_DATA_ROOT / "主图信息"
 DEBUG_DIR = TMP_ROOT / "debug" / "jd_main_image_snapshot"
 DEBUG = os.getenv("JD_MAIN_IMAGE_DEBUG", "0") == "1"

+# 京东商家后台门户（登录成功后应停留在这些主机上）。
+JD_PORTAL_HOSTS = frozenset({"shop.jd.com", "wares-jdm.jd.com"})
+# 京东风控/账号安全认证页：出现即代表登录态被判定异常，必须人工完成短信/人脸验证。
+JD_SECURITY_CHALLENGE_HOSTS = frozenset(
+    {"aq.jd.com", "safe.jd.com", "verify.jd.com", "risk.jd.com"}
+)
+# 密码登录后等待后台就绪的上限（京东会先加载登录页再异步跳转）。
+LOGIN_SETTLE_TIMEOUT_MS = 120_000
+# 无头模式下登录页长时间不跳转即视为登录失败，避免白等到超时。
+HEADLESS_LOGIN_STALL_SECONDS = 25.0
+
+
+def _url_summary(url: str) -> str:
+    """只保留协议/主机/路径。
+
+    京东认证页的查询串里带一次性 token（签名、trackId、returnUrl），
+    不能原样写进日志、运行记录和飞书告警。
+    """
+
+    parsed = urlparse(str(url))
+    if not parsed.hostname:
+        return "<unknown>"
+    return f"{parsed.scheme}://{parsed.netloc}{parsed.path}"
+
+
+def _security_challenge_host(url: str) -> str:
+    """返回命中的京东安全认证主机名，未命中返回空串。"""
+
+    hostname = (urlparse(str(url)).hostname or "").casefold()
+    return hostname if hostname in JD_SECURITY_CHALLENGE_HOSTS else ""
+
+
+def _security_challenge_error(url: str) -> RuntimeError:
+    """登录态被京东风控拦截时的统一可执行错误。"""
+
+    host = _security_challenge_host(url) or "京东安全页"
+    return RuntimeError(
+        f"京东触发安全认证（{host}）：登录态被判定需要人工验证，"
+        "账号密码登录无法自动完成短信/图形验证。请在能打开浏览器的机器上执行 "
+        "`uv run gyxx accounts login jd-shop --force` 完成验证，再重跑本步骤。"
+        f" 当前页面: {_url_summary(url)}"
+    )
+
+
+def _is_authenticated_portal(url: str) -> bool:
+    """当前 URL 是否已是可用的京东商家后台页面。"""
+
+    parsed = urlparse(str(url))
+    hostname = (parsed.hostname or "").casefold()
+    if hostname not in JD_PORTAL_HOSTS:
+        return False
+    return not _looks_like_login_url(url) and not _security_challenge_host(url)
+
 # 直接使用新版商品明细页，避免旧版别名在批量采集过程中偶发 ERR_CONNECTION_CLOSED。
 DETAIL_URL = "https://jdsz.jd.com/szweb/view/product/productDetail.html"
 SEARCH_OVERVIEW_URL = "https://jdsz.jd.com/squidweb/view/flow/searchOverview.html"
@@ -240,10 +293,13 @@ def _wait_for_product_detail(page, timeout_ms: int = 30_000) -> None:
        except Exception:
            last_url = ""

+        if _security_challenge_host(last_url):
+            raise _security_challenge_error(last_url)
+
        if _looks_like_login_url(last_url):
            raise RuntimeError(
                "京东商智登录态已失效：商品明细页跳转到登录页 "
-                f"{last_url}；请先刷新 jd-shop 账号登录态后再运行"
+                f"{_url_summary(last_url)}；请先刷新 jd-shop 账号登录态后再运行"
            )

        # A transient non-JD URL can be observed while the new tab settles;
@@ -261,14 +317,16 @@ def _wait_for_product_detail(page, timeout_ms: int = 30_000) -> None:
        except Exception:
            time.sleep(0.5)

+    if _security_challenge_host(last_url):
+        raise _security_challenge_error(last_url)
    if _looks_like_login_url(last_url):
        raise RuntimeError(
            "京东商智登录态已失效：商品明细页跳转到登录页 "
-            f"{last_url}；请先刷新 jd-shop 账号登录态后再运行"
+            f"{_url_summary(last_url)}；请先刷新 jd-shop 账号登录态后再运行"
        )
    raise RuntimeError(
        "京东商品明细页未加载完成，未找到商品ID搜索框；"
-        f"当前 URL: {last_url or '<unknown>'}"
+        f"当前 URL: {_url_summary(last_url)}"
    )


@@ -287,10 +345,12 @@ def _wait_for_search_overview(page, timeout_ms: int = 30_000) -> None:
            last_url = str(page.url)
        except Exception:
            last_url = ""
+        if _security_challenge_host(last_url):
+            raise _security_challenge_error(last_url)
        if _looks_like_login_url(last_url):
            raise RuntimeError(
                "京东商智登录态已失效：搜索数据概览跳转到登录页 "
-                f"{last_url}；请先刷新 jd-shop 账号登录态后再运行"
+                f"{_url_summary(last_url)}；请先刷新 jd-shop 账号登录态后再运行"
            )
        try:
            if page.locator("input.jmt-core-input").first.is_visible(timeout=800):
@@ -303,7 +363,7 @@ def _wait_for_search_overview(page, timeout_ms: int = 30_000) -> None:
            time.sleep(0.5)
    raise RuntimeError(
        "京东搜索数据概览页未加载完成，未找到 SKU 筛选输入框；"
-        f"当前 URL: {last_url or '<unknown>'}"
+        f"当前 URL: {_url_summary(last_url)}"
    )


@@ -698,6 +758,30 @@ def _ensure_page(context, page):
    return context.new_page()


+def _page_reports_no_data(page) -> bool:
+    """商品明细表是否明确渲染了「暂无数据」空态。
+
+    商智商品明细只展示窗口内有曝光/成交的商品；某款当日零曝光时查询结果
+    为空，这是正常业务空值。老逻辑把它当成「解析失败」，一款无数据就会让
+    整周工作流失败、其余款式也写不进飞书和 PG（2026-10-18 即此故障）。
+    """
+
+    for selector in (
+        ".jmt-base-table-empty .jmt-empty-description-color",
+        ".jmt-base-table-empty",
+    ):
+        try:
+            locator = page.locator(selector).first
+            if not locator.is_visible(timeout=800):
+                continue
+            text = locator.inner_text(timeout=800) or ""
+        except Exception:
+            continue
+        if "暂无数据" in text:
+            return True
+    return False
+
+
 def collect_style(context, page, style_name: str, spus: list[str], output_dir: Path) -> dict[str, Any]:
    """采集单款: 明细页取 SPU 树与主图 -> 搜索概览取曝光/点击 -> 聚合下载写 JSON。"""
    if not spus:
@@ -722,6 +806,20 @@ def collect_style(context, page, style_name: str, spus: list[str], output_dir: P
            snapshot_path.parent.mkdir(parents=True, exist_ok=True)
            snapshot_path.write_text(page.content(), encoding="utf-8")
            print(f"  [DEBUG] snapshot: {snapshot_path}")
+        if _page_reports_no_data(page):
+            empty = {
+                "style_name": style_name,
+                "platform": "jd",
+                "collected_at": datetime.now().isoformat(timespec="seconds"),
+                "spu_count": 0,
+                "raw_sku_count": 0,
+                "skipped": True,
+                "skip_reason": "京东商智当日无曝光/成交数据（查询结果为空）",
+                "records": [],
+            }
+            write_style_outputs(empty, output_dir)
+            print(f"  [SKIP] {style_name}: 京东商智当日无数据（暂无数据），记录空结果并继续")
+            return empty
        raise RuntimeError(f"款 {style_name} 明细页未解析到 SKU 行")

    metrics = collect_sku_search_metrics(context, [r["sku_id"] for r in sku_rows])
@@ -757,6 +855,65 @@ def parse_styles(value: str | None) -> list[dict]:
    return [g for g in groups if g["style"] in wanted]


+def wait_for_portal_session(page, *, headless: bool) -> None:
+    """等密码登录后回到商家后台，识别登录页与京东风控认证页。
+
+    2026-10-18 的实际故障就是停在 ``aq.jd.com/certified`` 认证页：
+    老逻辑只判断 ``shop.jd.com in url``，报错时把带一次性 token 的完整 URL
+    打进运行记录，运维无法判断该做什么。这里把三种终态分开处理：
+
+    - 回到后台：继续采集；
+    - 登录页：无头模式下短暂等待异步跳转，仍不跳转即判定登录失败；
+    - 安全认证页：无头模式直接抛可执行错误（无法自动完成短信/图形验证），
+      有界面模式给人工留出完成验证的时间。
+    """
+
+    deadline = time.monotonic() + LOGIN_SETTLE_TIMEOUT_MS / 1000
+    login_stuck_since: float | None = None
+    notice_printed = False
+    last_url = ""
+
+    while time.monotonic() < deadline:
+        try:
+            last_url = str(page.url)
+        except Exception:
+            last_url = ""
+
+        if _is_authenticated_portal(last_url):
+            page.wait_for_timeout(1500)
+            return
+
+        challenge = _security_challenge_host(last_url)
+        if challenge:
+            if headless:
+                raise _security_challenge_error(last_url)
+            if not notice_printed:
+                print(
+                    f"  [需要操作] 京东安全认证页（{challenge}）已打开，"
+                    "请在浏览器内完成短信/图形验证"
+                )
+                notice_printed = True
+        elif _looks_like_login_url(last_url):
+            login_stuck_since = login_stuck_since or time.monotonic()
+            if headless and time.monotonic() - login_stuck_since >= HEADLESS_LOGIN_STALL_SECONDS:
+                break
+        else:
+            login_stuck_since = None
+
+        try:
+            page.wait_for_timeout(500)
+        except Exception:
+            time.sleep(0.5)
+
+    if _security_challenge_host(last_url):
+        raise _security_challenge_error(last_url)
+    raise RuntimeError(
+        "京东登录未完成：密码登录后没有回到商家后台，请执行 "
+        "`uv run gyxx accounts login jd-shop --force` 人工确认登录态后重跑；"
+        f"当前页面: {_url_summary(last_url)}"
+    )
+
+
 def run(args: argparse.Namespace) -> None:
    global _DATE_LABEL
    output_dir = managed_data_path(args.output)
@@ -781,12 +938,15 @@ def run(args: argparse.Namespace) -> None:
        # Step 1: 登录
        page.goto("https://shop.jd.com/jdm/home", wait_until="domcontentloaded", timeout=60000)
        page.wait_for_timeout(2000)
-        if "shop.jd.com" in page.url and not _looks_like_login_url(page.url):
+        if _is_authenticated_portal(page.url):
            print("[OK] 检测到京东已登录态")
        else:
+            if _security_challenge_host(page.url):
+                # 认证页上再提交一次账密只会重复触发风控，直接给出人工处置路径。
+                raise _security_challenge_error(page.url)
            step_login(page, args.shop, args.password)
-            if "shop.jd.com" not in page.url or _looks_like_login_url(page.url):
-                raise RuntimeError(f"京东登录未完成，当前 URL: {page.url}")
+            wait_for_portal_session(page, headless=args.headless)
+            print("[OK] 京东登录完成")

        # Step 2-3: 商智 -> 新版页面
        sz_page = step_enter_shangzhi(context, page)
@@ -45,6 +45,7 @@ from collection_progress import SpuProgress

 from jd_data_collector import (
    DEFAULT_CONFIG,
+    close_ai_helper_drawer_if_present as close_jd_ai_helper_if_present,
    step_login,
    step_enter_shangzhi,
    extract,
@@ -128,7 +129,7 @@ def _print_preflight() -> None:
        return
    collectable = daily["jd"]["collectable"]
    skipped = daily["jd"]["skipped"]
-    print(f"[JD] 飞书表格 待采 {len(collectable)} 款, 跳过 {len(skipped)} 款 (款式级缺 erp/sales,或 JD 缺商品链接ID)")
+    print(f"[JD] 飞书表格 待采 {len(collectable)} 款, 跳过 {len(skipped)} 款 (款式级缺 erp 或已由同系列归集,或 JD 缺商品链接ID)")
    for s in skipped:
        print(f"  - {s['style']}: {s['reason']}")

@@ -932,19 +933,40 @@ _JMT_CUSTOM_REACT_CLICK = r"""(args) => {
    const [ty, tm, targetDay] = args;
    const pad = (n) => String(n).padStart(2, '0');
    const ymRe = new RegExp('^' + ty + '\\u5e74' + pad(tm) + '\\u6708$');
+    // 商智新版「自定义」是双月日历: 每个月的面板里除了本月日期, 还会渲染
+    // 上下月的补位日期 (class 带 -diff-month)。同一个月面板里因此可能出现
+    // 两个同号日期 (如 10 月面板里的 9/30 补位格与 10/30 真身), 之前按
+    // 「第一个匹配」点击会落到补位格, 于是 echo 变成 2026-09-30。
+    // 这里先定位目标月份自己的面板, 再排除补位格与禁用格。
    let panel = null;
    for (const el of document.querySelectorAll('*')) {
        const t = (el.textContent || '').trim();
        if (!ymRe.test(t)) continue;
-        panel = el.closest('[class*="calendar"]') || el.parentElement.parentElement || el.parentElement;
-        break;
+        let cur = el.parentElement;
+        let depth = 0;
+        while (cur && depth < 6) {
+            if (cur.querySelector && cur.querySelector('td')) break;
+            cur = cur.parentElement;
+            depth++;
+        }
+        if (cur && cur.querySelector('td')) {
+            panel = cur;
+            break;
+        }
    }
    if (!panel) return {status: 'no_panel'};
-    for (const cell of panel.querySelectorAll('td, [class*="cell"]')) {
-        if ((cell.textContent || '').trim() !== targetDay) continue;
-        if (/disabled|off/i.test(String(cell.className))) continue;
+    const cells = Array.from(panel.querySelectorAll('td, [class*="cell"]')).filter((cell) => {
+        if ((cell.textContent || '').trim() !== targetDay) return false;
+        const cls = String(cell.className || '');
+        if (/diff-month|disabled|-off\b|other-month|prev-month|next-month/i.test(cls)) return false;
+        const rect = cell.getBoundingClientRect();
+        return rect.width > 0 && rect.height > 0;
+    });
+    if (!cells.length) return {status: 'no_cell'};
+    for (const cell of cells) {
        const nodes = [cell, ...cell.querySelectorAll('*'), cell.parentElement];
        for (const node of nodes) {
+            if (!node) continue;
            const propsKey = Object.keys(node).find((k) => k.startsWith('__reactProps$'));
            if (!propsKey) continue;
            const props = node[propsKey];
@@ -955,9 +977,8 @@ _JMT_CUSTOM_REACT_CLICK = r"""(args) => {
                return {status: 'ok'};
            }
        }
-        return {status: 'no_handler'};
    }
-    return {status: 'no_cell'};
+    return {status: 'no_handler'};
 }"""


