fix(后端): 区分 T+ 连接文件的缺失与字段错误

原先 OSError、ValueError、ValidationError 统一归并为「不可用
或字段不正确」，文件不存在和字段写错无法区分，排障困难。
拆分为两条消息分别指示故障类别，仍不回显文件内容与路径。

Co-Authored-By: Claude Code <noreply@anthropic.com>

.gitignore

@@ -1,5 +1,5 @@
 # AI辅助目录（整个目录忽略）
-ai-aux/
+.ai/
 
 # 依赖
 node_modules/
@@ -45,8 +45,9 @@ htmlcov/
 .mypy_cache/
 .ruff_cache/
 .tox/
-# 端到端测试产出的截图
-frontend/e2e/*.png
+# 端到端测试产出的截图（脚本以相对路径 e2e/failure.png 落盘，在不同 CWD
+# 下会落到 frontend/e2e/ 或 frontend/e2e/e2e/，用 ** 一并覆盖）
+frontend/e2e/**/*.png
 
 # 临时文件
 *.tmp
@@ -64,9 +65,14 @@ CLAUDE.md
 backend/uploads/
 # 数据库备份导出目录（⑨ 备份脚本产物）
 db/backups/
+# 客户发来的原始账套备份压缩包。
+/db/*.zip
 
 # 外部参考资料（2026-10-20）：T+12.3 数据字典压缩包与解压目录不入版本库
 docs/**/*.rar
 docs/T_12.3数据字典/
 # macOS 压缩包残留元数据
 __MACOSX/
+
+# AI 辅助生成文档的临时工作目录（脚本、预览截图），不入版本库
+.tmpdocx/

backend/.env.example

@@ -26,6 +26,23 @@ LOGIN_FAIL_LIMIT=5
 LOGIN_LOCK_MINUTES=15
 
 # ===== T+ 对接（待畅捷通确认后填写）=====
-T_PLUS_BASE_URL=
-T_PLUS_APP_KEY=
-T_PLUS_APP_SECRET=
+T_PLUS_ENABLED=false
+T_PLUS_HOST=127.0.0.1
+T_PLUS_PORT=1433
+T_PLUS_DATABASE=
+T_PLUS_USER=
+T_PLUS_PASSWORD=
+# 可显式指定项目外的只读连接 JSON；不从页面接收该路径。
+T_PLUS_CONNECTION_FILE=
+# 必须填写在客户 SQL Server 2008 R2 上验证过的驱动名称。
+# 本地 PoC 可复用已安装的 SQL Server 驱动；不自动安装或升级客户驱动。
+T_PLUS_ODBC_DRIVER=
+# 默认启用传输加密。旧实例 TLS 不兼容时，仅隔离开发环境可显式设为 false，
+# 记录原因并重新验收；程序不会自动降级。真实环境值后续按验收结果填写。
+T_PLUS_ENCRYPT=true
+# 内网自签证书可暂用 true；它不校验证书身份。部署可信证书后改为 false。
+T_PLUS_TRUST_SERVER_CERTIFICATE=true
+T_PLUS_CONNECT_TIMEOUT=10
+T_PLUS_QUERY_TIMEOUT=120
+# 可选：留空时使用仓库根目录下的 db/queries/tplus/
+T_PLUS_QUERY_DIR=

backend/README.md

@@ -83,8 +83,9 @@ app/
 | GET / POST | `/api/v1/base/products` | 产品列表（名称/规格模糊、类别/状态筛选）/ 新增（类别值域校验；名称+规格组合唯一） |
 | GET / PUT / DELETE | `/api/v1/base/products/{id}` | 详情 / 编辑 / 删除（被业务明细引用保护在业务表就位后生效） |
 | PUT | `/api/v1/base/products/{id}/status` | 启用 / 停用 |
-| GET / POST | `/api/v1/base/customers` | 客户列表（公司/负责人模糊、分类/归属筛选）/ 新增（分类值域；公司名+负责人唯一） |
+| GET / POST | `/api/v1/base/customers` | 客户列表（公司/负责人模糊、分类/归属筛选；`order_by=rating_score` 按评分排序，未评分垫底）/ 新增（分类值域；公司名+负责人唯一） |
 | GET / PUT / DELETE | `/api/v1/base/customers/{id}` | 详情 / 编辑（**不触归属**）/ 删除 |
+| PUT | `/api/v1/base/customers/{id}/rating` | 保存评级（评级计算器「保存评分」；只写 `rating_score`/`rating_time`，五项明细不落库 #25） |
 | PUT | `/api/v1/base/customers/{id}/ownership` | 归属变更（同事务写留痕；只影响变更后核算，不回溯历史 #19） |
 | GET | `/api/v1/base/customers/{id}/ownership-logs` | 归属变更留痕分页（join 人员出姓名） |
 
@@ -139,6 +140,19 @@ app/
 （`warning.enabled` / `warning.receipt.overdue.days` / `warning.approval.timeout.days`，
 V10 种子）；⑥ T+ 同步未建，同步相关条目以 `sync_available=false` 显式降级。
 
+### T+ 本地只读同步与管理
+
+T+ 数据库是 SQL Server 只读源库，业务库仍为 MySQL。Docker 恢复 PoC、备份检查和 D 盘目录约定见 `docs/T+本地恢复与同步开发手册.md`。
+
+- 连接配置在 `backend/.env` 的 `T_PLUS_*` 变量中，默认关闭，不配置时不影响主业务启动。
+- `uv run python -m app.cli tplus-stage --entity personnel` 按主键分页，将源行写入 `tplus_sync_records` staging，并记录批次和失败信息。水位表已预留，时间增量未启用，`--watermark` 会明确拒绝。
+- T+ 驱动通过 `uv sync --extra tplus` 安装到应用环境，ODBC 驱动名称必须显式配置；交付兼容目标为 SQL Server 2008 R2，2022 副本不能代替同版本验收。
+- 查询目录 `db/queries/tplus/` 随部署交付，或使用 `T_PLUS_QUERY_DIR` 指定受控位置；支持六类主表和三类明细。
+- T+ 字典中的候选字段在实际备份核验和客户确认前，不得直接覆盖岗位、工资标准、归属、业务状态或核算状态。
+- 管理 API `/api/v1/system/tplus` 全部限管理员：概况、批次/异常分页、按需报告、创建与失败重试。HTTP 只创建 QUEUED 任务；`uv run python -m app.cli tplus-worker --watch --connection-file <项目外JSON路径>` 独立处理。
+- 核心文件：`services/tplus/task_service.py`（幂等/重试）、`worker.py`（执行）、`relationship_audit.py`（字典与关联）、`query_service.py`（安全输出）、`core/tplus_lock.py`（跨提交执行锁）。原 CLI 读取与 worker 共用执行锁。
+- V15 / Alembic `e5f607a8b9c0` 增加范围、跳过数、请求键、父批次、请求人和核验结果。未启用时间增量与工资业务导入；完整运行步骤和异常处理见本地恢复手册。
+
 ## 常用命令
 
 ```bash
@@ -221,6 +235,22 @@ bcrypt 5.0 起对超过 72 字节的密码**直接抛 ValueError**（此前是
 - **留痕表的固化设计**：`ownership_change_logs` 与 `business_status_logs` 对客户/业务/人员 ID 均**无外键**（冗余固化，被引用方删除后记录仍完整可读）；RESTRICT 外键反而会把删除逼成 500。删除保护全部由服务层检查承担（人员→团队成员/客户归属引用；客户→业务引用；业务→发货/回款记录）。两张留痕表 append-only、不继承 TimestampMixin。
 - **金额契约**：schema 入参出参一律 `str`，小数超两位直接拒绝（不静默舍入）；服务层 `Decimal`，比率 `Fraction`；毛利润率出参为分数原样字符串（`"997/1000"`，同 M1 计算预览口径），前端原样展示禁止转 float。
 
+### 审批快照与方案版本化（2026-10-20 定稿，改动前先读）
+
+- **审批单带版本号与内容快照**：`approval_orders.version_no` 按 `(order_type, ref_id)` 从 1 递增，重提先撤回旧在途单、但**旧单快照不覆盖**；`snapshot_json` 统一信封 `{schema_version, order_type, captured_at, data}`，`data` 由 `approval_service.{allocation,settlement,adjustment}_content` 组装（Decimal 转 str、日期转 ISO），开关关闭的自动通过路径同样写快照。审批期间业务再改，复核者看到的仍是送审版本。
+- **两个 `version_no` 语义不同，别混用**：`business_secondary_allocations.version_no` = 方案内容第几版，**仅方案内容（参与人+比例）变化时 +1**；`approval_orders.version_no` = 同一对象第几次送审，**仅送审内容变化时 +1**。改备注不动金额 → 两者都不动；改金额 → 只有审批单版本号动；改方案比例 → 两者都动。
+- **送审内容未变不重提**：编辑业务时若**存在在途审批单**且其快照 `data` 与本次送审内容逐键相等（`approval_service.pending_matches_snapshot`，只比 `data` 不比 `captured_at`），则保留原单不撤回不重建；否则撤旧提新。方案内容未变时**不新增方案版本行**（`business_service._plan_changed` 按「人员+比例」集合比较，与行序无关）。
+- **快照的覆盖范围是「方案 + 钱 + 归属」**（业务编号/日期、客户、负责与协助人、销售额、成本额、方案参与人与比例、合计与余额），**不含产品明细行**。因此只改产品构成而金额不变时不会重提——分摊依据（谁分、分多少、基于多少钱）未变；产品明细由复核者按业务编号回业务页查看。
+- **列表不加载快照列**：快照可达上百 KB，审批单列表、工作台待办、工作台审批超时预警三处查询一律 `defer(ApprovalOrder.snapshot_json)`（与操作日志列表排除 `request_data` 同一理由，参考 `operation_log_repository.list_logs`）；详情走 `get_order` 按需读取。
+- **详情内容来源三态**：`content_source=snapshot`（送审快照）/ `live`（V11 前的历史单体无快照，只读回退当前内容并显式告警）/ `missing`（引用对象不存在且无快照）。**禁止回填历史单快照**——回填等于伪造提交时内容。
+- **二次分配方案 append-only**：`business_secondary_allocations` 行不物理删除，编辑全量替换与驳回清空只把旧行 `is_active=0`；唯一键为 `(business_id, version_no, personnel_id)`，同一版本内参与人唯一、跨版本可重复。`business_repository.list_allocations(active_only=True)` 是核算与详情回显的唯一取数口径；业务详情 `allocation_history` 返回全部版本供回溯。
+- **月结提交要锁行**：`settlement_service.submit` 取月结行必须 `for_update=True`——并发提交会让两次 `submit_order` 算出同一个审批单 `version_no`，送审次序失真。
+- **审批单 `version_no` 有唯一键兜底（V12）**：`uk_approval_orders_order_type_ref_version (order_type, ref_id, version_no)`——同一对象的第 N 次送审在库里有唯一一行。ORM 侧约束名必须与迁移 DDL 严格一致，否则 autogenerate 误判。
+- **取版本号必须是「当前读」（`approval_service._next_version`）**，两个禁忌：(a) 不能用普通 `MAX` 查询——请求级鉴权与业务共用同一 DbSession，鉴权的普通 SELECT 已把 REPEATABLE READ 快照锚定在请求开始时刻，父行锁只串行化写入时刻、**不刷新快照**，MAX 会读到旧值；(b) 不能写成 `select(func.max(...)).with_for_update()`——MySQL 官方明确「Locking clauses are not permitted in queries with aggregate functions or GROUP BY」，聚合上的 FOR UPDATE 语义无保证。正解：锁读该 ref 的现有版本行（`select(version_no)...with_for_update()`），最大值 Python 侧算（同 `no_generator` 套路）。
+- **版本号冲突的收敛口径**：`submit_order` 内不 catch、不 rollback（红线）。create 路径由外层编号重试循环整体重来（`business_service._NO_RETRY`）；编辑/月结提交路径由审计上下文统一回滚，再把 1062 翻译为 400「提交冲突（版本号生成重复），请重试」（同 `param_service` 版本号冲突模板）。唯一键是最后一道防线，不是取号正确性的前提。
+- **迁移前先修存量**：V11 把存量行 `version_no` 一律置 1，而 V11 前每次重提都新建行——同一 `(order_type, ref_id)` 存在多行 version_no=1，直接加唯一键必报 1062。V12 先重排「组内有重复」的分组（按 id 升序 1..n），其余分组不动（全量重排会改掉前端已展示过的历史版本号）。
+- **新增审批对象类型**须同时提供快照组装函数并接入 `_content_of` 的 live 回退分支，不要在各业务服务内联 JSON 结构。
+
 ### 工作台口径（M4 ⑧，改动前先读）
 
 - **只读聚合**：总览只做 SQL 聚合（SUM / COUNT DISTINCT），不复用报表的部门汇总（后者加载整月台账，过重）；人数用 `COUNT(DISTINCT personnel_id)`——二次分摊拆行会让行数虚高。
@@ -247,6 +277,17 @@ bcrypt 5.0 起对超过 72 字节的密码**直接抛 ValueError**（此前是
 - **汇总 = 明细合计**：团队/公司基数由台账分桶结果直接得出，报表侧不另算一遍；未筛选时 `团队基数 + 未分配绩效 = 全公司基数`；集成测试逐项断言。
 - **岗位筛选选项**：`GET /api/v1/base/business-roles/options` 仅要求登录（含停用岗位），供财务/领导在报表中按岗位筛选；岗位管理接口仍仅管理员。
 
+### 面向业务人员的文案（2026-10-23 定稿，改文案前先读）
+
+系统由财务与营销人员日常使用，**不是给开发者看的**。凡是会渲染到页面的文本（参数说明、错误提示、口径注释）一律按业务语言写：
+
+- **`param_registry.py` 的 `description` 是唯一显示源**，会直出到「计提参数」页参数名下方。写清「这个参数管什么、钱怎么变、什么时候该调它」，不写英文枚举名（`FULL`/`BALANCE`）、需求编号（`澄清 #10`）、内部阶段号（`M3`/`M4 ⑧`）、实现层精度约束（「禁止按 0.011111 截断存储」留在代码注释与测试里）。
+- **枚举值在页面上必须翻译**：前端 `params/index.vue` 的 `ENUM_VALUE_LABELS`（`FULL` 当月全额 / `BALANCE` 余额 / `INCLUDE` 含基本工资 / `DEDUCT` 扣除基本工资 / `EXCLUDE` 不计基本工资 / `1` 开 / `0` 关）与 `DIMENSION_LABELS`（`municipal_water` 市政水利 / `rail_transit` 轨道交通）。后端报错也不许把原始枚举值甩给用户 —— `param_service._preview_rate` 的产品类别校验用 `「市政水利、轨道交通」` 而不是 `municipal_water / rail_transit`。
+- **公式模板报错用中文模板名**：`calc_templates.TEMPLATES[code].name`，取不到才回退 code；文案要给出下一步（「请在岗位角色里改用业务员递减或信息员固定计提」），而不是只说「不支持」。
+- **`db/init/init-dev.sql` 的种子 description 必须与 registry 同源**：它只在数据卷为空时首次启动执行，已有库要跑 `ai-aux/preview/sync_param_desc.py`（读 `PARAM_REGISTRY` 逐条 UPDATE）才看得到新文案。`db/migration/V*.sql` 里的旧描述是历史记录，**不要回头改** —— 改了会让已部署库与脚本不一致。
+- **前端模板层禁止出现开发者语汇**：`澄清 / 待业务 / M3 / M4 / 一期 / 二阶段 / 暂口径 / 待确认 / 待接 / 暂不 / 待补` 只允许出现在代码注释里。改完用 grep 扫一遍 `frontend/src/views`。
+- **台账与预览里落库的说明文字（`rate_text` / `rate_note` / 聚合行 `snapshot_json.note`）同样受管**：这些会出现在绩效台账、个人明细、打印稿与计算预览上。写成「这笔为什么是这个比例、什么时候不再降」（如「回款 20 天，从起始 10‰ 按日递减（满 360 天后固定按 4‰ 计算）」），不写 `基数开关=1`、`#12 参数化`、`9.8 方案公式` 这类需求编号。**分数原文（`11/1125`）与百分比拼装（`聚合:20%+5%`）不算说明文字**，那是金额口径的可复核原文，不要为了「口语化」把它改成四舍五入的小数。
+
 ## 常见问题
 
 **起服务报 `APP_SECRET_KEY` 相关错误** —— `backend/.env` 未填或不足 32 字符。这是刻意的 fail-fast：空密钥签发 JWT 是安全问题，必须启动即失败。

backend/alembic/versions/7f4f7b918bc4_v12_审批单版本号唯一键.py

@@ -0,0 +1,71 @@
+"""v12 审批单版本号唯一键
+
+Revision ID: 7f4f7b918bc4
+Revises: f3a8c2d61b47
+Create Date: 2026-10-21 10:15:00.000000
+
+给 ``approval_orders`` 加 ``(order_type, ref_id, version_no)`` 唯一键，兜住
+「同一引用对象的第几次送审」在并发下算重。
+
+**加键前必须先重排存量**：V11 迁移用 ``ADD COLUMN version_no INT NOT NULL
+DEFAULT 1`` 把存量行一律置 1，而 V11 之前的每次重提都是**新建行**（撤回
+只改旧单 status，行不删）。于是任何有 V11 前审批历史的库里，同一
+``(order_type, ref_id)`` 存在多行且 ``version_no`` 全为 1——直接
+``ADD CONSTRAINT`` 必报 1062，迁移会停在半途。
+
+重排只针对「组内有重复」的分组（``COUNT(*) > COUNT(DISTINCT version_no)``），
+按 id 升序重排为 1..n；其余分组一个字节都不动——全量重排会把前端已展示
+过的历史版本号（``V{{ version_no }}``）改掉，属于用户可见的历史数据篡改。
+"""
+
+from collections.abc import Sequence
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "7f4f7b918bc4"
+down_revision: str | Sequence[str] | None = "f3a8c2d61b47"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_UK = "uk_approval_orders_order_type_ref_version"
+
+
+def upgrade() -> None:
+    """先重排存量重复的版本号，再加唯一键。"""
+    # 重排：只动「组内有重复版本号」的分组，按 id 升序编 1..n。
+    # 用派生表内联而非临时表——同一张临时表被 JOIN 两侧引用会报 1137
+    # "Can't reopen table"。
+    op.execute(
+        "UPDATE approval_orders ao "
+        "JOIN (SELECT id, ROW_NUMBER() OVER ("
+        "  PARTITION BY order_type, ref_id ORDER BY id) AS rn "
+        "  FROM approval_orders "
+        "  WHERE (order_type, ref_id) IN ("
+        "    SELECT order_type, ref_id FROM approval_orders "
+        "    GROUP BY order_type, ref_id "
+        "    HAVING COUNT(*) > COUNT(DISTINCT version_no))) t "
+        "ON ao.id = t.id "
+        "SET ao.version_no = t.rn"
+    )
+    # 幂等 guard：MySQL DDL 隐式提交，若加键成功而版本号未落盘，
+    # 重跑会报 1091（键已存在）。写法沿用 V9 先例。
+    op.execute(
+        "SET @uk_exists := (SELECT COUNT(*) FROM information_schema.TABLE_CONSTRAINTS "
+        "WHERE CONSTRAINT_SCHEMA = DATABASE() AND TABLE_NAME = 'approval_orders' "
+        f"AND CONSTRAINT_NAME = '{_UK}')"
+    )
+    op.execute(
+        "SET @ddl := IF(@uk_exists = 0, "
+        f"'ALTER TABLE approval_orders ADD CONSTRAINT {_UK} "
+        "UNIQUE (order_type, ref_id, version_no)', 'DO 0')"
+    )
+    op.execute("PREPARE stmt FROM @ddl")
+    op.execute("EXECUTE stmt")
+    op.execute("DEALLOCATE PREPARE stmt")
+
+
+def downgrade() -> None:
+    """删唯一键。唯一键本质是索引，DROP INDEX 不校验数据，存量是否已有
+    重复 version_no 与删索引无关；重排结果也不回滚（版本号保持 1..n 连续）。"""
+    op.execute(f"ALTER TABLE approval_orders DROP INDEX {_UK}")

backend/alembic/versions/90653814a421_v10_工作台预警规则参数.py

@@ -11,6 +11,7 @@ Create Date: 2026-10-19 17:55:48.892998
 
 「同步连续失败」规则依赖 ⑥ T+ 同步，模块未建前不设参数、不产生条目。
 """
+
 from collections.abc import Sequence
 from datetime import date
 
@@ -19,8 +20,8 @@ import sqlalchemy as sa
 
 
 # revision identifiers, used by Alembic.
-revision: str = '90653814a421'
-down_revision: str | Sequence[str] | None = 'ba0217f5091b'
+revision: str = "90653814a421"
+down_revision: str | Sequence[str] | None = "ba0217f5091b"
 branch_labels: str | Sequence[str] | None = None
 depends_on: str | Sequence[str] | None = None
 

backend/alembic/versions/c3a9e1d7b2f4_v13_客户评级落库.py

@@ -0,0 +1,51 @@
+"""v13 客户评级落库
+
+Revision ID: c3a9e1d7b2f4
+Revises: 7f4f7b918bc4
+Create Date: 2026-10-23 10:30:00.000000
+
+评级计算器从独立工具页并入客户管理（行内弹窗），评分由「不落库」改为
+保存为客户档案属性：``rating_score`` 总分 + ``rating_time`` 评级时间。
+两列均可空（null = 未评分，历史客户与新客户无需回填），人工维护字段，
+T+ 同步守则不覆盖（见 TPlusSyncMixin 的同步守则说明）。
+
+五项评分明细与命中等级都不落库：前者是打分过程的中间量，没有下游消费方；
+后者由分数按当前等级映射实时推导（等级映射是可变判定表，改了配置全部客户
+的等级随之更新），因此本迁移只落总分与评级时间两列。
+"""
+
+from collections.abc import Sequence
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "c3a9e1d7b2f4"
+down_revision: str | Sequence[str] | None = "7f4f7b918bc4"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    """customers 加评级两列，紧跟 remark 之后保持与 ORM 声明同序。
+
+    用 execute 手写而非 op.add_column：后者不带 AFTER 子句，两列会落到
+    表尾（T+ 预留列之后），与 ORM/init-dev 的声明顺序不一致。
+    """
+    op.execute(
+        "ALTER TABLE customers "
+        "ADD COLUMN rating_score TINYINT COMMENT "
+        "'评级总分（0-100，评级计算器保存；null=未评分；人工维护字段，T+ 同步不覆盖）' "
+        "AFTER remark"
+    )
+    op.execute(
+        "ALTER TABLE customers "
+        "ADD COLUMN rating_time DATETIME COMMENT "
+        "'最近评级时间（评级计算器保存评分时刷新；人工维护字段）' "
+        "AFTER rating_score"
+    )
+
+
+def downgrade() -> None:
+    """删除评级两列。评分数据随列删除，评级无下游引用、无悬空风险。"""
+    op.drop_column("customers", "rating_time")
+    op.drop_column("customers", "rating_score")

backend/alembic/versions/c7d2e5a81f30_v8_业务答复口径落实_审批开关默认_试用期参数_经理基数开关_台账分摊唯一键.py

@@ -34,10 +34,7 @@ depends_on = None
 def upgrade() -> None:
     """种子行级变更 + 台账唯一键扩展（无表结构新增/删除）。"""
     # 1. 审批开关默认值翻转（#5/#7：默认开启，业务可关闭）
-    op.execute(
-        "UPDATE param_definitions SET default_value = '1' "
-        "WHERE id IN (17, 18)"
-    )
+    op.execute("UPDATE param_definitions SET default_value = '1' WHERE id IN (17, 18)")
     # 时点化生效行：插新行不改历史（10-19 前仍按 0 解析）
     op.bulk_insert(
         sa.table(
@@ -188,6 +185,4 @@ def downgrade() -> None:
     )
 
     op.execute("DELETE FROM param_values WHERE id IN (23, 24)")
-    op.execute(
-        "UPDATE param_definitions SET default_value = '0' WHERE id IN (17, 18)"
-    )
+    op.execute("UPDATE param_definitions SET default_value = '0' WHERE id IN (17, 18)")

backend/alembic/versions/d4e5f6a7b8c9_tplus_sync_staging.py

@@ -0,0 +1,122 @@
+"""T+ 只读同步批次、错误、staging 和增量水位。
+
+Revision ID: d4e5f6a7b8c9
+Revises: c3a9e1d7b2f4
+"""
+
+from collections.abc import Sequence
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.mysql import BIGINT
+
+revision: str = "d4e5f6a7b8c9"
+down_revision: str | Sequence[str] | None = "c3a9e1d7b2f4"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    """增加 T+ 源更新时间和同步审计/staging 表。"""
+    for table in ("personnel", "products", "customers", "businesses", "shipments", "receipts"):
+        op.execute(
+            f"ALTER TABLE {table} ADD COLUMN source_updated_at DATETIME "
+            "COMMENT 'T+ 源数据更新时间（未确认时可为空）' AFTER synced_at"
+        )
+
+    op.create_table(
+        "tplus_sync_batches",
+        sa.Column("id", BIGINT(unsigned=True), autoincrement=True, nullable=False, comment="主键ID"),
+        sa.Column("batch_no", sa.String(length=32), nullable=False, comment="同步批次号"),
+        sa.Column("sync_mode", sa.String(length=16), nullable=False, comment="同步模式：FULL全量 INCREMENTAL增量"),
+        sa.Column("status", sa.String(length=16), nullable=False, comment="状态：RUNNING/SUCCEEDED/PARTIAL/FAILED"),
+        sa.Column("source_database", sa.String(length=128), nullable=True, comment="T+ 源数据库名"),
+        sa.Column("started_at", sa.DateTime(), nullable=False, comment="开始读取时间"),
+        sa.Column("finished_at", sa.DateTime(), nullable=True, comment="结束时间"),
+        sa.Column("watermark_from", sa.DateTime(), nullable=True, comment="增量起始水位"),
+        sa.Column("watermark_to", sa.DateTime(), nullable=True, comment="本批次结束水位"),
+        sa.Column("source_record_count", sa.Integer(), nullable=False, server_default="0", comment="源记录数"),
+        sa.Column("staged_record_count", sa.Integer(), nullable=False, server_default="0", comment="写入 staging 的记录数"),
+        sa.Column("failed_record_count", sa.Integer(), nullable=False, server_default="0", comment="失败记录数"),
+        sa.Column("error_message", sa.Text(), nullable=True, comment="批次错误摘要"),
+        sa.Column("create_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP"), comment="创建时间"),
+        sa.Column("update_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP"), comment="更新时间"),
+        sa.PrimaryKeyConstraint("id", name="pk_tplus_sync_batches"),
+        sa.UniqueConstraint("batch_no", name="uk_tplus_sync_batches_batch_no"),
+        mysql_engine="InnoDB", mysql_charset="utf8mb4", mysql_collate="utf8mb4_general_ci", comment="T+ 同步批次表（记录读取范围、结果和对账统计）",
+    )
+    op.create_index("idx_tplus_sync_batches_status", "tplus_sync_batches", ["status"])
+    op.create_index("idx_tplus_sync_batches_started_at", "tplus_sync_batches", ["started_at"])
+
+    op.create_table(
+        "tplus_sync_errors",
+        sa.Column("id", BIGINT(unsigned=True), autoincrement=True, nullable=False, comment="主键ID"),
+        sa.Column("batch_no", sa.String(length=32), nullable=False, comment="同步批次号"),
+        sa.Column("entity_type", sa.String(length=32), nullable=False, comment="业务实体类型"),
+        sa.Column("source_table", sa.String(length=128), nullable=False, comment="T+ 源表"),
+        sa.Column("source_id", sa.String(length=64), nullable=True, comment="T+ 源主键"),
+        sa.Column("source_line_no", sa.String(length=64), nullable=True, comment="T+ 明细行号或明细主键"),
+        sa.Column("error_code", sa.String(length=32), nullable=False, comment="错误编码"),
+        sa.Column("error_message", sa.Text(), nullable=False, comment="错误说明"),
+        sa.Column("retry_count", sa.Integer(), nullable=False, server_default="0", comment="已重试次数"),
+        sa.Column("create_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP"), comment="创建时间"),
+        sa.Column("update_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP"), comment="更新时间"),
+        sa.PrimaryKeyConstraint("id", name="pk_tplus_sync_errors"),
+        mysql_engine="InnoDB", mysql_charset="utf8mb4", mysql_collate="utf8mb4_general_ci", comment="T+ 同步错误表（记录失败行和重试信息）",
+    )
+    op.create_index("idx_tplus_sync_errors_batch_no", "tplus_sync_errors", ["batch_no"])
+    op.create_index("idx_tplus_sync_errors_entity_type", "tplus_sync_errors", ["entity_type"])
+
+    op.create_table(
+        "tplus_sync_records",
+        sa.Column("id", BIGINT(unsigned=True), autoincrement=True, nullable=False, comment="主键ID"),
+        sa.Column("batch_no", sa.String(length=32), nullable=False, comment="同步批次号"),
+        sa.Column("entity_type", sa.String(length=32), nullable=False, comment="业务实体类型"),
+        sa.Column("source_table", sa.String(length=128), nullable=False, comment="T+ 源表"),
+        sa.Column("source_id", sa.String(length=64), nullable=False, comment="T+ 源主键"),
+        sa.Column("source_no", sa.String(length=64), nullable=True, comment="T+ 原始编号"),
+        sa.Column("source_line_no", sa.String(length=64), nullable=True, comment="T+ 明细行号或明细主键"),
+        sa.Column("source_updated_at", sa.DateTime(), nullable=True, comment="T+ 源数据更新时间"),
+        sa.Column("row_hash", sa.String(length=64), nullable=False, comment="原始行 SHA-256"),
+        sa.Column("state", sa.String(length=16), nullable=False, server_default="PENDING", comment="状态：PENDING/APPLIED/SKIPPED/ERROR"),
+        sa.Column("payload_json", sa.JSON(), nullable=False, comment="原始行 JSON"),
+        sa.Column("create_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP"), comment="创建时间"),
+        sa.Column("update_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP"), comment="更新时间"),
+        sa.PrimaryKeyConstraint("id", name="pk_tplus_sync_records"),
+        sa.UniqueConstraint("entity_type", "source_id", "row_hash", name="uk_tplus_sync_records_entity_source_hash"),
+        mysql_engine="InnoDB", mysql_charset="utf8mb4", mysql_collate="utf8mb4_general_ci", comment="T+ 原始同步 staging 表（映射确认前只保存原始行，不直接改业务表）",
+    )
+    op.create_index("idx_tplus_sync_records_batch_no", "tplus_sync_records", ["batch_no"])
+    op.create_index("idx_tplus_sync_records_source_no", "tplus_sync_records", ["source_no"])
+    op.create_index("idx_tplus_sync_records_state", "tplus_sync_records", ["state"])
+
+    op.create_table(
+        "tplus_sync_watermarks",
+        sa.Column("id", BIGINT(unsigned=True), autoincrement=True, nullable=False, comment="主键ID"),
+        sa.Column("entity_type", sa.String(length=32), nullable=False, comment="业务实体类型"),
+        sa.Column("last_source_updated_at", sa.DateTime(), nullable=True, comment="最近源更新时间"),
+        sa.Column("last_source_id", sa.String(length=64), nullable=True, comment="同一时间点的最后源主键"),
+        sa.Column("status", sa.String(length=16), nullable=False, server_default="READY", comment="状态：READY/BLOCKED"),
+        sa.Column("create_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP"), comment="创建时间"),
+        sa.Column("update_time", sa.DateTime(), nullable=False, server_default=sa.text("CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP"), comment="更新时间"),
+        sa.PrimaryKeyConstraint("id", name="pk_tplus_sync_watermarks"),
+        sa.UniqueConstraint("entity_type", name="uk_tplus_sync_watermarks_entity_type"),
+        mysql_engine="InnoDB", mysql_charset="utf8mb4", mysql_collate="utf8mb4_general_ci", comment="T+ 增量同步水位表",
+    )
+
+
+def downgrade() -> None:
+    """删除 T+ 同步支撑表和源更新时间列。"""
+    op.drop_table("tplus_sync_watermarks")
+    op.drop_index("idx_tplus_sync_records_state", table_name="tplus_sync_records")
+    op.drop_index("idx_tplus_sync_records_source_no", table_name="tplus_sync_records")
+    op.drop_index("idx_tplus_sync_records_batch_no", table_name="tplus_sync_records")
+    op.drop_table("tplus_sync_records")
+    op.drop_index("idx_tplus_sync_errors_entity_type", table_name="tplus_sync_errors")
+    op.drop_index("idx_tplus_sync_errors_batch_no", table_name="tplus_sync_errors")
+    op.drop_table("tplus_sync_errors")
+    op.drop_index("idx_tplus_sync_batches_started_at", table_name="tplus_sync_batches")
+    op.drop_index("idx_tplus_sync_batches_status", table_name="tplus_sync_batches")
+    op.drop_table("tplus_sync_batches")
+    for table in ("receipts", "shipments", "businesses", "customers", "products", "personnel"):
+        op.drop_column(table, "source_updated_at")

backend/alembic/versions/e5f607a8b9c0_tplus_task_management.py

@@ -0,0 +1,118 @@
+"""V15 T+ 持久化任务、幂等和重试元数据。"""
+
+from collections.abc import Sequence
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.mysql import BIGINT
+
+revision: str = "e5f607a8b9c0"
+down_revision: str | Sequence[str] | None = "d4e5f6a7b8c9"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    """仅扩展管理元数据，不改原始暂存行或工资业务表。"""
+    op.add_column(
+        "tplus_sync_batches",
+        sa.Column("entity_types_json", sa.JSON(), comment="请求读取的实体类型；历史批次可为空"),
+    )
+    op.add_column(
+        "tplus_sync_batches",
+        sa.Column(
+            "skipped_record_count",
+            sa.Integer(),
+            nullable=False,
+            server_default="0",
+            comment="重复快照跳过条数",
+        ),
+    )
+    op.add_column(
+        "tplus_sync_batches",
+        sa.Column("parent_batch_no", sa.String(32), comment="重试来源批次号，保留原批次记录"),
+    )
+    op.add_column(
+        "tplus_sync_batches",
+        sa.Column("request_key", sa.String(36), comment="请求幂等键，空值用于历史和直接CLI任务"),
+    )
+    op.add_column(
+        "tplus_sync_batches",
+        sa.Column(
+            "requested_by", BIGINT(unsigned=True), comment="请求用户ID，CLI任务为空，无物理外键"
+        ),
+    )
+    op.add_column(
+        "tplus_sync_batches",
+        sa.Column("audit_result_json", sa.JSON(), comment="关联核验结果，只含系统字典和汇总"),
+    )
+    op.alter_column(
+        "tplus_sync_batches",
+        "started_at",
+        existing_type=sa.DateTime(),
+        nullable=True,
+        comment="开始执行时间（排队时为空）",
+    )
+    op.alter_column(
+        "tplus_sync_batches",
+        "sync_mode",
+        existing_type=sa.String(16),
+        nullable=False,
+        comment="任务方式：FULL全量 AUDIT关联核验 INCREMENTAL预留",
+    )
+    op.alter_column(
+        "tplus_sync_batches",
+        "status",
+        existing_type=sa.String(16),
+        nullable=False,
+        comment="状态：QUEUED/RUNNING/SUCCEEDED/PARTIAL/FAILED",
+    )
+    op.create_unique_constraint(
+        "uk_tplus_sync_batches_request_key", "tplus_sync_batches", ["request_key"]
+    )
+    op.create_index(
+        "idx_tplus_sync_batches_parent_status", "tplus_sync_batches", ["parent_batch_no", "status"]
+    )
+    op.execute(
+        "UPDATE tplus_sync_batches SET skipped_record_count=GREATEST(0, source_record_count-staged_record_count-failed_record_count) WHERE status='SUCCEEDED' AND sync_mode='FULL'"
+    )
+
+
+def downgrade() -> None:
+    """回滚前关闭 worker；原始暂存数据和历史批次仍保留。"""
+    op.execute(
+        "UPDATE tplus_sync_batches SET started_at=create_time, status='FAILED', finished_at=CURRENT_TIMESTAMP, error_message='任务管理版本已回滚' WHERE status='QUEUED'"
+    )
+    op.execute("UPDATE tplus_sync_batches SET started_at=create_time WHERE started_at IS NULL")
+    op.drop_index("idx_tplus_sync_batches_parent_status", table_name="tplus_sync_batches")
+    op.drop_constraint("uk_tplus_sync_batches_request_key", "tplus_sync_batches", type_="unique")
+    for name in (
+        "audit_result_json",
+        "requested_by",
+        "request_key",
+        "parent_batch_no",
+        "skipped_record_count",
+        "entity_types_json",
+    ):
+        op.drop_column("tplus_sync_batches", name)
+    op.alter_column(
+        "tplus_sync_batches",
+        "started_at",
+        existing_type=sa.DateTime(),
+        nullable=False,
+        comment="开始读取时间",
+    )
+    op.alter_column(
+        "tplus_sync_batches",
+        "sync_mode",
+        existing_type=sa.String(16),
+        nullable=False,
+        comment="同步模式：FULL全量 INCREMENTAL增量",
+    )
+    op.alter_column(
+        "tplus_sync_batches",
+        "status",
+        existing_type=sa.String(16),
+        nullable=False,
+        comment="状态：RUNNING/SUCCEEDED/PARTIAL/FAILED",
+    )

backend/alembic/versions/f3a8c2d61b47_v11_审批快照与方案版本化.py

@@ -0,0 +1,93 @@
+"""v11 审批快照与方案版本化
+
+Revision ID: f3a8c2d61b47
+Revises: 90653814a421
+Create Date: 2026-10-20 10:30:00.000000
+
+补齐 code-review 遗留 2 项：
+
+- 审批单加 ``version_no`` 与 ``snapshot_json``：重提产生新版本，送审内容
+  在提交时固化，审批期间业务再改不影响复核依据；
+- 二次分配方案行改 append-only 版本化：加 ``version_no`` / ``is_active``，
+  唯一键由 (business_id, personnel_id) 扩为
+  (business_id, version_no, personnel_id)，编辑替换与驳回清空只置历史、
+  不再物理删除。
+
+存量数据均落 ``version_no=1`` / ``is_active=1``：升级前的行就是当前生效
+版本，语义不变。审批单存量快照为空，详情接口按「当前内容回退 + 显式
+标注」处理，不做历史回填（回填会伪造提交时内容）。
+"""
+
+from collections.abc import Sequence
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "f3a8c2d61b47"
+down_revision: str | Sequence[str] | None = "90653814a421"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    """加审批快照列/索引与方案版本列，替换唯一键。"""
+    op.execute(
+        "ALTER TABLE approval_orders "
+        "ADD COLUMN version_no INT NOT NULL DEFAULT 1 "
+        "COMMENT '提交版本号（同一对象类型+引用对象从1递增；重提产生新版本，旧单快照不覆盖）' "
+        "AFTER refer_no"
+    )
+    op.execute(
+        "ALTER TABLE approval_orders "
+        "ADD COLUMN snapshot_json JSON NULL "
+        "COMMENT '提交时内容快照（信封：schema_version/order_type/captured_at/data；历史单可空）' "
+        "AFTER version_no"
+    )
+    op.execute("CREATE INDEX idx_approval_orders_type_ref ON approval_orders (order_type, ref_id)")
+
+    op.execute(
+        "ALTER TABLE business_secondary_allocations "
+        "ADD COLUMN version_no INT NOT NULL DEFAULT 1 "
+        "COMMENT '方案版本号（同一业务从1递增；编辑/重提产生新版本，旧版本仅置历史）' "
+        "AFTER business_id"
+    )
+    op.execute(
+        "ALTER TABLE business_secondary_allocations "
+        "ADD COLUMN is_active TINYINT NOT NULL DEFAULT 1 "
+        "COMMENT '是否当前生效版本：1生效 0历史（编辑替换/驳回清空后置0，行不删除）' "
+        "AFTER ratio"
+    )
+    op.execute(
+        "ALTER TABLE business_secondary_allocations "
+        "DROP INDEX uk_business_secondary_allocations_business_personnel, "
+        "ADD CONSTRAINT uk_business_secondary_allocations_business_version_personnel "
+        "UNIQUE (business_id, version_no, personnel_id)"
+    )
+    op.execute(
+        "CREATE INDEX idx_business_secondary_allocations_business_active "
+        "ON business_secondary_allocations (business_id, is_active)"
+    )
+    op.execute(
+        "ALTER TABLE business_secondary_allocations COMMENT = "
+        "'业务二次分配方案版本表（append-only；比例不要求合计满额，余额归属待澄清）'"
+    )
+
+
+def downgrade() -> None:
+    """回退结构（降级前须确认同业务同一参与人只存在于一个版本）。"""
+    op.execute(
+        "DROP INDEX idx_business_secondary_allocations_business_active "
+        "ON business_secondary_allocations"
+    )
+    op.execute(
+        "ALTER TABLE business_secondary_allocations "
+        "DROP INDEX uk_business_secondary_allocations_business_version_personnel, "
+        "ADD CONSTRAINT uk_business_secondary_allocations_business_personnel "
+        "UNIQUE (business_id, personnel_id)"
+    )
+    op.execute("ALTER TABLE business_secondary_allocations DROP COLUMN is_active")
+    op.execute("ALTER TABLE business_secondary_allocations DROP COLUMN version_no")
+
+    op.execute("DROP INDEX idx_approval_orders_type_ref ON approval_orders")
+    op.execute("ALTER TABLE approval_orders DROP COLUMN snapshot_json")
+    op.execute("ALTER TABLE approval_orders DROP COLUMN version_no")

backend/app/api/router.py

@@ -26,6 +26,7 @@ from app.api.v1.system import logs as system_logs
 from app.api.v1.system import params as system_params
 from app.api.v1.system import roles as system_roles
 from app.api.v1.system import settings as system_settings
+from app.api.v1.system import tplus as system_tplus
 from app.api.v1.system import users as system_users
 
 api_router = APIRouter()
@@ -47,6 +48,7 @@ system_router.include_router(system_settings.router)
 system_router.include_router(system_approvals.router)
 system_router.include_router(system_logs.operation_router)
 system_router.include_router(system_logs.login_router)
+system_router.include_router(system_tplus.router)
 v1_router.include_router(system_router)
 
 # 基础数据（需求 2.1）：/api/v1/base/*，业务操作登录即可访问

backend/app/api/v1/base/business_roles.py

@@ -12,7 +12,8 @@ from fastapi import APIRouter, Depends
 from app.api.deps import DbSession, get_current_user
 from app.core.response import ApiResponse, ok
 from app.models import User
-from app.repositories import business_role_repository
+from app.schemas.business_role import BusinessRoleOptionOut
+from app.services import business_role_service
 
 router = APIRouter(prefix="/business-roles", tags=["人员管理"])
 
@@ -20,21 +21,14 @@ router = APIRouter(prefix="/business-roles", tags=["人员管理"])
 CurrentUser = Annotated[User, Depends(get_current_user)]
 
 
-@router.get("/options", response_model=ApiResponse[list[dict]], summary="岗位下拉选项")
-def list_options(session: DbSession, _: CurrentUser) -> ApiResponse[list[dict]]:
+@router.get(
+    "/options",
+    response_model=ApiResponse[list[BusinessRoleOptionOut]],
+    summary="岗位下拉选项",
+)
+def list_options(session: DbSession, _: CurrentUser) -> ApiResponse[list[BusinessRoleOptionOut]]:
     """岗位选项（登录即可；含停用——人员档案可能仍引用停用岗位）。
 
-    岗位规模有限（内置 + 自定义），一次取足够上限，不做分页。
+    岗位规模有限（内置 + 自定义），一次取全量，不做分页。
     """
-    roles = business_role_repository.list_roles(session, offset=0, limit=1000)
-    return ok(
-        [
-            {
-                "id": role.id,
-                "role_code": role.role_code,
-                "role_name": role.role_name,
-                "status": role.status,
-            }
-            for role in roles
-        ]
-    )
+    return ok(business_role_service.list_options(session))

backend/app/api/v1/base/customers.py

@@ -21,6 +21,7 @@ from app.schemas.customer import (
     CustomerOut,
     CustomerOwnershipIn,
     CustomerQuery,
+    CustomerRatingIn,
     CustomerUpdateIn,
     OwnershipChangeLogOut,
 )
@@ -102,6 +103,31 @@ def delete_customer(
     return ok(message="客户已删除")
 
 
+@router.put(
+    "/{customer_id}/rating",
+    response_model=ApiResponse[CustomerOut],
+    summary="保存客户评级",
+)
+def rate_customer(
+    session: DbSession,
+    current_user: CurrentUser,
+    customer_id: Annotated[int, Path(ge=1)],
+    payload: CustomerRatingIn,
+    meta: RequestMetaDep,
+) -> ApiResponse[CustomerOut]:
+    """保存评级总分（评级计算器「保存评分」）。
+
+    评分是客户档案属性：存总分与评级时间，五项明细不落库（对照表 #25）；
+    编辑端点不含评分 —— 评分只走本端点，由计算器统一入口写入。
+    """
+    return ok(
+        customer_service.rate_customer(
+            session, customer_id, payload, operator=current_user, meta=meta
+        ),
+        message="评级已保存",
+    )
+
+
 @router.put(
     "/{customer_id}/ownership",
     response_model=ApiResponse[CustomerOut],

backend/app/api/v1/dashboard.py

@@ -1,4 +1,4 @@
-"""工作台接口（M4 ⑧）：四区块一次性聚合。
+"""工作台接口（M4 ⑧）：三区块聚合 + 最近动态分页。
 
 准入：管理员/财务/领导（与报表、审批接口一致；FR-033/036 的「全部用户」
 待业务权限清单 #6 答复后再放宽）。只读接口，不写操作日志。
@@ -12,22 +12,18 @@ from app.api.deps import DbSession, require_roles
 from app.core.constants import ROLE_ADMIN, ROLE_FINANCE, ROLE_LEADER
 from app.core.response import ApiResponse, ok
 from app.models import User
-from app.schemas.dashboard import DashboardOut
+from app.schemas.dashboard import ActivityPageOut, ActivityQuery, DashboardOut
 from app.services import dashboard_service
 
 router = APIRouter(prefix="/dashboard", tags=["工作台"])
 
 _DashboardUser = Annotated[
     User,
-    Depends(
-        require_roles(
-            ROLE_ADMIN, ROLE_FINANCE, ROLE_LEADER, message="无权访问工作台"
-        )
-    ),
+    Depends(require_roles(ROLE_ADMIN, ROLE_FINANCE, ROLE_LEADER, message="无权访问工作台")),
 ]
 
 
-@router.get("", response_model=ApiResponse[DashboardOut], summary="工作台四区块")
+@router.get("", response_model=ApiResponse[DashboardOut], summary="工作台三区块")
 def get_dashboard(
     session: DbSession,
     current_user: _DashboardUser,
@@ -35,10 +31,22 @@ def get_dashboard(
         str | None, Query(max_length=7, description="统计月份 YYYY-MM，默认当月")
     ] = None,
 ) -> ApiResponse[DashboardOut]:
-    """一次返回经营总览/待办/风险预警/最近动态（待办与预警实时派生）。
+    """一次返回经营总览/待办/风险预警（待办与预警实时派生）。
 
     待办按当前用户角色只返回其可操作节点（admin 可查看全部在途单）。
+    最近动态是分页列表，走 `GET /dashboard/activities`。
     """
-    return ok(
-        dashboard_service.get_dashboard(session, month, current_user=current_user)
-    )
+    return ok(dashboard_service.get_dashboard(session, month, current_user=current_user))
+
+
+@router.get("/activities", response_model=ApiResponse[ActivityPageOut], summary="最近动态（分页）")
+def list_dashboard_activities(
+    session: DbSession,
+    current_user: _DashboardUser,
+    query: Annotated[ActivityQuery, Query()],
+) -> ApiResponse[ActivityPageOut]:
+    """分页查询最近动态，可按模块筛选。
+
+    与聚合端点拆开：翻页只重取这张列表，不必重算总览/待办/预警。
+    """
+    return ok(dashboard_service.list_activities(session, query))

backend/app/api/v1/system/tplus.py

@@ -0,0 +1,71 @@
+"""管理员管理 T+ 只读任务；长任务交给独立 worker 执行。"""
+
+from typing import Annotated
+
+from fastapi import APIRouter, Path, Query
+
+from app.api.deps import AdminUser, DbSession, RequestMetaDep
+from app.core.response import ApiResponse, PageResult, ok
+from app.schemas.tplus import (
+    TPlusBatchDetailOut,
+    TPlusBatchOut,
+    TPlusBatchQuery,
+    TPlusCreateIn,
+    TPlusErrorOut,
+    TPlusErrorQuery,
+    TPlusOverviewOut,
+    TPlusRetryIn,
+)
+from app.services.tplus import query_service, task_service
+
+router = APIRouter(prefix="/tplus", tags=["系统管理 · T+ 数据同步"])
+BatchNumber = Annotated[str, Path(min_length=1, max_length=32, pattern=r"^[A-Za-z0-9_-]+$")]
+
+
+@router.get("/overview", response_model=ApiResponse[TPlusOverviewOut])
+def overview(session: DbSession, _: AdminUser):
+    """查询配置状态、任务统计和最近核验。"""
+    return ok(query_service.overview(session))
+
+
+@router.get("/batches", response_model=ApiResponse[PageResult[TPlusBatchOut]])
+def batches(session: DbSession, _: AdminUser, query: Annotated[TPlusBatchQuery, Query()]):
+    """分页查询批次元数据。"""
+    return ok(query_service.list_batches(session, query))
+
+
+@router.get("/batches/{batch_no}", response_model=ApiResponse[TPlusBatchDetailOut])
+def detail(session: DbSession, _: AdminUser, batch_no: BatchNumber):
+    """查询批次详情与受控的核验报告。"""
+    return ok(query_service.get_batch(session, batch_no))
+
+
+@router.get("/batches/{batch_no}/errors", response_model=ApiResponse[PageResult[TPlusErrorOut]])
+def errors(
+    session: DbSession,
+    _: AdminUser,
+    batch_no: BatchNumber,
+    query: Annotated[TPlusErrorQuery, Query()],
+):
+    """分页查询异常定位和固定错误说明。"""
+    return ok(query_service.list_errors(session, batch_no, query))
+
+
+@router.post("/batches", response_model=ApiResponse[TPlusBatchDetailOut])
+def create(session: DbSession, admin: AdminUser, meta: RequestMetaDep, payload: TPlusCreateIn):
+    """将读取或核验任务持久化排队。"""
+    batch = task_service.create_task(session, payload, task_service.TaskActor(admin, meta))
+    return ok(query_service.get_batch(session, batch.batch_no))
+
+
+@router.post("/batches/{batch_no}/retry", response_model=ApiResponse[TPlusBatchDetailOut])
+def retry(
+    session: DbSession,
+    admin: AdminUser,
+    meta: RequestMetaDep,
+    batch_no: BatchNumber,
+    payload: TPlusRetryIn,
+):
+    """新建重试批次，保留原批次与异常历史。"""
+    batch = task_service.retry_task(session, batch_no, payload, task_service.TaskActor(admin, meta))
+    return ok(query_service.get_batch(session, batch.batch_no))

backend/app/cli.py

@@ -11,13 +11,20 @@
 import argparse
 import getpass
 import sys
+import time
+from datetime import datetime
 
 from sqlalchemy import select
 
+from app.core.config import get_settings
 from app.core.constants import PASSWORD_MAX_BYTES, PASSWORD_MIN_BYTES, ROLE_ADMIN
 from app.core.db import dispose_engine, get_session_factory
 from app.core.security import hash_password
+from app.core.tplus_db import dispose_tplus_engine, get_tplus_session_factory
 from app.models import Role, User, UserRole
+from app.repositories.tplus_repository import get_batch
+from app.services.tplus.sync_service import TPlusSyncService
+from app.services.tplus.worker import run_once
 
 
 def _prompt_password() -> str:
@@ -80,6 +87,54 @@ def _create_admin(args: argparse.Namespace) -> int:
         return 0
 
 
+def _stage_tplus(args: argparse.Namespace) -> int:
+    """读取 T+ 源库并写入 MySQL staging。"""
+    target_factory = get_session_factory()
+    source_factory = get_tplus_session_factory()
+    watermark = datetime.fromisoformat(args.watermark) if args.watermark else None
+    with source_factory() as source_session, target_factory() as target_session:
+        batch = TPlusSyncService(source_session, target_session).run(
+            entity_types=args.entity,
+            watermark=watermark,
+            batch_size=args.batch_size,
+        )
+    print(
+        f"T+ staging 完成：批次={batch.batch_no} 状态={batch.status} "
+        f"源记录={batch.source_record_count} 新增={batch.staged_record_count} "
+        f"失败={batch.failed_record_count}"
+    )
+    return 0 if batch.status == "SUCCEEDED" else 1
+
+
+def _tplus_worker(args: argparse.Namespace) -> int:
+    """独立消费持久化任务；HTTP 进程不承担长时间读取。"""
+    if args.connection_file:
+        settings = get_settings()
+        settings.t_plus_enabled, settings.t_plus_connection_file = True, args.connection_file
+    if not 2 <= args.poll_seconds <= 60:
+        print("轮询间隔必须在 2 到 60 秒之间", file=sys.stderr)
+        return 1
+    factory = get_session_factory()
+    try:
+        while True:
+            with factory() as session:
+                batch_no = run_once(session)
+                batch = get_batch(session, batch_no) if batch_no else None
+                status = batch.status if batch else None
+            if batch_no:
+                print(f"任务完成：{batch_no} 状态={status}", flush=True)
+            if not args.watch:
+                return 0 if status in (None, "SUCCEEDED") else 1
+            if batch_no is None or status != "SUCCEEDED":
+                time.sleep(args.poll_seconds)
+    except KeyboardInterrupt:
+        print("T+ 读取服务已停止")
+        return 130
+    except Exception as exc:
+        print(f"T+ 读取服务停止（{type(exc).__name__}），请检查运行环境", file=sys.stderr)
+        return 1
+
+
 def main() -> int:
     """命令行入口。"""
     parser = argparse.ArgumentParser(
@@ -96,12 +151,49 @@ def main() -> int:
         help="密码。不传则交互式输入（推荐）——命令行传参会留在 shell 历史中",
     )
 
+    p_tplus = sub.add_parser("tplus-stage", help="读取 T+ 源库并写入原始 staging")
+    p_tplus.add_argument(
+        "--entity",
+        action="append",
+        choices=[
+            "personnel",
+            "products",
+            "customers",
+            "businesses",
+            "shipments",
+            "receipts",
+            "business_items",
+            "shipment_items",
+            "receipt_items",
+        ],
+        required=True,
+        help="实体类型，可重复指定",
+    )
+    p_tplus.add_argument(
+        "--watermark",
+        default=None,
+        help="增量起始时间，ISO 8601 格式；不传表示全量读取",
+    )
+    p_tplus.add_argument("--batch-size", type=int, default=500, help="每次从源库读取的行数")
+
+    p_worker = sub.add_parser("tplus-worker", help="执行已排队的 T+ 读取或关联核验")
+    modes = p_worker.add_mutually_exclusive_group()
+    modes.add_argument("--once", action="store_true", help="执行一个任务后退出（默认）")
+    modes.add_argument("--watch", action="store_true", help="持续消费手动排队任务")
+    p_worker.add_argument("--connection-file", default=None, help="显式指定项目外只读连接 JSON")
+    p_worker.add_argument("--poll-seconds", type=int, default=3, help="空闲轮询间隔，2–60 秒")
+
     args = parser.parse_args()
 
     try:
-        return _create_admin(args)
+        if args.command == "create-admin":
+            return _create_admin(args)
+        if args.command == "tplus-worker":
+            return _tplus_worker(args)
+        return _stage_tplus(args)
     finally:
         dispose_engine()
+        dispose_tplus_engine()
 
 
 if __name__ == "__main__":

backend/app/core/config.py

@@ -58,6 +58,23 @@ class Settings(BaseSettings):
     redis_password: str = ""
     redis_db: int = 0
 
+    # ===== T+ 只读 SQL Server =====
+    # T+ 数据库默认关闭，避免开发者未配置驱动时影响主业务启动。
+    t_plus_enabled: bool = False
+    t_plus_host: str = "127.0.0.1"
+    t_plus_port: int = 1433
+    t_plus_database: str = ""
+    t_plus_user: str = ""
+    t_plus_password: str = Field(default="", repr=False)
+    t_plus_connection_file: str = ""
+    # 部署前必须指定已在 SQL Server 2008 R2 实测的驱动，不自动选择最新版。
+    t_plus_odbc_driver: str = ""
+    t_plus_encrypt: bool = True
+    t_plus_trust_server_certificate: bool = True
+    t_plus_connect_timeout: int = 10
+    t_plus_query_timeout: int = Field(default=120, ge=1, le=600)
+    t_plus_query_dir: str = ""
+
     @field_validator("app_secret_key")
     @classmethod
     def _check_secret_key(cls, v: str) -> str:
@@ -92,6 +109,32 @@ class Settings(BaseSettings):
         auth = f":{self.redis_password}@" if self.redis_password else ""
         return f"redis://{auth}{self.redis_host}:{self.redis_port}/{self.redis_db}"
 
+    @property
+    def t_plus_database_url(self) -> URL:
+        """构造 T+ 只读 SQL Server 连接 URL。
+
+        连接 URL 只在启用 T+ 适配器时使用；密码由 SQLAlchemy URL 对象保存，
+        日志中调用 ``render_as_string()`` 时仍会默认隐藏密码。
+        ODBC 的分号和花括号由方言统一转义，调用方不得再手工拼接连接串。
+        """
+        query = {
+            "driver": self.t_plus_odbc_driver,
+            "Encrypt": "yes" if self.t_plus_encrypt else "no",
+            "TrustServerCertificate": "yes"
+            if self.t_plus_trust_server_certificate
+            else "no",
+            "Connection Timeout": str(self.t_plus_connect_timeout),
+        }
+        return URL.create(
+            drivername="mssql+pyodbc",
+            username=self.t_plus_user,
+            password=self.t_plus_password,
+            host=self.t_plus_host,
+            port=self.t_plus_port,
+            database=self.t_plus_database,
+            query=query,
+        )
+
     @property
     def upload_dir(self) -> Path:
         """上传文件根目录（公司 logo 等）：backend/uploads/。

backend/app/core/tplus_db.py

@@ -0,0 +1,115 @@
+"""T+ 只读 SQL Server 连接。
+
+T+ 是外部源库，业务接口不直接依赖它。连接采用延迟初始化，未配置或未启用
+时不会影响主业务服务启动；适配器执行查询时才建立连接池。
+"""
+
+import json
+from collections.abc import Generator
+from pathlib import Path
+
+from pydantic import BaseModel, ConfigDict, Field, SecretStr, ValidationError
+from sqlalchemy import Engine, create_engine, event
+from sqlalchemy.orm import Session, sessionmaker
+
+from app.core.config import Settings, get_settings
+
+_engine: Engine | None = None
+_session_factory: sessionmaker[Session] | None = None
+
+
+class _ConnectionFile(BaseModel):
+    """连接文件仅允许源库字段，密码不会出现在模型表示中。"""
+
+    model_config = ConfigDict(extra="forbid", strict=True)
+    host: str = Field(min_length=1, max_length=255)
+    port: int = Field(ge=1, le=65535)
+    database: str = Field(min_length=1, max_length=128)
+    user: str = Field(min_length=1, max_length=128)
+    password: SecretStr
+    driver: str = Field(min_length=1, max_length=128)
+
+
+def resolve_tplus_settings(settings: Settings) -> Settings:
+    """解析显式配置的只读连接，返回副本，不修改应用全局配置。"""
+    if not settings.t_plus_enabled:
+        raise RuntimeError("T+ 数据源未启用，请设置 T_PLUS_ENABLED=true")
+    resolved = settings
+    if settings.t_plus_connection_file:
+        try:
+            path = Path(settings.t_plus_connection_file)
+            if path.stat().st_size > 16384:
+                raise ValueError("连接文件过大")
+            config = _ConnectionFile.model_validate(
+                json.loads(path.read_text(encoding="utf-8-sig"))
+            )
+        except OSError:
+            # 消息只区分故障类别，不回显路径内容；路径本身可在应用配置中核对。
+            raise RuntimeError("T+ 只读连接文件不存在或不可读") from None
+        except (ValueError, ValidationError):
+            raise RuntimeError("T+ 只读连接文件字段不正确") from None
+        values = {
+            "t_plus_" + key: getattr(config, key) for key in ("host", "port", "database", "user")
+        }
+        values.update(
+            t_plus_password=config.password.get_secret_value(), t_plus_odbc_driver=config.driver
+        )
+        resolved = settings.model_copy(update=values)
+    if resolved.t_plus_user.strip().casefold() == "sa":
+        raise RuntimeError("应用必须使用 T+ 专用只读账号，不能使用 sa")
+    if not resolved.t_plus_database or not resolved.t_plus_user or not resolved.t_plus_odbc_driver:
+        raise RuntimeError("T+ 只读连接缺少数据库、账号或已验证的 ODBC 驱动")
+    return resolved
+
+
+def get_tplus_engine() -> Engine:
+    """获取 T+ 只读引擎；未启用或缺少 ODBC 驱动时给出可读错误。"""
+    global _engine, _session_factory
+    if _engine is not None:
+        return _engine
+
+    settings = resolve_tplus_settings(get_settings())
+
+    try:
+        _engine = create_engine(
+            settings.t_plus_database_url,
+            pool_pre_ping=True,
+            pool_recycle=1800,
+            pool_timeout=settings.t_plus_connect_timeout,
+            echo=False,
+            hide_parameters=True,
+        )
+    except ModuleNotFoundError as exc:
+        raise RuntimeError(
+            "T+ 连接需要 pyodbc（uv sync --extra tplus）和已验证的 SQL Server ODBC 驱动"
+        ) from exc
+    _session_factory = sessionmaker(_engine, expire_on_commit=False)
+
+    def configure_timeout(connection, _record):
+        connection.timeout = settings.t_plus_query_timeout
+
+    event.listen(_engine, "connect", configure_timeout)
+    return _engine
+
+
+def get_tplus_session_factory() -> sessionmaker[Session]:
+    """获取 T+ SQLAlchemy 会话工厂。"""
+    get_tplus_engine()
+    assert _session_factory is not None
+    return _session_factory
+
+
+def get_tplus_db() -> Generator[Session, None, None]:
+    """提供只读 T+ 会话；调用方不得 commit 或执行写操作。"""
+    factory = get_tplus_session_factory()
+    with factory() as session:
+        yield session
+
+
+def dispose_tplus_engine() -> None:
+    """释放 T+ 连接池。"""
+    global _engine, _session_factory
+    if _engine is not None:
+        _engine.dispose()
+        _engine = None
+        _session_factory = None

backend/app/core/tplus_lock.py

@@ -0,0 +1,48 @@
+"""同账套任务串行锁；独立持有 MySQL 连接，跨分页 commit 仍保持有效。"""
+
+import hashlib
+from contextlib import contextmanager
+
+from sqlalchemy import text
+from sqlalchemy.orm import Session
+
+
+class TPlusBusyError(RuntimeError):
+    """已有任务持有执行锁。"""
+
+
+def lock_name(session: Session) -> str:
+    """生成不含地址、账号或密码的固定长度锁名。"""
+    database = session.get_bind().url.database or "default"
+    return "ys:tplus:read:" + hashlib.sha256(database.encode()).hexdigest()[:24]
+
+
+@contextmanager
+def sync_lock(session: Session):
+    """只在取得锁后执行；释放失败时使连接失效，防止带锁回池。"""
+    engine = session.get_bind()
+    if engine.dialect.name != "mysql":
+        raise RuntimeError("T+ 执行协调需要 MySQL 命名锁")
+    connection = engine.connect()
+    acquired = False
+    try:
+        acquired = (
+            connection.execute(
+                text("SELECT GET_LOCK(:name, 0)"), {"name": lock_name(session)}
+            ).scalar_one()
+            == 1
+        )
+        if not acquired:
+            raise TPlusBusyError("已有 T+ 任务正在执行，请稍后重试")
+        yield
+    finally:
+        if acquired:
+            try:
+                released = connection.execute(
+                    text("SELECT RELEASE_LOCK(:name)"), {"name": lock_name(session)}
+                ).scalar_one()
+                if released != 1:
+                    connection.invalidate()
+            except Exception:
+                connection.invalidate()
+        connection.close()

backend/app/main.py

@@ -15,6 +15,7 @@ from app.core.db import dispose_engine
 from app.core.exception_handlers import register_exception_handlers
 from app.core.logging import setup_logging
 from app.core.redis_client import close_redis
+from app.core.tplus_db import dispose_tplus_engine
 
 logger = logging.getLogger(__name__)
 
@@ -35,6 +36,7 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
     )
     yield
     dispose_engine()
+    dispose_tplus_engine()
     close_redis()
     logger.info("应用已停止，连接池已释放")
 

backend/app/models/__init__.py

@@ -26,6 +26,12 @@ from app.models.role import Permission, Role, RolePermission
 from app.models.settlement import AdjustmentOrder, MonthlySettlement, SettlementItem
 from app.models.shipment import Shipment, ShipmentItem
 from app.models.system_setting import SystemSetting
+from app.models.tplus_sync import (
+    TPlusSyncBatch,
+    TPlusSyncError,
+    TPlusSyncRecord,
+    TPlusSyncWatermark,
+)
 from app.models.user import User, UserRole
 
 __all__ = [
@@ -60,6 +66,10 @@ __all__ = [
     "Shipment",
     "ShipmentItem",
     "SystemSetting",
+    "TPlusSyncBatch",
+    "TPlusSyncError",
+    "TPlusSyncRecord",
+    "TPlusSyncWatermark",
     "TimestampMixin",
     "User",
     "UserRole",

backend/app/models/approval.py

@@ -15,7 +15,16 @@ ref_id 为**固化无外键**（同 business_status_logs 先例）：order_type=
 
 from datetime import datetime
 
-from sqlalchemy import DateTime, ForeignKey, Index, String, text
+from sqlalchemy import (
+    JSON,
+    DateTime,
+    ForeignKey,
+    Index,
+    Integer,
+    String,
+    UniqueConstraint,
+    text,
+)
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.models.base import BIGINT_UNSIGNED, TINYINT_TYPE, Base, TimestampMixin, mysql_table_args
@@ -54,7 +63,17 @@ class ApprovalOrder(Base, TimestampMixin):
     __tablename__ = "approval_orders"
     __table_args__ = mysql_table_args(
         "审批单表（营销后勤提交→财务复核→领导审核；开关关闭时自动通过留痕）",
+        # 同一引用对象的第 N 次送审在库里有唯一一行：并发提单算重版本号时
+        # 由 DB 兜住（应用层取号走锁读，见 approval_service._next_version）
+        UniqueConstraint(
+            "order_type",
+            "ref_id",
+            "version_no",
+            name="uk_approval_orders_order_type_ref_version",
+        ),
         Index("idx_approval_orders_type_status", "order_type", "status"),
+        # 版本号按 (order_type, ref_id) 递增，须有对应索引避免全表 max 扫描
+        Index("idx_approval_orders_type_ref", "order_type", "ref_id"),
     )
 
     id: Mapped[int] = mapped_column(
@@ -76,6 +95,17 @@ class ApprovalOrder(Base, TimestampMixin):
     refer_no: Mapped[str] = mapped_column(
         String(32), nullable=False, comment="引用对象单号（冗余固化，列表页免联查）"
     )
+    version_no: Mapped[int] = mapped_column(
+        Integer,
+        nullable=False,
+        server_default="1",
+        comment="提交版本号（同一对象类型+引用对象从1递增；重提产生新版本，旧单快照不覆盖；"
+        "同一对象内唯一，由uk_approval_orders_order_type_ref_version强制）",
+    )
+    snapshot_json: Mapped[dict | None] = mapped_column(
+        JSON,
+        comment="提交时内容快照（信封：schema_version/order_type/captured_at/data；历史单可空）",
+    )
     status: Mapped[int] = mapped_column(
         TINYINT_TYPE,
         nullable=False,

backend/app/models/base.py

@@ -91,6 +91,9 @@ class TPlusSyncMixin:
         String(32), comment="同步批次号（同步模块写入）"
     )
     synced_at: Mapped[datetime | None] = mapped_column(DateTime, comment="最近同步时间")
+    source_updated_at: Mapped[datetime | None] = mapped_column(
+        DateTime, comment="T+ 源数据更新时间（未确认时可为空）"
+    )
 
 
 def mysql_table_args(comment: str, *extra: object) -> tuple[object, ...]:

backend/app/models/business.py

@@ -15,7 +15,17 @@
 from datetime import date, datetime
 from decimal import Decimal
 
-from sqlalchemy import Date, DateTime, ForeignKey, Index, Numeric, String, UniqueConstraint, text
+from sqlalchemy import (
+    Date,
+    DateTime,
+    ForeignKey,
+    Index,
+    Integer,
+    Numeric,
+    String,
+    UniqueConstraint,
+    text,
+)
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.models.base import (
@@ -42,27 +52,34 @@ class SecondaryAllocation(Base, TimestampMixin):
     - 分配比例**不要求合计 100%**：余额归属未定（澄清 #2），存储只落参与人
       与比例，响应以 hint 提示合计/余额，不做强制校验
     - T+ 无二次分配数据（2026-10-15 回馈）：参与人全为系统内人工维护
-    - **分摊计算不在本期**（澄清 #1 分配标的三种口径结果不同）：④ 核算时按
-      核算时刻读「当前生效行」做快照，天然隔离历史版本，故无版本表
+    - **版本行 append-only**：编辑全量替换与驳回清空都把旧行置
+      ``is_active=0`` 而非物理删除，历史方案可回溯；当前生效行由
+      ``is_active=1`` 唯一标识。核算仍按「当前生效行」取快照
+      （``performance_ledgers.snapshot_json.alloc_plan`` 固化），版本化只
+      补齐「方案事实留存」，不改变核算口径
 
     personnel 外键 **RESTRICT**：被方案引用的人员禁删——分摊金额待定的记录
     不允许静默失联（silently deleted）导致回档无法人工查验；personnel_name
-    冗余固化，人离职后方案回显仍有名。uk 锁 (business_id, personnel_id)，
-    同一合同的同一参与人仅一行，重复提交在服务层聚合。
+    冗余固化，人离职后方案回显仍有名。uk 锁
+    (business_id, version_no, personnel_id)，同一版本内同一参与人仅一行，
+    重复提交在服务层聚合；跨版本允许同一人再次出现。
     """
 
     __tablename__ = "business_secondary_allocations"
     __table_args__ = mysql_table_args(
         # 表/列注释避开「%」字符：SHOW CREATE TABLE 经驱动参数化转义为「%%」，
         # 会与 ORM 侧 comment 字面断言失配（全库既有注释同此惯例）
-        "业务二次分配方案明细表（合同建立时录入；比例不要求合计满额，余额归属待澄清）",
+        "业务二次分配方案版本表（append-only；比例不要求合计满额，余额归属待澄清）",
         UniqueConstraint(
             "business_id",
+            "version_no",
             "personnel_id",
-            name="uk_business_secondary_allocations_business_personnel",
+            name="uk_business_secondary_allocations_business_version_personnel",
         ),
         Index("idx_business_secondary_allocations_business_id", "business_id"),
         Index("idx_business_secondary_allocations_personnel_id", "personnel_id"),
+        # 取当前生效行的唯一查询形态：WHERE business_id=? AND is_active=1
+        Index("idx_business_secondary_allocations_business_active", "business_id", "is_active"),
     )
 
     id: Mapped[int] = mapped_column(
@@ -74,6 +91,12 @@ class SecondaryAllocation(Base, TimestampMixin):
         nullable=False,
         comment="业务ID（businesses.id；业务删除随删）",
     )
+    version_no: Mapped[int] = mapped_column(
+        Integer,
+        nullable=False,
+        server_default="1",
+        comment="方案版本号（同一业务从1递增；编辑/重提产生新版本，旧版本仅置历史）",
+    )
     personnel_id: Mapped[int] = mapped_column(
         BIGINT_UNSIGNED,
         ForeignKey("personnel.id", ondelete="RESTRICT"),
@@ -88,6 +111,12 @@ class SecondaryAllocation(Base, TimestampMixin):
         nullable=False,
         comment="分摊比例（0-100 两位小数；不要求合计满额，余额归属待澄清#2）",
     )
+    is_active: Mapped[int] = mapped_column(
+        TINYINT_TYPE,
+        nullable=False,
+        server_default="1",
+        comment="是否当前生效版本：1生效 0历史（编辑替换/驳回清空后置0，行不删除）",
+    )
 
 
 class Business(Base, TimestampMixin, TPlusSyncMixin):

backend/app/models/customer.py

@@ -63,6 +63,14 @@ class Customer(Base, TimestampMixin, TPlusSyncMixin):
         comment="归属负责营销人员ID（手动修改不回溯历史，变更留痕见 ownership_change_logs）",
     )
     remark: Mapped[str | None] = mapped_column(String(255), comment="备注")
+    rating_score: Mapped[int | None] = mapped_column(
+        TINYINT_TYPE,
+        comment="评级总分（0-100，评级计算器保存；null=未评分；人工维护字段，T+ 同步不覆盖）",
+    )
+    rating_time: Mapped[datetime | None] = mapped_column(
+        DateTime,
+        comment="最近评级时间（评级计算器保存评分时刷新；人工维护字段）",
+    )
 
 
 class OwnershipChangeLog(Base):

backend/app/models/tplus_sync.py

@@ -0,0 +1,149 @@
+"""T+ 同步批次、错误和原始 staging 记录。"""
+
+from datetime import datetime
+
+from sqlalchemy import JSON, DateTime, Index, Integer, String, Text, UniqueConstraint
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.models.base import BIGINT_UNSIGNED, Base, TimestampMixin, mysql_table_args
+
+
+class TPlusSyncBatch(Base, TimestampMixin):
+    """一次 T+ 全量或增量同步的审计记录。"""
+
+    __tablename__ = "tplus_sync_batches"
+    __table_args__ = mysql_table_args(
+        "T+ 同步批次表（记录读取范围、结果和对账统计）",
+        UniqueConstraint("batch_no", name="uk_tplus_sync_batches_batch_no"),
+        UniqueConstraint("request_key", name="uk_tplus_sync_batches_request_key"),
+        Index("idx_tplus_sync_batches_status", "status"),
+        Index("idx_tplus_sync_batches_started_at", "started_at"),
+        Index("idx_tplus_sync_batches_parent_status", "parent_batch_no", "status"),
+    )
+
+    id: Mapped[int] = mapped_column(
+        BIGINT_UNSIGNED, primary_key=True, autoincrement=True, comment="主键ID"
+    )
+    batch_no: Mapped[str] = mapped_column(String(32), nullable=False, comment="同步批次号")
+    sync_mode: Mapped[str] = mapped_column(
+        String(16), nullable=False, comment="任务方式：FULL全量 AUDIT关联核验 INCREMENTAL预留"
+    )
+    status: Mapped[str] = mapped_column(
+        String(16), nullable=False, comment="状态：QUEUED/RUNNING/SUCCEEDED/PARTIAL/FAILED"
+    )
+    source_database: Mapped[str | None] = mapped_column(String(128), comment="T+ 源数据库名")
+    started_at: Mapped[datetime | None] = mapped_column(
+        DateTime, comment="开始执行时间（排队时为空）"
+    )
+    finished_at: Mapped[datetime | None] = mapped_column(DateTime, comment="结束时间")
+    watermark_from: Mapped[datetime | None] = mapped_column(DateTime, comment="增量起始水位")
+    watermark_to: Mapped[datetime | None] = mapped_column(DateTime, comment="本批次结束水位")
+    source_record_count: Mapped[int] = mapped_column(
+        Integer, nullable=False, server_default="0", comment="源记录数"
+    )
+    staged_record_count: Mapped[int] = mapped_column(
+        Integer, nullable=False, server_default="0", comment="写入 staging 的记录数"
+    )
+    failed_record_count: Mapped[int] = mapped_column(
+        Integer, nullable=False, server_default="0", comment="失败记录数"
+    )
+    error_message: Mapped[str | None] = mapped_column(Text, comment="批次错误摘要")
+    entity_types_json: Mapped[list[str] | None] = mapped_column(
+        JSON, comment="请求读取的实体类型；历史批次可为空"
+    )
+    skipped_record_count: Mapped[int] = mapped_column(
+        Integer, nullable=False, server_default="0", comment="重复快照跳过条数"
+    )
+    parent_batch_no: Mapped[str | None] = mapped_column(
+        String(32), comment="重试来源批次号，保留原批次记录"
+    )
+    request_key: Mapped[str | None] = mapped_column(
+        String(36), comment="请求幂等键，空值用于历史和直接CLI任务"
+    )
+    requested_by: Mapped[int | None] = mapped_column(
+        BIGINT_UNSIGNED, comment="请求用户ID，CLI任务为空，无物理外键"
+    )
+    audit_result_json: Mapped[dict | None] = mapped_column(
+        JSON, comment="关联核验结果，只含系统字典和汇总"
+    )
+
+
+class TPlusSyncError(Base, TimestampMixin):
+    """单条 T+ 数据读取、清洗或 staging 失败记录。"""
+
+    __tablename__ = "tplus_sync_errors"
+    __table_args__ = mysql_table_args(
+        "T+ 同步错误表（记录失败行和重试信息）",
+        Index("idx_tplus_sync_errors_batch_no", "batch_no"),
+        Index("idx_tplus_sync_errors_entity_type", "entity_type"),
+    )
+
+    id: Mapped[int] = mapped_column(
+        BIGINT_UNSIGNED, primary_key=True, autoincrement=True, comment="主键ID"
+    )
+    batch_no: Mapped[str] = mapped_column(String(32), nullable=False, comment="同步批次号")
+    entity_type: Mapped[str] = mapped_column(String(32), nullable=False, comment="业务实体类型")
+    source_table: Mapped[str] = mapped_column(String(128), nullable=False, comment="T+ 源表")
+    source_id: Mapped[str | None] = mapped_column(String(64), comment="T+ 源主键")
+    source_line_no: Mapped[str | None] = mapped_column(String(64), comment="T+ 明细行号或明细主键")
+    error_code: Mapped[str] = mapped_column(String(32), nullable=False, comment="错误编码")
+    error_message: Mapped[str] = mapped_column(Text, nullable=False, comment="错误说明")
+    retry_count: Mapped[int] = mapped_column(
+        Integer, nullable=False, server_default="0", comment="已重试次数"
+    )
+
+
+class TPlusSyncRecord(Base, TimestampMixin):
+    """T+ 原始行 staging，保存映射未定前的可追溯数据。"""
+
+    __tablename__ = "tplus_sync_records"
+    __table_args__ = mysql_table_args(
+        "T+ 原始同步 staging 表（映射确认前只保存原始行，不直接改业务表）",
+        UniqueConstraint(
+            "entity_type", "source_id", "row_hash", name="uk_tplus_sync_records_entity_source_hash"
+        ),
+        Index("idx_tplus_sync_records_batch_no", "batch_no"),