@@ -1006,13 +1027,33 @@ def _select_custom_date_new_version(page, target) -> bool:
    点工具条「自定义」→ 在双月日历里对目标日连续触发两次 React onClick
    (起止同日 = 单日区间) → echo 显示目标日期即成功.
    """
+    # 商智会随机弹出 AI 助手抽屉 / 引导层, 全屏 jd-modal-wrap 会拦截
+    # 日期工具条的点击; 先尝试收起再点「自定义」。
+    try:
+        page.keyboard.press("Escape")
+        page.wait_for_timeout(300)
+    except Exception:
+        pass
+    try:
+        close_jd_ai_helper_if_present(page)
+    except Exception:
+        pass
    try:
        tag = page.get_by_text("自定义", exact=True).first
        tag.wait_for(state="visible", timeout=8000)
        tag.click(timeout=8000)
    except Exception as exc:
-        print(f"    [WARN] 新版「自定义」入口点击失败: {exc}")
-        return False
+        print(f"    [WARN] 新版「自定义」入口点击失败({str(exc)[:120]}), 改用 DOM 事件点击")
+        clicked = page.evaluate(
+            r"""() => {
+                const el = document.querySelector('li[data-event-content="当前时间_自定义"]');
+                if (!el) return false;
+                el.dispatchEvent(new MouseEvent('click', {bubbles: true, cancelable: true}));
+                return true;
+            }"""
+        )
+        if not clicked:
+            return False
    page.wait_for_timeout(1200)

    for _ in range(2):
@@ -1437,7 +1437,8 @@ def load_jd_product_references(
            """
            SELECT style_name, jd_spus
            FROM dim_style
-            WHERE jd_spus IS NOT NULL AND cardinality(jd_spus) > 0
+            WHERE jsonb_typeof(jd_spus) = 'array'
+              AND jsonb_array_length(jd_spus) > 0
            """
        )
        rows = cursor.fetchall()
@@ -433,20 +433,9 @@ def run_collect(args, target_date, platforms, failed):
            failed=failed,
            historical=is_historical_target(target_date),
        )
-    # 评论入库 (DY 默认分支已采评论 -> 评论汇总_<YYYYMMDD>.json; JD 单独 --reviews-only 采)
-    # 幂等: product_reviews UNIQUE (platform, order_id) ON CONFLICT DO UPDATE, 重跑覆盖不双写.
-    # critical=False: 评论入库失败不影响后续 analyze/export/insert.
-    if not args.skip_platforms and not args.skip_reviews_import:
-        review_plats = [p for p in platforms if p in ("jd", "dy")]
-        if review_plats:
-            run_step_with_retry(
-                "import_product_reviews",
-                [sys.executable, str(runtime_script("import_product_reviews.py")),
-                 "--platforms", ",".join(review_plats)],
-                dry_run=args.dry_run,
-                critical=False,
-                failed=failed,
-            )
+    # 评论入库已下线 (2026-11-01): 差评由 shop.jd/dy/tmall_negative_reviews.daily
+    # 三个独立工作流采集, 消费方统一读视图 v_negative_reviews。
+    # product_reviews 停写并归档为 product_reviews_archived_20261101。