+        Index("idx_tplus_sync_records_source_no", "source_no"),
+        Index("idx_tplus_sync_records_state", "state"),
+    )
+
+    id: Mapped[int] = mapped_column(
+        BIGINT_UNSIGNED, primary_key=True, autoincrement=True, comment="主键ID"
+    )
+    batch_no: Mapped[str] = mapped_column(String(32), nullable=False, comment="同步批次号")
+    entity_type: Mapped[str] = mapped_column(String(32), nullable=False, comment="业务实体类型")
+    source_table: Mapped[str] = mapped_column(String(128), nullable=False, comment="T+ 源表")
+    source_id: Mapped[str] = mapped_column(String(64), nullable=False, comment="T+ 源主键")
+    source_no: Mapped[str | None] = mapped_column(String(64), comment="T+ 原始编号")
+    source_line_no: Mapped[str | None] = mapped_column(String(64), comment="T+ 明细行号或明细主键")
+    source_updated_at: Mapped[datetime | None] = mapped_column(DateTime, comment="T+ 源更新时间")
+    row_hash: Mapped[str] = mapped_column(String(64), nullable=False, comment="原始行 SHA-256")
+    state: Mapped[str] = mapped_column(
+        String(16),
+        nullable=False,
+        server_default="PENDING",
+        comment="状态：PENDING/APPLIED/SKIPPED/ERROR",
+    )
+    payload_json: Mapped[dict] = mapped_column(JSON, nullable=False, comment="原始行 JSON")
+
+
+class TPlusSyncWatermark(Base, TimestampMixin):
+    """按实体保存增量同步水位。"""
+
+    __tablename__ = "tplus_sync_watermarks"
+    __table_args__ = mysql_table_args(
+        "T+ 增量同步水位表",
+        UniqueConstraint("entity_type", name="uk_tplus_sync_watermarks_entity_type"),
+    )
+
+    id: Mapped[int] = mapped_column(
+        BIGINT_UNSIGNED, primary_key=True, autoincrement=True, comment="主键ID"
+    )
+    entity_type: Mapped[str] = mapped_column(String(32), nullable=False, comment="业务实体类型")
+    last_source_updated_at: Mapped[datetime | None] = mapped_column(
+        DateTime, comment="最近源更新时间"
+    )
+    last_source_id: Mapped[str | None] = mapped_column(String(64), comment="同一时间点的最后源主键")
+    status: Mapped[str] = mapped_column(
+        String(16), nullable=False, server_default="READY", comment="状态：READY/BLOCKED"
+    )

backend/app/repositories/business_repository.py

@@ -5,9 +5,7 @@
 models/customer.py 定义）。
 """
 
-from datetime import date
-
-from sqlalchemy import Row, Select, delete, exists, func, or_, select
+from sqlalchemy import Row, Select, delete, exists, func, or_, select, update
 from sqlalchemy.orm import Session, aliased
 
 from app.models import (
@@ -23,6 +21,8 @@ from app.models import (
 )
 from app.models.customer import OWNER_TYPE_BUSINESS
 from app.repositories import ownership_log_repository
+from app.repositories.ownership_log_repository import OwnershipLogRef
+from app.schemas.business import BusinessQuery
 
 
 def get_by_id(session: Session, business_id: int, *, for_update: bool = False) -> Business | None:
@@ -77,50 +77,32 @@ def get_with_names(
     return session.execute(stmt).first()
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    keyword: str | None,
-    status: int | None,
-    customer_id: int | None,
-    owner_personnel_id: int | None,
-    date_from: date | None,
-    date_to: date | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: BusinessQuery) -> Select:
     """业务列表的筛选条件（调用方须已 join Customer）。"""
-    if keyword:
+    if query.keyword:
         # 业务编号/公司名称双列 OR 模糊；contains(autoescape) 同 query_utils.fuzzy
         # 的转义口径（fuzzy 为单列 where 语义，双列 OR 在此直接组合）
         stmt = stmt.where(
             or_(
-                Business.business_no.contains(keyword, autoescape=True),
-                Customer.company_name.contains(keyword, autoescape=True),
+                Business.business_no.contains(query.keyword, autoescape=True),
+                Customer.company_name.contains(query.keyword, autoescape=True),
             )
         )
-    if status is not None:
-        stmt = stmt.where(Business.status == status)
-    if customer_id is not None:
-        stmt = stmt.where(Business.customer_id == customer_id)
-    if owner_personnel_id is not None:
-        stmt = stmt.where(Business.owner_personnel_id == owner_personnel_id)
-    if date_from is not None:
-        stmt = stmt.where(Business.business_date >= date_from)
-    if date_to is not None:
-        stmt = stmt.where(Business.business_date <= date_to)
+    if query.status is not None:
+        stmt = stmt.where(Business.status == query.status)
+    if query.customer_id is not None:
+        stmt = stmt.where(Business.customer_id == query.customer_id)
+    if query.owner_personnel_id is not None:
+        stmt = stmt.where(Business.owner_personnel_id == query.owner_personnel_id)
+    if query.date_from is not None:
+        stmt = stmt.where(Business.business_date >= query.date_from)
+    if query.date_to is not None:
+        stmt = stmt.where(Business.business_date <= query.date_to)
     return stmt
 
 
 def list_businesses(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    keyword: str | None,
-    status: int | None,
-    customer_id: int | None,
-    owner_personnel_id: int | None,
-    date_from: date | None,
-    date_to: date | None,
+    session: Session, query: BusinessQuery
 ) -> list[tuple[Business, str, str | None, str | None]]:
     """分页查询业务（附客户名/负责人名/协助人名），按 id 倒序。"""
     owner = aliased(Personnel)
@@ -128,49 +110,24 @@ def list_businesses(
     stmt = select(
         Business, Customer.company_name, owner.personnel_name, assistant.personnel_name
     ).join(Customer, Customer.id == Business.customer_id)
-    stmt = _apply_filters(
-        stmt,
-        keyword=keyword,
-        status=status,
-        customer_id=customer_id,
-        owner_personnel_id=owner_personnel_id,
-        date_from=date_from,
-        date_to=date_to,
-    )
+    stmt = _apply_filters(stmt, query)
     stmt = (
         stmt.outerjoin(owner, owner.id == Business.owner_personnel_id)
         .outerjoin(assistant, assistant.id == Business.assistant_personnel_id)
         .order_by(Business.id.desc())
-        .offset(offset)
-        .limit(limit)
+        .offset(query.offset)
+        .limit(query.page_size)
     )
     return list(session.execute(stmt).all())
 
 
-def count_businesses(
-    session: Session,
-    *,
-    keyword: str | None,
-    status: int | None,
-    customer_id: int | None,
-    owner_personnel_id: int | None,
-    date_from: date | None,
-    date_to: date | None,
-) -> int:
+def count_businesses(session: Session, query: BusinessQuery) -> int:
     stmt = (
         select(func.count())
         .select_from(Business)
         .join(Customer, Customer.id == Business.customer_id)
     )
-    stmt = _apply_filters(
-        stmt,
-        keyword=keyword,
-        status=status,
-        customer_id=customer_id,
-        owner_personnel_id=owner_personnel_id,
-        date_from=date_from,
-        date_to=date_to,
-    )
+    stmt = _apply_filters(stmt, query)
     return int(session.execute(stmt).scalar_one())
 
 
@@ -199,43 +156,75 @@ def replace_items(session: Session, business_id: int, rows: list[BusinessItem])
     session.flush()
 
 
-def list_allocations(session: Session, business_id: int) -> list[SecondaryAllocation]:
-    """二次分配方案（详情回显；空列表=未设置方案）。"""
+def list_allocations(
+    session: Session, business_id: int, *, active_only: bool = True
+) -> list[SecondaryAllocation]:
+    """二次分配方案（详情回显与核算取数；默认仅当前生效版本）。"""
+    stmt = select(SecondaryAllocation).where(SecondaryAllocation.business_id == business_id)
+    if active_only:
+        stmt = stmt.where(SecondaryAllocation.is_active == 1)
     return list(
         session.execute(
-            select(SecondaryAllocation)
-            .where(SecondaryAllocation.business_id == business_id)
-            .order_by(SecondaryAllocation.id)
+            stmt.order_by(SecondaryAllocation.version_no, SecondaryAllocation.id)
         ).scalars()
     )
 
 
+def list_allocation_versions(session: Session, business_id: int) -> list[SecondaryAllocation]:
+    """方案全部版本行（含历史；业务详情「方案历史」用，按版本正序）。"""
+    return list_allocations(session, business_id, active_only=False)
+
+
 def replace_allocations(
     session: Session, business_id: int, rows: list[SecondaryAllocation]
 ) -> None:
-    """全量替换二次分配方案（不 commit，事务边界在服务层）。
+    """全量替换二次分配方案：旧生效行置历史，新行按下一版本号插入。
 
-    本表有 uk(business_id, personnel_id)，**必须**沿用先删后 flush 再插的
-    防御写法：同 flush 内 INSERT 先于 DELETE 处理，直接删加插会撞唯一键。
+    不 commit（事务边界在服务层）。版本行 append-only——编辑与清空都不
+    物理删除，驳回后仍可回溯被否决方案的参与人与比例。空 ``rows`` 表示
+    只清空生效态（不产生新版本）。
     """
-    session.execute(
-        delete(SecondaryAllocation).where(SecondaryAllocation.business_id == business_id)
+    deactivate_allocations(session, business_id)
+    if not rows:
+        return
+    next_version = (
+        session.execute(
+            select(func.coalesce(func.max(SecondaryAllocation.version_no), 0)).where(
+                SecondaryAllocation.business_id == business_id
+            )
+        ).scalar_one()
+        + 1
     )
-    session.flush()
     for row in rows:
         row.business_id = business_id
+        row.version_no = next_version
+        row.is_active = 1
         session.add(row)
     session.flush()
 
 
-def delete_allocations(session: Session, business_id: int) -> None:
-    """清空业务的二次分配方案（审批驳回/撤回时由 approval_service 调用）。
+def deactivate_allocations(session: Session, business_id: int) -> None:
+    """把业务的当前生效方案置为历史（审批驳回/编辑替换时调用；行保留）。
 
-    被否决的方案不得以「当前生效行」身份留在表里（M3 核算快照会取到）；
-    方案事实由审批单留痕与操作日志 request_data 承载。
+    被否决的方案不得以「当前生效行」身份继续被核算取到；``is_active=0``
+    后核算侧自然读不到，而历史版本仍可在业务详情回溯。不 commit。
+
+    显式声明 ``synchronize_session="fetch"``：Core UPDATE 默认走 auto 策略
+    （优先 evaluate，本例的条件恰好能在 Python 侧求值，故实测也会刷新在途
+    对象）。显式写 fetch 是为了不把「在途对象被刷新」这件事押在 evaluate
+    能否解析条件上——条件一旦复杂化，auto 会退化为不刷新，同一会话里已
+    加载的方案行就会停留在 ``is_active=1`` 的过期内存值（查询虽有
+    is_active 过滤，但行对象本身仍是「生效态」，被序列化或参与判断就是
+    错的）。
     """
     session.execute(
-        delete(SecondaryAllocation).where(SecondaryAllocation.business_id == business_id)
+        update(SecondaryAllocation)
+        .where(
+            SecondaryAllocation.business_id == business_id,
+            SecondaryAllocation.is_active == 1,
+        )
+        .values(is_active=0)
+        .execution_options(synchronize_session="fetch")
     )
     session.flush()
 
@@ -281,18 +270,17 @@ def list_ownership_logs(
 ) -> list[Row]:
     """某业务的归属变更留痕（共享实现，owner_type/business_id 参数化）。"""
     return ownership_log_repository.list_by_ref(
-        session,
-        owner_type=OWNER_TYPE_BUSINESS,
-        ref_column=OwnershipChangeLog.business_id,
-        ref_id=business_id,
-        offset=offset,
-        limit=limit,
+        session, _business_log_ref(business_id), offset=offset, limit=limit
     )
 
 
 def count_ownership_logs(session: Session, business_id: int) -> int:
-    return ownership_log_repository.count_by_ref(
-        session,
+    return ownership_log_repository.count_by_ref(session, _business_log_ref(business_id))
+
+
+def _business_log_ref(business_id: int) -> OwnershipLogRef:
+    """业务维度的归属留痕锚点。"""
+    return OwnershipLogRef(
         owner_type=OWNER_TYPE_BUSINESS,
         ref_column=OwnershipChangeLog.business_id,
         ref_id=business_id,

backend/app/repositories/business_role_repository.py

@@ -7,6 +7,7 @@ from sqlalchemy import Select, func, or_, select
 from sqlalchemy.orm import Session
 
 from app.models import BusinessRole
+from app.schemas.business_role import BusinessRoleQuery
 
 # ==================== 读取 ====================
 
@@ -39,41 +40,48 @@ def list_by_codes(session: Session, role_codes: list[str]) -> dict[str, Business
     return {role.role_code: role for role in rows}
 
 
-def list_roles(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    keyword: str | None = None,
-    status: int | None = None,
-) -> list[BusinessRole]:
+def list_roles(session: Session, query: BusinessRoleQuery) -> list[BusinessRole]:
     """分页查询岗位（含停用）。排序 (sort_order, id)，与系统角色口径一致。"""
-    stmt = _apply_filters(select(BusinessRole), keyword=keyword, status=status)
-    stmt = stmt.order_by(BusinessRole.sort_order, BusinessRole.id).offset(offset).limit(limit)
+    stmt = _apply_filters(select(BusinessRole), query)
+    stmt = (
+        stmt.order_by(BusinessRole.sort_order, BusinessRole.id)
+        .offset(query.offset)
+        .limit(query.page_size)
+    )
     return list(session.execute(stmt).scalars())
 
 
-def count_roles(session: Session, *, keyword: str | None = None, status: int | None = None) -> int:
+def count_roles(session: Session, query: BusinessRoleQuery) -> int:
     """统计符合条件的岗位总数，供分页使用。"""
-    stmt = _apply_filters(select(func.count(BusinessRole.id)), keyword=keyword, status=status)
+    stmt = _apply_filters(select(func.count(BusinessRole.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
-def _apply_filters(stmt: Select, *, keyword: str | None, status: int | None) -> Select:
+def list_all(session: Session) -> list[BusinessRole]:
+    """全部岗位（含停用，不分页）——岗位下拉选项的取数口径。
+
+    岗位规模有限（内置 + 自定义），一次取全量，不受分页上限约束；
+    排序与 list_roles 一致（sort_order, id），下拉与列表相对次序相同。
+    """
+    stmt = select(BusinessRole).order_by(BusinessRole.sort_order, BusinessRole.id)
+    return list(session.execute(stmt).scalars())
+
+
+def _apply_filters(stmt: Select, query: BusinessRoleQuery) -> Select:
     """应用通用筛选条件。列表与计数共用，避免两处口径不一致。
 
     关键词走 ``contains(autoescape=True)``（转义 LIKE 通配符）——岗位编码含下划线，
     不转义时 ``_`` 会当单字符通配执行。
     """
-    if keyword:
+    if query.keyword:
         stmt = stmt.where(
             or_(
-                BusinessRole.role_code.contains(keyword, autoescape=True),
-                BusinessRole.role_name.contains(keyword, autoescape=True),
+                BusinessRole.role_code.contains(query.keyword, autoescape=True),
+                BusinessRole.role_name.contains(query.keyword, autoescape=True),
             )
         )
-    if status is not None:
-        stmt = stmt.where(BusinessRole.status == status)
+    if query.status is not None:
+        stmt = stmt.where(BusinessRole.status == query.status)
     return stmt
 
 

backend/app/repositories/customer_repository.py

@@ -9,7 +9,9 @@ from sqlalchemy.orm import Session
 from app.models import Customer, OwnershipChangeLog, Personnel
 from app.models.customer import OWNER_TYPE_CUSTOMER
 from app.repositories import ownership_log_repository
+from app.repositories.ownership_log_repository import OwnershipLogRef
 from app.repositories.query_utils import fuzzy
+from app.schemas.customer import CustomerQuery
 
 # ==================== 读取 ====================
 
@@ -45,49 +47,27 @@ def get_by_company_owner(session: Session, company_name: str, owner_name: str) -
     return session.execute(stmt).scalar_one_or_none()
 
 
-def list_customers(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    keyword: str | None = None,
-    customer_type: str | None = None,
-    owner_personnel_id: int | None = None,
-) -> list[Row]:
+def list_customers(session: Session, query: CustomerQuery) -> list[Row]:
     """分页查询客户，行内容为 (客户, 归属人员姓名)。
 
     归属人员外键非空但统一用 outerjoin：人员行不可考（被物理删除的脏数据）
     不应让客户行整个查不出来（同 personnel_repository 的韧性口径）。
-    排序按 id 倒序：客户无排序字段，新增的靠前符合列表页习惯。
+    排序键由查询白名单决定（见 ``CustomerQuery.order_by``）：
+    id 倒序 = 新建档靠前，rating_score 按分数排、未评分垫底。
     """
-    stmt = _apply_filters(
-        _base_select(),
-        keyword=keyword,
-        customer_type=customer_type,
-        owner_personnel_id=owner_personnel_id,
-    )
-    stmt = stmt.order_by(Customer.id.desc()).offset(offset).limit(limit)
+    stmt = _apply_filters(_base_select(), query)
+    stmt = _apply_order(stmt, query)
+    stmt = stmt.offset(query.offset).limit(query.page_size)
     return list(session.execute(stmt).all())
 
 
-def count_customers(
-    session: Session,
-    *,
-    keyword: str | None = None,
-    customer_type: str | None = None,
-    owner_personnel_id: int | None = None,
-) -> int:
+def count_customers(session: Session, query: CustomerQuery) -> int:
     """统计符合条件的客户总数，供分页使用。
 
     筛选条件全部落在主表上，计数无需 join（列表与计数的筛选口径由
     ``_apply_filters`` 单点保证）。
     """
-    stmt = _apply_filters(
-        select(func.count(Customer.id)),
-        keyword=keyword,
-        customer_type=customer_type,
-        owner_personnel_id=owner_personnel_id,
-    )
+    stmt = _apply_filters(select(func.count(Customer.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
@@ -101,9 +81,7 @@ def list_ownership_logs(
     """查某客户的归属变更留痕（共享实现，owner_type/customer_id 参数化）。"""
     return ownership_log_repository.list_by_ref(
         session,
-        owner_type=OWNER_TYPE_CUSTOMER,
-        ref_column=OwnershipChangeLog.customer_id,
-        ref_id=customer_id,
+        _customer_log_ref(customer_id),
         offset=offset,
         limit=limit,
     )
@@ -111,8 +89,12 @@ def list_ownership_logs(
 
 def count_ownership_logs(session: Session, *, customer_id: int) -> int:
     """统计某客户的归属留痕总数，供分页使用。"""
-    return ownership_log_repository.count_by_ref(
-        session,
+    return ownership_log_repository.count_by_ref(session, _customer_log_ref(customer_id))
+
+
+def _customer_log_ref(customer_id: int) -> OwnershipLogRef:
+    """客户维度的归属留痕锚点。"""
+    return OwnershipLogRef(
         owner_type=OWNER_TYPE_CUSTOMER,
         ref_column=OwnershipChangeLog.customer_id,
         ref_id=customer_id,
@@ -126,20 +108,14 @@ def _base_select() -> Select:
     )
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    keyword: str | None,
-    customer_type: str | None,
-    owner_personnel_id: int | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: CustomerQuery) -> Select:
     """应用通用筛选条件。列表与计数共用，避免两处口径不一致。"""
-    if keyword:
-        stmt = stmt.where(_keyword_clause(keyword))
-    if customer_type:
-        stmt = stmt.where(Customer.customer_type == customer_type)
-    if owner_personnel_id is not None:
-        stmt = stmt.where(Customer.owner_personnel_id == owner_personnel_id)
+    if query.keyword:
+        stmt = stmt.where(_keyword_clause(query.keyword))
+    if query.customer_type:
+        stmt = stmt.where(Customer.customer_type == query.customer_type)
+    if query.owner_personnel_id is not None:
+        stmt = stmt.where(Customer.owner_personnel_id == query.owner_personnel_id)
     return stmt
 
 
@@ -157,6 +133,23 @@ def _keyword_clause(keyword: str) -> ColumnElement[bool]:
     )
 
 
+def _apply_order(stmt: Select, query: CustomerQuery) -> Select:
+    """应用排序。排序键已由 schema 白名单收口，此处只做键 → 列的映射。
+
+    MySQL 把 NULL 视为最小值：ASC 时未评分（null）会排最前，但「没打分」
+    不该压在「打了低分」之上 —— ASC 分支用 ``is_(None)`` 布尔键把 null
+    挪到末尾；DESC 分支 null 天然在最后，无需处理。尾部再按 id 倒序兜底，
+    保证同分行的次序稳定（分页不跳行）。
+    """
+    if query.order_by == "rating_score":
+        score = Customer.rating_score
+        if query.order == "asc":
+            stmt = stmt.order_by(score.is_(None), score.asc())
+        else:
+            stmt = stmt.order_by(score.desc())
+    return stmt.order_by(Customer.id.desc())
+
+
 # ==================== 写入 ====================
 
 

backend/app/repositories/dashboard_repository.py

@@ -8,7 +8,7 @@
 from datetime import date, datetime
 
 from sqlalchemy import exists, func, or_, select
-from sqlalchemy.orm import Session
+from sqlalchemy.orm import Session, defer
 
 from app.models import (
     ApprovalOrder,
@@ -34,20 +34,25 @@ _PENDING_STATUSES = (
 
 # ==== 经营绩效总览 ====
 
+
 def sum_month_performance(session: Session, month: str) -> dict:
     """当月单笔生效行合计与人员数（拆行求和不失真，人数需 DISTINCT）。"""
-    return session.execute(
-        select(
-            func.coalesce(func.sum(PerformanceLedger.performance_amount), 0).label(
-                "performance_total"
-            ),
-            func.coalesce(func.sum(PerformanceLedger.payable_amount), 0).label("payable_total"),
-            func.count(func.distinct(PerformanceLedger.personnel_id)).label("person_count"),
-        ).where(
-            PerformanceLedger.settlement_month == month,
-            *_ACTIVE_SINGLE,
+    return (
+        session.execute(
+            select(
+                func.coalesce(func.sum(PerformanceLedger.performance_amount), 0).label(
+                    "performance_total"
+                ),
+                func.coalesce(func.sum(PerformanceLedger.payable_amount), 0).label("payable_total"),
+                func.count(func.distinct(PerformanceLedger.personnel_id)).label("person_count"),
+            ).where(
+                PerformanceLedger.settlement_month == month,
+                *_ACTIVE_SINGLE,
+            )
         )
-    ).mappings().one()
+        .mappings()
+        .one()
+    )
 
 
 def performance_trend(session: Session, months: list[str]) -> dict[str, object]:
@@ -66,9 +71,7 @@ def performance_trend(session: Session, months: list[str]) -> dict[str, object]:
     return dict(rows)
 
 
-def receipt_trend(
-    session: Session, *, start_date: date, end_date: date
-) -> dict[str, object]:
+def receipt_trend(session: Session, *, start_date: date, end_date: date) -> dict[str, object]:
     """近 N 个月回款合计（按到账月，含未核算回款——与台账口径不同，页面注明）。
 
     过滤走 ``receipt_date`` 原始列范围（可用索引）；``DATE_FORMAT`` 只用于
@@ -107,18 +110,22 @@ def get_month_settlement(session: Session, month: str) -> MonthlySettlement | No
 
 # ==== 待办事项 ====
 
+
 def list_pending_approvals(
     session: Session, *, statuses: tuple[int, ...], limit: int
 ) -> list[ApprovalOrder]:
     """待处理审批单（按提交时间升序——最久的排前面）。
 
     ``statuses`` 由调用方按当前用户角色解析（待我审批），空元组返回空列表。
+
+    排除 snapshot_json 列（快照可达上百 KB，待办只取单号/类型/提交人/时间）。
     """
     if not statuses:
         return []
     return list(
         session.execute(
             select(ApprovalOrder)
+            .options(defer(ApprovalOrder.snapshot_json))
             .where(ApprovalOrder.status.in_(statuses))
             .order_by(ApprovalOrder.create_time.asc(), ApprovalOrder.id.asc())
             .limit(limit)
@@ -157,15 +164,14 @@ def list_unassigned_receipts(session: Session, *, limit: int) -> list:
 def count_unassigned_receipts(session: Session) -> int:
     return int(
         session.execute(
-            select(func.count())
-            .select_from(Receipt)
-            .where(Receipt.shipment_id.is_(None))
+            select(func.count()).select_from(Receipt).where(Receipt.shipment_id.is_(None))
         ).scalar_one()
     )
 
 
 # ==== 风险预警 ====
 
+
 def list_overdue_shipments(session: Session, *, cutoff: date, limit: int) -> list:
     """发货后超过阈值且该批次无任何回款的发货单（回款逾期规则）。
 
@@ -210,10 +216,14 @@ def count_overdue_shipments(session: Session, *, cutoff: date) -> int:
 def list_overdue_approvals(
     session: Session, *, cutoff: datetime, limit: int
 ) -> list[ApprovalOrder]:
-    """提交后超过阈值仍未处理的审批单（审批超时规则）。"""
+    """提交后超过阈值仍未处理的审批单（审批超时规则）。
+
+    排除 snapshot_json 列（同待办：预警只用单号/类型/提交人/时间）。
+    """
     return list(
         session.execute(
             select(ApprovalOrder)
+            .options(defer(ApprovalOrder.snapshot_json))
             .where(
                 ApprovalOrder.status.in_(_PENDING_STATUSES),
                 ApprovalOrder.create_time < cutoff,

backend/app/repositories/login_log_repository.py

@@ -1,5 +1,7 @@
 """登录日志数据访问。"""
 
+from dataclasses import dataclass
+
 from sqlalchemy import Select, func, select
 from sqlalchemy.orm import Session
 
@@ -10,68 +12,73 @@ from app.repositories.query_utils import LogFilters, apply_log_filters
 _USER_AGENT_MAX_LEN = 255
 
 
-def add(
-    session: Session,
-    *,
-    user_id: int | None,
-    username: str,
-    ip: str | None,
-    user_agent: str | None,
-    success: bool,
-    fail_reason: str | None = None,
-) -> None:
+@dataclass(frozen=True, slots=True)
+class LoginLogEntry:
+    """一条待写入的登录日志（全字段 DTO）。
+
+    与 ``OperationLogEntry`` 同款做法：写入字段平铺传递时，任何一处漏传
+    都只表现为日志字段缺失、不报错（success 默认 0 的语义尤其隐蔽）。
+    """
+
+    user_id: int | None
+    username: str
+    success: bool
+    ip: str | None = None
+    user_agent: str | None = None
+    fail_reason: str | None = None
+
+
+@dataclass(frozen=True, slots=True)
+class LoginLogListQuery:
+    """登录日志列表的查询条件（筛选 + 分页）。"""
+
+    filters: LogFilters
+    offset: int = 0
+    limit: int = 20
+    ip: str | None = None
+
+
+def add(session: Session, entry: LoginLogEntry) -> None:
     """写入一条登录日志。只 flush 不 commit。
 
     **绝不记录密码** —— 明文与哈希都不写。日志是最容易泄漏凭据的地方。
 
-    :param user_id: 账号不存在时传 None（列本身可空）
-    :param username: 用户名原样记录，便于账号不存在时也能追溯
+    :param entry.user_id: 账号不存在时为 None（列本身可空）
+    :param entry.username: 用户名原样记录，便于账号不存在时也能追溯
     """
     session.add(
         LoginLog(
-            user_id=user_id,
-            username=username,
-            ip=ip,
-            user_agent=user_agent[:_USER_AGENT_MAX_LEN] if user_agent else None,
-            success=1 if success else 0,
-            fail_reason=fail_reason,
+            user_id=entry.user_id,
+            username=entry.username,
+            ip=entry.ip,
+            user_agent=entry.user_agent[:_USER_AGENT_MAX_LEN] if entry.user_agent else None,
+            success=1 if entry.success else 0,
+            fail_reason=entry.fail_reason,
         )
     )
     session.flush()
 
 
-def list_logs(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    filters: LogFilters,
-    ip: str | None = None,
-) -> list[LoginLog]:
+def list_logs(session: Session, query: LoginLogListQuery) -> list[LoginLog]:
     """分页查询登录日志。
 
     **按 id 倒序而非 create_time**：后者是秒级精度，同一秒内的多条记录排序
     不确定，翻页会重复或遗漏（与用户列表同一理由）。
     """
-    stmt = _apply_filters(select(LoginLog), filters=filters, ip=ip)
-    stmt = stmt.order_by(LoginLog.id.desc()).offset(offset).limit(limit)
+    stmt = _apply_filters(select(LoginLog), query)
+    stmt = stmt.order_by(LoginLog.id.desc()).offset(query.offset).limit(query.limit)
     return list(session.execute(stmt).scalars())
 
 
-def count_logs(
-    session: Session,
-    *,
-    filters: LogFilters,
-    ip: str | None = None,
-) -> int:
+def count_logs(session: Session, query: LoginLogListQuery) -> int:
     """统计符合条件的登录日志总数，供分页使用。"""
-    stmt = _apply_filters(select(func.count(LoginLog.id)), filters=filters, ip=ip)
+    stmt = _apply_filters(select(func.count(LoginLog.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
-def _apply_filters(stmt: Select, *, filters: LogFilters, ip: str | None) -> Select:
+def _apply_filters(stmt: Select, query: LoginLogListQuery) -> Select:
     """应用筛选条件（公共部分复用 query_utils，与操作日志同口径）。"""
-    stmt = apply_log_filters(stmt, LoginLog, filters)
-    if ip:
-        stmt = stmt.where(LoginLog.ip == ip)
+    stmt = apply_log_filters(stmt, LoginLog, query.filters)
+    if query.ip:
+        stmt = stmt.where(LoginLog.ip == query.ip)
     return stmt

backend/app/repositories/operation_log_repository.py

@@ -107,15 +107,18 @@ def get_by_id(session: Session, log_id: int) -> OperationLog | None:
     return session.get(OperationLog, log_id)
 
 
-def list_logs(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    filters: LogFilters,
-    module: str | None = None,
-    action: str | None = None,
-) -> list[OperationLog]:
+@dataclass(frozen=True, slots=True)
+class OperationLogListQuery:
+    """操作日志列表的查询条件（筛选 + 分页）。"""
+
+    filters: LogFilters
+    offset: int = 0
+    limit: int = 20
+    module: str | None = None
+    action: str | None = None
+
+
+def list_logs(session: Session, query: OperationLogListQuery) -> list[OperationLog]:
     """分页查询操作日志。
 
     **按 id 倒序而非 create_time**：后者是秒级精度，同一秒内的多条记录排序
@@ -127,41 +130,34 @@ def list_logs(
     """
     stmt = _apply_filters(
         select(OperationLog).options(defer(OperationLog.request_data)),
-        filters=filters,
-        module=module,
-        action=action,
+        query,
     )
-    stmt = stmt.order_by(OperationLog.id.desc()).offset(offset).limit(limit)
+    stmt = stmt.order_by(OperationLog.id.desc()).offset(query.offset).limit(query.limit)
     return list(session.execute(stmt).scalars())
 
 
-def count_logs(
-    session: Session,
-    *,
-    filters: LogFilters,
-    module: str | None = None,
-    action: str | None = None,
-) -> int:
+def count_logs(session: Session, query: OperationLogListQuery) -> int:
     """统计符合条件的日志总数，供分页使用。"""
-    stmt = _apply_filters(
-        select(func.count(OperationLog.id)), filters=filters, module=module, action=action
-    )
+    stmt = _apply_filters(select(func.count(OperationLog.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    filters: LogFilters,
-    module: str | None,
-    action: str | None,
-) -> Select:
+def distinct_modules(session: Session) -> list[str]:
+    """全部模块名（去重、按名称升序）：工作台「最近动态」的模块筛选项。
+
+    交给 SQL 的 DISTINCT，不在应用层聚合 —— 应用层去重要把整列拉进内存。
+    """
+    stmt = select(OperationLog.module).distinct().order_by(OperationLog.module)
+    return [module for module in session.execute(stmt).scalars() if module]
+
+
+def _apply_filters(stmt: Select, query: OperationLogListQuery) -> Select:
     """应用筛选条件（公共部分复用 query_utils，与登录日志同口径）。"""
-    stmt = apply_log_filters(stmt, OperationLog, filters)
-    if module:
-        stmt = stmt.where(OperationLog.module == module)
-    if action:
-        stmt = stmt.where(OperationLog.action == action)
+    stmt = apply_log_filters(stmt, OperationLog, query.filters)
+    if query.module:
+        stmt = stmt.where(OperationLog.module == query.module)
+    if query.action:
+        stmt = stmt.where(OperationLog.action == query.action)
     return stmt
 
 

backend/app/repositories/ownership_log_repository.py

@@ -5,18 +5,32 @@ owner_type 与引用列由调用方参数化 —— 此前客户、业务两侧
 「被删人员留痕仍可读」「翻页稳定双键倒序」等口径单点维护。
 """
 
+from dataclasses import dataclass
+from typing import Any
+
 from sqlalchemy import Row, func, select
 from sqlalchemy.orm import Session, aliased
 
 from app.models import OwnershipChangeLog, Personnel
 
 
+@dataclass(frozen=True, slots=True)
+class OwnershipLogRef:
+    """归属留痕的引用锚点（维度类型 + 引用列 + 引用 ID）。
+
+    引用列是 SQLAlchemy 列表达式（如 ``OwnershipChangeLog.customer_id``），
+    不是可序列化值 —— 故用冻结 dataclass 而非 Pydantic 模型。
+    """
+
+    owner_type: int
+    ref_column: Any
+    ref_id: int
+
+
 def list_by_ref(
     session: Session,
+    ref: OwnershipLogRef,
     *,
-    owner_type: int,
-    ref_column,
-    ref_id: int,
     offset: int,
     limit: int,
 ) -> list[Row]:
@@ -32,7 +46,7 @@ def list_by_ref(
         select(OwnershipChangeLog, old_person.personnel_name, new_person.personnel_name)
         .outerjoin(old_person, OwnershipChangeLog.old_owner_personnel_id == old_person.id)
         .outerjoin(new_person, OwnershipChangeLog.new_owner_personnel_id == new_person.id)
-        .where(OwnershipChangeLog.owner_type == owner_type, ref_column == ref_id)
+        .where(OwnershipChangeLog.owner_type == ref.owner_type, ref.ref_column == ref.ref_id)
         .order_by(OwnershipChangeLog.create_time.desc(), OwnershipChangeLog.id.desc())