 def collect_daily_skip_report(date_str: str) -> Path | None:
@@ -675,8 +664,6 @@ def main() -> int:
    parser.add_argument("--platforms", default="jd,dy,tm", help="逗号分隔: jd,dy,tm")
    parser.add_argument("--serial-platforms", action="store_true", help="三平台串行跑")
    parser.add_argument("--skip-insert", action="store_true", help="只导出，不插入飞书")
-    parser.add_argument("--skip-reviews-import", action="store_true",
-                        help="跳过评论入库 (默认 daily collect 后跑 import_product_reviews)")
    parser.add_argument("--dry-run", action="store_true")
    parser.add_argument(
        "--erp-mode",
@@ -1,192 +0,0 @@
-"""评论采集编排器: 默认只跑 DY 采集 + 入库 PG。
-
-JD 评价采集代码已就绪 (jd_product_data_collector.py --reviews-only),
-但 JD 评价表只有 sku_id 缺 SPU, 暂时无法反查到款式, 所以编排器默认不跑 JD。
-等 dim_jd_sku_spu 映射表补齐后可加 --platforms jd,dy。
-
-DY 主流程 (dy_product_scraping.py 默认分支) 已挂载 _collect_comment_summary,
-daily --target-date 会顺带采评论; 本编排器用 --reviews-only 只跑评论 (跳过单品列表/
-下载/人群画像), 用于历史补采或重跑。
-入库调 import_product_reviews.py, ON CONFLICT (platform, order_id) 幂等覆盖。
-
-用法:
-    .venv/Scripts/python.exe orchestrate_review_collection.py
-    .venv/Scripts/python.exe orchestrate_review_collection.py --target-date 2026-07-13
-    .venv/Scripts/python.exe orchestrate_review_collection.py --platforms jd,dy  # JD 待 SPU 映射补齐
-    .venv/Scripts/python.exe orchestrate_review_collection.py --skip-collect   # 只入库
-    .venv/Scripts/python.exe orchestrate_review_collection.py --dry-run
-"""
-from __future__ import annotations
-
-import argparse
-import os
-import subprocess
-import sys
-from datetime import datetime, timedelta
-from pathlib import Path
-
-from collect_retry_utils import FailedCollector
-from runtime_paths import runtime_script
-
-from gyxx_flow.adapters import environment_for_child_script
-
-
-PROJECT_ROOT = Path(__file__).resolve().parent
-MAX_RUN_RETRIES = 2
-RUN_SLEEP_SECONDS = 5
-
-
-def default_target_date() -> str:
-    return (datetime.now().date() - timedelta(days=1)).strftime("%Y-%m-%d")
-
-
-def validate_date(value: str) -> str:
-    try:
-        datetime.strptime(value, "%Y-%m-%d")
-    except ValueError as exc:
-        raise argparse.ArgumentTypeError("date must be YYYY-MM-DD") from exc
-    return value
-
-
-def run_step(name, command, dry_run=False, critical=True):
-    """执行子命令。critical=True 失败抛 RuntimeError, False 仅 [WARN] 继续。"""
-    printable = " ".join(str(c) for c in command)
-    print(f"[RUN] {name}: {printable}")
-    if dry_run:
-        return {"name": name, "code": 0, "dry_run": True}
-    try:
-        proc = subprocess.run(
-            command,
-            cwd=PROJECT_ROOT,
-            text=True,
-            encoding="utf-8",
-            errors="replace",
-            env=environment_for_child_script(command[1], os.environ),
-        )
-    except (FileNotFoundError, OSError) as exc:
-        msg = f"{name} {type(exc).__name__}: {exc}"
-        if critical:
-            raise RuntimeError(msg) from exc
-        print(f"[WARN] {msg}")
-        return {"name": name, "code": 126, "error": msg}
-
-    if proc.returncode:
-        msg = f"{name} failed with exit code {proc.returncode}"
-        if critical:
-            raise RuntimeError(msg)
-        print(f"[WARN] {msg}")
-        return {"name": name, "code": proc.returncode, "error": msg}
-    print(f"[OK] {name}")
-    return {"name": name, "code": proc.returncode}
-
-
-def run_step_with_retry(name, command, dry_run=False, critical=True,
-                        failed: FailedCollector | None = None,
-                        max_retries: int = MAX_RUN_RETRIES,
-                        sleep_seconds: float = RUN_SLEEP_SECONDS) -> dict:
-    if dry_run:
-        return run_step(name, command, dry_run=True, critical=critical)
-    last_result: dict = {"name": name, "code": 0}
-    total = max_retries + 1
-    for attempt in range(1, total + 1):
-        result = run_step(name, command, dry_run=False, critical=False)
-        last_result = result
-        if result.get("code") == 0:
-            if attempt > 1:
-                print(f"[OK] {name} 第 {attempt}/{total} 次重试成功", flush=True)
-            return result
-        err = result.get("error", "unknown")
-        if attempt >= total:
-            if failed is not None:
-                failed.add_page(name, f"{err} (attempts={attempt})", attempts=attempt)
-            if critical:
-                raise RuntimeError(f"{name} failed after {attempt} attempts: {err}")
-            print(f"[WARN] {name} 失败 (已重试 {max_retries} 次, {err}), 继续", flush=True)
-            return result
-        print(f"[RETRY] {name} 第 {attempt}/{total} 次失败, {sleep_seconds}s 后重试...", flush=True)
-        import time
-        time.sleep(sleep_seconds)
-    return last_result
-
-
-def build_collect_command(plat: str, target_date: str) -> list[str]:
-    if plat == "jd":
-        # JD --reviews-only: 只跑评价采集 (好/中/差全采), 跳过商智单品明细
-        return [sys.executable, str(runtime_script("jd_product_data_collector.py")),
-                "--reviews-only", "--target-date", target_date]
-    if plat == "dy":
-        # DY --reviews-only: 登录后直接跳评论页, 跳过单品列表/下载/人群画像
-        return [sys.executable, str(runtime_script("dy_product_scraping.py")),
-                "--reviews-only", "--target-date", target_date]
-    raise ValueError(f"unknown platform: {plat}")
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description="评论采集编排器 (JD + DY)")
-    parser.add_argument("--target-date", type=validate_date, default=None,
-                        help="采集指定日期 YYYY-MM-DD (默认昨天)")
-    parser.add_argument("--platforms", default="dy",
-                        help="平台, 逗号分隔 (默认 dy; JD 待 SPU 映射补齐, 可手动 jd,dy)")
-    parser.add_argument("--skip-collect", action="store_true",
-                        help="跳过采集, 只跑 import_product_reviews.py")
-    parser.add_argument("--skip-import", action="store_true",
-                        help="跳过入库, 只跑采集")
-    parser.add_argument("--dry-run", action="store_true",
-                        help="只打印命令不执行")
-    args = parser.parse_args()
-
-    target_date = args.target_date or default_target_date()
-    platforms = [p.strip() for p in args.platforms.split(",") if p.strip()]
-    for p in platforms:
-        if p not in ("jd", "dy"):
-            print(f"[ERR] 未知平台: {p} (当前评论采集仅支持 jd, dy)")
-            return 2
-
-    today_str = datetime.now().strftime("%Y-%m-%d")
-    failed = FailedCollector("reviews", today_str)
-
-    print("=" * 60)
-    print("  评论采集编排器 (默认 DY)")
-    print("=" * 60)
-    print(f"  Target date: {target_date}")
-    print(f"  Platforms:   {platforms}")
-    print(f"  Skip collect: {args.skip_collect}")
-    print(f"  Skip import: {args.skip_import}")
-    print(f"  Dry run:     {args.dry_run}")
-    print("=" * 60 + "\n")
-
-    # Stage 1: 采集 (串行, Chrome profile 互锁)
-    if not args.skip_collect:
-        for plat in platforms:
-            cmd = build_collect_command(plat, target_date)
-            run_step_with_retry(
-                f"collect_{plat}", cmd,
-                dry_run=args.dry_run, critical=False, failed=failed,
-            )
-    else:
-        print("[INFO] --skip-collect: 跳过采集阶段\n")
-
-    # Stage 2: 入库
-    if not args.skip_import:
-        import_cmd = [sys.executable, str(runtime_script("import_product_reviews.py")),
-                      "--platforms", ",".join(platforms)]
-        if args.dry_run:
-            import_cmd.append("--dry-run")
-        run_step_with_retry(
-            "import_reviews", import_cmd,
-            dry_run=args.dry_run, critical=False, failed=failed,
-        )
-    else:
-        print("[INFO] --skip-import: 跳过入库阶段\n")
-
-    # 失败清单
-    failed_path = failed.write()
-    if failed_path:
-        print(f"\n[FAIL_LOG] {failed.summary_line()} -> {failed_path}")
-        return 1
-    print("\n[OK] 评论采集编排完成, 无失败")
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
@@ -0,0 +1,344 @@
+"""Weekly points-exchange ranking for private-domain material exchange.
+
+The data-hub points-exchange endpoint returns one row per exchanged SKU with a
+``quantity`` field. This workflow aggregates those rows into one ranking of
+items (商品) for a closed time window and posts the ranking to Feishu.
+"""
+
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+import os
+import re
+import sys
+from dataclasses import dataclass
+from datetime import date, datetime, timedelta
+from decimal import Decimal, InvalidOperation
+from zoneinfo import ZoneInfo
+
+from gyxx_flow.adapters import (
+    ANALYZER_NOTIFICATION_APP_PROFILE,
+    resolve_notification_route,
+    send_lark_bot_message,
+)
+from gyxx_flow.adapters.data_interaction import query_points_exchange
+
+WORKFLOW_ID = "product.points_exchange.weekly"
+DEFAULT_WINDOW_DAYS = 7
+MAX_WINDOW_DAYS = 92
+MESSAGE_CHUNK_SIZE = 40
+STDOUT_TOP_ROWS = 20
+SHANGHAI = ZoneInfo("Asia/Shanghai")
+_SAFE_DATE = re.compile(r"^\d{4}-\d{2}-\d{2}$")
+
+
+@dataclass(frozen=True, slots=True)
+class ExchangeItem:
+    """One exchanged item aggregated across every SKU of that item."""
+
+    item_id: str
+    title: str
+    quantity: Decimal
+    sku_count: int
+
+
+def _parse_date(value: str, field: str) -> date:
+    try:
+        parsed = date.fromisoformat(value.strip())
+    except (AttributeError, TypeError, ValueError) as exc:
+        raise ValueError(f"{field} must use the YYYY-MM-DD format: {value!r}") from exc
+    if not _SAFE_DATE.fullmatch(value.strip()):
+        raise ValueError(f"{field} must use the YYYY-MM-DD format: {value!r}")
+    return parsed
+
+
+def _business_date() -> date:
+    value = os.environ.get("GYXX_BUSINESS_DATE", "").strip()
+    if value and _SAFE_DATE.fullmatch(value):
+        return _parse_date(value, "GYXX_BUSINESS_DATE")
+    return datetime.now(SHANGHAI).date()
+
+
+def _previous_full_week(today: date) -> tuple[date, date]:
+    """Return the Monday/Sunday pair of the last completed natural week."""
+
+    start = today - timedelta(days=today.weekday() + DEFAULT_WINDOW_DAYS)
+    return start, start + timedelta(days=DEFAULT_WINDOW_DAYS - 1)
+
+
+def resolve_window(
+    start_value: str | None,
+    end_value: str | None,
+) -> tuple[date, date]:
+    """Resolve one closed date window, defaulting to last week's Mon-Sun."""
+
+    if start_value and end_value:
+        start = _parse_date(start_value, "--from")
+        end = _parse_date(end_value, "--to")
+    elif start_value:
+        start = _parse_date(start_value, "--from")
+        end = start + timedelta(days=DEFAULT_WINDOW_DAYS - 1)
+    elif end_value:
+        end = _parse_date(end_value, "--to")
+        start = end - timedelta(days=DEFAULT_WINDOW_DAYS - 1)
+    else:
+        start, end = _previous_full_week(_business_date())
+
+    if end < start:
+        raise ValueError(f"window end {end} is before start {start}")
+    span = (end - start).days + 1
+    if span > MAX_WINDOW_DAYS:
+        raise ValueError(
+            f"window spans {span} days; the supported maximum is {MAX_WINDOW_DAYS}"
+        )
+    return start, end
+
+
+def _quantity(row: dict[str, object], index: int) -> Decimal:
+    value = row.get("quantity")
+    if value is None or isinstance(value, bool):
+        raise ValueError(f"points-exchange row {index} has no quantity")
+    try:
+        quantity = Decimal(str(value).strip())
+    except (InvalidOperation, ValueError) as exc:
+        raise ValueError(f"points-exchange row {index} has invalid quantity") from exc
+    if not quantity.is_finite():
+        raise ValueError(f"points-exchange row {index} has non-finite quantity")
+    return quantity
+
+
+def _text_field(row: dict[str, object], *names: str) -> str:
+    for name in names:
+        value = row.get(name)
+        if value is not None and str(value).strip():
+            return " ".join(str(value).split())
+    return ""
+
+
+def aggregate_items(rows: list[dict[str, object]]) -> list[ExchangeItem]:
+    """Aggregate SKU rows into an item ranking ordered by quantity desc."""
+
+    titles: dict[str, dict[str, Decimal]] = {}
+    sku_ids: dict[str, set[str]] = {}
+    totals: dict[str, Decimal] = {}
+    for index, row in enumerate(rows, start=1):
+        quantity = _quantity(row, index)
+        item_id = _text_field(row, "itemId", "item_id")
+        title = _text_field(row, "title", "item_name", "name") or "未命名商品"
+        sku = _text_field(row, "skuId", "sku_id", "skuCode", "sku_code")
+        if not item_id:
+            item_id = f"sku:{sku}" if sku else f"row:{index}"
+        if quantity <= 0:
+            continue
+        totals[item_id] = totals.get(item_id, Decimal("0")) + quantity
+        titles.setdefault(item_id, {})
+        titles[item_id][title] = titles[item_id].get(title, Decimal("0")) + quantity
+        sku_ids.setdefault(item_id, set())
+        if sku:
+            sku_ids[item_id].add(sku)
+
+    items = [
+        ExchangeItem(
+            item_id=item_id,
+            title=max(titles[item_id].items(), key=lambda pair: (pair[1], pair[0]))[0],
+            quantity=quantity,
+            sku_count=len(sku_ids.get(item_id, ())),
+        )
+        for item_id, quantity in totals.items()
+    ]
+    return sorted(items, key=lambda item: (-item.quantity, item.title, item.item_id))
+
+
+def _quantity_text(quantity: Decimal) -> str:
+    return format(quantity.normalize(), "f")
+
+
+def _card(
+    items: list[ExchangeItem],
+    *,
+    start: date,
+    end: date,
+    total_quantity: Decimal,
+    item_count: int,
+    page: int,
+    pages: int,
+) -> dict[str, object]:
+    summary = (
+        f"**周期**：{start.isoformat()} ~ {end.isoformat()}\n"
+        f"**兑换商品**：{item_count} 个 · **兑换总量**：{_quantity_text(total_quantity)}\n"
+    )
+    if items:
+        top = items[0]
+        summary += f"**TOP1**：{top.title} · {_quantity_text(top.quantity)}"
+    if pages > 1:
+        summary += f"\n第 {page}/{pages} 条，本条 {len(items)} 个商品"
+    return {
+        "schema": "2.0",
+        "config": {"update_multi": True, "width_mode": "fill"},
+        "header": {
+            "title": {"tag": "plain_text", "content": "私域积分兑换周报"},
+            "subtitle": {
+                "tag": "plain_text",
+                "content": "积分兑换物资消耗排行 · 按兑换数量从高到低",
+            },
+            "template": "blue",
+        },
+        "body": {
+            "direction": "vertical",
+            "padding": "12px 12px 20px 12px",
+            "vertical_spacing": "8px",
+            "elements": [
+                {
+                    "tag": "markdown",
+                    "element_id": "summary",
+                    "content": summary,
+                },
+                {
+                    "tag": "table",
+                    "element_id": "exchange_ranking",
+                    "page_size": 10,
+                    "row_height": "auto",
+                    "row_max_height": "124px",
+                    "freeze_first_column": True,
+                    "columns": [
+                        {
+                            "name": "rank",
+                            "display_name": "排名",
+                            "data_type": "text",
+                            # Feishu rejects any column narrower than 80px.
+                            "width": "80px",
+                            "horizontal_align": "center",
+                        },
+                        {
+                            "name": "title",
+                            "display_name": "商品",
+                            "data_type": "text",
+                            "width": "auto",
+                        },
+                        {
+                            "name": "sku_count",
+                            "display_name": "SKU 数",
+                            "data_type": "text",
+                            "width": "80px",
+                            "horizontal_align": "right",
+                        },
+                        {
+                            "name": "quantity",
+                            "display_name": "兑换数量",
+                            "data_type": "text",
+                            "width": "88px",
+                            "horizontal_align": "right",
+                        },
+                    ],
+                    "rows": [
+                        {
+                            "rank": str(rank),
+                            "title": item.title,
+                            "sku_count": str(item.sku_count),
+                            "quantity": _quantity_text(item.quantity),
+                        }
+                        for rank, item in enumerate(items, start=1)
+                    ],
+                },
+            ],
+        },
+    }
+
+
+def _parse_args(argv: list[str] | None = None) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="汇总私域积分兑换商品的消耗数量排行并发送飞书周报。",
+    )
+    parser.add_argument(
+        "--from",
+        dest="start_date",
+        help="窗口起始日（YYYY-MM-DD），缺省取上一个完整自然周的周一",
+    )
+    parser.add_argument(
+        "--to",
+        dest="end_date",
+        help="窗口结束日（YYYY-MM-DD），缺省为起始日 +6 天",
+    )
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="只打印排行，不发送飞书消息",
+    )
+    return parser.parse_args(argv)
+
+
+def run(argv: list[str] | None = None) -> int:
+    arguments = _parse_args(argv)
+    start, end = resolve_window(arguments.start_date, arguments.end_date)
+    rows = query_points_exchange(
+        start_time=f"{start.isoformat()} 00:00:00",
+        end_time=f"{end.isoformat()} 23:59:59",
+    )
+    items = aggregate_items(rows)
+    total_quantity = sum((item.quantity for item in items), Decimal("0"))
+    print(
+        f"积分兑换查询完成：{start.isoformat()} ~ {end.isoformat()}，"
+        f"明细 {len(rows)} 条，商品 {len(items)} 个，兑换总量 {_quantity_text(total_quantity)}"
+    )
+    for rank, item in enumerate(items[:STDOUT_TOP_ROWS], start=1):
+        print(
+            f"{rank:>3}. {item.title}\tSKU {item.sku_count}\t"
+            f"{_quantity_text(item.quantity)}"
+        )
+
+    if not items:
+        print("该周期没有积分兑换记录，不发送飞书周报。")
+        return 0
+    if arguments.dry_run:
+        print("dry-run：跳过飞书发送。")
+        return 0
+
+    route = resolve_notification_route(WORKFLOW_ID, ())
+    if not route.enabled:
+        print("该工作流的飞书通知路由已停用。")
+        return 0
+    if not route.open_ids:
+        raise RuntimeError("积分兑换周报通知路由没有配置接收人")
+
+    pages = (len(items) + MESSAGE_CHUNK_SIZE - 1) // MESSAGE_CHUNK_SIZE
+    window_key = f"{start.isoformat()}~{end.isoformat()}"
+    for page in range(pages):
+        chunk = items[page * MESSAGE_CHUNK_SIZE : (page + 1) * MESSAGE_CHUNK_SIZE]
+        card = _card(
+            chunk,
+            start=start,
+            end=end,
+            total_quantity=total_quantity,
+            item_count=len(items),
+            page=page + 1,
+            pages=pages,
+        )
+        card_bytes = json.dumps(
+            card, ensure_ascii=False, sort_keys=True, separators=(",", ":")
+        ).encode("utf-8")
+        chunk_hash = hashlib.sha256(card_bytes).hexdigest()[:20]
+        idempotency_key = f"{WORKFLOW_ID}:{window_key}:{page + 1}:{chunk_hash}"
+        for open_id in route.open_ids:
+            send_lark_bot_message(
+                user_id=open_id,
+                content=card,
+                msg_type="interactive",
+                idempotency_key=idempotency_key,
+                profile=route.app_profile or ANALYZER_NOTIFICATION_APP_PROFILE,
+            )
+    print(f"积分兑换周报已发送：{len(items)} 个商品，{pages} 条飞书消息。")
+    return 0
+
+
+def main() -> int:
+    try:
+        return run()
+    except Exception as exc:
+        print(f"积分兑换周报工作流失败：{exc}", file=sys.stderr)
+        return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
@@ -0,0 +1,442 @@
+"""Refresh Hong Kong Anpei public inventory in the Feishu spreadsheet."""
+
+from __future__ import annotations
+
+import json
+import sys
+import time
+from collections.abc import Mapping
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+from zoneinfo import ZoneInfo
+
+from gyxx_flow.adapters.jushuitan_inventory import (
+    InventoryRecord,
+    JushuitanInventoryClient,
+)
+from gyxx_flow.modules.product_commerce.lark_cli_runtime import run_lark_cli
+
+_WORKFLOW_ID = "product.hk_anpei_public_inventory.daily"
+_CONFIG_PATH = (
+    Path(__file__).resolve().parent
+    / "config"
+    / "hk_anpei_public_inventory.json"
+)
+_REPOSITORY_ROOT = Path(__file__).resolve().parents[4]
+_TIMEZONE = ZoneInfo("Asia/Shanghai")
+_MAX_SHEET_ROWS = 200
+_WRITE_INTERVAL_SECONDS = 0.25
+
+
+class InventorySpreadsheetError(RuntimeError):
+    """Safe-to-log error for invalid spreadsheet configuration or state."""
+
+
+def _load_config() -> dict[str, Any]:
+    try:
+        config = json.loads(_CONFIG_PATH.read_text(encoding="utf-8"))
+    except (OSError, UnicodeError, json.JSONDecodeError) as exc:
+        raise InventorySpreadsheetError(
+            "Unable to read the public inventory workflow configuration"
+        ) from exc
+    if not isinstance(config, dict):
+        raise InventorySpreadsheetError(
+            "Public inventory configuration must be an object"
+        )
+
+    target = config.get("target")
+    if not isinstance(target, dict):
+        raise InventorySpreadsheetError(
+            "Public inventory configuration has no spreadsheet target"
+        )
+    for key in (
+        "spreadsheet_token",
+        "sheet_id",
+        "sheet_name",
+        "sku_column",
+        "inventory_column",
+        "product_code_column",
+    ):
+        if not str(target.get(key) or "").strip():
+            raise InventorySpreadsheetError(
+                f"Spreadsheet target is missing {key}"
+            )
+    for key in ("first_data_row", "last_data_row", "expected_sku_count"):
+        value = target.get(key)
+        if isinstance(value, bool) or not isinstance(value, int) or value < 1:
+            raise InventorySpreadsheetError(
+                f"Spreadsheet target has an invalid {key}"
+            )
+    if target["last_data_row"] - target["first_data_row"] + 1 != target[
+        "expected_sku_count"
+    ]:
+        raise InventorySpreadsheetError(
+            "Spreadsheet target rows do not match the frozen SKU count"
+        )
+
+    sku_items = config.get("skus")
+    if not isinstance(sku_items, list) or not sku_items:
+        raise InventorySpreadsheetError(
+            "Public inventory configuration has no SKUs"
+        )
+    if len(sku_items) != target["expected_sku_count"]:
+        raise InventorySpreadsheetError(
+            "Configured SKU count does not match the spreadsheet target"
+        )
+    seen_skus: set[str] = set()
+    seen_product_codes: set[str] = set()
+    for item in sku_items:
+        if not isinstance(item, dict):
+            raise InventorySpreadsheetError(
+                "Public inventory SKU metadata is invalid"
+            )
+        sku = str(item.get("sku") or "").strip()
+        product_code = str(item.get("jushuitan_sku_id") or "").strip()
+        if not sku or sku in seen_skus:
+            raise InventorySpreadsheetError(
+                "Public inventory configuration contains a blank or duplicate SKU"
+            )
+        if not product_code or product_code in seen_product_codes:
+            raise InventorySpreadsheetError(
+                "Public inventory configuration contains a blank or duplicate "
+                "Jushuitan product code"
+            )
+        seen_skus.add(sku)
+        seen_product_codes.add(product_code)
+    return config
+
+
+def _sheet_args(config: Mapping[str, Any]) -> list[str]:
+    target = config["target"]
+    return [
+        "--as",
+        "user",
+        "--spreadsheet-token",
+        str(target["spreadsheet_token"]),
+        "--sheet-id",
+        str(target["sheet_id"]),
+        "--format",
+        "json",
+    ]
+
+
+def _run_sheets(args: list[str], *, operation: str) -> dict[str, Any]:
+    try:
+        return run_lark_cli(
+            args,
+            cwd=_REPOSITORY_ROOT,
+            timeout=180,
+        )
+    except Exception as exc:
+        raise InventorySpreadsheetError(
+            f"Feishu spreadsheet {operation} failed"
+        ) from exc
+
+
+def _cell_value(cell: object) -> object:
+    if not isinstance(cell, Mapping):
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet returned an invalid cell"
+        )
+    return cell.get("value")
+
+
+def _read_sheet_rows(
+    config: Mapping[str, Any],
+    sku_items: list[Mapping[str, Any]],
+) -> dict[str, dict[str, Any]]:
+    target = config["target"]
+    result = _run_sheets(
+        [
+            "sheets",
+            "+cells-get",
+            *_sheet_args(config),
+            "--range",
+            f"{target['sku_column']}1:{target['product_code_column']}{_MAX_SHEET_ROWS}",
+            "--include",
+            "value",
+        ],
+        operation="cells-get",
+    )
+    data = result.get("data")
+    ranges = data.get("ranges") if isinstance(data, Mapping) else None
+    if (
+        not isinstance(data, Mapping)
+        or data.get("has_more") is not False
+        or not isinstance(ranges, list)
+        or len(ranges) != 1
+    ):
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet response is incomplete or invalid"
+        )
+
+    sheet_range = ranges[0]
+    if not isinstance(sheet_range, Mapping):
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet range response is invalid"
+        )
+    cells = sheet_range.get("cells")
+    row_indices = sheet_range.get("row_indices")
+    col_indices = sheet_range.get("col_indices")
+    required_columns = {
+        str(target["sku_column"]),
+        str(target["inventory_column"]),
+        str(target["product_code_column"]),
+    }
+    if (
+        not isinstance(cells, list)
+        or not isinstance(row_indices, list)
+        or not isinstance(col_indices, list)
+        or len(cells) != len(row_indices)
+        or not required_columns.issubset({str(col) for col in col_indices})
+    ):
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet response does not cover the configured columns"
+        )
+    column_positions = {str(col): index for index, col in enumerate(col_indices)}
+    row_values: dict[int, list[object]] = {}
+    for row_number, values in zip(row_indices, cells, strict=True):
+        if (
+            isinstance(row_number, bool)
+            or not isinstance(row_number, int)
+            or not isinstance(values, list)
+            or len(values) != len(col_indices)
+        ):
+            raise InventorySpreadsheetError(
+                "Feishu spreadsheet returned malformed row coordinates"
+            )
+        row_values[row_number] = values
+
+    first_row = int(target["first_data_row"])
+    last_row = int(target["last_data_row"])
+    header = row_values.get(1)
+    if header is None:
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet is missing its header row"
+        )
+    expected_headers = {
+        str(target["sku_column"]): "sku",
+        str(target["inventory_column"]): "仓库现有库存",
+        str(target["product_code_column"]): "商品编码（光影）",
+    }
+    for column, expected_header in expected_headers.items():
+        actual_header = str(
+            _cell_value(header[column_positions[column]]) or ""
+        ).strip()
+        if actual_header != expected_header:
+            raise InventorySpreadsheetError(
+                f"Feishu spreadsheet header in column {column} does not match"
+            )
+
+    rows: dict[str, dict[str, Any]] = {}
+    for row_number in range(first_row, last_row + 1):
+        values = row_values.get(row_number)
+        if values is None:
+            raise InventorySpreadsheetError(
+                "Feishu spreadsheet is missing an expected inventory row"
+            )
+        sku = str(
+            _cell_value(values[column_positions[str(target["sku_column"])]]) or ""
+        ).strip()
+        product_code = str(
+            _cell_value(
+                values[column_positions[str(target["product_code_column"])] ]
+            )
+            or ""
+        ).strip()
+        if not sku or sku in rows:
+            raise InventorySpreadsheetError(
+                "Feishu spreadsheet contains a blank or duplicate SKU"
+            )
+        rows[sku] = {
+            "row": row_number,
+            "product_code": product_code,
+        }
+
+    expected_by_sku = {
+        str(item["sku"]).strip(): str(item["jushuitan_sku_id"]).strip()
+        for item in sku_items
+    }
+    if set(rows) != set(expected_by_sku):
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet SKU rows differ from the frozen workbook list"
+        )
+    for sku, row in rows.items():
+        if row["product_code"] != expected_by_sku[sku]:
+            raise InventorySpreadsheetError(
+                f"Feishu spreadsheet product code does not match SKU {sku} "
+                f"at row {row['row']}"
+            )
+    return rows
+
+
+def _read_written_inventory(
+    config: Mapping[str, Any],
+    expected: list[object],
+) -> None:
+    target = config["target"]
+    column = str(target["inventory_column"])
+    first_row = int(target["first_data_row"])
+    last_row = int(target["last_data_row"])
+    result = _run_sheets(
+        [
+            "sheets",
+            "+cells-get",
+            *_sheet_args(config),
+            "--range",
+            f"{column}{first_row}:{column}{last_row}",
+            "--include",
+            "value",
+        ],
+        operation="inventory readback",
+    )
+    data = result.get("data")
+    ranges = data.get("ranges") if isinstance(data, Mapping) else None
+    if (
+        not isinstance(data, Mapping)
+        or data.get("has_more") is not False
+        or not isinstance(ranges, list)
+        or len(ranges) != 1
+    ):
+        raise InventorySpreadsheetError(
+            "Feishu inventory readback is incomplete or invalid"
+        )
+    sheet_range = ranges[0]
+    if not isinstance(sheet_range, Mapping):
+        raise InventorySpreadsheetError(
+            "Feishu inventory readback range is invalid"
+        )
+    cells = sheet_range.get("cells")
+    row_indices = sheet_range.get("row_indices")
+    if (
+        not isinstance(cells, list)
+        or not isinstance(row_indices, list)
+        or row_indices != list(range(first_row, last_row + 1))
+        or len(cells) != len(expected)
+    ):
+        raise InventorySpreadsheetError(
+            "Feishu inventory readback does not cover every target row"
+        )
+    actual: list[object] = []
+    for row in cells:
+        if not isinstance(row, list) or len(row) != 1:
+            raise InventorySpreadsheetError(
+                "Feishu inventory readback has an invalid cell shape"
+            )
+        actual.append(_cell_value(row[0]))
+    for offset, (actual_value, expected_value) in enumerate(
+        zip(actual, expected, strict=True)
+    ):
+        if expected_value == "":
+            matched = actual_value in (None, "")
+        else:
+            try:
+                matched = int(str(actual_value).strip()) == int(expected_value)
+            except (TypeError, ValueError):
+                matched = False
+        if not matched:
+            raise InventorySpreadsheetError(
+                "Feishu inventory readback did not match the submitted value "
+                f"at row {first_row + offset}"
+            )
+
+
+def _write_inventory(
+    config: Mapping[str, Any],
+    sku_items: list[Mapping[str, Any]],
+    inventories: Mapping[str, InventoryRecord],
+    rows: Mapping[str, Mapping[str, Any]],
+) -> tuple[int, bool, int]:
+    target = config["target"]
+    first_row = int(target["first_data_row"])
+    last_row = int(target["last_data_row"])
+    by_row = sorted(rows.items(), key=lambda entry: int(entry[1]["row"]))
+    if [int(row["row"]) for _, row in by_row] != list(
+        range(first_row, last_row + 1)
+    ):
+        raise InventorySpreadsheetError(
+            "Feishu spreadsheet SKU rows are not a contiguous target range"
+        )
+    sku_config = {str(item["sku"]): item for item in sku_items}
+    expected_values: list[object] = []
+    missing_count = 0
+    for sku, _ in by_row:
+        product_code = str(sku_config[sku]["jushuitan_sku_id"])
+        inventory = inventories.get(product_code)
+        if inventory is None:
+            # An omitted Jushuitan row is unknown inventory, so clear stale data.
+            expected_values.append("")
+            missing_count += 1
+        else:
+            expected_values.append(inventory.public_available_qty)
+
+    cells = [[{"value": value}] for value in expected_values]
+    result = _run_sheets(
+        [
+            "sheets",
+            "+cells-set",
+            *_sheet_args(config),
+            "--range",
+            f"{target['inventory_column']}{first_row}:"
+            f"{target['inventory_column']}{last_row}",
+            "--cells",
+            json.dumps(cells, ensure_ascii=False, separators=(",", ":")),
+        ],
+        operation="inventory update",
+    )
+    if result.get("acceptance_skipped") is True:
+        return 0, True, missing_count
+    _read_written_inventory(config, expected_values)
+    return len(expected_values), False, missing_count
+
+
+def run() -> dict[str, object]:
+    config = _load_config()
+    sku_items = config["skus"]
+    jushuitan_sku_ids = [str(item["jushuitan_sku_id"]) for item in sku_items]
+
+    # Validate the complete target and ERP response before modifying Feishu.
+    rows = _read_sheet_rows(config, sku_items)
+    inventories = JushuitanInventoryClient().query(jushuitan_sku_ids)
+    if not inventories:
+        raise InventorySpreadsheetError(
+            "Jushuitan returned no inventory rows for the frozen product-code list; "
+            "refusing to clear the Feishu inventory column"
+        )
+    written, acceptance_skipped, inventory_missing = _write_inventory(
+        config,
+        sku_items,
+        inventories,
+        rows,
+    )
+    return {
+        "workflow_id": _WORKFLOW_ID,
+        "target_sheet": str(config["target"]["sheet_name"]),
+        "sku_count": len(sku_items),
+        "inventory_found": len(inventories),
+        "inventory_missing": inventory_missing,
+        "cells_written": written,
+        "acceptance_skipped": acceptance_skipped,
+        "queried_at": datetime.now(_TIMEZONE).strftime("%Y-%m-%d %H:%M:%S"),
+    }
+
+
+def main() -> int:
+    try:
+        print(json.dumps(run(), ensure_ascii=False, separators=(",", ":")))
+        return 0
+    except Exception as exc:
+        print(
+            json.dumps(
+                {"workflow_id": _WORKFLOW_ID, "error": str(exc)},
+                ensure_ascii=False,
+                separators=(",", ":"),
+            ),
+            file=sys.stderr,
+        )
+        return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
@@ -8,7 +8,8 @@