         .offset(offset)
         .limit(limit)
@@ -40,11 +54,11 @@ def list_by_ref(
     return list(session.execute(stmt).all())
 
 
-def count_by_ref(session: Session, *, owner_type: int, ref_column, ref_id: int) -> int:
+def count_by_ref(session: Session, ref: OwnershipLogRef) -> int:
     """统计某引用对象的归属留痕总数，供分页使用。"""
     stmt = select(func.count(OwnershipChangeLog.id)).where(
-        OwnershipChangeLog.owner_type == owner_type,
-        ref_column == ref_id,
+        OwnershipChangeLog.owner_type == ref.owner_type,
+        ref.ref_column == ref.ref_id,
     )
     return int(session.execute(stmt).scalar_one())
 

backend/app/repositories/param_repository.py

@@ -10,6 +10,7 @@ from sqlalchemy import Select, delete, func, select
 from sqlalchemy.orm import Session
 
 from app.models import ParamChangeBatch, ParamDefinition, ParamValue
+from app.schemas.param import ParamBatchQuery, ParamHistoryQuery
 
 # ==================== 读取 ====================
 
@@ -97,27 +98,27 @@ def list_definitions_by_codes(
 
 
 def list_history(
-    session: Session,
-    param_id: int,
-    *,
-    offset: int,
-    limit: int,
-    dimension_value: str | None = None,
+    session: Session, param_id: int, query: ParamHistoryQuery
 ) -> list[tuple[ParamValue, ParamChangeBatch]]:
-    """单参数的版本历史（含批次信息），最新在前。"""
-    stmt = _history_stmt(session, param_id, dimension_value=dimension_value)
-    return [(value, batch) for value, batch in session.execute(stmt.offset(offset).limit(limit))]
+    """单参数的版本历史（含批次信息），最新在前。
+
+    ``param_id`` 是「查哪个参数」的主锚点（由路径参数 param_code 解析而来），
+    不属于列表筛选项，故仍以位置参数传入；筛选与分页走 ``query``。
+    """
+    stmt = _history_stmt(param_id, query)
+    return [
+        (value, batch)
+        for value, batch in session.execute(stmt.offset(query.offset).limit(query.page_size))
+    ]
 
 
-def count_history(session: Session, param_id: int, *, dimension_value: str | None = None) -> int:
+def count_history(session: Session, param_id: int, query: ParamHistoryQuery) -> int:
     """统计单参数的历史条目数。"""
-    stmt = select(func.count()).select_from(
-        _history_stmt(session, param_id, dimension_value=dimension_value).subquery()
-    )
+    stmt = select(func.count()).select_from(_history_stmt(param_id, query).subquery())
     return int(session.execute(stmt).scalar_one())
 
 
-def _history_stmt(session: Session, param_id: int, *, dimension_value: str | None) -> Select:
+def _history_stmt(param_id: int, query: ParamHistoryQuery) -> Select:
     """历史查询基语句：条目 join 批次（版本号/操作人），最新在前。"""
     stmt = (
         select(ParamValue, ParamChangeBatch)
@@ -125,8 +126,8 @@ def _history_stmt(session: Session, param_id: int, *, dimension_value: str | Non
         .where(ParamValue.param_id == param_id)
         .order_by(ParamValue.effective_date.desc(), ParamValue.id.desc())
     )
-    if dimension_value is not None:
-        stmt = stmt.where(ParamValue.dimension_value == dimension_value)
+    if query.dimension_value is not None:
+        stmt = stmt.where(ParamValue.dimension_value == query.dimension_value)
     return stmt
 
 
@@ -150,41 +151,30 @@ def get_batch_by_id(
     return session.get(ParamChangeBatch, batch_id)
 
 
-def list_batches(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    start_date: date | None = None,
-    end_date: date | None = None,
-) -> list[ParamChangeBatch]:
+def list_batches(session: Session, query: ParamBatchQuery) -> list[ParamChangeBatch]:
     """分页查询变更批次（按创建时间倒序）。"""
-    stmt = _batch_filters(select(ParamChangeBatch), start_date=start_date, end_date=end_date)
+    stmt = _batch_filters(select(ParamChangeBatch), query)
     return list(
         session.execute(
             stmt.order_by(ParamChangeBatch.create_time.desc(), ParamChangeBatch.id.desc())
-            .offset(offset)
-            .limit(limit)
+            .offset(query.offset)
+            .limit(query.page_size)
         ).scalars()
     )
 
 
-def count_batches(
-    session: Session, *, start_date: date | None = None, end_date: date | None = None
-) -> int:
+def count_batches(session: Session, query: ParamBatchQuery) -> int:
     """统计符合条件的批次总数。"""
-    stmt = _batch_filters(
-        select(func.count(ParamChangeBatch.id)), start_date=start_date, end_date=end_date
-    )
+    stmt = _batch_filters(select(func.count(ParamChangeBatch.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
-def _batch_filters(stmt: Select, *, start_date: date | None, end_date: date | None) -> Select:
+def _batch_filters(stmt: Select, query: ParamBatchQuery) -> Select:
     """批次筛选（生效日期区间，含边界）。列表与计数共用。"""
-    if start_date is not None:
-        stmt = stmt.where(ParamChangeBatch.effective_date >= start_date)
-    if end_date is not None:
-        stmt = stmt.where(ParamChangeBatch.effective_date <= end_date)
+    if query.start_date is not None:
+        stmt = stmt.where(ParamChangeBatch.effective_date >= query.start_date)
+    if query.end_date is not None:
+        stmt = stmt.where(ParamChangeBatch.effective_date <= query.end_date)
     return stmt
 
 

backend/app/repositories/personnel_repository.py

@@ -9,6 +9,7 @@ from sqlalchemy.orm import Session, aliased
 
 from app.models import BusinessRole, Customer, Personnel, PersonnelPlan, PersonnelRegion
 from app.repositories.query_utils import fuzzy
+from app.schemas.personnel import PersonnelQuery
 
 # 自引用别名：manager_id 指向 personnel.id，同一张表自联取「直属经理姓名」
 _manager = aliased(Personnel)
@@ -30,44 +31,23 @@ def get_with_names(session: Session, personnel_id: int) -> Row | None:
     return session.execute(stmt).first()
 
 
-def list_personnel(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    keyword: str | None = None,
-    status: int | None = None,
-    business_role_id: int | None = None,
-) -> list[Row]:
+def list_personnel(session: Session, query: PersonnelQuery) -> list[Row]:
     """分页查询人员，行内容为 (人员, 岗位名称, 直属经理姓名)。
 
     排序按 id 倒序：人员无排序字段，新增的靠前符合列表页习惯。
     """
-    stmt = _apply_filters(
-        _base_select(), keyword=keyword, status=status, business_role_id=business_role_id
-    )
-    stmt = stmt.order_by(Personnel.id.desc()).offset(offset).limit(limit)
+    stmt = _apply_filters(_base_select(), query)
+    stmt = stmt.order_by(Personnel.id.desc()).offset(query.offset).limit(query.page_size)
     return list(session.execute(stmt).all())
 
 
-def count_personnel(
-    session: Session,
-    *,
-    keyword: str | None = None,
-    status: int | None = None,
-    business_role_id: int | None = None,
-) -> int:
+def count_personnel(session: Session, query: PersonnelQuery) -> int:
     """统计符合条件的人员总数，供分页使用。
 
     筛选条件全部落在主表上，计数无需 join（列表与计数的筛选口径由
     ``_apply_filters`` 单点保证）。
     """
-    stmt = _apply_filters(
-        select(func.count(Personnel.id)),
-        keyword=keyword,
-        status=status,
-        business_role_id=business_role_id,
-    )
+    stmt = _apply_filters(select(func.count(Personnel.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
@@ -154,20 +134,14 @@ def _base_select() -> Select:
     )
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    keyword: str | None,
-    status: int | None,
-    business_role_id: int | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: PersonnelQuery) -> Select:
     """应用通用筛选条件。列表与计数共用，避免两处口径不一致。"""
-    if keyword:
-        stmt = fuzzy(stmt, Personnel.personnel_name, keyword)
-    if status is not None:
-        stmt = stmt.where(Personnel.status == status)
-    if business_role_id is not None:
-        stmt = stmt.where(Personnel.business_role_id == business_role_id)
+    if query.keyword:
+        stmt = fuzzy(stmt, Personnel.personnel_name, query.keyword)
+    if query.status is not None:
+        stmt = stmt.where(Personnel.status == query.status)
+    if query.business_role_id is not None:
+        stmt = stmt.where(Personnel.business_role_id == query.business_role_id)
     return stmt
 
 

backend/app/repositories/product_repository.py

@@ -7,6 +7,7 @@ from sqlalchemy import Select, func, or_, select
 from sqlalchemy.orm import Session
 
 from app.models import BusinessItem, Product, ShipmentItem
+from app.schemas.product import ProductQuery
 
 # ==================== 读取 ====================
 
@@ -34,50 +35,20 @@ def get_by_name_spec(session: Session, product_name: str, spec_model: str) -> Pr
     return session.execute(stmt).scalar_one_or_none()
 
 
-def list_products(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    keyword: str | None = None,
-    product_category: str | None = None,
-    status: int | None = None,
-) -> list[Product]:
+def list_products(session: Session, query: ProductQuery) -> list[Product]:
     """分页查询产品（含停用）。按 id 升序，与 M1 岗位列表的次序口径一致。"""
-    stmt = _apply_filters(
-        select(Product),
-        keyword=keyword,
-        product_category=product_category,
-        status=status,
-    )
-    stmt = stmt.order_by(Product.id).offset(offset).limit(limit)
+    stmt = _apply_filters(select(Product), query)
+    stmt = stmt.order_by(Product.id).offset(query.offset).limit(query.page_size)
     return list(session.execute(stmt).scalars())
 
 
-def count_products(
-    session: Session,
-    *,
-    keyword: str | None = None,
-    product_category: str | None = None,
-    status: int | None = None,
-) -> int:
+def count_products(session: Session, query: ProductQuery) -> int:
     """统计符合条件的产品总数，供分页使用。"""
-    stmt = _apply_filters(
-        select(func.count(Product.id)),
-        keyword=keyword,
-        product_category=product_category,
-        status=status,
-    )
+    stmt = _apply_filters(select(func.count(Product.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    keyword: str | None,
-    product_category: str | None,
-    status: int | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: ProductQuery) -> Select:
     """应用通用筛选条件。列表与计数共用，避免两处口径不一致。
 
     关键词对名称/规格两列取 OR 命中。转义与 query_utils.fuzzy 同一机制
@@ -86,17 +57,17 @@ def _apply_filters(
     写 contains —— M1 business_role_repository 的双列关键词同款写法。
     规格型号常含 ``_``（如 DN300_II），不转义会当单字符通配执行。
     """
-    if keyword:
+    if query.keyword:
         stmt = stmt.where(
             or_(
-                Product.product_name.contains(keyword, autoescape=True),
-                Product.spec_model.contains(keyword, autoescape=True),
+                Product.product_name.contains(query.keyword, autoescape=True),
+                Product.spec_model.contains(query.keyword, autoescape=True),
             )
         )
-    if product_category:
-        stmt = stmt.where(Product.product_category == product_category)
-    if status is not None:
-        stmt = stmt.where(Product.status == status)
+    if query.product_category:
+        stmt = stmt.where(Product.product_category == query.product_category)
+    if query.status is not None:
+        stmt = stmt.where(Product.status == query.status)
     return stmt
 
 

backend/app/repositories/rating_repository.py

@@ -32,8 +32,9 @@ def list_all(session: Session, *, for_update: bool = False) -> list[CustomerRati
 def get_by_id(session: Session, grade_id: int) -> CustomerRatingGrade | None:
     """按主键查询。
 
-    等级表无下游引用（评级是纯计算器不落评分记录），无需 ``for_update``
-    锁行 —— product_repository 锁行防的是并发引用悬空，此处不存在该风险。
+    等级表无外键引用（客户档案只存总分，等级由分数按当前映射实时推导，
+    不存等级 id），无需 ``for_update`` 锁行 —— product_repository 锁行防的
+    是并发引用悬空，此处不存在该风险：删掉一个等级，只是以后不再判到它。
     """
     return session.get(CustomerRatingGrade, grade_id)
 
@@ -60,6 +61,10 @@ def create(session: Session, grade: CustomerRatingGrade) -> CustomerRatingGrade:
 
 
 def delete(session: Session, grade: CustomerRatingGrade) -> None:
-    """物理删除等级。等级表无下游引用，调用方无需引用保护校验。"""
+    """物理删除等级。
+
+    等级表无外键引用（客户档案只存总分，不存等级 id），调用方无需引用保护
+    校验：删掉等级只是「以后不再判到它」，已存的总分不受影响。
+    """
     session.delete(grade)
     session.flush()

backend/app/repositories/receipt_repository.py

@@ -14,6 +14,7 @@ from sqlalchemy import Select, func, select
 from sqlalchemy.orm import Session
 
 from app.models import Business, Receipt, Shipment
+from app.schemas.receipt import ReceiptQuery
 
 
 def get_by_id(session: Session, receipt_id: int) -> Receipt | None:
@@ -43,72 +44,42 @@ def get_shipment(session: Session, shipment_id: int) -> Shipment | None:
     return session.get(Shipment, shipment_id)
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    business_id: int | None,
-    date_from: date | None,
-    date_to: date | None,
-    is_barter: int | None,
-    settlement_status: int | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: ReceiptQuery) -> Select:
     """回款列表的筛选条件（全部落在 Receipt 自身列上）。"""
-    if business_id is not None:
-        stmt = stmt.where(Receipt.business_id == business_id)
-    if date_from is not None:
-        stmt = stmt.where(Receipt.receipt_date >= date_from)
-    if date_to is not None:
-        stmt = stmt.where(Receipt.receipt_date <= date_to)
-    if is_barter is not None:
-        stmt = stmt.where(Receipt.is_barter == is_barter)
-    if settlement_status is not None:
-        stmt = stmt.where(Receipt.settlement_status == settlement_status)
+    if query.business_id is not None:
+        stmt = stmt.where(Receipt.business_id == query.business_id)
+    if query.date_from is not None:
+        stmt = stmt.where(Receipt.receipt_date >= query.date_from)
+    if query.date_to is not None:
+        stmt = stmt.where(Receipt.receipt_date <= query.date_to)
+    if query.is_barter is not None:
+        stmt = stmt.where(Receipt.is_barter == query.is_barter)
+    # settlement_status 是 M3 核算防重锚点（需规 2.3.3），必须保持具名字段
+    if query.settlement_status is not None:
+        stmt = stmt.where(Receipt.settlement_status == query.settlement_status)
     return stmt
 
 
 def list_receipts(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    business_id: int | None,
-    date_from: date | None,
-    date_to: date | None,
-    is_barter: int | None,
-    settlement_status: int | None,
+    session: Session, query: ReceiptQuery
 ) -> list[tuple[Receipt, str, str | None, date | None]]:
     """分页查询回款（附业务号/发货单号/发货日），按 id 倒序。"""
     stmt = _apply_filters(
         select(Receipt, Business.business_no, Shipment.shipment_no, Shipment.ship_date)
         .join(Business, Business.id == Receipt.business_id)
         .outerjoin(Shipment, Shipment.id == Receipt.shipment_id),
-        business_id=business_id,
-        date_from=date_from,
-        date_to=date_to,
-        is_barter=is_barter,
-        settlement_status=settlement_status,
+        query,
+    )
+    return list(
+        session.execute(
+            stmt.order_by(Receipt.id.desc()).offset(query.offset).limit(query.page_size)
+        ).all()
     )
-    return list(session.execute(stmt.order_by(Receipt.id.desc()).offset(offset).limit(limit)).all())
 
 
-def count_receipts(
-    session: Session,
-    *,
-    business_id: int | None,
-    date_from: date | None,
-    date_to: date | None,
-    is_barter: int | None,
-    settlement_status: int | None,
-) -> int:
+def count_receipts(session: Session, query: ReceiptQuery) -> int:
     """回款总数（筛选条件与列表共用同一 apply_filters，防两侧口径漂移）。"""
-    stmt = _apply_filters(
-        select(func.count()).select_from(Receipt),
-        business_id=business_id,
-        date_from=date_from,
-        date_to=date_to,
-        is_barter=is_barter,
-        settlement_status=settlement_status,
-    )
+    stmt = _apply_filters(select(func.count()).select_from(Receipt), query)
     return int(session.execute(stmt).scalar_one())
 
 

backend/app/repositories/report_repository.py

@@ -389,9 +389,7 @@ def _bucket_condition(bucket: DayBucket) -> tuple:
     return tuple(conds)
 
 
-def list_payment_detail(
-    session: Session, query: PaymentDetailQuery, bucket: DayBucket, *, offset: int, limit: int
-) -> list:
+def list_payment_detail(session: Session, query: PaymentDetailQuery, bucket: DayBucket) -> list:
     """区间下钻：单笔生效行联回款表（行级展示回款金额与批次号）。
 
     bucket 由服务层解析（未注册编码 400），仓储只按区间拼条件。
@@ -408,7 +406,7 @@ def list_payment_detail(
         .where(*_payment_filters(query), *_bucket_condition(bucket))
     )
     stmt = stmt.order_by(PerformanceLedger.receipt_date.desc(), PerformanceLedger.id.desc())
-    stmt = stmt.offset(offset).limit(limit)
+    stmt = stmt.offset(query.offset).limit(query.page_size)
     return list(session.execute(stmt).all())
 
 

backend/app/repositories/role_repository.py

@@ -10,6 +10,7 @@ from sqlalchemy import Select, func, or_, select
 from sqlalchemy.orm import Session
 
 from app.models import Role, UserRole
+from app.schemas.role import RoleQuery
 
 # ==================== 读取 ====================
 
@@ -47,44 +48,37 @@ def get_all_roles(session: Session) -> list[Role]:
     return list(session.execute(stmt).scalars())
 
 
-def list_roles(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    keyword: str | None = None,
-    status: int | None = None,
-) -> list[Role]:
+def list_roles(session: Session, query: RoleQuery) -> list[Role]:
     """分页查询角色（管理列表：含停用角色）。
 
     排序与下拉口径一致（sort_order, id），保持同一套相对次序。
     """
-    stmt = _apply_filters(select(Role), keyword=keyword, status=status)
-    stmt = stmt.order_by(Role.sort_order, Role.id).offset(offset).limit(limit)
+    stmt = _apply_filters(select(Role), query)
+    stmt = stmt.order_by(Role.sort_order, Role.id).offset(query.offset).limit(query.page_size)
     return list(session.execute(stmt).scalars())
 
 
-def count_roles(session: Session, *, keyword: str | None = None, status: int | None = None) -> int:
+def count_roles(session: Session, query: RoleQuery) -> int:
     """统计符合条件的角色总数，供分页使用。"""
-    stmt = _apply_filters(select(func.count(Role.id)), keyword=keyword, status=status)
+    stmt = _apply_filters(select(func.count(Role.id)), query)
     return int(session.execute(stmt).scalar_one())
 
 
-def _apply_filters(stmt: Select, *, keyword: str | None, status: int | None) -> Select:
+def _apply_filters(stmt: Select, query: RoleQuery) -> Select:
     """应用通用筛选条件。列表与计数共用，避免两处口径不一致。
 
     关键词匹配走 `contains(autoescape=True)`（转义 LIKE 通配符）——
     角色编码普遍含下划线，直接拼接会让 `_` 当单字符通配执行。
     """
-    if keyword:
+    if query.keyword:
         stmt = stmt.where(
             or_(
-                Role.role_code.contains(keyword, autoescape=True),
-                Role.role_name.contains(keyword, autoescape=True),
+                Role.role_code.contains(query.keyword, autoescape=True),
+                Role.role_name.contains(query.keyword, autoescape=True),
             )
         )
-    if status is not None:
-        stmt = stmt.where(Role.status == status)
+    if query.status is not None:
+        stmt = stmt.where(Role.status == query.status)
     return stmt
 
 

backend/app/repositories/settlement_repository.py

@@ -7,7 +7,16 @@ from app.models import MonthlySettlement, SettlementItem
 from app.schemas.settlement import SettlementQuery
 
 
-def get_by_id(session: Session, settlement_id: int) -> MonthlySettlement | None:
+def get_by_id(
+    session: Session, settlement_id: int, *, for_update: bool = False
+) -> MonthlySettlement | None:
+    """月结行；``for_update`` 供提交审批路径加行锁用。
+
+    提交要算审批单版本号（同一月结的第几次送审），并发提交会让两次
+    算出同一个版本号；锁住月结行使并发提交串行化。
+    """
+    if for_update:
+        return session.get(MonthlySettlement, settlement_id, with_for_update=True)
     return session.get(MonthlySettlement, settlement_id)
 
 

backend/app/repositories/shipment_repository.py

@@ -8,13 +8,13 @@
   业务级汇总仍收口 business_service._recalc_totals
 """
 
-from datetime import date
 from decimal import Decimal
 
 from sqlalchemy import Select, delete, func, select
 from sqlalchemy.orm import Session
 
 from app.models import Business, Receipt, Shipment, ShipmentItem
+from app.schemas.shipment import ShipmentQuery
 
 
 def get_by_id(session: Session, shipment_id: int) -> Shipment | None:
@@ -31,59 +31,37 @@ def get_with_business_no(session: Session, shipment_id: int) -> tuple[Shipment,
     ).first()
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    business_id: int | None,
-    ship_date_from: date | None,
-    ship_date_to: date | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: ShipmentQuery) -> Select:
     """发货列表的筛选条件（business_id 空=全量，一期按业务维度查看为主）。"""
-    if business_id is not None:
-        stmt = stmt.where(Shipment.business_id == business_id)
-    if ship_date_from is not None:
-        stmt = stmt.where(Shipment.ship_date >= ship_date_from)
-    if ship_date_to is not None:
-        stmt = stmt.where(Shipment.ship_date <= ship_date_to)
+    if query.business_id is not None:
+        stmt = stmt.where(Shipment.business_id == query.business_id)
+    if query.ship_date_from is not None:
+        stmt = stmt.where(Shipment.ship_date >= query.ship_date_from)
+    if query.ship_date_to is not None:
+        stmt = stmt.where(Shipment.ship_date <= query.ship_date_to)
     return stmt
 
 
-def list_shipments(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    business_id: int | None,
-    ship_date_from: date | None,
-    ship_date_to: date | None,
-) -> list[tuple[Shipment, str]]:
+def list_shipments(session: Session, query: ShipmentQuery) -> list[tuple[Shipment, str]]:
     """分页查询发货单（附业务号），按 id 倒序。"""
     stmt = _apply_filters(
         select(Shipment, Business.business_no).join(Business, Business.id == Shipment.business_id),
-        business_id=business_id,
-        ship_date_from=ship_date_from,
-        ship_date_to=ship_date_to,
+        query,
     )
     return list(
-        session.execute(stmt.order_by(Shipment.id.desc()).offset(offset).limit(limit)).all()
+        session.execute(
+            stmt.order_by(Shipment.id.desc()).offset(query.offset).limit(query.page_size)
+        ).all()
     )
 
 
-def count_shipments(
-    session: Session,
-    *,
-    business_id: int | None,
-    ship_date_from: date | None,
-    ship_date_to: date | None,
-) -> int:
+def count_shipments(session: Session, query: ShipmentQuery) -> int:
     """发货单总数（筛选条件与列表共用同一 apply_filters，防两侧口径漂移）。"""
     stmt = _apply_filters(
         select(func.count())
         .select_from(Shipment)
         .join(Business, Business.id == Shipment.business_id),
-        business_id=business_id,
-        ship_date_from=ship_date_from,
-        ship_date_to=ship_date_to,
+        query,
     )
     return int(session.execute(stmt).scalar_one())
 

backend/app/repositories/tplus_repository.py

@@ -0,0 +1,149 @@
+"""T+ 管理库访问；只 flush，事务由 service 管理。"""
+
+from sqlalchemy import func, select, update
+from sqlalchemy.orm import Session, defer
+
+from app.models import TPlusSyncBatch, TPlusSyncError
+
+
+def get_batch(
+    session: Session, batch_no: str, *, for_update: bool = False, with_result: bool = False
+):
+    """按批次号查询；普通查询不加载核验大字段。"""
+    statement = select(TPlusSyncBatch).where(TPlusSyncBatch.batch_no == batch_no)
+    if not with_result:
+        statement = statement.options(defer(TPlusSyncBatch.audit_result_json))
+    if for_update:
+        statement = statement.with_for_update().execution_options(populate_existing=True)
+    return session.scalar(statement)
+
+
+def get_request(session: Session, key: str, *, for_update: bool = False):
+    """锁读用于唯一约束竞争后的当前读，避免旧事务快照看不到新任务。"""
+    statement = select(TPlusSyncBatch).where(TPlusSyncBatch.request_key == key)
+    if for_update:
+        statement = statement.with_for_update().execution_options(populate_existing=True)
+    return session.scalar(statement)
+
+
+def active_retry(session: Session, parent_no: str):
+    """调用方先锁父批次，再用当前读判断是否已有未完成重试。"""
+    return session.scalar(
+        select(TPlusSyncBatch)
+        .where(
+            TPlusSyncBatch.parent_batch_no == parent_no,
+            TPlusSyncBatch.status.in_(("QUEUED", "RUNNING")),
+        )
+        .with_for_update()
+    )
+
+
+def increment_retry_errors(session: Session, parent_no: str) -> None:
+    """创建新重试时对原异常累计重试次数。"""
+    session.execute(
+        update(TPlusSyncError)
+        .where(TPlusSyncError.batch_no == parent_no)
+        .values(retry_count=TPlusSyncError.retry_count + 1)
+    )
+    session.flush()
+
+
+def next_queued(session: Session):
+    """调用方持有账套执行锁后领取最早排队任务。"""
+    return session.scalar(
+        select(TPlusSyncBatch)
+        .where(TPlusSyncBatch.status == "QUEUED")
+        .order_by(TPlusSyncBatch.id)
+        .limit(1)
+        .with_for_update()
+    )
+
+
+def running_batches(session: Session) -> list[TPlusSyncBatch]:
+    """只在持有执行锁时使用，不能依据时间戳判断任务已经停止。"""
+    return list(
+        session.scalars(
+            select(TPlusSyncBatch).where(TPlusSyncBatch.status == "RUNNING").with_for_update()
+        ).all()
+    )
+
+
+def list_batches(session: Session, query):
+    """分页只读元数据，核验报告和原异常不随列表加载。"""
+    filters = []
+    if query.status:
+        filters.append(TPlusSyncBatch.status == query.status)
+    if query.task_type:
+        filters.append(
+            TPlusSyncBatch.sync_mode == ("AUDIT" if query.task_type == "AUDIT" else "FULL")
+        )
+    total = session.scalar(select(func.count()).select_from(TPlusSyncBatch).where(*filters))
+    rows = session.scalars(
+        select(TPlusSyncBatch)
+        .where(*filters)
+        .options(defer(TPlusSyncBatch.audit_result_json), defer(TPlusSyncBatch.error_message))
+        .order_by(TPlusSyncBatch.id.desc())
+        .offset(query.offset)
+        .limit(query.page_size)
+    ).all()
+    return rows, total or 0
+
+
+def active_retry_parents(session: Session, numbers: list[str]) -> set[str]:
+    """整页一次获取未完成的重试，避免逐行查询。"""
+    if not numbers:
+        return set()
+    return set(
+        session.scalars(
+            select(TPlusSyncBatch.parent_batch_no).where(
+                TPlusSyncBatch.parent_batch_no.in_(numbers),
+                TPlusSyncBatch.status.in_(("QUEUED", "RUNNING")),
+            )
+        ).all()
+    )
+
+
+def list_errors(session: Session, number: str, query):
+    """只读取当前批次的异常定位元数据。"""
+    filters = [TPlusSyncError.batch_no == number]
+    if query.entity_type:
+        filters.append(TPlusSyncError.entity_type == query.entity_type)
+    total = session.scalar(select(func.count()).select_from(TPlusSyncError).where(*filters))
+    rows = session.scalars(
+        select(TPlusSyncError)
+        .where(*filters)
+        .options(defer(TPlusSyncError.error_message))
+        .order_by(TPlusSyncError.id.desc())
+        .offset(query.offset)
+        .limit(query.page_size)
+    ).all()
+    return rows, total or 0
+
+
+def status_counts(session: Session) -> dict[str, int]:
+    """汇总任务表，不扫描原始快照。"""
+    return dict(
+        session.execute(
+            select(TPlusSyncBatch.status, func.count()).group_by(TPlusSyncBatch.status)
+        ).all()
+    )
+
+
+def latest_audit(session: Session):
+    """最近完成的核验；返回批次供查询层过滤报告。"""
+    return session.scalar(
+        select(TPlusSyncBatch)
+        .where(
+            TPlusSyncBatch.sync_mode == "AUDIT",
+            TPlusSyncBatch.finished_at.is_not(None),
+        )
+        .order_by(TPlusSyncBatch.id.desc())
+        .limit(1)
+    )
+
+
+def total_staged_records(session: Session) -> int:
+    """从批次元数据汇总累计新增数，避免扫描原始快照。"""
+    return (
+        session.scalar(select(func.coalesce(func.sum(TPlusSyncBatch.staged_record_count), 0))) or 0
+    )

backend/app/repositories/user_repository.py

@@ -10,6 +10,7 @@ from sqlalchemy.orm import Session, selectinload
 
 from app.models import Role, User, UserRole
 from app.repositories.query_utils import fuzzy
+from app.schemas.user import UserQuery
 
 
 def _base_query() -> Select[tuple[User]]:
@@ -48,16 +49,7 @@ def touch_last_login(session: Session, user: User, moment: datetime) -> None:
 # ==================== 列表查询 ====================
 
 
-def list_users(
-    session: Session,
-    *,
-    offset: int,
-    limit: int,
-    username: str | None = None,
-    real_name: str | None = None,
-    status: int | None = None,
-    role_code: str | None = None,
-) -> list[User]:
+def list_users(session: Session, query: UserQuery) -> list[User]:
     """分页查询用户列表。
 
     **按 id 倒序而非 create_time**：create_time 是秒级精度，批量导入的用户
@@ -65,48 +57,35 @@ def list_users(
     id 是自增主键，天然唯一且与创建顺序一致。
     """
     stmt = _base_query()
-    stmt = _apply_filters(stmt, username=username, real_name=real_name, status=status)
-    if role_code:
+    stmt = _apply_filters(stmt, query)
+    if query.role_code:
         # 按角色筛选需 join 关联表；角色是字典数据，数量极少，join 代价可忽略
-        stmt = stmt.join(UserRole).join(Role).where(Role.role_code == role_code)
-    stmt = stmt.order_by(User.id.desc()).offset(offset).limit(limit)
+        stmt = stmt.join(UserRole).join(Role).where(Role.role_code == query.role_code)
+    stmt = stmt.order_by(User.id.desc()).offset(query.offset).limit(query.page_size)
     return list(session.execute(stmt).scalars().unique())
 
 
-def count_users(
-    session: Session,
-    *,
-    username: str | None = None,
-    real_name: str | None = None,
-    status: int | None = None,
-    role_code: str | None = None,
-) -> int:
+def count_users(session: Session, query: UserQuery) -> int:
     """统计符合条件的用户总数，供分页使用。"""
     stmt = select(func.count(User.id))
-    stmt = _apply_filters(stmt, username=username, real_name=real_name, status=status)
-    if role_code:
-        stmt = stmt.join(UserRole).join(Role).where(Role.role_code == role_code)
+    stmt = _apply_filters(stmt, query)
+    if query.role_code:
+        stmt = stmt.join(UserRole).join(Role).where(Role.role_code == query.role_code)
     return int(session.execute(stmt).scalar_one())
 
 
-def _apply_filters(
-    stmt: Select,
-    *,
-    username: str | None,
-    real_name: str | None,
-    status: int | None,
-) -> Select:
+def _apply_filters(stmt: Select, query: UserQuery) -> Select:
     """应用通用筛选条件。列表与计数共用，避免两处口径不一致。
 
     模糊匹配统一走 `query_utils.fuzzy`（转义 LIKE 通配符）—— 用户名普遍
     含下划线，直接拼接会让 `_` 当单字符通配执行、误配无关账号。
     """
-    if username:
-        stmt = fuzzy(stmt, User.username, username)
-    if real_name:
-        stmt = fuzzy(stmt, User.real_name, real_name)
-    if status is not None:
-        stmt = stmt.where(User.status == status)
+    if query.username:
+        stmt = fuzzy(stmt, User.username, query.username)
+    if query.real_name:
+        stmt = fuzzy(stmt, User.real_name, query.real_name)
+    if query.status is not None:
+        stmt = stmt.where(User.status == query.status)
     return stmt
 
 

backend/app/schemas/approval.py

@@ -1,6 +1,7 @@
 """审批管理的请求/响应模型（2026-10-15 复核审批补充需求）。"""
 
 from datetime import datetime
+from typing import Any, Literal
 
 from pydantic import BaseModel, ConfigDict, Field, model_validator
 
@@ -31,6 +32,7 @@ class ApprovalOrderOut(BaseModel):
     order_type: int
     ref_id: int
     refer_no: str
+    version_no: int