 款式来源 (2026-06-27 改造):
  - --styles 显式传: 用 CLI 参数指定的款式列表
-  - --styles 留空: 动态从项目数据库读 (get_main_image_styles: 天猫分组 + 主图飞书多维表格地址 同时存在)
+  - --styles 留空: 动态从项目数据库读 (get_main_image_styles: 身份齐全的天猫款式)
+    · 主图飞书表地址不参与采集门：缺地址的款照常抓图并入库，写表环节跳过

 数据落点: data/主图信息/<date>/<款式>/wanxiang_creative_main_images.json
 插入幂等: 飞书按 (时间, 平台, 图片) 更新，PG 按日期/款式/平台/图片覆盖。
@@ -112,7 +113,7 @@ def main() -> int:
    parser.add_argument("--skip-collect", action="store_true", help="跳过采集,只跑 insert")
    parser.add_argument("--skip-insert", action="store_true", help="只采集,不推飞书")
    parser.add_argument("--styles", default="",
-                        help="只采集/插入指定款式,逗号分隔 (传空=动态从项目数据库读天猫 + 主图飞书多维表格地址)")
+                        help="只采集/插入指定款式,逗号分隔 (传空=动态从项目数据库读身份齐全的天猫款式)")
    parser.add_argument("--headless", action="store_true",
                        help="采集时用无头浏览器 (服务器/无人值守场景)")
    parser.add_argument("--max-pages", type=int, default=100,
@@ -128,11 +129,11 @@ def main() -> int:
        target_date = last_sunday.strftime("%Y-%m-%d")
    py = sys.executable

-    # --styles 为空时动态从项目数据库读 (天猫分组 + 主图飞书多维表格地址 同时存在)
+    # --styles 为空时动态从项目数据库读（身份齐全即可，主图地址不参与采集门）
    if not args.styles:
        bitable_styles = load_dynamic_styles()
        args.styles = ",".join(bitable_styles)
-        print(f"[PREFLIGHT] 飞书多维表格动态读主图款式 {len(bitable_styles)} 款")
+        print(f"[PREFLIGHT] 动态读主图款式 {len(bitable_styles)} 款（缺主图地址的款只入库不写表）")
    else:
        print(f"[PREFLIGHT] --styles 显式指定: {args.styles}")

@@ -27,7 +27,6 @@ from runtime_paths import LOG_ROOT, artifact_relative_path, managed_data_path
 # 复用 TM insert 脚本的工具函数
 from scripts.insert_main_image_records import (
    DEFAULT_DATA_ROOT,
-    MAIN_IMAGE_MAP,
    batch_create_records,
    date_to_month_week,
    discover_fields,
@@ -121,6 +120,11 @@ def process_style(style_key: str, cfg: dict, date_str: str, data_root: Path, dry
            row["_pg_record"] = pg_record
            pg_records.append(pg_record)

+    # 缺主图飞书表地址：照常入库，只是不写表（与商品经营日报缺销量表同口径）
+    if not str(cfg.get("base_token") or "").strip() or not str(cfg.get("table_id") or "").strip():
+        print(f"  [SKIP-TABLE] {style_key}: 未配置主图飞书表地址，仅入库")
+        return result
+
    try:
        fields = discover_fields(cfg["base_token"], cfg["table_id"])
    except Exception as e:
@@ -223,11 +227,14 @@ def main():
    try:
        config = get_bitable_main_image_map()
        if not config:
-            raise RuntimeError("飞书动态配置为空")
+            raise RuntimeError("dim_style 未返回任何带主图表地址的款式")
    except Exception as exc:
-        print(f"[WARN] 动态读取主图配置失败,使用本地文件回退: {exc}", file=sys.stderr)
-        config = read_json(MAIN_IMAGE_MAP)
-        config.pop("_comment", None)
+        # 主图表地址唯一来源是 dim_style；读不到即失败，不再回退本地 JSON。
+        print(
+            f"[FAIL] 主图配置读取失败，拒绝回退本地 JSON: {exc}",
+            file=sys.stderr,
+        )
+        return 1

    data_root = managed_data_path(args.data_root)
    date_dir = data_root / args.date
@@ -262,9 +269,12 @@ def main():
    for style_dir_name in target_styles:
        key = normalize_key(style_dir_name)
        if key not in norm_config:
-            print(f"[SKIP] {style_dir_name}: config 里找不到 (lowercase 匹配)")
-            continue
-        orig_key, cfg = norm_config[key]
+            # 缺主图飞书表地址：照常采图入库，只是不写表
+            orig_key = style_dir_name
+            cfg = {"display_name": style_dir_name, "platform": "京东", "enabled": True}
+            print(f"[NO-TABLE] {style_dir_name}: 未配置主图飞书表地址，仅入库")
+        else:
+            orig_key, cfg = norm_config[key]
        if not cfg.get("enabled", True):
            print(f"[SKIP] {orig_key}: enabled=false")
            continue
@@ -30,7 +30,6 @@ from runtime_paths import (
    managed_data_path,
 )

-MAIN_IMAGE_MAP = PROJECT_ROOT / "bitable_main_image_map.json"
 DEFAULT_DATA_ROOT = RAW_DATA_ROOT / "主图信息"

 LARK_CLI = (
@@ -41,10 +40,14 @@ LARK_CLI = (


 def _print_preflight() -> None:
-    """启动期打印: 飞书表格里本周要插主图记录的款式数 (仅要求 主图飞书多维表格地址 非空)。"""
+    """启动期打印: 采集目标款数，以及其中能写飞书表的款数。"""
    try:
        styles = get_main_image_styles()
-        print(f"[主图-insert] 飞书表格 周汇总目标 {len(styles)} 款 (天猫 + 主图飞书多维表格地址)")
+        write_targets = get_bitable_main_image_map()
+        print(
+            f"[主图-insert] 采集目标 {len(styles)} 款；"
+            f"可写飞书表 {len(write_targets)} 款（其余仅入库）"
+        )
    except Exception as exc:
        print(f"[WARN] 飞书读取失败, 跳过 preflight: {exc}", file=sys.stderr)

@@ -492,6 +495,11 @@ def process_product(style_key: str, cfg: dict, date_str: str, data_root: Path, d
            row["_pg_record"] = pg_record
            pg_records.append(pg_record)

+    # 缺主图飞书表地址：照常入库，只是不写表（与商品经营日报缺销量表同口径）
+    if not str(cfg.get("base_token") or "").strip() or not str(cfg.get("table_id") or "").strip():
+        print(f"  [SKIP-TABLE] {style_key}: 未配置主图飞书表地址，仅入库")
+        return result
+
    # discover field ids
    try:
        fields = discover_fields(cfg["base_token"], cfg["table_id"])
@@ -598,11 +606,14 @@ def main():
    try:
        config = get_bitable_main_image_map()
        if not config:
-            raise RuntimeError("飞书动态配置为空")
+            raise RuntimeError("dim_style 未返回任何带主图表地址的款式")
    except Exception as exc:
-        print(f"[WARN] 动态读取主图表配置失败，使用本地文件回退: {exc}", file=sys.stderr)
-        config = read_json(MAIN_IMAGE_MAP)
-        config.pop("_comment", None)
+        # 主图表地址唯一来源是 dim_style；读不到即失败，不再回退本地 JSON。
+        print(
+            f"[FAIL] 主图表配置读取失败，拒绝回退本地 JSON: {exc}",
+            file=sys.stderr,
+        )
+        return 1
    data_root = managed_data_path(args.data_root)
    date_dir = data_root / args.date
    if not date_dir.exists():
@@ -640,9 +651,12 @@ def main():
    for style_dir_name in target_styles:
        key = normalize_key(style_dir_name)
        if key not in norm_config:
-            print(f"[SKIP] {style_dir_name}: config 里找不到 (lowercase 匹配)")
-            continue
-        orig_key, cfg = norm_config[key]
+            # 缺主图飞书表地址：照常采图入库，只是不写表
+            orig_key = style_dir_name
+            cfg = {"display_name": style_dir_name, "platform": "天猫", "enabled": True}
+            print(f"[NO-TABLE] {style_dir_name}: 未配置主图飞书表地址，仅入库")
+        else:
+            orig_key, cfg = norm_config[key]
        if not cfg.get("enabled", True):
            print(f"[SKIP] {orig_key}: enabled=false")
            continue
@@ -0,0 +1,349 @@
+"""一次性迁移：把天猫广告指标从 ``product_daily_metrics.raw_data`` 搬到独立表。
+
+背景：天猫广告指标原先写在 ``product_daily_metrics.raw_data.tm_ad_metrics``
+JSONB 节点里，与经营日报共用一行；经营数据重新同步时需要靠 JSONB 节点级合并
+才能保住投放数据。2026-11-01 起改为写 ``tm_product_ad_daily_metrics``（与京东、
+抖音同构），本脚本负责把历史节点回填进新表，并在校验通过后清除旧节点。
+
+行为：
+- 默认 dry-run：只统计源数据规模、金额合计、款式匹配情况，不写库。
+- ``--execute``：先备份源节点到 JSON（``--backup-dir``，默认运行时 TMP_ROOT），
+  再按 ``(stat_date, product_id)`` 幂等 upsert 到新表。
+- ``--clear-source``：仅在行数/金额校验一致、备份写盘成功后，才执行
+  ``raw_data - 'tm_ad_metrics'`` 清除节点；任何一项不通过都不清理并返回非零。
+
+用法::
+
+    python scripts/migrate_tm_ad_metrics_to_table.py
+    python scripts/migrate_tm_ad_metrics_to_table.py --execute
+    python scripts/migrate_tm_ad_metrics_to_table.py --execute --clear-source
+
+连接信息来自 PG_* 环境变量（与 db/__init__.py 同一套）。
+"""
+# ruff: noqa: E402
+
+from __future__ import annotations
+
+import argparse
+import gzip
+import json
+import sys
+from datetime import date, datetime
+from decimal import Decimal, InvalidOperation
+from pathlib import Path
+from typing import Any, Iterator
+
+PROJECT_ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(PROJECT_ROOT))
+
+import db  # noqa: E402
+from runtime_paths import TMP_ROOT  # noqa: E402
+
+SOURCE_PREDICATE = (
+    "platform = 'tm' AND jsonb_typeof(raw_data -> 'tm_ad_metrics') = 'object'"
+)
+BATCH_SIZE = 500
+
+
+def _decimal(value: Any) -> Decimal:
+    if value is None:
+        return Decimal("0")
+    text = str(value).strip().replace(",", "")
+    if not text:
+        return Decimal("0")
+    try:
+        return Decimal(text)
+    except InvalidOperation:
+        return Decimal("0")
+
+
+def _iter_source_rows(
+    conn, date_from: date | None, date_to: date | None, batch_size: int
+) -> Iterator[list[dict[str, Any]]]:
+    """按 (stat_date, product_id) keyset 分页读源节点，避免一次性载入全部行。"""
+
+    clauses = [SOURCE_PREDICATE]
+    params: list[Any] = []
+    if date_from:
+        clauses.append("stat_date >= %s")
+        params.append(date_from)
+    if date_to:
+        clauses.append("stat_date <= %s")
+        params.append(date_to)
+    where = " AND ".join(clauses)
+
+    last: tuple[date, str] | None = None
+    while True:
+        page_where = where
+        page_params = list(params)
+        if last is not None:
+            page_where += " AND (stat_date, product_id) > (%s, %s)"
+            page_params.extend([last[0], last[1]])
+        sql = f"""
+            SELECT stat_date, product_id, product_name,
+                   raw_data -> 'tm_ad_metrics' AS node,
+                   source_file
+              FROM product_daily_metrics
+             WHERE {page_where}
+             ORDER BY stat_date, product_id
+             LIMIT {batch_size}
+        """
+        with conn.cursor() as cur:
+            cur.execute(sql, page_params)
+            page = cur.fetchall()
+        if not page:
+            return
+        yield [
+            {
+                "stat_date": row[0],
+                "product_id": str(row[1]),
+                "product_name": row[2],
+                "node": row[3] or {},
+                "source_file": row[4],
+            }
+            for row in page
+        ]
+        last = (page[-1][0], str(page[-1][1]))
+        if len(page) < batch_size:
+            return
+
+
+def _to_records(rows: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    """把 JSONB 节点转成 upsert 函数接受的记录形态。"""
+
+    records = []
+    for row in rows:
+        node = dict(row["node"])
+        metrics = {
+            key: value
+            for key, value in node.items()
+            if key not in {"style_name", "style_match_method"}
+        }
+        records.append(
+            {
+                "stat_date": row["stat_date"],
+                "product_id": row["product_id"],
+                "product_name": row["product_name"],
+                "ad_metrics": metrics,
+                # 节点里已有的款式映射作为兜底；能按 tm_spus/名称重新匹配的以重算为准。
+                "style_name": node.get("style_name"),
+                "style_match_method": node.get("style_match_method"),
+                "source_file": row["source_file"],
+            }
+        )
+    return records
+
+
+def _write_backup(rows: list[dict[str, Any]], backup_dir: Path) -> Path:
+    backup_dir.mkdir(parents=True, exist_ok=True)
+    stamp = datetime.now().strftime("%Y%m%d-%H%M%S")
+    path = backup_dir / f"tm_ad_metrics_backup_{stamp}.json.gz"
+    payload = [
+        {
+            "stat_date": row["stat_date"].isoformat(),
+            "product_id": row["product_id"],
+            "product_name": row["product_name"],
+            "source_file": row["source_file"],
+            "tm_ad_metrics": row["node"],
+        }
+        for row in rows
+    ]
+    with gzip.open(path, "wt", encoding="utf-8") as handle:
+        json.dump(payload, handle, ensure_ascii=False, default=str)
+    return path
+
+
+def _target_totals(conn, date_from: date | None, date_to: date | None) -> tuple[int, Decimal]:
+    clauses = ["TRUE"]
+    params: list[Any] = []
+    if date_from:
+        clauses.append("stat_date >= %s")
+        params.append(date_from)
+    if date_to:
+        clauses.append("stat_date <= %s")
+        params.append(date_to)
+    with conn.cursor() as cur:
+        cur.execute(
+            f"""
+            SELECT COUNT(*), COALESCE(SUM(spend), 0)
+              FROM tm_product_ad_daily_metrics
+             WHERE {' AND '.join(clauses)}
+            """,
+            params,
+        )
+        count, spend = cur.fetchone()
+    return int(count), Decimal(str(spend))
+
+
+def _source_count(conn, date_from: date | None, date_to: date | None) -> int:
+    clauses = [SOURCE_PREDICATE]
+    params: list[Any] = []
+    if date_from:
+        clauses.append("stat_date >= %s")
+        params.append(date_from)
+    if date_to:
+        clauses.append("stat_date <= %s")
+        params.append(date_to)
+    with conn.cursor() as cur:
+        cur.execute(
+            f"SELECT COUNT(*) FROM product_daily_metrics WHERE {' AND '.join(clauses)}",
+            params,
+        )
+        return int(cur.fetchone()[0])
+
+
+def _parse_date(value: str | None) -> date | None:
+    return date.fromisoformat(value) if value else None
+
+
+def _missing_target_columns(conn) -> list[str]:
+    """dry-run 只读检查目标表是否已具备款式/指标列。"""
+
+    required = {"style_name", "style_match_method", "metrics", "source_row_number"}
+    with conn.cursor() as cur:
+        cur.execute(
+            """
+            SELECT column_name FROM information_schema.columns
+             WHERE table_name = 'tm_product_ad_daily_metrics'
+            """
+        )
+        existing = {row[0] for row in cur.fetchall()}
+    if not existing:
+        return sorted(required | {"<表不存在>"})
+    return sorted(required - existing)
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="迁移天猫广告指标到独立表")
+    parser.add_argument("--date-from", help="只迁移该日期起 YYYY-MM-DD")
+    parser.add_argument("--date-to", help="只迁移到该日期 YYYY-MM-DD")
+    parser.add_argument("--execute", action="store_true", help="确认写入 PostgreSQL")
+    parser.add_argument(
+        "--clear-source",
+        action="store_true",
+        help="校验通过后清除 product_daily_metrics 里的旧 JSONB 节点",
+    )
+    parser.add_argument(
+        "--backup-dir", type=Path, help="旧节点备份目录（默认运行时 TMP_ROOT）"
+    )
+    args = parser.parse_args(argv)
+
+    date_from = _parse_date(args.date_from)
+    date_to = _parse_date(args.date_to)
+    backup_dir = args.backup_dir or TMP_ROOT
+
+    with db.get_conn() as conn:
+        total_rows = _source_count(conn, date_from, date_to)
+        if total_rows == 0:
+            print("[INFO] 没有可迁移的天猫广告节点，退出")
+            return 0
+
+        tally = {
+            "rows": 0,
+            "spend": Decimal("0"),
+            "deal": Decimal("0"),
+            "missing_style": 0,
+            "unknown_style": 0,
+            "min_date": None,
+            "max_date": None,
+        }
+        all_rows: list[dict[str, Any]] = []
+        resolved_style: dict[tuple[str, str], str | None] = {}
+
+        style_by_product_id, style_names = db._load_platform_style_catalog(conn, "tm_spus")
+        for batch in _iter_source_rows(conn, date_from, date_to, BATCH_SIZE):
+            all_rows.extend(batch)
+            for record in _to_records(batch):
+                node_style = str(record.get("style_name") or "").strip()
+                if not node_style:
+                    tally["missing_style"] += 1
+                key = (record["product_id"], str(record.get("product_name") or ""))
+                if key not in resolved_style:
+                    resolved_style[key] = db._resolve_style_match(
+                        record, style_by_product_id, style_names
+                    )[0]
+                if not (resolved_style[key] or node_style):
+                    tally["unknown_style"] += 1
+                tally["rows"] += 1
+                tally["spend"] += _decimal(record["ad_metrics"].get("spend"))
+                tally["deal"] += _decimal(
+                    record["ad_metrics"].get("attributed_deal_amount")
+                )
+                stat_date = record["stat_date"]
+                tally["min_date"] = (
+                    stat_date if tally["min_date"] is None else min(tally["min_date"], stat_date)
+                )
+                tally["max_date"] = (
+                    stat_date if tally["max_date"] is None else max(tally["max_date"], stat_date)
+                )
+
+        print(
+            f"[INFO] 源节点 {tally['rows']} 行（{tally['min_date']} ~ {tally['max_date']}），"
+            f"花费 {tally['spend']:.2f}，广告归因成交 {tally['deal']:.2f}"
+        )
+        print(
+            f"[INFO] 款式：源节点缺 style_name {tally['missing_style']} 行；"
+            f"按 tm_spus/名称仍无法确定 {tally['unknown_style']} 行"
+        )
+
+        if not args.execute:
+            existing_count, _ = _target_totals(conn, date_from, date_to)
+            missing = _missing_target_columns(conn)
+            print(
+                f"[INFO] 目标表现有 {existing_count} 行（同日期区间）；"
+                "DRY-RUN 未写库，确认后添加 --execute"
+            )
+            print(
+                "[INFO] 目标表待补列："
+                + (", ".join(missing) if missing else "无（结构已就绪）")
+            )
+            return 0
+
+        db.ensure_tmall_product_ad_daily_table(conn)
+
+        backup_path = _write_backup(all_rows, backup_dir)
+        print(f"[OK] 旧节点已备份：{backup_path}")
+
+        written = 0
+        for offset in range(0, len(all_rows), BATCH_SIZE):
+            records = _to_records(all_rows[offset : offset + BATCH_SIZE])
+            written += db.upsert_tmall_product_ad_daily_metrics(conn, records)
+        print(f"[OK] 已写入 tm_product_ad_daily_metrics：{written} 行")
+
+        target_count, target_spend = _target_totals(conn, date_from, date_to)
+        ok = target_count == tally["rows"] and target_spend == tally["spend"]
+        print(
+            f"[CHECK] 行数 {target_count}/{tally['rows']}，"
+            f"花费 {target_spend:.2f}/{tally['spend']:.2f} -> "
+            f"{'一致' if ok else '不一致'}"
+        )
+        if not ok:
+            print("[FAIL] 校验不一致，保留旧 JSONB 节点，请人工核对后重跑", file=sys.stderr)
+            return 1
+
+        if not args.clear_source:
+            print("[INFO] 未指定 --clear-source，旧节点保留")
+            return 0
+
+        with conn.cursor() as cur:
+            cur.execute(
+                f"""
+                UPDATE product_daily_metrics
+                   SET raw_data = COALESCE(raw_data, '{{}}'::jsonb) - 'tm_ad_metrics'
+                 WHERE {SOURCE_PREDICATE}
+                   {"AND stat_date >= %s" if date_from else ""}
+                   {"AND stat_date <= %s" if date_to else ""}
+                """,
+                [value for value in (date_from, date_to) if value is not None],
+            )
+            cleared = cur.rowcount
+        remaining = _source_count(conn, date_from, date_to)
+        print(f"[OK] 已清除旧节点：{cleared} 行，剩余 {remaining} 行")
+        if remaining:
+            print("[FAIL] 仍有残留节点，请重跑 --clear-source", file=sys.stderr)
+            return 1
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
@@ -23,10 +23,7 @@ try:
        collect_monthly_erp_metrics,
        monthly_report_interval,
    )
-    from .config.style_config_loader import (
-        StyleConfigLoader,
-        has_complete_style_identity,
-    )
+    from .config.style_config_loader import StyleConfigLoader
    from .lark_cli_runtime import run_lark_cli
 except ImportError:  # Direct script execution from the module root.
    from collect_erp_monthly_metrics import (
@@ -35,10 +32,7 @@ except ImportError:  # Direct script execution from the module root.
        collect_monthly_erp_metrics,
        monthly_report_interval,
    )
-    from config.style_config_loader import (
-        StyleConfigLoader,
-        has_complete_style_identity,
-    )
+    from config.style_config_loader import StyleConfigLoader
    from lark_cli_runtime import run_lark_cli