     status: int
     submitter_id: int | None = None
     submitter_name: str
@@ -38,9 +40,21 @@ class ApprovalOrderOut(BaseModel):
 
 
 class ApprovalOrderDetailOut(ApprovalOrderOut):
-    """审批单详情：附节点时间线。"""
+    """审批单详情：附节点时间线与审批内容。
+
+    ``content_source`` 语义：``snapshot``=提交时快照（可信）；
+    ``live``=历史单无快照、回退读当前业务内容（仅供参考）；
+    ``missing``=引用对象已不存在且无快照。
+    """
 
     steps: list[ApprovalStepOut] = Field(default_factory=list)
+    snapshot: dict[str, Any] | None = Field(
+        default=None,
+        description="提交时内容快照（信封：schema_version/order_type/captured_at/data）",
+    )
+    content_source: Literal["snapshot", "live", "missing"] = Field(
+        default="missing", description="审批内容来源：提交时快照/当前内容回退/缺失"
+    )
 
 
 class ApprovalQuery(PageQuery):
@@ -48,7 +62,7 @@ class ApprovalQuery(PageQuery):
 
     keyword: str | None = Field(default=None, max_length=32, description="审批单号/引用单号，模糊")
     status: int | None = Field(default=None, ge=1, le=5, description="状态筛选")
-    order_type: int | None = Field(default=None, ge=1, le=2, description="对象类型筛选")
+    order_type: int | None = Field(default=None, ge=1, le=3, description="对象类型筛选")
 
 
 class ApprovalActionIn(BaseModel):

backend/app/schemas/business.py

@@ -180,6 +180,14 @@ class SecondaryAllocationOut(BaseModel):
     ratio: str
 
 
+class AllocationVersionOut(SecondaryAllocationOut):
+    """方案版本行（含历史；业务详情抽屉回溯用）。"""
+
+    version_no: int
+    is_active: int = Field(description="1当前生效 0历史版本")
+    create_time: datetime
+
+
 def _ensure_allocations_unique(items: list["SecondaryAllocationIn"]) -> None:
     """同一业务内参与人去重：uk(business_id, personnel_id) 的前置友好提示
     （让 400 文案说「重复参与人」而不是裸唯一键冲突）。"""
@@ -191,13 +199,16 @@ def _ensure_allocations_unique(items: list["SecondaryAllocationIn"]) -> None:
 
 
 class BusinessDetailOut(BusinessOut):
-    """业务详情：附产品明细行与二次分配方案。"""
+    """业务详情：附产品明细行、当前方案与方案版本历史。"""
 
     items: list[BusinessItemOut] = Field(default_factory=list)
     secondary_allocations: list[SecondaryAllocationOut] = Field(default_factory=list)
     allocation_hint: str | None = Field(
         default=None, description="方案合计与余额提示（余额归属待澄清#2，仅展示）"
     )
+    allocation_history: list[AllocationVersionOut] = Field(
+        default_factory=list, description="方案全部版本（含历史；被驳回方案亦保留）"
+    )
 
 
 class BusinessCreateIn(BaseModel):

backend/app/schemas/business_role.py

@@ -27,6 +27,19 @@ class BusinessRoleOut(BaseModel):
     is_builtin: int = 0
 
 
+class BusinessRoleOptionOut(BaseModel):
+    """岗位下拉选项（人员档案与报表筛选共用）。
+
+    与 ``BusinessRoleOut`` 分开：下拉只需渲染用的最小字段集，不必顺带
+    暴露计提模板等管理字段。字段名直出蛇形（前端按同名取值）。
+    """
+
+    id: int = Field(description="岗位ID")
+    role_code: str = Field(description="岗位编码")
+    role_name: str = Field(description="岗位名称")
+    status: int = Field(description="状态：1启用 0禁用")
+
+
 class TemplateParamOut(BaseModel):
     """模板声明的参数（供岗位表单展示，含当前口径说明）。"""
 

backend/app/schemas/customer.py

@@ -1,6 +1,7 @@
 """客户管理的请求/响应模型（需求 2.1.3）。"""
 
 from datetime import datetime
+from typing import Literal
 
 from pydantic import BaseModel, ConfigDict, Field
 
@@ -28,6 +29,8 @@ class CustomerOut(BaseModel):
     owner_personnel_id: int | None = None
     owner_personnel_name: str | None = None
     remark: str | None = None
+    rating_score: int | None = None
+    rating_time: datetime | None = None
 
 
 class CustomerCreateIn(BaseModel):
@@ -90,13 +93,32 @@ class CustomerUpdateIn(BaseModel):
 
 
 class CustomerQuery(PageQuery):
-    """客户列表查询条件（字段名蛇形直出，与前端 api 封装同名对应）。"""
+    """客户列表查询条件（字段名蛇形直出，与前端 api 封装同名对应）。
+
+    排序用 ``Literal`` 白名单而不是自由字符串：``order_by`` 直接拼进
+    ``ORDER BY``，自由值等于把注入面开到排序键上；客户列表只有两个
+    有意义的排序键，白名单足够。
+    """
 
     keyword: str | None = Field(
         default=None, max_length=64, description="公司名称或客户负责人，模糊匹配"
     )
     customer_type: str | None = Field(default=None, max_length=32, description="客户分类筛选")
     owner_personnel_id: int | None = Field(default=None, ge=1, description="归属人员ID筛选")
+    order_by: Literal["id", "rating_score"] = Field(
+        default="id", description="排序键：id 建档序 / rating_score 评级总分"
+    )
+    order: Literal["asc", "desc"] = Field(default="desc", description="排序方向")
+
+
+class CustomerRatingIn(BaseModel):
+    """保存客户评级入参（评级计算器「保存评分」）。
+
+    score 是五项之和的整数（0-100）；五项明细不落库（评级记录只存总分，
+    见对照表 #25 的定位：评分是档案属性而非审计事件）。
+    """
+
+    score: int = Field(ge=0, le=100, description="评级总分（0-100 整数）")
 
 
 class CustomerOwnershipIn(BaseModel):

backend/app/schemas/dashboard.py

@@ -1,9 +1,13 @@
 """工作台的请求/响应模型（M4 ⑧）。
 
-四区块一次返回（前端一次请求渲染整页，避免 6-10 次往返）：
-经营绩效总览 / 待办事项 / 风险预警 / 最近动态。金额一律 Decimal
+聚合端点一次返回三区块（前端一次请求渲染整页，避免多次往返）：
+经营绩效总览 / 待办事项 / 风险预警。金额一律 Decimal
 （序列化为字符串），前端不做数值运算。
 
+**最近动态独立成分页端点**：它可能有很长（操作日志全量），塞进聚合
+响应既撑爆页面又只能截断前 20 条；拆开后翻页只重取这张列表，不必
+重算总览/待办/预警。列表与模块筛选项同源返回（一次请求给全）。
+
 待办与预警为**查询时实时派生**（不落库、不引调度器，见计划决策）；
 ⑥ T+ 同步未建前，同步相关条目用 `sync_available=false` 显式降级，
 不返回伪造的 0 条。
@@ -12,7 +16,9 @@
 from datetime import date, datetime
 from decimal import Decimal
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, ConfigDict, Field
+
+from app.core.response import PageQuery
 
 
 class TrendPoint(BaseModel):
@@ -116,11 +122,42 @@ class ActivityItem(BaseModel):
     create_time: datetime
 
 
+class ActivityQuery(PageQuery):
+    """最近动态查询条件：分页 + 模块筛选。
+
+    只支持模块筛选（工作台只需要这一个维度）；其余筛选项见日志查询页。
+    """
+
+    module: str | None = Field(default=None, max_length=64, description="模块，精确匹配")
+
+
+class ActivityPageOut(BaseModel):
+    """最近动态分页结果：当前页 + 总数 + 模块筛选项。
+
+    分页字段名沿用全站规范（list/total/pageNum/pageSize），与
+    `PageResult` 同形；额外带 `modules`（全量模块名），
+    供前端模块下拉使用 —— 不随当前页变化，避免翻页后选项增减。
+    """
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    records: list[ActivityItem] = Field(
+        default_factory=list, alias="list", serialization_alias="list", description="当前页数据"
+    )
+    total: int = Field(default=0, description="总记录数")
+    page_num: int = Field(
+        default=1, alias="pageNum", serialization_alias="pageNum", description="当前页码"
+    )
+    page_size: int = Field(
+        default=10, alias="pageSize", serialization_alias="pageSize", description="每页条数"
+    )
+    modules: list[str] = Field(default_factory=list, description="全部模块名（筛选项）")
+
+
 class DashboardOut(BaseModel):
-    """工作台四区块一次性响应。"""
+    """工作台三区块一次性响应（最近动态走独立分页端点）。"""
 
     month: str
     overview: OverviewBlock
     todos: TodoBlock
     warnings: WarningBlock
-    activities: list[ActivityItem] = Field(default_factory=list)

backend/app/schemas/report.py

@@ -264,7 +264,9 @@ class DepartmentReconciliation(BaseModel):
         description="聚合额 − 团队分成合计 − 全公司分成合计（分列舍入差）",
     )
     adjust_net_total: Decimal = Field(description="明细层调整净额合计（成员 + 未分配）")
-    payable_total: Decimal = Field(description="应发合计 = 成员应付 + 未分配应付 + 聚合行 + 调整净额")
+    payable_total: Decimal = Field(
+        description="应发合计 = 成员应付 + 未分配应付 + 聚合行 + 调整净额"
+    )
     team_base_total: Decimal = Field(description="展示团队基数合计")
     unassigned_performance_total: Decimal = Field(description="展示未分配人员绩效合计")
     other_team_base: Decimal = Field(

backend/app/schemas/tplus.py

@@ -0,0 +1,109 @@
+"""T+ 管理接口合同；只暴露范围、数量和受控核验结果。"""
+
+from datetime import datetime
+from typing import Literal
+from uuid import UUID
+
+from pydantic import BaseModel, Field, model_validator
+
+from app.core.response import PageQuery
+
+EntityType = Literal[
+    "personnel",
+    "products",
+    "customers",
+    "businesses",
+    "shipments",
+    "receipts",
+    "business_items",
+    "shipment_items",
+    "receipt_items",
+]
+TaskType = Literal["READ", "AUDIT"]
+TaskStatus = Literal["QUEUED", "RUNNING", "SUCCEEDED", "PARTIAL", "FAILED"]
+
+
+class TPlusCreateIn(BaseModel):
+    """创建长任务；幂等键是调用意图的一部分，必须由调用方保留。"""
+
+    task_type: TaskType = "READ"
+    entity_types: list[EntityType] = Field(default_factory=list, max_length=9)
+    request_key: UUID
+
+    @model_validator(mode="after")
+    def check_scope(self):
+        if self.task_type == "READ" and not self.entity_types:
+            raise ValueError("请选择至少一种读取数据")
+        if self.task_type == "AUDIT" and self.entity_types:
+            raise ValueError("关联核验不接受自定义读取范围")
+        if len(set(self.entity_types)) != len(self.entity_types):
+            raise ValueError("读取类型不能重复")
+        return self
+
+
+class TPlusRetryIn(BaseModel):
+    """失败批次重试不接受任意新范围。"""
+
+    request_key: UUID
+
+
+class TPlusBatchQuery(PageQuery):
+    """批次筛选，与读取状态和任务类型保持一致。"""
+
+    status: TaskStatus | None = None
+    task_type: TaskType | None = None
+
+
+class TPlusErrorQuery(PageQuery):
+    """异常只能在指定批次内查询。"""
+
+    entity_type: EntityType | None = None
+
+
+class TPlusBatchOut(BaseModel):
+    """批次列表不包含原始数据、源配置或数据库异常。"""
+
+    batch_no: str
+    task_type: TaskType
+    status: str
+    entity_types: list[str] | None
+    source_record_count: int
+    staged_record_count: int
+    skipped_record_count: int
+    failed_record_count: int
+    parent_batch_no: str | None
+    create_time: datetime
+    started_at: datetime | None
+    finished_at: datetime | None
+    can_retry: bool
+    error_summary: str | None = None
+
+
+class TPlusBatchDetailOut(TPlusBatchOut):
+    """详情按需携带受控的关联核验结果。"""
+
+    audit_result: dict | None = None
+
+
+class TPlusErrorOut(BaseModel):
+    """异常使用固定说明，源定位仅开放整数 ID。"""
+
+    id: int
+    entity_type: str
+    source_id: str | None
+    error_code: str
+    error_message: str
+    retry_count: int
+    create_time: datetime
+
+
+class TPlusOverviewOut(BaseModel):
+    """配置就绪不等同于源库连通或兼容验收通过。"""
+
+    is_configured: bool
+    configuration_message: str
+    status_counts: dict[str, int]
+    total_staged_records: int = 0
+    compatibility_status: str = "PENDING"
+    business_import_enabled: bool = False
+    latest_audit: dict | None = None

backend/app/services/approval_service.py

@@ -20,10 +20,11 @@
 
 import logging
 from collections.abc import Iterable
-from datetime import date
+from datetime import date, datetime
+from decimal import Decimal
 
 from sqlalchemy import func, select
-from sqlalchemy.orm import Session
+from sqlalchemy.orm import Session, defer
 
 from app.core.constants import (
     ACTION_APPROVE,
@@ -36,7 +37,7 @@ from app.core.constants import (
 )
 from app.core.exceptions import BadRequestError, ForbiddenError, NotFoundError
 from app.core.response import PageResult
-from app.models import ApprovalOrder, ApprovalStep, User
+from app.models import ApprovalOrder, ApprovalStep, Business, User
 from app.models.approval import (
     APPROVAL_ACTION_APPROVE,
     APPROVAL_ACTION_AUTO_APPROVE,
@@ -113,6 +114,7 @@ def submit_order(
     ref_id: int,
     refer_no: str,
     submitter: User,
+    snapshot: dict | None = None,
 ) -> ApprovalOrder:
     """提单（业务写路径内部调用；不 commit、不嵌套 audit）。
 
@@ -122,6 +124,10 @@ def submit_order(
     一并重做），update 路径由审计上下文回滚并经全局 handler 收敛 400——
     **任何情况下都不允许本函数内部 rollback 后部分重试**，那会把同事务
     的业务主行/明细/方案回滚掉而审批单单表重插，静默丢业务数据。
+
+    ``snapshot`` 是「提交时」的业务内容（由调用方组装，Decimal 已转 str）；
+    落库时包一层信封并固化采集时间——审批期间业务再改，复核者看到的仍是
+    送审内容。开关关闭的自动通过路径同样写快照，留痕不走样。
     """
     _withdraw_pending(session, order_type=order_type, ref_id=ref_id, operator=submitter)
 
@@ -135,6 +141,17 @@ def submit_order(
         order_type=order_type,
         ref_id=ref_id,
         refer_no=refer_no,
+        version_no=_next_version(session, order_type, ref_id),
+        snapshot_json=(
+            {
+                "schema_version": 1,
+                "order_type": order_type,
+                "captured_at": datetime.now().isoformat(timespec="seconds"),
+                "data": snapshot,
+            }
+            if snapshot is not None
+            else None
+        ),
         submitter_id=submitter.id,
         submitter_name=submitter.username,
     )
@@ -165,6 +182,35 @@ def submit_order(
     return order
 
 
+def pending_matches_snapshot(
+    session: Session, *, order_type: int, ref_id: int, snapshot: dict | None
+) -> bool:
+    """在途审批单的送审内容是否与给定快照一致（用于「内容未变不重提」）。
+
+    只比较快照信封里的 ``data`` 部分——``captured_at`` 每次采集都不同，
+    参与比较会让判断恒为不一致。V11 之前的历史单没有快照，一律视为不一致
+    （保守重提：宁可多提一次，不能漏审）。
+    """
+    if snapshot is None:
+        return False
+    pending = (
+        session.execute(
+            select(ApprovalOrder).where(
+                ApprovalOrder.order_type == order_type,
+                ApprovalOrder.ref_id == ref_id,
+                ApprovalOrder.status.in_(_PENDING_STATUSES),
+            )
+        )
+        .scalars()
+        .all()
+    )
+    for order in pending:
+        stored = order.snapshot_json or {}
+        if stored.get("data") == snapshot:
+            return True
+    return False
+
+
 def withdraw_pending_orders(
     session: Session, *, order_type: int, ref_id: int, operator: User
 ) -> int:
@@ -250,15 +296,15 @@ def act(
                 operator_name=operator.username,
             )
         )
-        # 驳回=方案未获批准：清空该业务的二次分配行，防止被否决的方案以
-        # 「当前生效行」身份留在表里、被 M3 核算快照算进工资；方案事实由
-        # 审批单留痕与操作日志 request_data 承载（分摊语义落定后如需保留
-        # 历史版本，在 ④ 设计时演进）。审批单的清理由调用方（act 内联）
-        # 委托 business_repository 完成 —— service 层互调、不触 commit。
+        # 驳回=方案未获批准：把该业务的生效方案行置历史（is_active=0，
+        # **不删行**——V11 起 append-only 版本化），防止被否决的方案以
+        # 「当前生效行」身份被 M3 核算快照算进工资；历史方案仍可在业务
+        # 详情 allocation_history 回溯。委托 business_repository 完成
+        # —— service 层互调、不触 commit。
         if rejected and order.order_type == 1:
             from app.repositories import business_repository
 
-            business_repository.delete_allocations(session, order.ref_id)
+            business_repository.deactivate_allocations(session, order.ref_id)
 
         # 审批结果联动（服务层回调，不 commit）：月结 通过=锁定/驳回=回草稿；
         # 调整单 通过=生效/驳回=作废。函数内延迟导入防循环依赖。
@@ -299,11 +345,18 @@ def act(
 
 
 def list_orders(session: Session, query: ApprovalQuery) -> PageResult[ApprovalOrderOut]:
-    """分页查询审批单（准入由 api 层依赖判定）。"""
+    """分页查询审批单（准入由 api 层依赖判定）。
+
+    排除 snapshot_json 列（月结快照含按人明细，可达数十~上百 KB）：列表出参
+    本就不含快照，整列加载会让翻页负担随快照体积线性放大；详情走
+    `get_order` 按需读取（与操作日志列表排除 request_data 同一理由，
+    见 operation_log_repository.list_logs）。
+    """
     filters = _order_filters(query)
     rows = (
         session.execute(
             select(ApprovalOrder)
+            .options(defer(ApprovalOrder.snapshot_json))
             .where(*filters)
             .order_by(ApprovalOrder.id.desc())
             .offset(query.offset)
@@ -360,8 +413,152 @@ def _withdraw_pending(session: Session, *, order_type: int, ref_id: int, operato
         logger.info("撤回在途审批单 %s 张 | ref=%s", len(pending), ref_id)
 
 
+def _next_version(session: Session, order_type: int, ref_id: int) -> int:
+    """同一对象类型+引用对象的下一个提交版本号（从 1 递增）。
+
+    **必须用锁读（当前读），不能用普通 MAX 查询**：普通 SELECT 在
+    REPEATABLE READ 下是快照读，读到的是本事务首次一致读的快照；请求级
+    鉴权（deps.get_current_user 与业务共用同一 DbSession）那次 SELECT 已把
+    快照锚定在请求开始时刻，此后即使父行锁等到并发事务提交，MAX 仍停在
+    旧快照上——两个请求会算出同一个 version_no。锁读是当前读，永远读最新
+    已提交数据。
+
+    **不能写成 select(func.max(...)).with_for_update()**：MySQL 官方明确
+    「Locking clauses are not permitted in queries with aggregate functions
+    or GROUP BY」，聚合结果上的 FOR UPDATE 语义无保证。改锁「该 ref 的现有
+    版本行」，最大值在 Python 侧算——同 no_generator 的既有套路。只取
+    version_no 单列，不把 snapshot_json 拉进内存。
+    """
+    versions = (
+        session.execute(
+            select(ApprovalOrder.version_no)
+            .where(
+                ApprovalOrder.order_type == order_type,
+                ApprovalOrder.ref_id == ref_id,
+            )
+            .with_for_update()
+        )
+        .scalars()
+        .all()
+    )
+    return (max(versions) if versions else 0) + 1
+
+
+def allocation_content(session: Session, business: Business) -> dict:
+    """二次分配方案的审批内容（提单快照与历史单回退共用同一结构）。
+
+    业务主行已存在但客户联查落空（客户被删等异常）时返回空字典——调用方
+    以 ``or None`` 归一，避免生成「有快照但内容为空」的信封。
+    """
+    from app.repositories import business_repository
+
+    row = business_repository.get_with_names(session, business.id)
+    if row is None:
+        return {}
+    _, customer_name, owner_name, assistant_name = row
+    allocations = business_repository.list_allocations(session, business.id)
+    total = sum((a.ratio for a in allocations), Decimal("0"))
+    return {
+        "business_no": business.business_no,
+        "business_date": business.business_date.isoformat() if business.business_date else None,
+        "customer_name": customer_name,
+        "owner_personnel_name": owner_name,
+        "assistant_personnel_name": assistant_name,
+        "sale_amount": _amount_text(business.total_sale_amount),
+        "cost_amount": _amount_text(business.total_cost_amount),
+        "allocations": [
+            {
+                "personnel_id": a.personnel_id,
+                "personnel_name": a.personnel_name,
+                "ratio": str(a.ratio),
+            }
+            for a in allocations
+        ],
+        "allocation_total": str(total.quantize(Decimal("0.01"))),
+        "allocation_balance": str((Decimal("100") - total).quantize(Decimal("0.01"))),
+    }
+
+
+def _amount_text(value: Decimal | None) -> str | None:
+    """金额文本化（Decimal 全程不转 float；None 保留为空，不伪造 0.00）。"""
+    return str(value) if value is not None else None
+
+
+def settlement_content(session: Session, settlement_id: int) -> dict | None:
+    """月结审批内容（汇总 + 按人明细；提单快照与历史单回退共用）。"""
+    from app.models import MonthlySettlement
+    from app.repositories import settlement_repository
+
+    settlement = session.get(MonthlySettlement, settlement_id)
+    if settlement is None:
+        return None
+    items = settlement_repository.list_items(session, settlement_id)
+    return {
+        "settlement_month": settlement.settlement_month,
+        "total_performance": _amount_text(settlement.total_performance),
+        "total_payable": _amount_text(settlement.total_payable),
+        "person_count": settlement.person_count,
+        "items": [
+            {
+                "personnel_name": item.personnel_name,
+                "role_code": item.role_code,
+                "calc_type": item.calc_type,
+                "performance_sum": str(item.performance_sum),
+                "payable_sum": str(item.payable_sum),
+                "basic_amount": _amount_text(item.basic_amount),
+                "salary_mode": item.salary_mode,
+                "net_amount": _amount_text(item.net_amount),
+                "is_probation": item.is_probation,
+            }
+            for item in items
+        ],
+    }
+
+
+def adjustment_content(session: Session, adjustment_id: int) -> dict | None:
+    """调整单审批内容（提单快照与历史单回退共用）。"""
+    from app.models import AdjustmentOrder
+
+    adjustment = session.get(AdjustmentOrder, adjustment_id)
+    if adjustment is None:
+        return None
+    return {
+        "order_no": adjustment.order_no,
+        "settlement_month": adjustment.settlement_month,
+        "personnel_name": adjustment.personnel_name,
+        "direction": adjustment.direction,
+        "amount": str(adjustment.amount),
+        "reason": adjustment.reason,
+    }
+
+
+def _content_of(session: Session, order: ApprovalOrder) -> tuple[dict | None, str]:
+    """审批内容出参：优先提交时快照，历史单回退当前内容并显式标注来源。"""
+    if order.snapshot_json:
+        return order.snapshot_json, "snapshot"
+
+    if order.order_type == 1:
+        business = session.get(Business, order.ref_id)
+        data = allocation_content(session, business) if business is not None else None
+    elif order.order_type == 2:
+        data = settlement_content(session, order.ref_id)
+    elif order.order_type == 3:
+        data = adjustment_content(session, order.ref_id)
+    else:
+        data = None
+
+    if not data:
+        return None, "missing"
+    return {
+        "schema_version": 1,
+        "order_type": order.order_type,
+        "captured_at": None,
+        "data": data,
+    }, "live"
+
+
 def _detail_of(session: Session, order: ApprovalOrder) -> ApprovalOrderDetailOut:
-    """出参组装（含 steps 时间线）。调用方须保证处于 commit 之前。"""
+    """出参组装（含 steps 时间线与审批内容）。调用方须保证处于 commit 之前。"""
     steps = (
         session.execute(
             select(ApprovalStep).where(ApprovalStep.order_id == order.id).order_by(ApprovalStep.id)
@@ -371,6 +568,7 @@ def _detail_of(session: Session, order: ApprovalOrder) -> ApprovalOrderDetailOut
     )
     detail = ApprovalOrderDetailOut.model_validate(order)
     detail.steps = [ApprovalStepOut.model_validate(s) for s in steps]
+    detail.snapshot, detail.content_source = _content_of(session, order)
     return detail
 
 

backend/app/services/auth_service.py

@@ -37,12 +37,14 @@ def _record(
     """写登录日志并立即提交。单独提交是刻意为之，见模块 docstring。"""
     login_log_repository.add(
         session,
-        user_id=user_id,
-        username=username,
-        ip=ip,
-        user_agent=user_agent,
-        success=success,
-        fail_reason=fail_reason,
+        login_log_repository.LoginLogEntry(
+            user_id=user_id,
+            username=username,
+            ip=ip,
+            user_agent=user_agent,
+            success=success,
+            fail_reason=fail_reason,
+        ),
     )
     session.commit()
 

backend/app/services/business_role_service.py

@@ -31,6 +31,7 @@ from app.models import BusinessRole, User
 from app.repositories import business_role_repository
 from app.schemas.business_role import (
     BusinessRoleCreateIn,
+    BusinessRoleOptionOut,
     BusinessRoleOut,
     BusinessRoleQuery,
     BusinessRoleUpdateIn,
@@ -67,16 +68,8 @@ def _ensure_template(calc_template: str) -> None:
 
 def list_roles(session: Session, query: BusinessRoleQuery) -> PageResult[BusinessRoleOut]:
     """分页查询岗位（含停用）。"""
-    roles = business_role_repository.list_roles(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        keyword=query.keyword,
-        status=query.status,
-    )
-    total = business_role_repository.count_roles(
-        session, keyword=query.keyword, status=query.status
-    )
+    roles = business_role_repository.list_roles(session, query)
+    total = business_role_repository.count_roles(session, query)
     return PageResult[BusinessRoleOut](
         records=[_to_out(role) for role in roles],
         total=total,
@@ -85,6 +78,23 @@ def list_roles(session: Session, query: BusinessRoleQuery) -> PageResult[Busines
     )
 
 
+def list_options(session: Session) -> list[BusinessRoleOptionOut]:
+    """岗位下拉选项（含停用，不分页）。
+
+    含停用是因为人员档案可能仍引用停用岗位（停用只是禁止新增引用，
+    不改历史归属）；岗位规模有限（内置 + 自定义），一次取全量。
+    """
+    return [
+        BusinessRoleOptionOut(
+            id=role.id,
+            role_code=role.role_code,
+            role_name=role.role_name,
+            status=role.status,
+        )
+        for role in business_role_repository.list_all(session)
+    ]
+
+
 def get_role(session: Session, role_id: int) -> BusinessRoleOut:
     """岗位详情。
 

backend/app/services/business_service.py

@@ -15,6 +15,8 @@
 """
 
 import logging
+from collections.abc import Callable
+from dataclasses import dataclass
 from datetime import date
 from decimal import ROUND_HALF_UP, Decimal
 from fractions import Fraction
@@ -54,6 +56,7 @@ from app.models.business import (
 from app.models.customer import OWNER_TYPE_BUSINESS
 from app.repositories import business_repository, customer_repository, personnel_repository
 from app.schemas.business import (
+    AllocationVersionOut,
     BusinessCreateIn,
     BusinessDetailOut,
     BusinessItemIn,
@@ -233,26 +236,8 @@ def _is_duplicate_entry(exc: IntegrityError) -> bool:
 
 def list_businesses(session: Session, query: BusinessQuery) -> PageResult[BusinessOut]:
     """分页查询业务（附客户名与归属人名）。"""
-    rows = business_repository.list_businesses(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        keyword=query.keyword,
-        status=query.status,
-        customer_id=query.customer_id,
-        owner_personnel_id=query.owner_personnel_id,
-        date_from=query.date_from,
-        date_to=query.date_to,
-    )
-    total = business_repository.count_businesses(
-        session,
-        keyword=query.keyword,
-        status=query.status,
-        customer_id=query.customer_id,
-        owner_personnel_id=query.owner_personnel_id,
-        date_from=query.date_from,
-        date_to=query.date_to,
-    )
+    rows = business_repository.list_businesses(session, query)
+    total = business_repository.count_businesses(session, query)
     return PageResult[BusinessOut](
         records=[_to_out(row) for row in rows],
         total=total,
@@ -281,6 +266,17 @@ def get_detail(session: Session, business_id: int) -> BusinessDetailOut:
         for a in allocations
     ]
     out.allocation_hint = _allocation_hint(allocations)
+    out.allocation_history = [
+        AllocationVersionOut(
+            personnel_id=a.personnel_id,
+            personnel_name=a.personnel_name,
+            ratio=str(a.ratio),
+            version_no=a.version_no,
+            is_active=a.is_active,
+            create_time=a.create_time,
+        )
+        for a in business_repository.list_allocation_versions(session, business_id)
+    ]
     return out
 
 
@@ -335,6 +331,20 @@ def _build_allocation_rows(
     return rows
 
 
+def _plan_changed(old_rows: list[SecondaryAllocation], new_rows: list[SecondaryAllocation]) -> bool:
+    """方案内容（参与人 + 比例）是否真的变了。
+
+    比较口径是**集合**（人员 ID 与其比例的组合），与提交顺序无关——送审
+    快照同样按库内生效行的顺序生成，若按顺序比较，用户只是把行拖了个
+    序就会被判成「变了」，进而无谓地新增版本与重提审批单。
+    """
+
+    def _key(row: SecondaryAllocation) -> tuple[int, Decimal]:
+        return (row.personnel_id, Decimal(row.ratio).quantize(Decimal("0.01")))
+
+    return {_key(r) for r in old_rows} != {_key(r) for r in new_rows}
+
+
 def _allocation_hint(allocations: list[SecondaryAllocation]) -> str | None:
     """方案合计/余额提示（不要求合计满额是需求原文；余额归属 = 澄清 #2 答复）。
 
@@ -433,6 +443,7 @@ def create_business(
                         ref_id=business.id,
                         refer_no=business.business_no,
                         submitter=operator,
+                        snapshot=approval_service.allocation_content(session, business) or None,
                     )
                 customer = customer_repository.get_by_id(session, business.customer_id)
                 out = _to_out((business, customer.company_name if customer else None, None, None))
@@ -507,28 +518,51 @@ def update_business(
         business_repository.replace_items(
             session, business_id, _build_item_rows(business_id, payload.items)
         )
-        # 方案全量替换（空数组=清空）；终态门槛已挡「修改影响已结算月份」的
-        # 前置场景（月结在 ④，快照固化后按同一哲学隔离）
-        old_ids = {
-            a.personnel_id for a in business_repository.list_allocations(session, business_id)
-        }
-        business_repository.replace_allocations(
+        # 方案：内容真的变了才动版本行（内容未变不新增版本，避免「只改备注」
+        # 也堆出一个方案版本）；空数组=清空=把生效行置历史
+        old_allocations = business_repository.list_allocations(session, business_id)
+        new_rows = _build_allocation_rows(
             session,
             business_id,
-            _build_allocation_rows(
-                session, business_id, payload.secondary_allocations, old_ids=old_ids
-            ),
+            payload.secondary_allocations,
+            old_ids={a.personnel_id for a in old_allocations},
         )
-        # 提单/撤单与方案同步生命周期：非空=提单（内含撤旧建新）；空=仅撤回
-        # 在途单 —— 否则旧在途单挂着已不存在的方案继续走审批
+        if not payload.secondary_allocations:
+            business_repository.replace_allocations(session, business_id, [])
+        elif _plan_changed(old_allocations, new_rows):
+            business_repository.replace_allocations(session, business_id, new_rows)
+        # 汇总先重算：送审快照里的销售额/成本额必须反映本次编辑后的值，
+        # 放到提单之后会固化成上一版本的金额
+        _recalc_totals(session, business)
+        # 提单/撤单与方案同步生命周期：
+        # - 方案为空 → 仅撤回在途单（旧在途单挂着已不存在的方案继续走审批）
+        # - 送审内容与在途单一致 → 保留原单，不撤回不重建（避免反复保存
+        #   刷出「撤回+新建」的审批留痕噪音）
+        # - 其余 → 提新单（内含撤旧建新）
         if payload.secondary_allocations:
-            approval_service.submit_order(
+            snapshot = approval_service.allocation_content(session, business) or None
+            if not approval_service.pending_matches_snapshot(
                 session,
                 order_type=APPROVAL_TYPE_SECONDARY_ALLOCATION,
                 ref_id=business_id,
-                refer_no=business.business_no,
-                submitter=operator,
-            )
+                snapshot=snapshot,
+            ):
+                try:
+                    approval_service.submit_order(
+                        session,
+                        order_type=APPROVAL_TYPE_SECONDARY_ALLOCATION,
+                        ref_id=business_id,
+                        refer_no=business.business_no,
+                        submitter=operator,
+                        snapshot=snapshot,
+                    )
+                except IntegrityError as exc:
+                    # version_no 唯一键的并发冲突（取号是锁读后此项理论不可达，
+                    # 留作 DB 兜底）：审计上下文会统一回滚，这里只把 1062
+                    # 翻译成可重试的业务提示，其余完整性错误原样上抛
+                    if _is_duplicate_entry(exc):
+                        raise BadRequestError("提交冲突（版本号生成重复），请重试") from exc
+                    raise
         else:
             approval_service.withdraw_pending_orders(
                 session,
@@ -536,7 +570,6 @@ def update_business(
                 ref_id=business_id,
                 operator=operator,
             )
-        _recalc_totals(session, business)
         # 锁定行上已赋值，同一身份映射实例的 _to_out 即反映新值；名称字段
         # 从首查 row 取（换客户时 _ensure_customer 已给出新公司名）
         out = _to_out((business, customer.company_name, owner_name, assistant_name))
@@ -699,6 +732,67 @@ def change_status(
         return out
 
 
+@dataclass(frozen=True, slots=True)
+class _BusinessOwnershipResolution:
+    """业务归属变更的解析结果（新归属人行 + 变更前固化 ID）。"""
+
+    owner: Personnel | None
+    assistant: Personnel | None
+    old_owner_id: int | None
+    old_assistant_id: int | None
+
+
+def _resolve_business_ownership(
+    session: Session, business: Business, payload: BusinessOwnershipIn
+) -> _BusinessOwnershipResolution:
+    """校验新归属人（须在职）并判定归属是否真的变化。
+
+    校验在审计上下文内执行 —— 失败也留痕。两者均未变化 → 400「归属未变化」。
+    """
+    # 归属变更的出参姓名来自校验函数的返回行（锁定读）：首查姓名是变更前
+    # 固化标量，新归属人名必须按新 ID 重取
+    owner = (
+        _ensure_personnel_active(session, payload.new_owner_personnel_id, "负责")
+        if payload.new_owner_personnel_id is not None
+        else None
+    )
+    assistant = (
+        _ensure_personnel_active(session, payload.new_assistant_personnel_id, "协助")
+        if payload.new_assistant_personnel_id is not None
+        else None
+    )
+    owner_changed = business.owner_personnel_id != payload.new_owner_personnel_id