 _MARKDOWN_LINK = re.compile(r"^\s*\[([^]]+)]\([^)]*\)\s*$")
@@ -324,13 +318,20 @@ def plan_erp_styles(
            missing.append(target.product_name)
            continue
        style_name, style_config = resolved
-        raw_codes = style_config.get("erp_codes") or ()
+        # 月度表按款明细：取款式建档时自己填的编码（series.own_erp_codes）。
+        # loader 把整条系列的编码并集挂到了锚点的 erp_codes 上，那是给采集端
+        # 当唯一取数单元用的；报表若照搬，同系列每个成员行都会把并集查一遍，
+        # 销量被重复计入。锚点自身没填编码时这里为空，按 0 写入即可。
+        series = style_config.get("series") or {}
+        raw_codes = series.get("own_erp_codes")
+        if raw_codes is None:
+            raw_codes = style_config.get("erp_codes") or ()
        codes = tuple(
            dict.fromkeys(
                str(code).strip() for code in raw_codes if str(code).strip()
            )
        )
-        if not codes or not has_complete_style_identity(style_config):
+        if not codes:
            missing.append(target.product_name)
            continue
        key = normalized_name(style_name)
@@ -174,7 +174,7 @@ def main(argv: list[str] | None = None) -> int:
    daily = get_daily_styles()
    tm_collectable = daily["tm"]["collectable"]
    tm_skipped = daily["tm"]["skipped"]
-    print(f"[TM] 飞书表格 待采 {len(tm_collectable)} 款, 跳过 {len(tm_skipped)} 款 (款式级缺 erp/sales,或 TM 缺商品链接ID)")
+    print(f"[TM] 飞书表格 待采 {len(tm_collectable)} 款, 跳过 {len(tm_skipped)} 款 (款式级缺 erp 或已由同系列归集,或 TM 缺商品链接ID)")
    for s in tm_skipped:
        print(f"  - {s['style']}: {s['reason']}")
    brands: dict[str, list[str]] = defaultdict(list)
@@ -26,7 +26,6 @@ from urllib.parse import quote
 from urllib.request import Request, urlopen

 from config.style_config_loader import (
-    get_bitable_main_image_map,
    get_main_image_styles,
    get_tm_style_id_map,
 )
@@ -87,12 +86,29 @@ _STATIC_STYLE_NAMES = [
 ]


+# 万相「主体名称/ID」搜索词覆盖（2026-10-30）：
+# 款式身份仍然只用 style_name —— 输出目录、飞书主图表、PG 记录都以款名为键；
+# 只有搜索框里填的词可以换，用于投放主体名与内部款名不一致的款。
+# 主搜索词无结果时按「原始款名 → 天猫商品ID」逐级兜底，不会因为改名少采集。
+STYLE_SEARCH_KEYWORDS = {
+    "黑曜石吐司包": "黑曜石吐司",
+}
+
+
+def search_keywords_for(style_name: str) -> tuple[str, ...]:
+    """返回按优先级排列的万相搜索词；首个元素是主搜索词。"""
+    keyword = STYLE_SEARCH_KEYWORDS.get(style_name, "").strip()
+    if not keyword or keyword == style_name:
+        return (style_name,)
+    return (keyword, style_name)
+
+
 def _get_style_names() -> list[str]:
-    """优先从飞书读取有主图表配置的款式列表，失败则回退硬编码。"""
+    """采集目标=身份齐全的天猫款式；主图飞书表地址只影响写表，不影响采集。"""
    try:
-        cfg = get_bitable_main_image_map()
+        cfg = get_main_image_styles()
        if cfg:
-            return sorted(cfg.keys())
+            return sorted(cfg)
    except Exception as exc:
        print(f"[WARN] 动态读取主图款式列表失败，使用硬编码回退: {exc}")
    return list(_STATIC_STYLE_NAMES)
@@ -1657,17 +1673,37 @@ def aggregate_rows(rows: list[dict[str, Any]]) -> dict[str, dict[str, Any]]:
    return aggregated


-def collect_style(page, style_name: str, max_pages: int, item_ids: list[str] | None = None) -> dict[str, Any]:
-    search_style_by_subject(page, style_name)
+def collect_style(
+    page,
+    style_name: str,
+    max_pages: int,
+    item_ids: list[str] | None = None,
+    search_keywords: tuple[str, ...] | list[str] | None = None,
+) -> dict[str, Any]:
+    # 搜索词可与款名不同（见 STYLE_SEARCH_KEYWORDS），但 style_name 始终是结果身份，
+    # 输出目录/飞书/PG 都用它，不受搜索词影响。
+    keywords = tuple(search_keywords) if search_keywords else search_keywords_for(style_name)
+    used_keyword = keywords[0]
+    if used_keyword != style_name:
+        print(f"  [SEARCH] 搜索词覆盖: {style_name} -> {used_keyword}")
+    search_style_by_subject(page, used_keyword)
    all_rows: list[dict[str, Any]] = []
    visited_signatures: set[str] = set()

-    # 先检查款式名搜索是否有结果，无结果时用商品ID回退
+    # 先检查主搜索词是否有结果，无结果时按「原始款名 → 商品ID」逐级回退
    first_check = extract_current_page_rows(page)
+    for fallback in keywords[1:]:
+        if first_check:
+            break
+        print(f"  [FALLBACK] 搜索词“{used_keyword}”无结果，改用“{fallback}”重搜")
+        search_style_by_subject(page, fallback)
+        used_keyword = fallback
+        first_check = extract_current_page_rows(page)
    if not first_check and item_ids:
        first_id = item_ids[0]
        print(f"  [FALLBACK] 款式名搜不到结果，用商品ID {first_id} 重搜")
        search_style_by_subject(page, first_id)
+        used_keyword = first_id

    for page_no in range(1, max_pages + 1):
        image_map = collect_current_page_image_map(page)
@@ -1693,6 +1729,7 @@ def collect_style(page, style_name: str, max_pages: int, item_ids: list[str] | N
    aggregated = aggregate_rows(all_rows)
    return {
        "style_name": style_name,
+        "search_keyword": used_keyword,
        "collected_at": datetime.now().isoformat(timespec="seconds"),
        "page_count": len(visited_signatures),
        "raw_row_count": len(all_rows),
@@ -1732,6 +1769,7 @@ def write_style_outputs(result: dict[str, Any], output_dir: Path) -> tuple[Path,
        f"# {style_name} 万相创意主图汇总",
        "",
        f"- 采集时间: {result.get('collected_at', '')}",
+        f"- 搜索词: {result.get('search_keyword') or style_name}",
        f"- 翻页数: {result.get('page_count', 0)}",
        f"- 原始行数: {result.get('raw_row_count', 0)}",
        f"- 聚合主图数: {result.get('creative_count', 0)}",
@@ -29,15 +29,19 @@ from runtime_paths import TMP_ROOT

 from gyxx_flow.adapters.browser import BrowserCookieStore
 from gyxx_flow.adapters.scrapling import ScraplingBrowser, is_browser_timeout_error
+from gyxx_flow.core.config import Settings
 from gyxx_flow.core.locks import LockManager, ResourceBusyError

 _CONFIG_NAME = "tmall-baibu.json"
+_LOGIN_PAGE_URL = "https://login.taobao.com/member/login.jhtml"
 _EXPORT_TIMEOUT_SECONDS = 240
 _EXPORT_MAX_ATTEMPTS = 3
 _EXPORT_RETRY_DELAYS_SECONDS = (5, 15)
 _PAGE_TIMEOUT_MS = 60_000
 _IMPORT_TIMEOUT_SECONDS = 300
+_IMPORT_ACCEPTED_AFTER_PROCESSING_SECONDS = 120
 _STATUS_SETTLE_TIMEOUT_SECONDS = 5
+_ACCOUNT_LOCK_TIMEOUT_SECONDS = 120
 _RESULT_REFRESH_INTERVAL_SECONDS = 8
 _DRAFT_ACTION_TIMEOUT_SECONDS = 30
 _DRAFT_NAVIGATION_TIMEOUT_SECONDS = 15
@@ -103,6 +107,7 @@ _LOGIN_SECOND_FACTOR_MARKERS = (
    "安全验证",
    "二次验证",
    "验证码",
+    "滑块",
 )


@@ -805,6 +810,41 @@ def _find_password_login_form(
    return locate()


+def _is_login_agreement_visible(page: Any) -> bool:
+    """Return whether Tmall's 服务协议及隐私保护 popup is blocking the login."""
+
+    for context in _page_contexts(page):
+        get_by_role = getattr(context, "get_by_role", None)
+        if not callable(get_by_role):
+            continue
+        try:
+            dialogs = get_by_role("dialog")
+            for index in range(min(dialogs.count(), 8)):
+                dialog = dialogs.nth(index)
+                try:
+                    if not dialog.is_visible():
+                        continue
+                    text = _normalized_page_text(dialog.inner_text(timeout=2_000))
+                except Exception:
+                    continue
+                if "服务协议" in text or "隐私" in text:
+                    return True
+        except Exception:
+            continue
+    return False
+
+
+def _click_login_agreement_accept(page: Any) -> bool:
+    for context in _page_contexts(page):
+        for candidate in _visible_text_candidates(context, "同意"):
+            try:
+                candidate.click(timeout=8_000)
+                return True
+            except Exception:
+                continue
+    return False
+
+
 def _login_if_needed(
    entry: BaibuEntry,
    page: Any,
@@ -817,6 +857,11 @@ def _login_if_needed(
        return False
    login_url = _is_login_url(str(getattr(page, "url", "")))
    form = _find_password_login_form(page, use_password_form=False)
+    if form is None and not login_url:
+        # A redirect can land on a scanner-tab seller login page (qianniu)
+        # whose URL carries no login marker; probe the password tab before
+        # deciding there is nothing to log in to.
+        form = _find_password_login_form(page, use_password_form=True)
    if not login_url and form is None:
        return False

@@ -852,10 +897,92 @@ def _login_if_needed(
        raise BaibuApplyError(
            f"{entry.name} 自动填写账号密码并提交登录失败（{type(exc).__name__}）"
        ) from exc
+    # The first password submit can pop the 服务协议及隐私保护 confirmation
+    # which swallows the submit; accept it and submit once more so the login
+    # actually runs.  Both popups and form re-renders are tolerated.
+    for _attempt in range(2):
+        wait_for_timeout = getattr(page, "wait_for_timeout", None)
+        if callable(wait_for_timeout):
+            wait_for_timeout(2_000)
+        else:
+            time.sleep(2)
+        if not _is_login_agreement_visible(page):
+            break
+        if not _click_login_agreement_accept(page):
+            break
+        print(f"[百补] {entry.name} 已确认登录服务协议弹窗，重新提交登录")
+        refreshed = _find_password_login_form(page, use_password_form=True)
+        if refreshed is None:
+            break
+        _context, username, secret_input, submit = refreshed
+        try:
+            username.fill(credentials.username)
+            secret_input.fill(credentials.password)
+            submit.click(timeout=15_000)
+        except Exception as exc:
+            raise BaibuApplyError(
+                f"{entry.name} 协议确认后重新提交登录失败（{type(exc).__name__}）"
+            ) from exc
    print(f"[百补] {entry.name} 已提交密码登录，等待店铺身份确认")
    return True


+def _goto_tolerant(page: Any, url: str) -> None:
+    """Navigate, tolerating an aborted load when a redirect takes over.
+
+    An unauthenticated apply URL is answered with a redirect chain that can
+    abort the original navigation from inside the browser; treat that as the
+    navigation landing wherever the chain pointed and let it settle.
+    """
+
+    try:
+        page.goto(url, wait_until="domcontentloaded", timeout=_PAGE_TIMEOUT_MS)
+    except Exception as exc:
+        if "ERR_ABORTED" not in str(exc):
+            raise
+    wait_for_timeout = getattr(page, "wait_for_timeout", None)
+    if callable(wait_for_timeout):
+        wait_for_timeout(_LOGIN_FORM_SETTLE_MILLISECONDS)
+    else:
+        time.sleep(_LOGIN_FORM_SETTLE_MILLISECONDS / 1000)
+
+
+def _explicit_password_login(
+    entry: BaibuEntry,
+    page: Any,
+    *,
+    manual_login: bool,
+) -> None:
+    """Drive the Tmall password form when the apply page hid its login state.
+
+    The myseller/qn apply SPA can sit on an SSO interstitial that never shows
+    a login URL, or render the cached shop name on an expired session.  In
+    both cases ``_login_if_needed`` sees nothing to do.  Re-entering the apply
+    URL redirects to the seller login page (qianniu) which carries the right
+    merchant context; only when that shows no form fall back to the plain
+    login page.  Login there, return to the apply URL, and confirm identity.
+    """
+
+    print(f"[百补] {entry.name} 显式重新进入报名页并走密码登录")
+    _goto_tolerant(page, entry.apply_url)
+    wait_for_timeout = getattr(page, "wait_for_timeout", None)
+    if callable(wait_for_timeout):
+        wait_for_timeout(_LOGIN_FORM_SETTLE_MILLISECONDS * 3)
+    else:
+        time.sleep(_LOGIN_FORM_SETTLE_MILLISECONDS / 1000 * 3)
+    login_attempted = _login_if_needed(entry, page, manual_login=manual_login)
+    if not login_attempted:
+        _goto_tolerant(page, _LOGIN_PAGE_URL)
+        login_attempted = _login_if_needed(entry, page, manual_login=manual_login)
+    _goto_tolerant(page, entry.apply_url)
+    _wait_for_authenticated(
+        entry,
+        page,
+        login_attempted=login_attempted,
+        manual_login=manual_login,
+    )
+
+
 def _click_text(page: Any, labels: Iterable[str]) -> bool:
    contexts = _page_contexts(page)
    for label in labels:
@@ -1040,7 +1167,10 @@ def _open_batch_import(
                f"{entry.name} 页面已有批量导入操作正在处理，已停止，避免重复提交"
            )
        time.sleep(0.5)
-    raise BaibuApplyError(f"{entry.name} 页面没有找到可用的「商品批量导入」入口")
+    raise BaibuApplyError(
+        f"{entry.name} 页面没有找到可用的「商品批量导入」入口"
+        "（活动可能已报名/已结束或会话无效，请人工打开报名页确认）"
+    )


 def _wait_import_result(
@@ -1062,6 +1192,7 @@ def _wait_import_result(
    last_processing_log = 0.0
    last_refresh = 0.0
    observed_new_operation = False
+    first_new_operation_at: float | None = None
    while time.monotonic() < deadline:
        text = _page_text(page)
        operation = _latest_batch_import_status(text)
@@ -1070,6 +1201,8 @@ def _wait_import_result(
            or operation.signature != previous_operation.signature
        ):
            observed_new_operation = True
+            if first_new_operation_at is None:
+                first_new_operation_at = time.monotonic()
            if operation.state == "success":
                print(
                    f"[百补] {entry.name} 页面已完成导入处理，"
@@ -1082,6 +1215,21 @@ def _wait_import_result(
                    f"忽略行结果并按导入提交成功：{operation.detail}"
                )
                return
+        if (
+            observed_new_operation
+            and first_new_operation_at is not None
+            and time.monotonic() - first_new_operation_at
+            >= _IMPORT_ACCEPTED_AFTER_PROCESSING_SECONDS
+        ):
+            # Tmall has already acknowledged the batch ("执行了商品批量导入
+            # 操作，正在处理中"); the submission itself is the business
+            # effect, so a slow per-row summary must not fail the run.
+            print(
+                f"[百补] {entry.name} 天猫已受理批量导入，"
+                f"{_IMPORT_ACCEPTED_AFTER_PROCESSING_SECONDS} 秒内未返回行结果，"
+                "按导入提交成功，行结果以报名页面为准"
+            )
+            return
        if observed_new_operation and any(marker in text for marker in _SUCCESS_MARKERS):
            print(f"[百补] {entry.name} 页面已确认导入提交，忽略行结果")
            return
@@ -1284,10 +1432,21 @@ def _complete_draft_products(page: Any, entry: BaibuEntry) -> None:
    if not entry.complete_drafts_after_import:
        return

-    if not _click_draft_tab(page):
-        count = _draft_count(_page_text(page))
-        if count is None and not _has_visible_text(page, ("完善商品",)):
-            raise BaibuApplyError(f"{entry.name} 未找到草稿页入口")
+    # Right after the import acknowledgement the SPA may still be hydrating;
+    # the draft tab, the draft count and the per-row action rarely all appear
+    # on the first probe, so retry briefly before declaring the page broken.
+    deadline = time.monotonic() + _DRAFT_NAVIGATION_TIMEOUT_SECONDS
+    while time.monotonic() < deadline:
+        if _click_draft_tab(page):
+            break
+        text = _page_text(page)
+        if _draft_count(text) is not None or _has_visible_text(page, ("完善商品",)):
+            break
+        if any(marker in text for marker in _DRAFT_EMPTY_MARKERS):
+            break
+        time.sleep(0.5)
+    else:
+        raise BaibuApplyError(f"{entry.name} 未找到草稿页入口")

    draft_count, action_count = _wait_for_draft_state(page, entry)
    if draft_count == 0:
@@ -1439,6 +1598,8 @@ def _has_account_cookie(
 def _persist_account_state(
    browser: ScraplingBrowser,
    entry: BaibuEntry,
+    *,
+    session_lock_held: bool = False,
 ) -> bool:
    """Publish a verified login to the account vault without leaking cookies."""

@@ -1463,8 +1624,12 @@ def _persist_account_state(
    )
    cookies = _account_cookies(cookies, required_domains)
    if not _has_account_cookie(cookies, required_names):
+        names = sorted({str(item.get("name", "")) for item in cookies})
+        preview = "、".join(names[:12]) if names else "无"
        raise BaibuApplyError(
-            f"{entry.name} 登录后未获得可用的账号 cookie，未更新账号登录态"
+            f"{entry.name} 登录后未获得可用的账号 cookie"
+            f"（域内 cookie：{preview}；要求含：{'、'.join(required_names)}），"
+            "未更新账号登录态"
        )

    raw_state = context.storage_state()
@@ -1472,23 +1637,33 @@ def _persist_account_state(
    filtered_state = _account_storage_state(state, cookies, required_domains)
    cookie_file, storage_state_file = paths
    store = BrowserCookieStore(cookie_file, storage_state_file)
-    lock_root = cookie_file.parent.parent / "locks"
-    try:
-        with LockManager(lock_root).acquire(
-            f"browser-account:{entry.account_id}",
-            owner=f"tmall-baibu:{entry.entry_id}",
-            timeout_seconds=0,
-        ):
+
+    def _save() -> None:
+        try:
            store.save_cookies(cookies)
            store.save_storage_state(filtered_state)
-    except ResourceBusyError as exc:
-        raise BaibuApplyError(
-            f"{entry.name} 账号登录态正在被其他进程更新，未覆盖账号登录态"
-        ) from exc
-    except (OSError, ValueError, TypeError) as exc:
-        raise BaibuApplyError(
-            f"{entry.name} 保存账号登录态失败（{type(exc).__name__}）"
-        ) from exc
+        except (OSError, ValueError, TypeError) as exc:
+            raise BaibuApplyError(
+                f"{entry.name} 保存账号登录态失败（{type(exc).__name__}）"
+            ) from exc
+
+    if session_lock_held:
+        # The whole browser session already holds the scheduler-side
+        # browser-account lock; re-acquiring here would self-deadlock.
+        _save()
+    else:
+        lock_root = Settings.from_env().data_root / "state" / "locks"
+        try:
+            with LockManager(lock_root).acquire(
+                f"browser-account:{entry.account_id}",
+                owner=f"tmall-baibu:{entry.entry_id}",
+                timeout_seconds=0,
+            ):
+                _save()
+        except ResourceBusyError as exc:
+            raise BaibuApplyError(
+                f"{entry.name} 账号登录态正在被其他进程更新，未覆盖账号登录态"
+            ) from exc
    print(f"[百补] 已将 {entry.account_id} 的已验证登录态同步到账号 vault")
    return True