+    assistant_changed = business.assistant_personnel_id != payload.new_assistant_personnel_id
+    if not owner_changed and not assistant_changed:
+        raise BadRequestError("归属未变化")
+    return _BusinessOwnershipResolution(
+        owner=owner,
+        assistant=assistant,
+        old_owner_id=business.owner_personnel_id,
+        old_assistant_id=business.assistant_personnel_id,
+    )
+
+
+def _business_ownership_log(
+    business: Business,
+    payload: BusinessOwnershipIn,
+    resolution: _BusinessOwnershipResolution,
+    operator: User,
+) -> OwnershipChangeLog:
+    """归属变更留痕行（负责/协助两个字段的前后值都记）。"""
+    return OwnershipChangeLog(
+        owner_type=OWNER_TYPE_BUSINESS,
+        business_id=business.id,
+        old_owner_personnel_id=resolution.old_owner_id,
+        new_owner_personnel_id=payload.new_owner_personnel_id,
+        old_assistant_personnel_id=resolution.old_assistant_id,
+        new_assistant_personnel_id=payload.new_assistant_personnel_id,
+        change_reason=payload.reason,
+        operator_id=operator.id,
+        operator_name=operator.username,
+    )
+
+
 def change_ownership(
     session: Session,
     business_id: int,
@@ -730,41 +824,12 @@ def change_ownership(
         if row is None:
             raise NotFoundError("业务不存在")
         business, _, old_owner_name, old_assistant_name = row
-        # 归属变更的出参姓名来自校验函数的返回行（锁定读）：首查姓名是变更前
-        # 固化标量，新归属人名必须按新 ID 重取
-        new_owner = (
-            _ensure_personnel_active(session, payload.new_owner_personnel_id, "负责")
-            if payload.new_owner_personnel_id is not None
-            else None
-        )
-        new_assistant = (
-            _ensure_personnel_active(session, payload.new_assistant_personnel_id, "协助")
-            if payload.new_assistant_personnel_id is not None
-            else None
-        )
+        resolution = _resolve_business_ownership(session, business, payload)
 
-        owner_changed = business.owner_personnel_id != payload.new_owner_personnel_id
-        assistant_changed = business.assistant_personnel_id != payload.new_assistant_personnel_id
-        if not owner_changed and not assistant_changed:
-            raise BadRequestError("归属未变化")
-
-        old_owner_id = business.owner_personnel_id
-        old_assistant_id = business.assistant_personnel_id
         business.owner_personnel_id = payload.new_owner_personnel_id
         business.assistant_personnel_id = payload.new_assistant_personnel_id
         business_repository.add_ownership_log(
-            session,
-            OwnershipChangeLog(
-                owner_type=OWNER_TYPE_BUSINESS,
-                business_id=business_id,
-                old_owner_personnel_id=old_owner_id,
-                new_owner_personnel_id=payload.new_owner_personnel_id,
-                old_assistant_personnel_id=old_assistant_id,
-                new_assistant_personnel_id=payload.new_assistant_personnel_id,
-                change_reason=payload.reason,
-                operator_id=operator.id,
-                operator_name=operator.username,
-            ),
+            session, _business_ownership_log(business, payload, resolution, operator)
         )
         # 出参组装：ID 在锁定行上已赋值（_to_out 读内存值即反映新值），姓名
         # 一侧「变更项用新 ID 的人名、未变项沿用首查名」——首查名是变更前快照
@@ -772,17 +837,19 @@ def change_ownership(
             (
                 business,
                 row[1],
-                new_owner.personnel_name if new_owner is not None else old_owner_name,
-                new_assistant.personnel_name if new_assistant is not None else old_assistant_name,
+                resolution.owner.personnel_name if resolution.owner is not None else old_owner_name,
+                resolution.assistant.personnel_name
+                if resolution.assistant is not None
+                else old_assistant_name,
             )
         )
         session.commit()
         logger.info(
             "调整业务归属 | business_id=%s | owner %s->%s | assistant %s->%s",
             business_id,
-            old_owner_id,
+            resolution.old_owner_id,
             payload.new_owner_personnel_id,
-            old_assistant_id,
+            resolution.old_assistant_id,
             payload.new_assistant_personnel_id,
         )
         return out
@@ -848,6 +915,94 @@ def list_status_logs(
     )
 
 
+@dataclass(frozen=True, slots=True)
+class _GrossProfitScan:
+    """毛利润逐行扫描结果（行 DTO、类别聚合桶、成本缺失计数）。"""
+
+    rows: list[GrossProfitRowOut]
+    category_agg: dict[str, list[Decimal]]
+    missing_count: int
+
+
+def _gross_profit_row(
+    item: BusinessItem, threshold: Fraction
+) -> tuple[GrossProfitRowOut, tuple[Decimal, Decimal] | None]:
+    """单行毛利润派生值（行 DTO + 类别聚合桶）。
+
+    聚合桶为 None 表示成本缺失行 —— 这类行不计入类别汇总，但仍要输出
+    派生列为 None 的行与缺失原因。阈值对两类行都要取（成本缺失也展示阈值）。
+    """
+    row_kwargs: dict = {
+        "item_id": item.id,
+        "product_category": item.product_category,
+        "product_name": item.product_name,
+        "spec_model": item.spec_model,
+        "quantity": str(item.quantity),
+        "sale_amount": str(item.sale_amount),
+        "threshold": str(threshold),
+    }
+    if item.cost_amount is None:
+        row_kwargs.update(
+            cost_amount=None,
+            profit_amount=None,
+            profit_rate=None,
+            converted_profit=None,
+            missing_reason="成本缺失，无法计算毛利润",
+        )
+        return GrossProfitRowOut(**row_kwargs), None
+
+    profit = item.sale_amount - item.cost_amount
+    rate = profit_rate(profit, item.sale_amount)
+    conv = converted_profit(item.sale_amount, rate, threshold)
+    row_kwargs.update(
+        cost_amount=str(item.cost_amount),
+        profit_amount=str(profit),
+        profit_rate=str(rate),
+        converted_profit=str(conv),
+        missing_reason=None,
+    )
+    return GrossProfitRowOut(**row_kwargs), (item.sale_amount, item.cost_amount)
+
+
+def _scan_gross_profit_rows(
+    session: Session, business_id: int, threshold_of: Callable[[str], Fraction]
+) -> _GrossProfitScan:
+    """逐行扫描业务明细：出行 DTO、类别聚合桶与成本缺失计数。
+
+    类别顺序沿用明细行的首次出现顺序（不排序）—— 前端汇总展示依赖该顺序。
+    """
+    rows: list[GrossProfitRowOut] = []
+    agg: dict[str, list[Decimal]] = {}
+    missing_count = 0
+    for item in business_repository.list_items(session, business_id):
+        row, bucket = _gross_profit_row(item, threshold_of(item.product_category))
+        rows.append(row)
+        if bucket is None:
+            missing_count += 1
+            continue
+        current = agg.setdefault(item.product_category, [Decimal("0"), Decimal("0")])
+        current[0] += bucket[0]
+        current[1] += bucket[1]
+    return _GrossProfitScan(rows=rows, category_agg=agg, missing_count=missing_count)
+
+
+def _gross_profit_category(
+    category: str, sale: Decimal, cost: Decimal, threshold: Fraction
+) -> GrossProfitCategoryOut:
+    """类别汇总行（派生口径与行级一致：利润率/折算复用 formula 纯函数）。"""
+    profit = sale - cost
+    rate = profit_rate(profit, sale)
+    return GrossProfitCategoryOut(
+        product_category=category,
+        sale_amount=str(sale),
+        cost_amount=str(cost),
+        profit_amount=str(profit),
+        profit_rate=str(rate),
+        threshold=str(threshold),
+        converted_profit=str(converted_profit(sale, rate, threshold)),
+    )
+
+
 def gross_profit(session: Session, business_id: int, as_of: date | None) -> GrossProfitOut:
     """毛利润数据视图（需求 2.2.4）：行级派生值 + 类别汇总，实时计算不落库。
 
@@ -866,64 +1021,15 @@ def gross_profit(session: Session, business_id: int, as_of: date | None) -> Gros
     def threshold_of(category: str) -> Fraction:
         return snapshot.fraction("product.threshold", product_category=category)
 
-    rows_out: list[GrossProfitRowOut] = []
-    missing_count = 0
-    # 类别聚合容器：{类别: [sale, cost]}，成本缺失行不计入该类别聚合
-    agg: dict[str, list[Decimal]] = {}
-    for item in business_repository.list_items(session, business_id):
-        threshold = threshold_of(item.product_category)
-        row_kwargs: dict = {
-            "item_id": item.id,
-            "product_category": item.product_category,
-            "product_name": item.product_name,
-            "spec_model": item.spec_model,
-            "quantity": str(item.quantity),
-            "sale_amount": str(item.sale_amount),
-            "threshold": str(threshold),
-        }
-        if item.cost_amount is None:
-            missing_count += 1
-            row_kwargs.update(
-                cost_amount=None,
-                profit_amount=None,
-                profit_rate=None,
-                converted_profit=None,
-                missing_reason="成本缺失，无法计算毛利润",
-            )
-        else:
-            profit = item.sale_amount - item.cost_amount
-            rate = profit_rate(profit, item.sale_amount)
-            conv = converted_profit(item.sale_amount, rate, threshold)
-            row_kwargs.update(
-                cost_amount=str(item.cost_amount),
-                profit_amount=str(profit),
-                profit_rate=str(rate),
-                converted_profit=str(conv),
-                missing_reason=None,
-            )
-            bucket = agg.setdefault(item.product_category, [Decimal("0"), Decimal("0")])
-            bucket[0] += item.sale_amount
-            bucket[1] += item.cost_amount
-        rows_out.append(GrossProfitRowOut(**row_kwargs))
-
+    scan = _scan_gross_profit_rows(session, business_id, threshold_of)
     categories = [
-        GrossProfitCategoryOut(
-            product_category=category,
-            sale_amount=str(sale),
-            cost_amount=str(cost),
-            profit_amount=str(sale - cost),
-            profit_rate=str(profit_rate(sale - cost, sale)),
-            threshold=str(threshold_of(category)),
-            converted_profit=str(
-                converted_profit(sale, profit_rate(sale - cost, sale), threshold_of(category))
-            ),
-        )
-        for category, [sale, cost] in agg.items()
+        _gross_profit_category(category, sale, cost, threshold_of(category))
+        for category, [sale, cost] in scan.category_agg.items()
     ]
     return GrossProfitOut(
         business_id=business_id,
         as_of=as_of,
-        rows=rows_out,
+        rows=scan.rows,
         categories=categories,
-        missing_count=missing_count,
+        missing_count=scan.missing_count,
     )

backend/app/services/calc_service.py

@@ -60,7 +60,7 @@ from app.services.param_service import parse_fraction, resolve_snapshot
 logger = logging.getLogger(__name__)
 
 # snapshot_json 里的策略点注记（口径检索锚点；#12 已参数化，开关 0 时留此注记）
-_NOTE_MANAGER_PENDING = "经理个人户单笔不计提（基数开关=0）：聚合口径在月结时由团队聚合计算"
+_NOTE_MANAGER_PENDING = "经理个人户单笔不计提（未计入分成基数）：绩效在月结时按团队与全公司分成计算"
 
 
 @dataclass(frozen=True, slots=True)
@@ -544,7 +544,7 @@ def _personal_rate(
     if customer_type == "enterprise":
         # 存储口径 ‰ → 纯比例（漏换算放大 1000 倍，与预览同防线）
         rate = parse_fraction(snapshot.get("enterprise.rate")) / 1000
-        return _RateDecision(rate, "企业户固定计提比例", None, False)
+        return _RateDecision(rate, "企业户：不按回款天数，统一按固定比例计提", None, False)
 
     template = role.calc_template if role else "TPL_SALESPERSON_DECAY"
     if template == "TPL_MANAGER_SHARE":
@@ -554,7 +554,7 @@ def _personal_rate(
         if snapshot.get("manager.personal.included") != "1":
             return _RateDecision(
                 Decimal("0"),
-                "经理个人户：12.5%+5% 聚合计提（月结时计算），单笔落 0",
+                "经理个人户：绩效按团队与全公司分成在月结时计算，单笔不计提",
                 None,
                 True,
             )
@@ -563,12 +563,12 @@ def _personal_rate(
         floor = parse_fraction(snapshot.get("salesperson.rate.floor"))
         boundary = int(parse_fraction(snapshot.get("salesperson.days.boundary")))
         rate = salesperson_rate(days, start=start, slope=slope, floor=floor, boundary_days=boundary)
-        note = f"经理个人户（基数开关=1）：按业务员递减 {start}‰ 起，{days} 天"
+        note = f"经理个人户（已开启计入分成基数）：按业务员递减，回款 {days} 天，起始 {start}‰"
         return _RateDecision(rate, note, None, False)
     if template == "TPL_INFO_SALES_FIXED":
         # 信息员基数=本人销售额（覆盖折算与阈值），‰ 换算同防线
         rate = parse_fraction(snapshot.get("info.rate")) / 1000
-        return _RateDecision(rate, "信息员：本人销售额 × 计提比例", None, False)
+        return _RateDecision(rate, "信息员：按本人销售额乘计提比例，不随回款天数递减", None, False)
     if template == "TPL_SALESPERSON_DECAY":
         start = parse_fraction(snapshot.get("salesperson.rate.start"))
         slope = parse_fraction(snapshot.get("salesperson.rate.slope"))
@@ -577,9 +577,16 @@ def _personal_rate(
         # 经 Fraction 转换再取整（INT 型参数入库时已校验分母为 1，取整无损）
         boundary = int(parse_fraction(snapshot.get("salesperson.days.boundary")))
         rate = salesperson_rate(days, start=start, slope=slope, floor=floor, boundary_days=boundary)
-        note = f"业务员递减：{start}‰ 起，{days} 天（分界 {boundary} 天，超界取 {floor}‰）"
+        note = (
+            f"回款 {days} 天，从起始 {start}‰ 按日递减"
+            f"（满 {boundary} 天后固定按 {floor}‰ 计算）"
+        )
         return _RateDecision(rate, note, None, False)
-    raise BadRequestError(f"公式模板 {template} 暂不支持核算")
+    # 提示给模板中文名而非编码：业务人员在岗位角色页看到的是「营销经理·团队与全公司分成」
+    from app.services.calc_templates import TEMPLATES
+
+    tpl_name = TEMPLATES[template].name if template in TEMPLATES else template
+    raise BadRequestError(f"「{tpl_name}」不支持单笔核算，请检查该人员的岗位设置")
 
 
 def _role_of(session: Session, person):

backend/app/services/customer_service.py

@@ -14,6 +14,8 @@
 """
 
 import logging
+from dataclasses import dataclass
+from datetime import datetime
 
 from sqlalchemy import Row
 from sqlalchemy.exc import IntegrityError
@@ -37,6 +39,7 @@ from app.schemas.customer import (
     CustomerOut,
     CustomerOwnershipIn,
     CustomerQuery,
+    CustomerRatingIn,
     CustomerUpdateIn,
     OwnershipChangeLogOut,
 )
@@ -60,6 +63,8 @@ def _to_out(customer: Customer, owner_name: str | None) -> CustomerOut:
         owner_personnel_id=customer.owner_personnel_id,
         owner_personnel_name=owner_name,
         remark=customer.remark,
+        rating_score=customer.rating_score,
+        rating_time=customer.rating_time,
     )
 
 
@@ -107,20 +112,8 @@ def _is_duplicate_entry(exc: IntegrityError) -> bool:
 
 def list_customers(session: Session, query: CustomerQuery) -> PageResult[CustomerOut]:
     """分页查询客户（附归属人员姓名）。"""
-    rows = customer_repository.list_customers(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        keyword=query.keyword,
-        customer_type=query.customer_type,
-        owner_personnel_id=query.owner_personnel_id,
-    )
-    total = customer_repository.count_customers(
-        session,
-        keyword=query.keyword,
-        customer_type=query.customer_type,
-        owner_personnel_id=query.owner_personnel_id,
-    )
+    rows = customer_repository.list_customers(session, query)
+    total = customer_repository.count_customers(session, query)
     return PageResult[CustomerOut](
         records=[_to_out(customer, owner_name) for customer, owner_name in rows],
         total=total,
@@ -274,6 +267,53 @@ def update_customer(
         return out
 
 
+def rate_customer(
+    session: Session,
+    customer_id: int,
+    payload: CustomerRatingIn,
+    *,
+    operator: User,
+    meta: RequestMeta | None = None,
+) -> CustomerOut:
+    """保存客户评级（评级计算器「保存评分」）。
+
+    评分是客户档案属性而非审计事件：只存总分与评级时间，五项明细与所用
+    等级配置不落库（对照表 #25 的定位口径）——等级映射是可变判定表，回溯
+    「当时命中的等级」需要快照设计，超出档案属性的范围；评级时间已足够
+    回答「这个分是什么时候打的」。等级由前端按当前映射实时算并展示。
+
+    :raises NotFoundError: 客户不存在
+    """
+    # 客户名进审计摘要，预取放在审计上下文之前（同 change_ownership 口径）
+    row = customer_repository.get_with_owner_name(session, customer_id)
+    customer_name = row[0].company_name if row else f"#{customer_id}"
+    with audit_service.audit(
+        session,
+        operator=operator,
+        record=AuditRecord(
+            module=MODULE_CUSTOMER,
+            action=ACTION_UPDATE,
+            summary=f"保存客户评级 {customer_name}：{payload.score} 分",
+            request_data=payload.model_dump(),
+        ),
+        meta=meta,
+    ):
+        if row is None:
+            raise NotFoundError("客户不存在")
+        customer, owner_name = row
+
+        customer.rating_score = payload.score
+        # 评级时间取应用层本地时间（库列 TZ=Asia/Shanghai，与 TimestampMixin
+        # 的服务器默认值同一时区口径），不用数据库 NOW()：行已在本会话
+        customer.rating_time = datetime.now()
+
+        # 响应在 commit 之前构造（同 user_service 顶部说明）
+        out = _to_out(customer, owner_name)
+        session.commit()
+        logger.info("保存客户评级 | customer_id=%s | score=%s", customer_id, payload.score)
+        return out
+
+
 def delete_customer(
     session: Session,
     customer_id: int,
@@ -317,6 +357,71 @@ def delete_customer(
         logger.info("删除客户 | customer_id=%s | company=%s", customer_id, company_name)
 
 
+@dataclass(frozen=True, slots=True)
+class _CustomerOwnershipPrefetch:
+    """客户归属变更的审计前只读预取（摘要要客户名与新旧归属人名）。"""
+
+    row: Row | None
+    new_person: Personnel | None
+    old_label: str
+    new_label: str
+    customer_name: str
+
+
+def _prefetch_customer_ownership(
+    session: Session, customer_id: int, payload: CustomerOwnershipIn
+) -> _CustomerOwnershipPrefetch:
+    """审计上下文之前的只读预取（预取不写库，失败路径也要有可读摘要）。"""
+    row = customer_repository.get_with_owner_name(session, customer_id)
+    old_person = (
+        personnel_repository.get_by_id(session, row[0].owner_personnel_id)
+        if row is not None and row[0].owner_personnel_id is not None
+        else None
+    )
+    new_person = (
+        personnel_repository.get_by_id(session, payload.new_owner_personnel_id)
+        if payload.new_owner_personnel_id is not None
+        else None
+    )
+    return _CustomerOwnershipPrefetch(
+        row=row,
+        new_person=new_person,
+        old_label=old_person.personnel_name if old_person else "无",
+        new_label=new_person.personnel_name if new_person else "无",
+        customer_name=row[0].company_name if row else f"#{customer_id}",
+    )
+
+
+def _validate_customer_ownership(
+    session: Session, customer: Customer, payload: CustomerOwnershipIn
+) -> int | None:
+    """校验新归属人（须在职）并判定归属是否变化，返回变更前归属人 ID。"""
+    if payload.new_owner_personnel_id is not None:
+        _ensure_owner_personnel(session, payload.new_owner_personnel_id)
+    old_owner_id = customer.owner_personnel_id
+    if old_owner_id == payload.new_owner_personnel_id:
+        raise BadRequestError("归属未变化")
+    return old_owner_id
+
+
+def _customer_ownership_log(
+    customer: Customer,
+    payload: CustomerOwnershipIn,
+    old_owner_id: int | None,
+    operator: User,
+) -> OwnershipChangeLog:
+    """客户归属变更留痕行。"""
+    return OwnershipChangeLog(
+        owner_type=OWNER_TYPE_CUSTOMER,
+        customer_id=customer.id,
+        old_owner_personnel_id=old_owner_id,
+        new_owner_personnel_id=payload.new_owner_personnel_id,
+        change_reason=payload.reason,
+        operator_id=operator.id,
+        operator_name=operator.username,
+    )
+
+
 def change_ownership(
     session: Session,
     customer_id: int,
@@ -335,57 +440,34 @@ def change_ownership(
     """
     # 只读预取在审计上下文之前：摘要需要客户名与新旧归属人名；校验仍在
     # 上下文内进行，失败同样要留下可追溯的记录
-    row = customer_repository.get_with_owner_name(session, customer_id)
-    old_person = (
-        personnel_repository.get_by_id(session, row[0].owner_personnel_id)
-        if row is not None and row[0].owner_personnel_id is not None
-        else None
-    )
-    new_person = (
-        personnel_repository.get_by_id(session, payload.new_owner_personnel_id)
-        if payload.new_owner_personnel_id is not None
-        else None
-    )
-    old_label = old_person.personnel_name if old_person else "无"
-    new_label = new_person.personnel_name if new_person else "无"
-    customer_name = row[0].company_name if row else f"#{customer_id}"
+    prefetch = _prefetch_customer_ownership(session, customer_id, payload)
     with audit_service.audit(
         session,
         operator=operator,
         record=AuditRecord(
             module=MODULE_CUSTOMER,
             action=ACTION_UPDATE,
-            summary=f"调整客户归属 {customer_name}：{old_label} → {new_label}",
+            summary=(
+                f"调整客户归属 {prefetch.customer_name}："
+                f"{prefetch.old_label} → {prefetch.new_label}"
+            ),
             request_data=payload.model_dump(exclude_none=True),
         ),
         meta=meta,
     ):
-        if row is None:
+        if prefetch.row is None:
             raise NotFoundError("客户不存在")
-        customer, _ = row
+        customer, _ = prefetch.row
 
-        if payload.new_owner_personnel_id is not None:
-            _ensure_owner_personnel(session, payload.new_owner_personnel_id)
-        old_owner_id = customer.owner_personnel_id
-        if old_owner_id == payload.new_owner_personnel_id:
-            raise BadRequestError("归属未变化")
+        old_owner_id = _validate_customer_ownership(session, customer, payload)
 
         # 主表更新与留痕插入同事务：只在本方法内一次 commit（见模块 docstring）
         customer.owner_personnel_id = payload.new_owner_personnel_id
         customer_repository.add_ownership_log(
-            session,
-            OwnershipChangeLog(
-                owner_type=OWNER_TYPE_CUSTOMER,
-                customer_id=customer.id,
-                old_owner_personnel_id=old_owner_id,
-                new_owner_personnel_id=payload.new_owner_personnel_id,
-                change_reason=payload.reason,
-                operator_id=operator.id,
-                operator_name=operator.username,
-            ),
+            session, _customer_ownership_log(customer, payload, old_owner_id, operator)
         )
         # 响应在 commit 之前构造（同 user_service 顶部说明）；归属人名以新值为准
-        out = _to_out(customer, new_person.personnel_name if new_person else None)
+        out = _to_out(customer, prefetch.new_person.personnel_name if prefetch.new_person else None)
         session.commit()
         logger.info(
             "调整客户归属 | customer_id=%s | old=%s | new=%s",

backend/app/services/dashboard_service.py

@@ -1,4 +1,4 @@
-"""工作台服务（M4 ⑧）：四区块实时聚合编排。
+"""工作台服务（M4 ⑧）：区块实时聚合编排 + 最近动态分页。
 
 设计决策（2026-10-19）：
 - **只读**：不落库、不写审计（读操作不入操作日志）；总览只做 SQL 聚合，
@@ -9,6 +9,8 @@
 - **⑥ 同步模块未启用**：同步失败待办、同步连续失败预警以
   ``sync_available=false`` 显式降级，不返回伪造的 0 条
 - 预警阈值走参数体系（``warning.*``），一次请求只解析一次快照
+- **最近动态独立分页**（2026-10-23）：操作日志可能很长，聚合响应里
+  只能塞前 20 条；拆成独立端点后翻页只重取这张列表
 """
 
 from datetime import date, datetime, timedelta
@@ -20,22 +22,23 @@ from app.core.exceptions import BadRequestError
 from app.core.month import validate_month
 from app.repositories import dashboard_repository, operation_log_repository, user_repository
 from app.repositories.query_utils import LogFilters
+from app.schemas.dashboard import ActivityItem, ActivityPageOut, ActivityQuery
 from app.services import approval_service
 
 # 工作台只展示摘要，完整列表跳转对应管理页
 _TODO_LIMIT = 5
 _WARNING_LIMIT = 20
-_ACTIVITY_LIMIT = 20
 _TREND_MONTHS = 6
 
 _ORDER_TYPE_LABELS = {1: "二次分配方案", 2: "月度结算", 3: "调整单"}
 
 
 def get_dashboard(session: Session, month: str | None = None, *, current_user) -> dict:
-    """四区块一次性出数（前端一次请求渲染整页）。
+    """三区块一次性出数（前端一次请求渲染整页）。
 
     待办按当前用户角色只展示其可操作节点（admin 可查看全部在途单）；
     风险预警仍是全局经营视角，不按人裁剪。
+    最近动态不在此返回 —— 它是分页列表，走 ``list_activities``。
     """
     resolved = _normalize_month(month)
     warning_params = _warning_params(session)
@@ -47,7 +50,6 @@ def get_dashboard(session: Session, month: str | None = None, *, current_user) -
         "overview": _overview(session, resolved),
         "todos": _todos(session, pending_statuses),
         "warnings": _warnings(session, warning_params),
-        "activities": _activities(session),
     }
 
 
@@ -112,8 +114,8 @@ def _overview(session: Session, month: str) -> dict:
             for item in months
         ],
         "note": (
-            "绩效按台账归属月（单笔生效行，二次分摊拆行求和不失真）；回款按到账月"
-            "（含未核算回款）；不含经理 12.5%+5% 聚合分成与调整单"
+            "绩效按台账归属月统计（二次分配按参与人拆分后求和，不重复不计漏）；"
+            "回款按到账月统计（含尚未核算的回款）；不含经理团队分成与调整单"
         ),
     }
 
@@ -168,9 +170,7 @@ def _warning_params(session: Session) -> dict:
     return {
         "enabled": snapshot.get("warning.enabled") == "1",
         "receipt_overdue_days": int(parse_fraction(snapshot.get("warning.receipt.overdue.days"))),
-        "approval_timeout_days": int(
-            parse_fraction(snapshot.get("warning.approval.timeout.days"))
-        ),
+        "approval_timeout_days": int(parse_fraction(snapshot.get("warning.approval.timeout.days"))),
     }
 
 
@@ -249,26 +249,38 @@ def _warnings(session: Session, params: dict) -> dict:
     }
 
 
-def _activities(session: Session) -> list[dict]:
-    """最近动态（FR-036）：以操作日志为唯一来源（写操作即留痕）。"""
-    logs = operation_log_repository.list_logs(
-        session,
-        offset=0,
-        limit=_ACTIVITY_LIMIT,
+def list_activities(session: Session, query: ActivityQuery) -> ActivityPageOut:
+    """最近动态分页（FR-036）：以操作日志为唯一来源（写操作即留痕）。
+
+    模块筛选项与列表同源返回（全量模块名，不随当前页变化），
+    前端模块下拉不必另取一次全量。
+    """
+    repo_query = operation_log_repository.OperationLogListQuery(
         filters=LogFilters(),
+        offset=query.offset,
+        limit=query.page_size,
+        module=query.module,
+    )
+    logs = operation_log_repository.list_logs(session, repo_query)
+    total = operation_log_repository.count_logs(session, repo_query)
+    return ActivityPageOut(
+        records=[
+            ActivityItem(
+                id=log.id,
+                module=log.module,
+                action=log.action,
+                summary=log.summary,
+                username=log.username,
+                success=log.success,
+                create_time=log.create_time,
+            )
+            for log in logs
+        ],
+        total=total,
+        page_num=query.page_num,
+        page_size=query.page_size,
+        modules=operation_log_repository.distinct_modules(session),
     )
-    return [
-        {
-            "id": log.id,
-            "module": log.module,
-            "action": log.action,
-            "summary": log.summary,
-            "username": log.username,
-            "success": log.success,
-            "create_time": log.create_time,
-        }
-        for log in logs
-    ]
 
 
 def _dec(value) -> Decimal | None:

backend/app/services/log_service.py

@@ -22,18 +22,15 @@ from app.schemas.log import (
 
 def list_operation_logs(session: Session, query: OperationLogQuery) -> PageResult[OperationLogOut]:
     """分页查询操作日志（列表不含 request_data，详情走 `get_operation_log`）。"""
-    filters = _filters_from(query)
-    logs = operation_log_repository.list_logs(
-        session,
+    repo_query = operation_log_repository.OperationLogListQuery(
+        filters=_filters_from(query),
         offset=query.offset,
         limit=query.page_size,
-        filters=filters,
         module=query.module,
         action=query.action,
     )
-    total = operation_log_repository.count_logs(
-        session, filters=filters, module=query.module, action=query.action
-    )
+    logs = operation_log_repository.list_logs(session, repo_query)
+    total = operation_log_repository.count_logs(session, repo_query)
     return PageResult[OperationLogOut](
         records=[OperationLogOut.model_validate(log) for log in logs],
         total=total,
@@ -55,15 +52,14 @@ def get_operation_log(session: Session, log_id: int) -> OperationLogDetailOut:
 
 def list_login_logs(session: Session, query: LoginLogQuery) -> PageResult[LoginLogOut]:
     """分页查询登录日志。"""
-    filters = _filters_from(query)
-    logs = login_log_repository.list_logs(
-        session,
+    repo_query = login_log_repository.LoginLogListQuery(
+        filters=_filters_from(query),
         offset=query.offset,
         limit=query.page_size,
-        filters=filters,
         ip=query.ip,
     )
-    total = login_log_repository.count_logs(session, filters=filters, ip=query.ip)
+    logs = login_log_repository.list_logs(session, repo_query)
+    total = login_log_repository.count_logs(session, repo_query)
     return PageResult[LoginLogOut](
         records=[LoginLogOut.model_validate(log) for log in logs],
         total=total,

backend/app/services/param_registry.py

@@ -3,6 +3,10 @@
 注册表是**迁移种子与运行期校验的唯一来源**：编码、默认值、取值域在此定义
 一次；新增参数 = 改这里 + 生成迁移种子（两者一致性由测试锁定）。
 值一律以 Fraction 可解析字符串表示（如 "4/360"，禁 0.011111 —— 见算法红线）。
+
+`description` 会直接显示在「计提参数」页的参数名下方，**面向业务人员撰写**：
+说清这个参数管什么、钱怎么变、什么时候该调它。不写英文枚举名、需求编号、
+内部阶段号，也不写实现层面的精度约束（那些留在代码注释与测试里）。
 """
 
 from dataclasses import dataclass
@@ -43,7 +47,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="10",
             value_min="0",
             value_max=None,
-            description="线性递减段起始值：比例 = 起始 − 回款天数 × 斜率（≤分界天数）",
+            description="回款当天能拿到的最高计提比例。回款每晚一天，按日递减斜率往下走，直到分界天数为止。",
             sort_order=1,
         ),
         ParamSpec(
@@ -57,7 +61,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="4/360",
             value_min="0",
             value_max=None,
-            description="精确分数 4/360‰/天（=1/90），禁止按 0.011111 截断存储",
+            description="回款每延后一天，计提比例下降多少。按分数填写（如 4/360）可保持精确，填小数会引入四舍五入误差。",
             sort_order=2,
         ),
         ParamSpec(
@@ -71,7 +75,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="4",
             value_min="0",
             value_max=None,
-            description="回款天数超过分界天数后固定取此比例（≥361 天 = 4‰）",
+            description="回款天数超过分界天数后，计提比例不再下降，固定按这个值计算，即最低计提比例。",
             sort_order=3,
         ),
         ParamSpec(
@@ -85,7 +89,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="360",
             value_min="0",
             value_max=None,
-            description="≤分界天数按公式线性递减（360 天=6‰）；超过则取下限",
+            description="回款天数在这个范围内按日递减；超过这个天数就固定按下限计提。默认 360 天。",
             sort_order=4,
         ),
         ParamSpec(
@@ -99,7 +103,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="12.5",
             value_min="0",
             value_max="100",
-            description="∑本人带领业务员绩效 × 此比例（限本团队）",
+            description="经理按本团队业务员的绩效总额提取的分成比例，只统计自己带的业务员。",
             sort_order=5,
         ),
         ParamSpec(
@@ -113,7 +117,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="5",
             value_min="0",
             value_max="100",
-            description="∑全部业务员绩效 × 此比例（全公司）",
+            description="经理按全公司业务员的绩效总额提取的分成比例，与本团队分成叠加计算。",
             sort_order=6,
         ),
         ParamSpec(
@@ -127,7 +131,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="3",
             value_min="0",
             value_max=None,
-            description="所有企业户适用；2026-08-13 起由 2‰ 提高至 3‰",