@@ -1498,6 +1673,33 @@ def _apply_entry(
    template: Path,
    *,
    manual_login: bool = False,
+) -> None:
+    # Hold the scheduler-side account browser lock for the whole session.
+    # The resident keepalive heartbeats the same profile on an interval and
+    # its timeout-0 acquire skips busy accounts, so this lock is what stops
+    # the heartbeat from killing the import mid-run (TargetClosedError /
+    # silent page death).  Members must use the same lock directory as the
+    # scheduler (Settings.data_root/state/locks), not one derived from the
+    # vault paths.
+    locks = LockManager(Settings.from_env().data_root / "state" / "locks")
+    try:
+        with locks.acquire(
+            f"browser-account:{entry.account_id}",
+            owner=f"tmall-baibu:{entry.entry_id}",
+            timeout_seconds=_ACCOUNT_LOCK_TIMEOUT_SECONDS,
+        ):
+            _apply_entry_locked(entry, template, manual_login=manual_login)
+    except ResourceBusyError as exc:
+        raise BaibuApplyError(
+            f"{entry.name} 账号浏览器正被保活/登录占用，请稍后重试"
+        ) from exc
+
+
+def _apply_entry_locked(
+    entry: BaibuEntry,
+    template: Path,
+    *,
+    manual_login: bool = False,
 ) -> None:
    browser = _start_browser()
    try:
@@ -1516,14 +1718,36 @@ def _apply_entry(
            page,
            manual_login=manual_login,
        )
-        _wait_for_authenticated(
-            entry,
-            page,
-            login_attempted=login_attempted,
-            manual_login=manual_login,
-        )
+        try:
+            _wait_for_authenticated(
+                entry,
+                page,
+                login_attempted=login_attempted,
+                manual_login=manual_login,
+            )
+        except BaibuApplyError:
+            if login_attempted or manual_login:
+                raise
+            # The apply page may sit on an SSO interstitial without a login
+            # URL, so the passive check above never triggers the credentials;
+            # fall back to the explicit password login once.
+            _explicit_password_login(entry, page, manual_login=manual_login)
+            login_attempted = True
        browser.persist_bound_state()
-        _persist_account_state(browser, entry)
+        try:
+            _persist_account_state(browser, entry, session_lock_held=True)
+        except BaibuApplyError as exc:
+            # A failed vault refresh can also mean the shop name was rendered
+            # from cache on top of a guest session; retry once through the
+            # explicit login before giving up on the vault refresh.  The
+            # refresh only caches this session for later runs and must never
+            # block the import submission itself.
+            print(f"[WARN] {exc}；尝试显式重新登录后重试", file=sys.stderr)
+            try:
+                _explicit_password_login(entry, page, manual_login=manual_login)
+                _persist_account_state(browser, entry, session_lock_held=True)
+            except BaibuApplyError as retry_exc:
+                print(f"[WARN] {retry_exc}；本次继续执行报名导入", file=sys.stderr)
        print(f"[百补] 已确认账号 {entry.account_id}，打开 {entry.name} 报名入口")
        previous_operation = _open_batch_import(page, entry)
        time.sleep(1)
Author	SHA1	Message	Date
wangyunlong	6f1c193f4b	feat(product-commerce): 款式主键统一 style_id 并下线评价采集采集端与 data-hub 统一用 style_id (= dim_style.id) 关联款式，style_name 降级为显示冗余。款式主键： - 14 张采集表加 style_id BIGINT + style_match_method + _style_id_idx 索引 - 一律逻辑外键：不建 FOREIGN KEY，早期 10 个 _style_id_fk 在 schema.sql 里幂等 DROP - loader 下发 dim_style.id；db 层 attach_style_ids / resolve_style_ids_by_product 兜底 - 新增 backfill_style_id.py 三段回填（款名 / 日报按商品 ID / 广告表按 spu_id） - 新增视图 v_style_id_coverage、v_style_unmatched_product、v_style_unmatched_source - 实库补录 199 个商品 ID 到 67 款 spus；日报 style_id 覆盖 49.1% -> 79.0%， tm 广告 99.4% / jd 92.4% / dy 83.2%，其余 11 张 100% 评价采集下线： - product_reviews 归档为 product_reviews_archived_20261101 - 差评改由 shop.{tmall,jd,dy}_negative_reviews.daily 采集，消费方读 v_negative_reviews - 删除 import_product_reviews.py / orchestrate_review_collection.py 及相关命令与绑定 - dy_product_scraping 默认流程摘除评论采集，保留 --reviews-only 天猫广告： - 迁移到独立表 tm_product_ad_daily_metrics，不再写 product_daily_metrics.raw_data 同批附带： - 光合上传支持天猫新店 new_bag（仅黑曜石系列） - 京东商智双月日历排除 -diff-month 补位格 - 万相主体搜索词覆盖 STYLE_SEARCH_KEYWORDS - ERP 部分编码失败时用已成功编码销量兜底，避免整款归零	2026-11-01 16:43:27 +08:00
wangyunlong	622a609800	feat(product-commerce): 系列聚合去重与采集开关两层语义配置来源收敛 - 系列平台编码 tm/jd/dy/jd_self 与 9 类飞书地址改为整系列并集去重后挂锚点，子款编码保留在 series.platform_ids 备查；多表冲突保留最高优先级并告警 - 写表脚本（当日/天猫主图/京东主图）去掉本地 JSON 回退，地址缺失或读取失败直接 fail closed，避免库与表地址不一致被静默掩盖 - 测试注入 records_provider 时禁用配置缓存读写，防止假数据写入生产快照采集开关（dim_style.collect_enabled） - 锚点为系列总开关：false 则整系列不采 - 锚点为 true 时，子款自身 false 仅把自己的 ERP 编码排除出聚合，其余子款照常 - 排除结果记录在 series.erp_excluded 并在运行时提示聚水潭全店铺日报防错 - 列布局 fail closed、行 td 数校验、满页判定 + 总页数兜底、值域交叉校验 - 新增单日总量闸：超近 14 天中位数 8 倍拒绝落库其他 - 新增电商费用归集配置 ad_cost_rollup.json 与积分兑换周报采集 - 店铺情报、供应链到货同步、通知路由与相关配置/文档同步更新 - 忽略本地 dsh workbench 脚手架（含 node_modules）	2026-10-30 11:49:31 +08:00
wangyunlong	2f31d5872f	chore(product): 废弃并删除 db/sync_sku_master.py 与上一并废弃的 sync_dim_style.py 同类问题，且更彻底： - 身份列自投影：脚本读 runtime_workflow_configs()（现已直读 dim_style），再把 jd_spus/dy_spus/tm_spus/jd_self_skus/erp_codes 全量 EXCLUDED 覆盖写回 dim_style。对「同系列只填主款平台编码」的子款，它会走动态配置兜底把旧商品 ID 反写，绕过系列共享规则 - 唯一还写的 bitable_sku_master_url/base/table 是 sales_bitable_url 的副本：生产库 24 行与 bitable_weekly_* 完全相同、无一行独有，且采集端、data-hub 后端、运维脚本均无任何读取点（data-hub 审计文档标注「flow 只写，全生态不读」）删除后款式身份的唯一写入口是 data-hub「产品生命进程」建档页。 bitable_sku_master_* 三列保留在库中不删（存量值不动，需要时可低成本恢复）。 - 删除 src/gyxx_flow/modules/product_commerce/db/sync_sku_master.py（170 行） - 移除 config/runtime-bindings.json 绑定（cdp_port 22071）与 source-manifests 条目 - 测试计数同步：PRODUCTION_PYTHON 64 -> 63，集成绑定 157 -> 156 - docs/dynamic-workflow-config.md 合并两个废弃脚本的说明验证：tests 根级套件 11 failed / 820 passed / 1 skipped，与改动前一致。	2026-10-28 14:51:44 +08:00
wangyunlong	423f8d1f75	chore(product): 废弃并删除 db/sync_dim_style.py 采集端运行时身份已改为直接读 dim_style，这个脚本的投影方向变成"自己写自己"：既补不了旧 cmt_styles 台账数据，upsert 又是 EXCLUDED 全量覆盖，跑一次有可能把空值写回 dim_style。款式新增与改款统一走 data-hub「产品生命进程」建档页。 - 删除 src/gyxx_flow/modules/product_commerce/db/sync_dim_style.py（390 行） - 移除 config/runtime-bindings.json 绑定（cdp_port 22070）与 source-manifests 条目 - 测试计数同步：PRODUCTION_PYTHON 65 -> 64，集成绑定 158 -> 157 - docs/dynamic-workflow-config.md 补废弃说明；运行时路径图与表说明由 cmt_styles 更正为 dim_style - db/__init__.py 表来源说明同步验证：tests 根级套件 11 failed / 820 passed，与改动前完全一致（失败均为清单 sha256 与既有环境问题，非本次引入）。	2026-10-28 14:39:50 +08:00
wangyunlong	ee6fd4170e	feat(product): 取款主源切到 dim_style，平台编码与销量表按系列共享产品生命进程建档（dim_style）成为款式身份唯一权威，采集端不再依赖 workflow_dynamic_styles + cmt_styles 双份维护： - workflow_config 运行时改读 dim_style（ERP 编码、四平台商品 ID、销量表、系列归属），cmt_styles 降级为 dim_style 不存在时的兜底；source_record_id 前缀由 cmt_style 改为 lifecycle_style - merge_lifecycle_style_records 把 series_name / is_series_anchor / sales_bitable_url 写进运行时记录，建档地址优先于动态配置 - StyleConfigLoader 新增 _apply_series_sharing：按系列标注各平台编码的承担款，子款继承主款销量表，但不复制平台商品编码，避免同一份访客/加购被同系列多款重复入账 - ERP 三道门（ERP 昨日采集、全店铺日报、月度销量表）只校验款式编码；平台编码缺失时提示「同系列某款已在采，本款只采 ERP 销量」 - CACHE_SCHEMA_VERSION 5 -> 6，旧缓存自动失效系（系列）级配置结果：ERP 门可采款数 33 -> 53，全店铺日报取款计划 42 -> 198。	2026-10-28 14:20:59 +08:00
wangyunlong	3cf7048140	fix(content-marketing): 笔记标题入库前剥除话题链接段与裸 URL 登记人常把抖音「复制分享文案」整段粘进发布笔记标题列，含 [话题](url) 链接段（个别为 {url} 花括号包裹）和裸 URL。新增 clean_note_title：剥链接后非空则采用；整段皆链接时取方括号内话题文本去重拼接；仍为空则保留原文兜底，避免标题落空。	2026-10-27 10:10:33 +08:00
wangyunlong	47ed6ec3f6	fix(content-marketing): 笔记归属款改挂 dim_style 主档 id data-hub 按 dim_style（系列锚点款）口径消费 cmt_notes_master.style_id，直接写 cmt_styles.id 在款式主档重排后变成悬挂 id，笔记在系统中消失。新增 db.get_dim_style_id，采集写入（feishu_comment_batch / sync_notes_master）与指标回填（sync_metrics_to_cmt_notes）统一切换为 dim_style 优先、按款名解析；新款未收录款式库时回退 cmt_styles.id 并告警，待收录后归一。	2026-10-27 09:58:27 +08:00
wangyunlong	b1861f69cd	chore(workbench): 移除 dsh 工作台集成工作台以 deepseek-harness 为 agent 底座，把现有工作流以 dsh 插件接入 Web UI。该路线已停用，移除整个 workbench/ 与配套文档，避免留下点了启动不起来的死入口。 - 删除 workbench/：宿主插件 gyxx-workbench.mjs、侧边栏面板前端、cordis 配置模板、启动脚本、独立面板与 node 冒烟测试 - 删除 docs/workbench.md 与 README 的「智能工作台」章节 - .gitignore 去掉 workbench 本地产物两条规则 - 基线结构测试的 docs 白名单同步移除 workbench.md 工作流能力不受影响：gyxx console、调度器、运行记录与诊断接口均未改动。	2026-10-19 11:19:31 +08:00
wangyunlong	2ef635be0e	feat(ops): 执行日志与 cookie 失效遥测入库新增 PostgreSQL 运维审计层，本地 run.json 仍是权威源，数据库做跨机器可查、可长期留存的关系型镜像。数据面： - ops_workflow_runs / ops_workflow_run_steps：每次运行与每次步骤尝试一行，保留状态、模式、耗时、exit_code、错误与 journal 路径 - ops_account_cookie_events：登录态观测事件流，只存 cookie 指纹，不落明文值 - ops_account_cookie_lifetimes / ops_account_cookie_stats：一份 cookie 的实际寿命与账号级平均/中位/最短/最长，用于发现失效规律；同时记录 cookie 自带的声明过期时间，可与实际寿命对比区分自然到期与服务端提前吊销采集点： - RunIndex.index_journal 镜像运行记录，覆盖 CLI、控制台与调度器全部路径 - CookieGate 的 gate 判定与 AccountKeepaliveManager 的周期探活（默认 10 分钟）共同提供 cookie 观测，探活决定失效时间的精度可用性： - 新增 gyxx audit runs\|cookies\|cookie-stats\|schema，支持 --json - 首次写入自动幂等建表，建表用 advisory lock 串行化，避免并发 DDL 死锁 - 审计写入永不抛出异常：连续失败熔断冷却并落 logs/ops-audit 日志，数据库不可用不影响工作流执行	2026-10-19 11:01:14 +08:00
wangyunlong	cf85985753	chore: 忽略 .mimosa/.workbuddy 本地工具状态目录	2026-10-14 18:40:11 +08:00
wangyunlong	e5761bd828	feat(deploy): Windows 自动登录自启脚本与运行手册补章 autologon.ps1 交互输密码启用/撤销 AutoAdminLogon（DefaultPassword 仅 Administrators 可读），补齐「开机自动登录 → Startup keepalive → 控制台+ 调度器」自启链唯一缺口；采集依赖 CDP 交互会话，不能用服务/SYSTEM 任务替代。runbook 增补开机自启与重启验收章节。	2026-10-14 18:40:11 +08:00
wangyunlong	9feb188d93	fix(cookie-gate): 必需 cookie 改全量校验并收紧各账号必需名单 required_cookie_names 原语义是「任一必需名命中即有效」，单个匿名 cookie （如小红书 a1）就能把游客会话判为有效；改为全部必需名都存在、存活且域名匹配才算通过。随之按实际登录产物收紧名单：jd 去 pin、erp321 留 u_id+u_co_id、小红书改 customer-sso-sid+solar.beaker.session.id、星图仅 sid_guard。cookie_gate 用例冻结时钟改相对时钟防过期漂移。	2026-10-14 18:38:03 +08:00
wangyunlong	7b305b0b12	fix(content): 蝉妈妈导出标签页崩溃自愈与 cookie 文件格式对齐导出点击后 tab 死亡（渲染进程被清理）时：页面级死亡先同 context 重开新页（facade 新增 reopen_page），context 也死亡才整套重建驱动，按 profile → cookie 文件 → 账密顺序恢复会话，两种自愈各限一次防死循环。 cookie 文件改存裸列表，与 vault 同步及适配器 BrowserCookieStore 三方格式对齐（字典格式会被适配器拒绝加载）；adapters 暴露 is_browser_closed_error 供模块识别目标已关闭类异常。	2026-10-14 18:35:44 +08:00
wangyunlong	6cd563eab0	fix(product): 天猫百补报名登录链路三类阻断兜底 vault 假登录态（SPA 缓存店铺名误判已登录）与 SSO 中间态停留时改走显式账密登录兜底；登录页识别服务协议/隐私弹窗并先处理，二次验证标记补充滑块；报名导入增加账号级锁防并发挤兑。	2026-10-14 18:33:32 +08:00
wangyunlong	0cc4fafc7c	fix(product): 聚水潭 allsku 模板隐藏列错位致 daily 销量全 0 2026-10-11 模板重置把 38 列置隐藏后可见列整体左移，固定索引 45/57 落到「今日实退货量/今日仅退款量」。daily 改表头名匹配「昨日销量/昨日退货量」（weekly 暂留老索引待实测）；backfill 恢复自定义列可见（window._Cols + _RptRefresh），并把表头索引按可见列换算成行内位置防 td 缺位错读。	2026-10-14 18:31:23 +08:00
wangyunlong	9a91ce80e4	feat(product): 聚水潭销售主题分析按日采集工作流 collect_erp_sales_theme_daily.py 采集 bi.erp321.com「报表→销售主题分析→按日统计」，异步导出 + OSS 直连兜底，固定品牌+店铺口径，写 erp_sales_theme_daily_metrics （db 层建表与 upsert 同步，每日一行幂等覆盖）。注册 product.erp_sales_theme_daily.collect 手动命令，复用 erp-account 登录绑定新增 CDP 22167；目录/集成计数 60→61，新脚本纳入 CI 聚焦 lint 清单。	2026-10-14 18:29:15 +08:00
wangyunlong	97a486e97d	feat(product): 补充香港安培公开库存采集脚本与聚水潭库存适配 public_sku_inventory_sheet 与 jushuitan_inventory 适配器、hk_anpei 采集配置，支撑 product.hk_anpei_public_inventory.daily 工作流。	2026-10-09 13:45:56 +08:00
wangyunlong	36fe7b81f8	feat(cookie-gate): 补充 cookie_gate 网关模块与 deploy/watchdog 保活脚本 cookie_gate.py(账号三级门控/阻断记录 pending_reruns/恢复重跑管理器)为此前接线提交遗漏的模块文件；deploy/watchdog 为控制台与调度器的进程外保活脚本。	2026-10-09 13:45:56 +08:00
wangyunlong	dc8be88e60	chore(config): 工作流配置、清单与文档同步 workflows/schedules/commands/notification-routing 与 source-manifests 四模块清单同步登录统一化与新工作流(product.hk_anpei_public_inventory.daily 等)； docs 运维手册、控制台与耗时分布文档、CI 关键脚本清单、AGENTS 指南更新；测试与模块注册对齐。	2026-10-09 13:45:30 +08:00
wangyunlong	90282ae766	fix(product): 千川任务中心账户选择页兜底与 jd_spus jsonb 查询修复千川任务中心深链缺账户上下文时会被重定向到账户选择页，按 .account-list-item 卡片自动选中目标广告主后继续下载；jd_spus 列改 jsonb 后查询改用 jsonb_typeof/jsonb_array_length 防类型报错。	2026-10-09 13:45:30 +08:00
wangyunlong	df11e195b7	feat(product): 京东新版商智竞店采集改造入口跳转、两段式日期选择、搜索定位与表格解析适配新版商智；配合类目下拉与 XHR 拦截兜底。	2026-10-09 13:45:30 +08:00
wangyunlong	b0e1b5066d	feat(content): 自营三平台指标入库管线完善 sync_self_video_metrics 按平台隔离入库与字段口径调整，配套自营指标验收测试更新。	2026-10-09 13:45:30 +08:00
wangyunlong	1e9ddffd2f	fix(collect): 账号浏览器与小红书采集强制直连目标站点均为国内站，Clash 代理出口被蝉妈妈/小红书 WAF 拉黑(TLS 直接 reset，页面与 API 报 ERR_CONNECTION_CLOSED)。账号保活、扫码登录浏览器与小红书 ARK 采集统一加 --no-proxy-server 直连。	2026-10-09 13:45:29 +08:00
wangyunlong	41f448420d	fix(chanmama): 导出确认弹窗被浮层清理误删与代理出口拉黑修复浮层清理只删可见浮层——页面常驻 90+ 个 display:none 弹窗模板(含导出确认框模板)，误删后 Vue 只在游离节点上"打开"弹窗，页面上仅剩灰色遮罩与滚动锁，导出等待 1320s 必超时；导出等待改为逐秒弹窗状态机(可点即点/点不到先关/卡死强删并补点导出)，导出失败落截图与 HTML 诊断；浏览器强制直连并切换真 Chrome 非 stealthy 模式(StealthySession 下蝉妈妈前端会静默压制弹窗渲染)；导出按钮原生可信点击优先，防 @click 绑定在内层元素时 JS 合成点击无效。	2026-10-09 13:45:29 +08:00
wangyunlong	130f06b320	feat(cookie-gate): 调度器恢复重跑接线与控制台健康自检调度器维护循环挂载 CookieGateRecoveryManager，登录态恢复后按原业务日期自动重跑被阻断工作流并飞书确认；工作流级 required_accounts 声明接入目录解析；采集器抛 CookieExpiredError 退出 75；控制台新增 /api/health 自检端点与进程内看门狗，配合 deploy/watchdog 双层保活；安装脚本同步。	2026-10-09 13:45:29 +08:00
wangyunlong	78ea4641ce	refactor(login): 移除脚本内扫码重登，登录统一交给账号 vault 与 cookie-gate 删除 relogin_*/friday_relogin_parallel 子进程及其测试；批量补采、评论采集、星图/蒲公英/B站/抖音自营与抖店竞品脚本不再内嵌扫码等待，登录态失效统一由 cookie-gate 阻断并走 gyxx accounts login 恢复；runtime-bindings 与模块注册同步移除 content.relogin.weekly。	2026-10-09 13:45:19 +08:00