+            description="企业户客户的回款统一按这个比例计提，不看回款天数。2026-08-13 起由 2‰ 调整为 3‰。",
             sort_order=7,
         ),
         ParamSpec(
@@ -141,7 +145,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="1",
             value_min="0",
             value_max=None,
-            description="信息员绩效 = 本人销售额 × 此比例",
+            description="信息员的绩效按本人销售额乘这个比例计算，不随回款天数递减。",
             sort_order=8,
         ),
         ParamSpec(
@@ -155,7 +159,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="10",
             value_min="0",
             value_max=None,
-            description="预付款统一按此比例计提",
+            description="预付款到账即按这个比例计提绩效，不必等到全额回款。",
             sort_order=9,
         ),
         ParamSpec(
@@ -169,7 +173,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="80",
             value_min="0",
             value_max="100",
-            description="以物抵顶回款绩效按应发绩效的此比例折算",
+            description="以物抵顶的回款，按应发绩效的这个比例折算，即打折后的金额参与绩效。",
             sort_order=10,
         ),
         ParamSpec(
@@ -183,7 +187,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="60",
             value_min="0",
             value_max="100",
-            description="单项业务回款额达销售额（不含运费）此比例后起付；达标前累计",
+            description="单项业务的回款累计达到销售额（不含运费）的这个比例后才开始发放绩效；达标之前只累计、不发放。",
             sort_order=11,
         ),
         ParamSpec(
@@ -197,7 +201,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="FULL",
             value_min=None,
             value_max=None,
-            description="FULL=按 R×R 乘当月全额应发绩效；BALANCE=乘余额（语义待 M3 前确认）",
+            description="选择支付权重乘在哪个基数上：当月全额 = 乘本月全部应发绩效；余额 = 只乘尚未发放的部分。",
             sort_order=12,
         ),
         ParamSpec(
@@ -211,7 +215,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="30",
             value_min="0",
             value_max="100",
-            description="折算毛利润额 = 销售额 ×(毛利润率 − 阈值)；市政水利 30%、轨交 20%",
+            description="折算毛利润额 = 销售额 ×（毛利润率 − 阈值），低于阈值的部分不参与绩效。市政水利默认 30%，轨道交通默认 20%。",
             sort_order=13,
         ),
         ParamSpec(
@@ -225,7 +229,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="INCLUDE",
             value_min=None,
             value_max=None,
-            description="INCLUDE含基本工资 / DEDUCT扣除基本工资 / EXCLUDE不计",
+            description="算月工资时怎么处理基本工资：含基本工资 = 与绩效一起发；扣除基本工资 = 从绩效里扣掉已发的基本工资；不计 = 月工资只发绩效。",
             sort_order=14,
         ),
         ParamSpec(
@@ -239,7 +243,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="18000",
             value_min="0",
             value_max=None,
-            description="试用期年累计绩效超过此值部分年底一次性结算",
+            description="试用期绩效先不发放；年内累计超过这个金额的部分，年底一次性结算。",
             sort_order=15,
         ),
         ParamSpec(
@@ -253,11 +257,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="1",
             value_min=None,
             value_max=None,
-            description=(
-                "0=二次分配方案提交即生效（系统自动通过全节点留痕，留痕不断链）；"
-                "1=提交后进入财务复核→领导审核审批流。"
-                "默认开启（澄清 #5 业务答复 2026-10-19），业务可关闭"
-            ),
+            description="二次分配方案提交后是否要审批：开启 = 先经财务复核、再经领导审核，通过后才生效；关闭 = 提交即生效，系统自动记录各节点留痕。",
             sort_order=17,
         ),
         ParamSpec(
@@ -271,11 +271,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="1",
             value_min=None,
             value_max=None,
-            description=(
-                "月度结算审批链开关（④ 月结提交时消费）："
-                "0=结算生成即生效，1=走财务复核→领导审核审批流。"
-                "默认开启（澄清 #7 业务答复 2026-10-19，月结涉及工资发放从严），业务可关闭"
-            ),
+            description="月度结算提交后是否要审批：开启 = 先经财务复核、再经领导审核；关闭 = 生成即生效。月结涉及工资发放，默认开启。",
             sort_order=18,
         ),
         ParamSpec(
@@ -289,11 +285,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="0",
             value_min=None,
             value_max=None,
-            description=(
-                "0=调整单提交即生效（系统自动通过留痕）；1=走财务复核→领导审核审批流。"
-                "调整单是已锁定月份补发/扣减的唯一通道（澄清 #9：走同一审批链，"
-                "默认值业务未明确，暂按直接生效，业务可开启）"
-            ),
+            description="调整单提交后是否要审批：开启 = 先经财务复核、再经领导审核；关闭 = 提交即生效。调整单是已锁定月份补发或扣减的唯一通道。",
             sort_order=19,
         ),
         ParamSpec(
@@ -307,11 +299,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="0",
             value_min=None,
             value_max=None,
-            description=(
-                "经理 12.5%+5% 聚合时是否计入经理本人名下经理个人户业绩（澄清 #12 "
-                "业务答复 2026-10-19：作为参数配置、手动调整是否纳入）："
-                "0=不计入（默认，暂口径） 1=计入"
-            ),
+            description="经理提取团队分成时，是否把经理本人名下的业绩（经理个人户）也算进分成基数。默认不计入。",
             sort_order=20,
         ),
         # ===== 预警规则（M4 ⑧ 工作台，澄清 #10「做成参数化规则便于调整」）=====
@@ -327,7 +315,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="1",
             value_min=None,
             value_max=None,
-            description="工作台风险预警区块总开关：0=关闭（不扫描不展示） 1=开启（默认）",
+            description="工作台风险预警区块的总开关：关闭后不再扫描、也不展示预警条目。默认开启。",
             sort_order=21,
         ),
         ParamSpec(
@@ -341,9 +329,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="90",
             value_min="1",
             value_max=None,
-            description=(
-                "发货后超过该天数仍未回款的发货批次触发预警（澄清 #10 建议默认 90 天）"
-            ),
+            description="发货后超过这么多天仍未收到回款的发货批次，会出现在工作台的预警列表里。默认 90 天。",
             sort_order=22,
         ),
         ParamSpec(
@@ -357,9 +343,7 @@ PARAM_REGISTRY: dict[str, ParamSpec] = {
             default_value="3",
             value_min="1",
             value_max=None,
-            description=(
-                "待审批单提交后超过该天数未处理触发预警（澄清 #10 建议默认 3 天）"
-            ),
+            description="审批单提交后超过这么多天还没处理，会出现在工作台的预警列表里。默认 3 天。",
             sort_order=23,
         ),
     ]

backend/app/services/param_service.py

@@ -6,7 +6,7 @@ ParamSnapshot 是 M3 核算与计算预览共用的输入契约：按业务日
 """
 
 import logging
-from collections.abc import Iterable
+from collections.abc import Callable, Iterable
 from dataclasses import dataclass
 from datetime import date
 from decimal import Decimal
@@ -44,7 +44,7 @@ from app.schemas.param import (
     ParamPreviewOut,
     ParamUsedOut,
 )
-from app.services import audit_service, formula
+from app.services import audit_service, calc_templates, formula
 from app.services.audit_service import AuditRecord, RequestMeta
 from app.services.formula import to_decimal
 from app.services.no_generator import next_no
@@ -379,16 +379,8 @@ def get_history(
     definition = param_repository.get_definition_by_code(session, param_code)
     if definition is None:
         raise NotFoundError("参数不存在")
-    rows = param_repository.list_history(
-        session,
-        definition.id,
-        offset=query.offset,
-        limit=query.page_size,
-        dimension_value=query.dimension_value,
-    )
-    total = param_repository.count_history(
-        session, definition.id, dimension_value=query.dimension_value
-    )
+    rows = param_repository.list_history(session, definition.id, query)
+    total = param_repository.count_history(session, definition.id, query)
     return PageResult[ParamHistoryOut](
         records=[
             ParamHistoryOut(
@@ -410,16 +402,8 @@ def get_history(
 
 def list_batches(session: Session, query: ParamBatchQuery) -> PageResult[ParamBatchOut]:
     """变更批次分页（按创建时间倒序）。"""
-    batches = param_repository.list_batches(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        start_date=query.start_date,
-        end_date=query.end_date,
-    )
-    total = param_repository.count_batches(
-        session, start_date=query.start_date, end_date=query.end_date
-    )
+    batches = param_repository.list_batches(session, query)
+    total = param_repository.count_batches(session, query)
     return PageResult[ParamBatchOut](
         records=[
             ParamBatchOut(
@@ -509,42 +493,89 @@ def compare(session: Session, *, from_date: date, to_date: date) -> list[ParamDi
     return out
 
 
-def _preview_rate(
-    session: Session,
-    payload: ParamPreviewIn,
-    snapshot: ParamSnapshot,
-    days: int,
-    take,
-) -> tuple[Fraction, str, Decimal | None]:
+@dataclass(frozen=True, slots=True)
+class _PreviewContext:
+    """预览上下文（会话、入参、参数取值闭包）。
+
+    ``take`` 闭包同时负责取数与记录「用了哪些参数」——**调用顺序即
+    ``params_used`` 的展示顺序**，拆分时不得调换各段的先后。
+    """
+
+    session: Session
+    payload: ParamPreviewIn
+    take: Callable[[str], Fraction]
+
+
+@dataclass(frozen=True, slots=True)
+class _PreviewRateDecision:
+    """计提比例判定（回款天数、纯比例、人读说明、绩效基数覆盖值）。"""
+
+    days: int
+    rate: Fraction
+    rate_text: str
+    base_override: Decimal | None
+
+
+@dataclass(frozen=True, slots=True)
+class _PreviewEarning:
+    """折算与绩效（毛利润率、折算额、绩效额）。"""
+
+    profit_rate: Fraction
+    converted: Decimal
+    performance: Decimal
+
+
+@dataclass(frozen=True, slots=True)
+class _PreviewBarter:
+    """以物抵顶（是否适用、折算后绩效、计发基数）。"""
+
+    applied: bool
+    after_barter: Decimal | None
+    base_amount: Decimal
+
+
+@dataclass(frozen=True, slots=True)
+class _PreviewPayment:
+    """支付环节（起付判定、R 权重、权重、应付金额）。"""
+
+    start_ok: bool
+    weight_r: int
+    weight: Fraction
+    payable: Decimal
+
+
+def _preview_rate(ctx: _PreviewContext, days: int) -> tuple[Fraction, str, Decimal | None]:
     """按岗位模板分派计提比例，返回（纯比例, 人读说明, 绩效基数覆盖值）。
 
     基数覆盖值非 None 表示该模板的绩效基数不是折算毛利润额（如信息员按
     「本人销售额」计提），预览层直接采用并跳过折算与阈值。
     聚合模板（营销经理 12.5%+5%）依赖月度团队数据，不支持单笔预览。
     """
+    payload = ctx.payload
+    take = ctx.take
     if payload.customer_type == "enterprise":
         # 存储口径为 ‰ 数（如 "3"），除以 1000 转纯比例 —— performance_amount
         # 按纯比例相乘，漏换算会把绩效金额放大 1000 倍（与业务员分支同防线）
         rate = take("enterprise.rate") / 1000
-        return rate, "企业户固定计提比例", None
+        return rate, "企业户：不按回款天数，统一按固定比例计提", None
 
     role_code = payload.business_role
     template = "TPL_SALESPERSON_DECAY"
     if role_code:
-        role = business_role_repository.get_by_code(session, role_code)
+        role = business_role_repository.get_by_code(ctx.session, role_code)
         if role is None:
-            raise BadRequestError(f"岗位 {role_code!r} 不存在")
+            raise BadRequestError(f"岗位「{role_code}」不存在，请到岗位角色里确认")
         template = role.calc_template
 
     if template == "TPL_MANAGER_SHARE":
         raise BadRequestError(
-            "营销经理绩效为团队聚合口径（本团队 12.5% + 全公司 5%），不支持单笔预览"
+            "营销经理的绩效按团队与全公司分成计算，不单笔试算；请到月度结算里查看"
         )
     if template == "TPL_INFO_SALES_FIXED":
         # ‰ 存储口径 → 纯比例，同企业户分支（不换算则绩效放大 1000 倍）
         rate = take("info.rate") / 1000
         # 信息员的基数是本人销售额（无折算与阈值），覆盖默认基数
-        return rate, "信息员：本人销售额 × 计提比例", Decimal(payload.revenue)
+        return rate, "信息员：按本人销售额乘计提比例，不随回款天数递减", Decimal(payload.revenue)
     if template == "TPL_SALESPERSON_DECAY":
         start = take("salesperson.rate.start")
         slope = take("salesperson.rate.slope")
@@ -553,12 +584,84 @@ def _preview_rate(
         rate = formula.salesperson_rate(
             days, start=start, slope=slope, floor=floor, boundary_days=boundary
         )
+        # 说明写成业务话：说清这笔为什么是这个比例、什么时候不再降
         return (
             rate,
-            f"业务员递减：{to_decimal(start)}‰ 起，{days} 天（分界 {boundary} 天，超界取 {to_decimal(floor)}‰）",
+            f"回款 {days} 天，从起始 {to_decimal(start)}‰ 按日递减"
+            f"（满 {boundary} 天后固定按 {to_decimal(floor)}‰ 计算）",
             None,
         )
-    raise BadRequestError(f"公式模板 {template} 暂不支持预览")
+    # 提示里给模板中文名而不是编码：编码是内部标识，业务人员在岗位角色页看不到它
+    tpl_name = calc_templates.TEMPLATES[template].name if template in calc_templates.TEMPLATES else template
+    raise BadRequestError(
+        f"「{tpl_name}」不支持单笔试算，请在岗位角色里改用业务员递减或信息员固定计提"
+    )
+
+
+def _preview_rate_decision(ctx: _PreviewContext) -> _PreviewRateDecision:
+    """产品类别校验 + 回款天数 + 计提比例判定。"""
+    payload = ctx.payload
+    if payload.product_category and payload.product_category not in PRODUCT_CATEGORIES:
+        # 提示里用中文类别名：前端下拉显示的是「市政水利」，报错却回英文编码会让人对不上
+        raise BadRequestError(
+            f"产品类别「{payload.product_category}」不存在，可选：市政水利、轨道交通"
+        )
+    days = formula.payment_days(payload.ship_date, payload.receipt_date)
+    rate, rate_text, base_override = _preview_rate(ctx, days)
+    return _PreviewRateDecision(
+        days=days, rate=rate, rate_text=rate_text, base_override=base_override
+    )
+
+
+def _preview_earning(ctx: _PreviewContext, decision: _PreviewRateDecision) -> _PreviewEarning:
+    """毛利润率 → 折算额（阈值）→ 绩效额（比例）。"""
+    payload = ctx.payload
+    revenue = Decimal(payload.revenue)
+    profit = Decimal(payload.profit)
+    profit_rate = formula.profit_rate(profit, revenue)
+    if decision.base_override is not None:
+        # 基数覆盖（信息员按本人销售额计提）：不走折算与阈值
+        converted = decision.base_override
+    else:
+        threshold = ctx.take("product.threshold")
+        converted = formula.converted_profit(revenue, profit_rate, threshold)
+    performance = formula.performance_amount(converted, decision.rate)
+    return _PreviewEarning(profit_rate=profit_rate, converted=converted, performance=performance)
+
+
+def _preview_barter(ctx: _PreviewContext, performance: Decimal) -> _PreviewBarter:
+    """以物抵顶折算（按应发绩效的固定比例折算，比例来自参数）。"""
+    applied = ctx.payload.barter
+    after_barter: Decimal | None = None
+    if applied:
+        discount = ctx.take("barter.discount")
+        after_barter = formula.apply_barter(performance, discount)
+    base_amount = after_barter if applied else performance
+    return _PreviewBarter(applied=applied, after_barter=after_barter, base_amount=base_amount)
+
+
+def _preview_payment(
+    ctx: _PreviewContext, received: Decimal, revenue: Decimal, base_amount: Decimal
+) -> _PreviewPayment:
+    """起付比例判定 + R 权重 + 应付金额。"""
+    start_ratio = ctx.take("payment.start.ratio")
+    start_ok = formula.start_payment_ok(received, revenue, start_ratio)
+    # R 钳制已提取为纯函数 clamp_weight_r（④ 核算与预览共用同一实现）
+    weight_r = formula.clamp_weight_r(received, revenue)
+    weight = formula.payment_weight(weight_r)
+    payable = to_decimal(Fraction(base_amount) * weight)
+    return _PreviewPayment(start_ok=start_ok, weight_r=weight_r, weight=weight, payable=payable)
+
+
+def _preview_probation_note(ctx: _PreviewContext) -> str | None:
+    """试用期说明（绩效照算不发放，年累计超阈值的部分年底一次性结算）。"""
+    if not ctx.payload.probation:
+        return None
+    annual = ctx.take("probation.annual.threshold")
+    return (
+        f"试用期人员：绩效照常计算但不发放；年内累计超过 {to_decimal(annual)} 元的部分，"
+        "年底一次性结清。"
+    )
 
 
 def preview(session: Session, payload: ParamPreviewIn) -> ParamPreviewOut:
@@ -566,7 +669,6 @@ def preview(session: Session, payload: ParamPreviewIn) -> ParamPreviewOut:
     # 参数按回款到账日取值（对照表 #11）
     snapshot = resolve_snapshot(session, payload.receipt_date)
     revenue = Decimal(payload.revenue)
-    profit = Decimal(payload.profit)
     received = Decimal(payload.received_amount)
     used: list[ParamUsedOut] = []
 
@@ -589,55 +691,29 @@ def preview(session: Session, payload: ParamPreviewIn) -> ParamPreviewOut:
         )
         return parse_fraction(raw)
 
-    if payload.product_category and payload.product_category not in PRODUCT_CATEGORIES:
-        raise BadRequestError(
-            f"产品类别 {payload.product_category!r} 不存在，可选：municipal_water / rail_transit"
-        )
-
-    days = formula.payment_days(payload.ship_date, payload.receipt_date)
-    rate, rate_text, base_override = _preview_rate(session, payload, snapshot, days, take)
-
-    profit_rate = formula.profit_rate(profit, revenue)
-    if base_override is not None:
-        # 基数覆盖（信息员按本人销售额计提）：不走折算与阈值
-        converted = base_override
-    else:
-        threshold = take("product.threshold")
-        converted = formula.converted_profit(revenue, profit_rate, threshold)
-    performance = formula.performance_amount(converted, rate)
-
-    barter_applied = payload.barter
-    after_barter: Decimal | None = None
-    if barter_applied:
-        discount = take("barter.discount")
-        after_barter = formula.apply_barter(performance, discount)
-
-    start_ratio = take("payment.start.ratio")
-    start_ok = formula.start_payment_ok(received, revenue, start_ratio)
-    # R 钳制已提取为纯函数 clamp_weight_r（④ 核算与预览共用同一实现）
-    weight_r = formula.clamp_weight_r(received, revenue)
-    weight = formula.payment_weight(weight_r)
-    base_amount = after_barter if barter_applied else performance
-    payable = to_decimal(Fraction(base_amount) * weight)
-
-    probation_note: str | None = None
-    if payload.probation:
-        annual = take("probation.annual.threshold")
-        probation_note = f"试用期绩效照算不发放；年累计超 {to_decimal(annual)} 元部分年底一次性结算"
+    ctx = _PreviewContext(session=session, payload=payload, take=take)
+    # 各段顺序固定 —— take 的调用顺序即 params_used 的展示顺序
+    decision = _preview_rate_decision(ctx)
+    earning = _preview_earning(ctx, decision)
+    barter = _preview_barter(ctx, earning.performance)
+    payment = _preview_payment(ctx, received, revenue, barter.base_amount)
+    probation_note = _preview_probation_note(ctx)
 
     return ParamPreviewOut(
-        payment_days=days,
-        rate=str(rate),
-        rate_text=rate_text,
-        profit_rate=str(profit_rate),
-        converted_profit=str(converted),
-        performance_amount=str(performance),
-        barter_applied=barter_applied,
-        performance_after_barter=str(after_barter) if after_barter is not None else None,
-        start_payment_ok=start_ok,
-        payment_weight_r=weight_r,
-        payment_weight=str(weight),
-        payable_amount=str(payable),
+        payment_days=decision.days,
+        rate=str(decision.rate),
+        rate_text=decision.rate_text,
+        profit_rate=str(earning.profit_rate),
+        converted_profit=str(earning.converted),
+        performance_amount=str(earning.performance),
+        barter_applied=barter.applied,
+        performance_after_barter=(
+            str(barter.after_barter) if barter.after_barter is not None else None
+        ),
+        start_payment_ok=payment.start_ok,
+        payment_weight_r=payment.weight_r,
+        payment_weight=str(payment.weight),
+        payable_amount=str(payment.payable),
         probation_note=probation_note,
         params_used=used,
     )

backend/app/services/personnel_service.py

@@ -70,20 +70,8 @@ def ensure_active(session: Session, personnel_id: int, *, label: str) -> Personn
 
 def list_personnel(session: Session, query: PersonnelQuery) -> PageResult[PersonnelOut]:
     """分页查询人员（含离职）。"""
-    rows = personnel_repository.list_personnel(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        keyword=query.keyword,
-        status=query.status,
-        business_role_id=query.business_role_id,
-    )
-    total = personnel_repository.count_personnel(
-        session,
-        keyword=query.keyword,
-        status=query.status,
-        business_role_id=query.business_role_id,
-    )
+    rows = personnel_repository.list_personnel(session, query)
+    total = personnel_repository.count_personnel(session, query)
     return PageResult[PersonnelOut](
         records=[
             _to_out(personnel, role_name, manager_name)

backend/app/services/product_service.py

@@ -99,20 +99,8 @@ def _raise_if_duplicate(exc: IntegrityError) -> None:
 
 def list_products(session: Session, query: ProductQuery) -> PageResult[ProductOut]:
     """分页查询产品（含停用）。"""
-    products = product_repository.list_products(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        keyword=query.keyword,
-        product_category=query.product_category,
-        status=query.status,
-    )
-    total = product_repository.count_products(
-        session,
-        keyword=query.keyword,
-        product_category=query.product_category,
-        status=query.status,
-    )
+    products = product_repository.list_products(session, query)
+    total = product_repository.count_products(session, query)
     return PageResult[ProductOut](
         records=[_to_out(product) for product in products],
         total=total,

backend/app/services/receipt_service.py

@@ -23,6 +23,7 @@
 """
 
 import logging
+from dataclasses import dataclass
 from datetime import date
 from decimal import Decimal
 
@@ -39,7 +40,7 @@ from app.core.constants import (
 from app.core.db_errors import is_duplicate_entry
 from app.core.exceptions import BadRequestError, NotFoundError
 from app.core.response import PageResult
-from app.models import Receipt
+from app.models import Business, Receipt
 from app.models.business import BUSINESS_STATUS_ACTIVE
 from app.models.receipt import SETTLEMENT_DONE, SETTLEMENT_PENDING
 from app.repositories import business_repository, receipt_repository
@@ -129,24 +130,8 @@ def _ensure_business_writable(session: Session, business_id: int, action: str):
 
 def list_receipts(session: Session, query: ReceiptQuery) -> PageResult[ReceiptOut]:
     """分页查询回款（附业务号/发货单号与 payment_days 派生值）。"""
-    rows = receipt_repository.list_receipts(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        business_id=query.business_id,
-        date_from=query.date_from,
-        date_to=query.date_to,
-        is_barter=query.is_barter,
-        settlement_status=query.settlement_status,
-    )
-    total = receipt_repository.count_receipts(
-        session,
-        business_id=query.business_id,
-        date_from=query.date_from,
-        date_to=query.date_to,
-        is_barter=query.is_barter,
-        settlement_status=query.settlement_status,
-    )
+    rows = receipt_repository.list_receipts(session, query)
+    total = receipt_repository.count_receipts(session, query)
     return PageResult[ReceiptOut](
         records=[_to_out(row) for row in rows],
         total=total,
@@ -166,6 +151,57 @@ def get_detail(session: Session, receipt_id: int) -> ReceiptOut:
     return _to_out(row)
 
 
+@dataclass(frozen=True, slots=True)
+class _CreatedReceipt:
+    """建行结果（出参 + 固化的 ID/批次号，commit 后不再触碰 ORM 属性）。"""
+
+    out: ReceiptOut
+    receipt_id: int
+    batch_no: str
+
+
+def _create_receipt_once(
+    session: Session,
+    payload: ReceiptCreateIn,
+    business: Business | None,
+    batch_no: str,
+) -> _CreatedReceipt:
+    """一次建行尝试：准入校验 → 建行 → 重算合计 → 出参。
+
+    **不 commit**：提交与回滚由调用方（重试循环）统一控制，保证「一次尝试
+    = 一个事务边界」。
+    """
+    if business is None:
+        raise NotFoundError("业务不存在")
+    if business.status != BUSINESS_STATUS_ACTIVE:
+        raise BadRequestError("业务已完成或已终止，禁止新增回款")
+    # 月结锁定拦截（2026-10-15 补充需求）：比需求原文「提示」取硬，
+    # 一期安全侧（策略点注释）；未指定日期时以当天月近似
+    from app.services import settlement_service
+
+    _lock_month = (payload.receipt_date or date.today()).strftime("%Y-%m")
+    settlement_service.ensure_month_not_locked(session, _lock_month)
+    shipment_no, ship_date = _ensure_shipment_of_business(
+        session, payload.shipment_id, payload.business_id
+    )
+    receipt = receipt_repository.create(
+        session,
+        Receipt(
+            receipt_batch_no=batch_no,
+            business_id=payload.business_id,
+            shipment_id=payload.shipment_id,
+            receipt_date=payload.receipt_date,
+            amount=Decimal(payload.amount),
+            is_barter=payload.is_barter,
+            settlement_status=SETTLEMENT_PENDING,
+            remark=payload.remark,
+        ),
+    )
+    _recalc_totals(session, business)
+    out = _to_out((receipt, business.business_no, shipment_no, ship_date))
+    return _CreatedReceipt(out=out, receipt_id=receipt.id, batch_no=batch_no)
+
+
 def create_receipt(
     session: Session,
     payload: ReceiptCreateIn,
@@ -206,35 +242,7 @@ def create_receipt(
         # 锁业务行与生成批次号——成功/最终失败各只记一条审计，无递归
         for attempt in range(_NO_RETRY + 1):
             try:
-                if business is None:
-                    raise NotFoundError("业务不存在")
-                if business.status != BUSINESS_STATUS_ACTIVE:
-                    raise BadRequestError("业务已完成或已终止，禁止新增回款")
-                # 月结锁定拦截（2026-10-15 补充需求）：比需求原文「提示」取硬，
-                # 一期安全侧（策略点注释）；未指定日期时以当天月近似
-                from app.services import settlement_service
-
-                _lock_month = (payload.receipt_date or date.today()).strftime("%Y-%m")
-                settlement_service.ensure_month_not_locked(session, _lock_month)
-                shipment_no, ship_date = _ensure_shipment_of_business(
-                    session, payload.shipment_id, payload.business_id
-                )
-                receipt = receipt_repository.create(
-                    session,
-                    Receipt(
-                        receipt_batch_no=batch_no,
-                        business_id=payload.business_id,
-                        shipment_id=payload.shipment_id,
-                        receipt_date=payload.receipt_date,
-                        amount=Decimal(payload.amount),
-                        is_barter=payload.is_barter,
-                        settlement_status=SETTLEMENT_PENDING,
-                        remark=payload.remark,
-                    ),
-                )
-                _recalc_totals(session, business)
-                out = _to_out((receipt, business.business_no, shipment_no, ship_date))
-                new_id, new_no = receipt.id, batch_no  # 固化：commit 后不碰 ORM
+                created = _create_receipt_once(session, payload, business, batch_no)
                 session.commit()
                 break
             except IntegrityError as exc:
@@ -249,11 +257,11 @@ def create_receipt(
 
         logger.info(
             "新增回款 | receipt_id=%s | no=%s | business_id=%s",
-            new_id,
-            new_no,
+            created.receipt_id,
+            created.batch_no,
             payload.business_id,
         )
-        return out
+        return created.out
 
 
 def update_receipt(

backend/app/services/report_export.py

@@ -36,7 +36,6 @@ _DATE_FORMAT = "YYYY-MM-DD"
 _FORMULA_PREFIXES = ("=", "+", "-", "@", "\t")
 
 
-
 @dataclass(frozen=True, slots=True)
 class _Column:
     """列定义：key 取行 dict 字段，kind 决定单元格类型与格式。"""
@@ -403,7 +402,13 @@ def _write_department_document(
     company_share_count = int(company.get("company_share_count") or 0)
 
     _write_document_header(
-        sheet, len(_DEPARTMENT_WIDTHS), company_name, title, filters, template_version, operator_name
+        sheet,
+        len(_DEPARTMENT_WIDTHS),
+        company_name,
+        title,
+        filters,
+        template_version,
+        operator_name,
     )
 
     _write_block(
@@ -439,16 +444,12 @@ def _write_department_document(
                 "manager_name": "合计",
                 "member_count": sum(team["member_count"] for team in teams),
                 "team_base": _sum_amount(team["team_base"] for team in teams),
-                "team_share_amount": _sum_amount(
-                    team["team_share_amount"] for team in teams
-                ),
+                "team_share_amount": _sum_amount(team["team_share_amount"] for team in teams),
                 "team_subtotal": _sum_amount(team["team_subtotal"] for team in teams),
                 # 未生成月结时全部为 None，合计也留空（不能显示 0.00 误导）
                 "frozen_amount": (
                     _sum_amount(
-                        team["frozen_amount"]
-                        for team in teams
-                        if team["frozen_amount"] is not None
+                        team["frozen_amount"] for team in teams if team["frozen_amount"] is not None
                     )
                     if any(team["frozen_amount"] is not None for team in teams)
                     else None

backend/app/services/report_service.py

@@ -103,10 +103,7 @@ def _person_row(row: PerformanceLedger, role_names: dict[str, str]) -> dict:
 
 def _role_names(session) -> dict[str, str]:
     """岗位编码 → 名称一次查询（明细与部门行共用，避免 N+1）。"""
-    return {
-        r.role_code: r.role_name
-        for r in session.execute(select(BusinessRole)).scalars()
-    }
+    return {r.role_code: r.role_name for r in session.execute(select(BusinessRole)).scalars()}
 
 
 def _dec(value) -> Decimal | None:
@@ -178,11 +175,7 @@ def department_summary(session, query) -> dict:
         extra_ids = {
             target
             for pid in adjustments
-            if (
-                target := _adjustment_target(
-                    pid, member_map, manager_of, manager_role, include_mp
-                )
-            )
+            if (target := _adjustment_target(pid, member_map, manager_of, manager_role, include_mp))
             is not None
         } - set(manager_ids)
         if extra_ids:
@@ -193,9 +186,7 @@ def department_summary(session, query) -> dict:
         for team in teams_all
         if query.manager_id is None or team["manager_id"] == query.manager_id
     ]
-    unassigned_all = _unassigned_rows(
-        member_map, manager_of, manager_role, adjustments, include_mp
-    )
+    unassigned_all = _unassigned_rows(member_map, manager_of, manager_role, adjustments, include_mp)
     # 「仅看某经理团队」时未分配桶不随展示：全公司无经理人员与该团队无关
     unassigned = [] if query.manager_id is not None else unassigned_all
     company_share_amount = _company_share(company_base, share.company_pct)
@@ -335,9 +326,7 @@ def _reconciliation(
         (m["adjust_net"] for team in teams for m in team["members"]), Decimal("0")
     ) + sum((row["adjust_net"] for row in unassigned), Decimal("0"))
     team_base_total = sum((team["team_base"] for team in teams), Decimal("0"))
-    unassigned_performance = sum(
-        (row["performance_sum"] for row in unassigned), Decimal("0")
-    )
+    unassigned_performance = sum((row["performance_sum"] for row in unassigned), Decimal("0"))
     payable_total = (
         team_member_payable + unassigned_payable + aggregate_payable_total + adjust_net_total
     )
@@ -378,9 +367,7 @@ def _reconciliation(
 
 def _team_share(team_base: Decimal, team_pct: Fraction) -> Decimal:
     """12.5% 团队分成：全公司部分传 0，仅取 team_pct 项。"""
-    return manager_team_share(
-        team_base, Decimal("0"), team_pct=team_pct, company_pct=Fraction(0)
-    )
+    return manager_team_share(team_base, Decimal("0"), team_pct=team_pct, company_pct=Fraction(0))
 
 
 def _company_share(company_base: Decimal, company_pct: Fraction) -> Decimal:
@@ -483,12 +470,8 @@ def _team_relations(
     referenced = {p.manager_id for p in people if p.manager_id is not None}
     missing = referenced - seed
     if missing:
-        people.extend(
-            session.execute(select(Personnel).where(Personnel.id.in_(missing))).scalars()
-        )
-    role_by_id = {
-        r.id: r.role_code for r in session.execute(select(BusinessRole)).scalars()
-    }
+        people.extend(session.execute(select(Personnel).where(Personnel.id.in_(missing))).scalars())
+    role_by_id = {r.id: r.role_code for r in session.execute(select(BusinessRole)).scalars()}
     manager_of = {p.id: p.manager_id for p in people}
     manager_role = {p.id: role_by_id.get(p.business_role_id, "") for p in people}
     person_info = {
@@ -619,9 +602,7 @@ def _member_rows(
                 )
             )
         personal_in_team = (
-            pid == manager_id
-            and include_manager_personal
-            and member["has_manager_personal"]
+            pid == manager_id and include_manager_personal and member["has_manager_personal"]
         )
         if personal_in_team:
             rows.append(
@@ -893,9 +874,7 @@ def _group_rows(rows) -> list[dict]:
 def payment_detail(session, query) -> dict:
     """区间下钻分页（bucket 未注册 → 400）。"""
     bucket = _bucket_by_code(query.bucket)
-    rows = report_repository.list_payment_detail(
-        session, query, bucket, offset=query.offset, limit=query.page_size
-    )
+    rows = report_repository.list_payment_detail(session, query, bucket)
     return {
         "list": [_payment_row(r) for r in rows],
         "total": report_repository.count_payment_detail(session, query, bucket),

backend/app/services/role_service.py

@@ -68,14 +68,8 @@ def list_roles(session: Session, query: RoleQuery) -> PageResult[RoleManageOut]:
     与用户表单下拉（`/users/meta/roles`）口径不同：**管理列表含停用角色**，
     下拉只给启用项。
     """
-    roles = role_repository.list_roles(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        keyword=query.keyword,
-        status=query.status,
-    )
-    total = role_repository.count_roles(session, keyword=query.keyword, status=query.status)
+    roles = role_repository.list_roles(session, query)
+    total = role_repository.count_roles(session, query)
     counts = role_repository.count_users_by_role(session, [role.id for role in roles])
     return PageResult[RoleManageOut](
         records=[_to_out(role, counts.get(role.id, 0)) for role in roles],

backend/app/services/settlement_service.py

@@ -13,11 +13,13 @@ from datetime import date
 from decimal import ROUND_HALF_UP, Decimal
 
 from sqlalchemy import select
+from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
 from app.core.constants import (
     MODULE_SETTLEMENT,
 )
+from app.core.db_errors import is_duplicate_entry
 from app.core.exceptions import BadRequestError, NotFoundError
 from app.models import (
     MonthlySettlement,
@@ -248,19 +250,30 @@ def submit(
         ),
         meta=meta,
     ):
-        settlement = settlement_repository.get_by_id(session, settlement_id)
+        settlement = settlement_repository.get_by_id(session, settlement_id, for_update=True)
         if settlement is None:
             raise NotFoundError("月结不存在")
         if settlement.status not in (SETTLEMENT_STATUS_DRAFT, SETTLEMENT_STATUS_REJECTED):
             raise BadRequestError("仅草稿或已驳回状态可提交")
 
-        order = approval_service.submit_order(
-            session,
-            order_type=_SETTLEMENT_ORDER_TYPE,
-            ref_id=settlement.id,
-            refer_no=f"JS{settlement.settlement_month.replace('-', '')}",
-            submitter=operator,
-        )
+        try:
+            order = approval_service.submit_order(
+                session,
+                order_type=_SETTLEMENT_ORDER_TYPE,
+                ref_id=settlement.id,
+                refer_no=f"JS{settlement.settlement_month.replace('-', '')}",
+                submitter=operator,
+                # 月结明细在生成时已固化；提交时快照让复核者看到的是送审版本，
+                # 而不是审批期间若发生重生成后的内容
+                snapshot=approval_service.settlement_content(session, settlement.id),
+            )
+        except IntegrityError as exc:
+            # version_no 唯一键的并发冲突（提交前已锁月结行，此项理论不可达，
+            # 留作 DB 兜底）：审计上下文会统一回滚，这里只把 1062 翻译成
+            # 可重试的业务提示，其余完整性错误原样上抛
+            if is_duplicate_entry(exc):
+                raise BadRequestError("提交冲突（版本号生成重复），请重试") from exc
+            raise
         settlement.approval_order_no = order.order_no
         if order.status == SETTLEMENT_STATUS_APPROVED:
             mark_locked(session, settlement, operator=operator)
@@ -410,9 +423,10 @@ def _aggregate_row(
                 "members": member_names,
                 "params_as_of": str(share.param_date),
                 "manager_personal_included": 1 if share.include_manager_personal else 0,
+                # 说明写给业务人员看：说清这一笔的基数是怎么凑出来的
                 "note": (
-                    f"#12 参数化：基数{'含' if share.include_manager_personal else '不含'}"
-                    "经理个人户业绩；9.8 方案公式"
+                    f"团队分成：基数{'含' if share.include_manager_personal else '不含'}"
+                    "经理本人名下业绩，按本团队与全公司两级比例合计计提"
                 ),
             },
         ),
@@ -573,6 +587,7 @@ def _wire_adjustment_approval(session: Session, order, operator: User) -> None:
         ref_id=order.id,
         refer_no=order.order_no,
         submitter=operator,
+        snapshot=approval_service.adjustment_content(session, order.id),
     )
     order.status = (
         ADJUSTMENT_STATUS_EFFECTIVE

backend/app/services/shipment_service.py

@@ -157,20 +157,8 @@ def _is_duplicate_entry(exc: IntegrityError) -> bool:
 
 def list_shipments(session: Session, query: ShipmentQuery) -> PageResult[ShipmentOut]:
     """分页查询发货单（附业务号；business_id 可空=全量，一期按业务维度查看为主）。"""
-    rows = shipment_repository.list_shipments(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        business_id=query.business_id,
-        ship_date_from=query.ship_date_from,
-        ship_date_to=query.ship_date_to,
-    )
-    total = shipment_repository.count_shipments(
-        session,
-        business_id=query.business_id,
-        ship_date_from=query.ship_date_from,
-        ship_date_to=query.ship_date_to,
-    )
+    rows = shipment_repository.list_shipments(session, query)
+    total = shipment_repository.count_shipments(session, query)
     return PageResult[ShipmentOut](
         records=[_to_out(row) for row in rows],
         total=total,

backend/app/services/tplus/README.md

@@ -0,0 +1,23 @@
+# T+ 只读同步
+
+> AI 摘要：读取外部 SQL Server 的原始快照，并管理持久化任务。关键文件为 task_service、worker、sync_service、relationship_audit 和 query_service；管理员 API 为 `/api/v1/system/tplus`。
+
+## 职责与边界
+
+- `task_service` 创建/重试任务，控制请求幂等和事务，保留原批次与异常；HTTP 不执行源库长查询。
+- `worker.run_once` 取得 MySQL 执行锁后恢复中断任务、领取一个任务；读取和关联核验共用锁。
+- `sync_service` 主键分页读取白名单实体，每页保存原始快照，相同源 ID 和行哈希去重。
+- `relationship_audit` 执行 `db/queries/tplus/audit/queries.sql`，只返回系统字典和数量；按收款方向与单据类型核对关联。
+- `query_service` 分页元数据，按需加载白名单报告；不返回原始 payload、驱动异常或源配置。
+
+## 主要入口
+
+管理员 API 提供概况、批次/异常分页、详情、创建与失败重试。CLI 为 `python -m app.cli tplus-worker --once/--watch`；原 `tplus-stage --entity personnel` 可继续直接读取。
+
+连接配置、启动命令和异常处理参见项目 `docs/T+本地恢复与同步开发手册.md`。后端与 worker 必须指向同一 MySQL 业务库和 T+ 账套，查询文件随部署交付。
+
+应用与兼容性脚本共用 `Settings.t_plus_database_url` 和连接文件校验；分号、花括号由 SQLAlchemy ODBC 方言转义。默认启用加密，失败不自动降级；既有私有配置需要按运行手册核对。管理页面不返回原始数据，但暂存 JSON 的生产敏感字段保护仍待 S1 决策。
+
+## 验收限制
+
+当前只写 MySQL 暂存与管理表，不写工资业务表、不推进时间水位。关联、成本、日期和负向事件的业务规则待客户确认。开发验收先以备份源 10.50.1600 为基准，生产完整版本后补；2022 副本结果不能替代该基准的实测。

backend/app/services/tplus/__init__.py

@@ -0,0 +1,6 @@
+"""T+ 只读读取、字段映射和同步 staging。"""
+
+from app.services.tplus.reader import TPlusReader
+from app.services.tplus.sync_service import TPlusSyncService
+
+__all__ = ["TPlusReader", "TPlusSyncService"]

backend/app/services/tplus/mapping.py

@@ -0,0 +1,136 @@
+"""T+ 字段映射元数据。
+
+这里记录候选字段和确认状态，不把字典中的候选直接当成业务契约。
+高风险口径在客户确认前保持 ``PENDING``，同步器只允许写入原始 staging。
+"""
+
+from dataclasses import dataclass
+from enum import StrEnum
+
+
+class MappingStatus(StrEnum):
+    """字段映射的证据状态。"""
+
+    CONFIRMED = "CONFIRMED"
+    CANDIDATE = "CANDIDATE"
+    MISSING = "MISSING"
+    PENDING = "PENDING"
+
+
+@dataclass(frozen=True, slots=True)
+class TPlusMapping:
+    """一个实体的源表、候选字段和目标表说明。"""
+
+    entity_type: str
+    source_table: str
+    source_fields: tuple[str, ...]
+    target_table: str
+    status: MappingStatus
+    unresolved: tuple[str, ...] = ()
+    version: str = "0.2"
+    schema_evidence: str = "ceshi20261026.bak，2026-10-27 恢复副本结构核验"
+
+
+TPLUS_MAPPINGS: tuple[TPlusMapping, ...] = (
+    TPlusMapping(
+        "personnel",
+        "AA_Person",
+        ("id", "code", "name", "disabled", "mobilePhoneNo", "officePhoneNo", "iddepartment"),
+        "personnel",
+        MappingStatus.CANDIDATE,
+        ("工号是否等于 code", "经理隶属关系", "入职日期", "试用期"),
+    ),
+    TPlusMapping(
+        "products",
+        "AA_Inventory",
+        ("id", "code", "name", "specification", "idunit", "idinventoryclass"),
+        "products",
+        MappingStatus.CANDIDATE,
+        ("库存数量语义", "参考价来源", "系统产品分类"),
+    ),
+    TPlusMapping(
+        "customers",
+        "AA_Partner",
+        ("id", "code", "name", "Contact", "representative", "disabled", "updated"),
+        "customers",
+        MappingStatus.CANDIDATE,
+        ("客户唯一键", "改名/合并规则", "系统客户分类"),
+    ),
+    TPlusMapping(
+        "businesses",
+        "SA_SaleOrder",
+        ("id", "code", "voucherdate", "idcustomer", "idclerk", "idbusinesstype", "updated"),
+        "businesses",
+        MappingStatus.PENDING,
+        ("成本口径", "业务负责人覆盖规则", "含税口径"),
+    ),
+    TPlusMapping(
+        "shipments",
+        "SA_SaleDelivery",
+        (
+            "id",
+            "code",
+            "voucherdate",
+            "SendDate",
+            "deliveryDate",
+            "auditeddate",
+            "SourceVoucherCode",
+            "updated",
+        ),
+        "shipments",
+        MappingStatus.PENDING,
+        ("发货日期", "订单关联路径", "退货/红冲规则", "运费来源"),
+    ),
+    TPlusMapping(
+        "receipts",
+        "ARAP_ReceivePayment",
+        (
+            "id",
+            "code",
+            "voucherDate",
+            "auditedDate",
+            "createdTime",
+            "SaleOrderCode",
+            "SaleOrderID",
+            "ts",
+        ),
+        "receipts",
+        MappingStatus.PENDING,
+        ("到账日期", "收款方式", "核销状态", "回款与发货关联"),
+    ),
+)
+
+DETAIL_MAPPINGS: tuple[TPlusMapping, ...] = (
+    TPlusMapping(
+        "business_items",
+        "SA_SaleOrder_b",
+        ("id", "idSaleOrderDTO", "LatestCost"),
+        "business_items",
+        MappingStatus.PENDING,
+        ("成本口径", "含税口径"),
+    ),
+    TPlusMapping(
+        "shipment_items",
+        "SA_SaleDelivery_b",
+        ("ID", "idSaleDeliveryDTO", "saleOrderId"),
+        "shipment_items",
+        MappingStatus.PENDING,
+        ("发货归属", "红冲和退货"),
+    ),
+    TPlusMapping(
+        "receipt_items",
+        "ARAP_ReceivePayment_b",
+        ("id", "idArapReceivePaymentDTO", "saleorderid"),
+        "receipts",
+        MappingStatus.PENDING,
+        ("回款拆分", "核销语义", "增量水位"),
+    ),
+)
+
+
+def get_mapping(entity_type: str) -> TPlusMapping:
+    """按实体类型取得映射；未知实体直接拒绝。"""
+    for mapping in (*TPLUS_MAPPINGS, *DETAIL_MAPPINGS):
+        if mapping.entity_type == entity_type:
+            return mapping
+    raise KeyError(f"未知 T+ 实体类型：{entity_type}")

backend/app/services/tplus/query_service.py

@@ -0,0 +1,122 @@
+"""同步管理只读查询；只公开受控元数据，不公开源数据和原异常。"""
+
+from app.core.exceptions import NotFoundError
+from app.core.response import PageResult
+from app.core.tplus_db import resolve_tplus_settings
+from app.repositories import tplus_repository as repository
+from app.schemas.tplus import (
+    TPlusBatchDetailOut,
+    TPlusBatchOut,
+    TPlusErrorOut,
+    TPlusOverviewOut,
+)
+from app.services.tplus import task_service
+from app.services.tplus.mapping import DETAIL_MAPPINGS, TPLUS_MAPPINGS
+from app.services.tplus.relationship_audit import public_report
+
+ERROR_MESSAGES = {
+    "READ_ERROR": "源数据读取失败，请检查只读连接和字段映射",
+    "STAGE_ERROR": "原始数据保存失败，请核对字段类型后重试",
+    "INTERRUPTED": "执行进程中断，可重新读取，已保存数据会自动去重",
+    "EXECUTION_ERROR": "任务执行失败，请检查只读连接和字段映射",
+    "TYPE_UNRESOLVED": "关联核验所需的单据类型缺失或不唯一",
+}
+_ENTITY_TYPES = {item.entity_type for item in (*TPLUS_MAPPINGS, *DETAIL_MAPPINGS)}
+
+
+def batch_out(batch, active_parents: set[str]) -> TPlusBatchOut:
+    """只使用元数据字段组装输出，历史批次范围保持未知。"""
+    entities = batch.entity_types_json
+    scope_valid = (
+        isinstance(entities, list)
+        and bool(entities)
+        and all(isinstance(item, str) and item in _ENTITY_TYPES for item in entities)
+    )
+    return TPlusBatchOut(
+        batch_no=batch.batch_no,
+        task_type="AUDIT" if batch.sync_mode == "AUDIT" else "READ",
+        status=batch.status,
+        entity_types=entities if scope_valid else None,
+        source_record_count=batch.source_record_count,
+        staged_record_count=batch.staged_record_count,
+        skipped_record_count=batch.skipped_record_count,
+        failed_record_count=batch.failed_record_count,
+        parent_batch_no=batch.parent_batch_no,
+        create_time=batch.create_time,
+        started_at=batch.started_at,
+        finished_at=batch.finished_at,
+        can_retry=batch.status in ("FAILED", "PARTIAL")
+        and (batch.sync_mode == "AUDIT" or (batch.sync_mode == "FULL" and scope_valid))
+        and batch.batch_no not in active_parents,
+        error_summary="任务有异常，请查看异常明细"
+        if batch.status in ("FAILED", "PARTIAL")
+        else None,
+    )
+
+
+def get_batch(session, number: str) -> TPlusBatchDetailOut:
+    """按需加载报告，未找到时统一返回 404。"""
+    batch = repository.get_batch(session, number, with_result=True)
+    if batch is None:
+        raise NotFoundError("同步批次不存在")
+    out = batch_out(batch, repository.active_retry_parents(session, [number]))
+    return TPlusBatchDetailOut(
+        **out.model_dump(), audit_result=public_report(batch.audit_result_json)
+    )
+
+
+def list_batches(session, query):
+    """分页批次并批量查询重试状态。"""
+    rows, total = repository.list_batches(session, query)
+    active = repository.active_retry_parents(session, [row.batch_no for row in rows])
+    return PageResult[TPlusBatchOut](
+        records=[batch_out(row, active) for row in rows],
+        total=total,
+        page_num=query.page_num,
+        page_size=query.page_size,
+    )
+
+
+def _safe_id(value) -> str | None:
+    if isinstance(value, str) and value.isascii() and value.isdigit() and len(value) <= 10:
+        return value if int(value) <= 2147483647 else None
+    return None
+
+
+def list_errors(session, number, query):
+    """错误按编码生成固定说明，避免驱动异常泄漏连接参数。"""
+    if repository.get_batch(session, number) is None:
+        raise NotFoundError("同步批次不存在")
+    rows, total = repository.list_errors(session, number, query)
+    records = [
+        TPlusErrorOut(
+            id=row.id,
+            entity_type=row.entity_type if row.entity_type in _ENTITY_TYPES else "system",
+            source_id=_safe_id(row.source_id),
+            error_code=row.error_code if row.error_code in ERROR_MESSAGES else "UNKNOWN",
+            error_message=ERROR_MESSAGES.get(row.error_code, "任务未完成，请核对配置后重试"),
+            retry_count=row.retry_count,
+            create_time=row.create_time,
+        )
+        for row in rows
+    ]
+    return PageResult[TPlusErrorOut](
+        records=records, total=total, page_num=query.page_num, page_size=query.page_size
+    )
+
+
+def overview(session) -> TPlusOverviewOut:
+    """只校验配置，不在页面查询期间建立 T+ 连接。"""
+    try:
+        resolve_tplus_settings(task_service.get_settings())
+        configured, message = True, "只读配置已就绪，实际连接结果以任务执行为准"
+    except RuntimeError:
+        configured, message = False, "T+ 只读配置未就绪，请联系管理员配置"
+    latest = repository.latest_audit(session)
+    return TPlusOverviewOut(
+        is_configured=configured,
+        configuration_message=message,
+        status_counts=repository.status_counts(session),
+        total_staged_records=repository.total_staged_records(session),
+        latest_audit=public_report(latest.audit_result_json) if latest else None,
+    )

backend/app/services/tplus/reader.py

@@ -0,0 +1,95 @@
+"""T+ 六类源表的只读读取器。"""
+
+from collections.abc import Iterator
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+from sqlalchemy import BigInteger, Integer, bindparam, text
+from sqlalchemy.orm import Session
+
+from app.services.tplus.mapping import TPLUS_MAPPINGS, get_mapping
+
+_ENTITY_QUERY_FILES = {
+    "personnel": "personnel.sql",
+    "products": "products.sql",
+    "customers": "customers.sql",
+    "businesses": "businesses.sql",
+    "shipments": "shipments.sql",
+    "receipts": "receipts.sql",
+    "business_items": "business_items.sql",
+    "shipment_items": "shipment_items.sql",
+    "receipt_items": "receipt_items.sql",
+}
+
+
+class TPlusReader:
+    """执行白名单 SQL 文件，禁止调用方传入任意表名或 SQL。"""
+
+    def __init__(self, session: Session, query_dir: Path | None = None) -> None:
+        self.session = session
+        self.query_dir = query_dir or self._default_query_dir()
+
+    @staticmethod
+    def _default_query_dir() -> Path:
+        return Path(__file__).resolve().parents[4] / "db" / "queries" / "tplus"
+
+    def iter_entity(
+        self,
+        entity_type: str,
+        *,
+        watermark: datetime | None = None,
+        batch_size: int = 500,
+    ) -> Iterator[dict[str, Any]]:
+        """按实体流式读取，水位只通过绑定参数传入。"""
+        if entity_type not in _ENTITY_QUERY_FILES:
+            raise KeyError(f"不支持的 T+ 实体类型：{entity_type}")
+        if batch_size < 1 or batch_size > 5000:
+            raise ValueError("batch_size 必须在 1 到 5000 之间")
+        if watermark is not None:
+            raise ValueError("增量水位语义尚未确认，当前只支持全量读取和哈希去重")
+        get_mapping(entity_type)
+        sql_path = self.query_dir / _ENTITY_QUERY_FILES[entity_type]
+        if not sql_path.is_file():
+            raise FileNotFoundError(f"缺少 T+ 查询文件：{sql_path}")
+
+        statement = text(sql_path.read_text(encoding="utf-8")).bindparams(
+            bindparam("batch_size", type_=Integer()), bindparam("after_id", type_=BigInteger())
+        )
+        after_id = -2147483649  # 所有源表主键均已核验为 SQL Server int，包含其最小值。
+        while True:
+            result = self.session.execute(
+                statement, {"batch_size": batch_size, "after_id": after_id}
+            )
+            try:
+                rows = [dict(row) for row in result.mappings().all()]
+            finally:
+                result.close()
+            if not rows:
+                return
+            next_id = int(rows[-1]["id"])
+            if next_id <= after_id:
+                raise ValueError("T+ 源主键未递增，停止读取以避免重复循环")
+            yield from rows
+            after_id = next_id
+
+    def fetch_entity(
+        self,
+        entity_type: str,
+        *,
+        watermark: datetime | None = None,
+        max_rows: int = 5000,
+    ) -> list[dict[str, Any]]:
+        """读取有限数量的实体行，供核验命令和单元测试使用。"""
+        if max_rows < 1:
+            raise ValueError("max_rows 必须大于 0")
+        rows: list[dict[str, Any]] = []
+        for row in self.iter_entity(entity_type, watermark=watermark):
+            rows.append(row)
+            if len(rows) >= max_rows:
+                break
+        return rows
+
+    def inventory(self) -> tuple[str, ...]:
+        """返回当前适配器支持的实体类型。"""
+        return tuple(mapping.entity_type for mapping in TPLUS_MAPPINGS)

backend/app/services/tplus/relationship_audit.py

@@ -0,0 +1,135 @@
+"""读取实际系统字典并统计带单据类型的收款关联，不推断计提口径。"""
+
+import re
+from datetime import datetime
+from pathlib import Path
+
+from sqlalchemy import text
+from sqlalchemy.orm import Session
+
+from app.core.logging import scrub
+from app.services.tplus.reader import TPlusReader
+
+_COLUMNS = {
+    "source_version": ("source_version",),
+    "voucher_types": ("id", "code", "name"),
+    "state_enums": ("id", "code", "name", "enum_name"),
+    "receipt_kinds": ("voucher_type_code", "voucher_type_name", "receive_flag", "row_count"),
+    "receipt_line_kinds": ("voucher_type_id", "voucher_type_name", "receive_flag", "row_count"),
+    "sales_links": (
+        "row_count",
+        "delivery_matches",
+        "detail_matches",
+        "direct_order_matches",
+        "detail_order_matches",
+        "unambiguous_order_matches",
+        "invalid_order_ids",
+        "conflicting_order_ids",
+    ),
+    "order_coverage": ("row_count", "without_order", "single_order", "multiple_orders"),
+}
+PENDING_CONFIRMATIONS = (
+    ("cost_basis", "绩效成本取值", "订单最新成本或出库成本的类型、时点和含税口径需业务确认"),
+    ("receipt_date", "到账日", "单据日、审核日和明细 arrivalDate 的业务含义需确认"),
+    ("ship_date", "发货日", "回款天数应使用哪个发货日期需确认"),
+    ("allocation", "多单分摊", "一笔收款涉及多张销货单或订单时，不能自动把整笔金额记给一单"),
+    ("negative_events", "退货与红冲", "负数、冲销及已结算业务的调整方式需确认"),
+)
+
+
+def public_report(raw: dict | None) -> dict | None:
+    """历史报告也按当前字段白名单输出，禁止透传未知嵌套数据。"""
+    if not isinstance(raw, dict):
+        return None
+    result = {
+        key: TPlusRelationshipAudit._safe_value(raw.get(key))
+        for key in (
+            "schema_version",
+            "captured_at",
+            "source_version",
+            "is_target_engine",
+            "relationship_status",
+        )
+    }
+    result["compatibility_status"] = "PENDING"
+    for section, columns in _COLUMNS.items():
+        if section in ("source_version", "voucher_types"):
+            continue
+        value = raw.get(section)
+        if isinstance(value, list):
+            result[section] = [
+                {key: TPlusRelationshipAudit._safe_value(row.get(key)) for key in columns}
+                for row in value[:200]
+                if isinstance(row, dict)
+            ]
+        elif isinstance(value, dict):
+            result[section] = {
+                key: TPlusRelationshipAudit._safe_value(value.get(key)) for key in columns
+            }
+        else:
+            result[section] = None
+    result["pending_confirmations"] = [
+        {"key": k, "title": t, "reason": r} for k, t, r in PENDING_CONFIRMATIONS
+    ]
+    return result
+
+
+class TPlusRelationshipAudit:
+    """执行固定核验查询；结果只包含系统字典和数量，不包含客户记录。"""
+
+    def __init__(self, session: Session, query_dir: Path | None = None) -> None:
+        self.session = session
+        root = query_dir or TPlusReader._default_query_dir()
+        source = (root / "audit" / "queries.sql").read_text(encoding="utf-8")
+        sections = re.split(r"(?m)^-- @query (\w+)\s*$", source)
+        self.queries = dict(zip(sections[1::2], sections[2::2], strict=True))
+        if set(self.queries) != set(_COLUMNS):
+            raise ValueError("关联核验查询清单不完整")
+
+    def fetch(self, name: str, **params) -> list[dict]:
+        """只接受登记的查询名与绑定参数，过滤非合同字段。"""
+        if name not in self.queries:
+            raise ValueError("不支持的核验查询")
+        result = self.session.execute(text(self.queries[name]), params)
+        try:
+            return [
+                {key: self._safe_value(row.get(key)) for key in _COLUMNS[name]}
+                for row in result.mappings().all()
+            ]
+        finally:
+            result.close()
+
+    @staticmethod
+    def _safe_value(value):
+        if value is None or isinstance(value, (int, bool)):
+            return value
+        return scrub(re.sub(r"[\x00-\x1f]", " ", str(value)))[:200]
+
+    def run(self) -> dict:
+        """生成事实报告；类型缺失或有歧义时不执行该关联链路。"""
+        version = self.fetch("source_version")[0]["source_version"]
+        types = self.fetch("voucher_types")
+        report = {
+            "schema_version": 1,
+            "captured_at": datetime.now().isoformat(),
+            "source_version": version,
+            "is_target_engine": str(version).startswith("10.50."),
+            "compatibility_status": "PENDING",
+            "state_enums": self.fetch("state_enums"),
+            "receipt_kinds": self.fetch("receipt_kinds"),
+            "receipt_line_kinds": self.fetch("receipt_line_kinds"),
+            "sales_links": None,
+            "order_coverage": None,
+            "relationship_status": "UNRESOLVED",
+            "relationship_basis": "主表按 isReceiveFlag=1（是否收款）筛选；明细按 SM_VoucherType 的 SA04 类型关联。主表 idvouchertype 是来源类型，不能据此筛本单类型。",
+            "pending_confirmations": [
+                {"key": k, "title": t, "reason": r} for k, t, r in PENDING_CONFIRMATIONS
+            ],
+        }
+        sale = [r["id"] for r in types if r["code"] == "SA04"]
+        if len(sale) == 1:
+            params = {"sale_delivery_type_id": sale[0], "receive_flag": 1}
+            report["sales_links"] = self.fetch("sales_links", **params)[0]
+            report["order_coverage"] = self.fetch("order_coverage", **params)[0]
+            report["relationship_status"] = "OBSERVED"
+        return report

backend/app/services/tplus/staging.py

@@ -0,0 +1,123 @@
+"""T+ 原始行 staging 服务。"""
+
+import hashlib
+import json
+from datetime import date, datetime
+from decimal import Decimal
+from typing import Any
+
+from sqlalchemy import insert, select, tuple_
+from sqlalchemy.orm import Session
+
+from app.models import TPlusSyncRecord
+
+
+def _json_default(value: Any) -> str:
+    """将日期、Decimal 等源字段转换为稳定 JSON 文本。"""
+    if isinstance(value, (date, datetime)):
+        return value.isoformat()
+    if isinstance(value, Decimal):
+        return str(value)
+    if isinstance(value, (bytes, bytearray, memoryview)):
+        return "0x" + bytes(value).hex()
+    raise TypeError("T+ 源行包含无法序列化的数据类型")
+
+
+def stage_row(
+    session: Session,
+    *,
+    batch_no: str,
+    entity_type: str,
+    source_table: str,
+    row: dict[str, Any],
+) -> tuple[TPlusSyncRecord, bool]:
+    """幂等写入单条原始行；返回记录和是否新插入。"""
+    values = prepare_row(batch_no, entity_type, source_table, row)
+    existing = session.execute(
+        select(TPlusSyncRecord).where(
+            TPlusSyncRecord.entity_type == entity_type,
+            TPlusSyncRecord.source_id == values["source_id"],
+            TPlusSyncRecord.row_hash == values["row_hash"],
+        )
+    ).scalar_one_or_none()
+    if existing is not None:
+        return existing, False
+
+    record = TPlusSyncRecord(**values)
+    session.add(record)
+    session.flush()
+    return record, True
+
+
+def prepare_row(batch_no: str, entity_type: str, source_table: str, row: dict) -> dict:
+    """规范化原始值，生成与 JSON 列序列化一致的哈希。"""
+    source_id = str(row["id"]) if row.get("id") is not None else ""
+    if not source_id:
+        raise ValueError("T+ 源行缺少 id，无法建立幂等键")
+    source_no = row.get("code")
+    payload = json.dumps(
+        row, ensure_ascii=False, sort_keys=True, default=_json_default, allow_nan=False
+    )
+    return {
+        "batch_no": batch_no,
+        "entity_type": entity_type,
+        "source_table": source_table,
+        "source_id": source_id,
+        "source_no": str(source_no).strip() if source_no is not None else None,
+        "source_line_no": str(row.get("line_no")) if row.get("line_no") is not None else None,
+        "source_updated_at": _source_datetime(row),
+        "row_hash": hashlib.sha256(payload.encode("utf-8")).hexdigest(),
+        "payload_json": json.loads(payload),
+    }
+
+
+def stage_rows(session: Session, batch_no: str, mapping, rows: list[dict]) -> tuple[int, list]:
+    """批量去重和插入；数据库拒绝整批时逐条隔离，保留有效行。"""
+    records, errors = {}, []
+    for row in rows:
+        try:
+            values = prepare_row(batch_no, mapping.entity_type, mapping.source_table, row)
+            records[(values["source_id"], values["row_hash"])] = values
+        except (ValueError, TypeError) as exc:
+            errors.append((row, type(exc).__name__))
+    if not records:
+        return 0, errors
+    existing = set(
+        session.execute(
+            select(TPlusSyncRecord.source_id, TPlusSyncRecord.row_hash).where(
+                TPlusSyncRecord.entity_type == mapping.entity_type,
+                tuple_(TPlusSyncRecord.source_id, TPlusSyncRecord.row_hash).in_(list(records)),
+            )
+        ).all()
+    )
+    pending = [v for k, v in records.items() if k not in existing]
+    if not pending:
+        return 0, errors
+    try:
+        with session.begin_nested():
+            session.execute(insert(TPlusSyncRecord), pending)
+        return len(pending), errors
+    except Exception:
+        # 唯一约束竞争或坏行触发回滚后，再逐条处理，不能丢掉同批有效行。
+        inserted = 0
+        for row in rows:
+            try:
+                with session.begin_nested():
+                    _, is_new = stage_row(
+                        session,
+                        batch_no=batch_no,
+                        entity_type=mapping.entity_type,
+                        source_table=mapping.source_table,
+                        row=row,
+                    )
+                inserted += int(is_new)
+            except Exception as exc:
+                if not any(old is row for old, _ in errors):
+                    errors.append((row, type(exc).__name__))
+        return inserted, errors
+
+
+def _source_datetime(row: dict[str, Any]) -> datetime | None:
+    """只接收真实时间值；SQL Server rowversion 不能当作日期。"""
+    value = row.get("source_updated_at") or row.get("updated")
+    return value if isinstance(value, datetime) else None

backend/app/services/tplus/sync_service.py

@@ -0,0 +1,131 @@
+"""T+ 只读快照同步：批量 staging、幂等去重和失败隔离。"""
+
+from collections.abc import Iterable
+from datetime import datetime
+from itertools import batched
+from pathlib import Path
+
+from sqlalchemy import func, select
+from sqlalchemy.orm import Session
+
+from app.core.config import get_settings
+from app.core.tplus_lock import sync_lock
+from app.models import TPlusSyncBatch, TPlusSyncError
+from app.services.tplus.mapping import TPlusMapping, get_mapping
+from app.services.tplus.reader import TPlusReader
+from app.services.tplus.staging import stage_rows
+from app.services.tplus.task_service import fail_batch, new_batch_no, recover_running
+
+
+class TPlusSyncService:
+    """按分页提交原始快照，业务表始终不参与本服务。"""
+
+    def __init__(self, source_session: Session, target_session: Session) -> None:
+        self.source_session = source_session
+        self.target_session = target_session
+        configured = get_settings().t_plus_query_dir
+        self.reader = TPlusReader(source_session, Path(configured) if configured else None)
+
+    def run(
+        self,
+        *,
+        entity_types: Iterable[str],
+        watermark: datetime | None = None,
+        batch_no: str | None = None,
+        batch_size: int = 500,
+    ) -> TPlusSyncBatch:
+        """执行可重复的全量读取，提交每页进度；字段语义未定时拒绝时间增量。"""
+        entities = tuple(dict.fromkeys(entity_types))
+        if not entities or not 1 <= batch_size <= 5000:
+            raise ValueError("至少指定一个实体，batch_size 必须在 1 到 5000 之间")
+        if watermark is not None:
+            raise ValueError("增量水位语义尚未确认，当前只支持全量读取和哈希去重")
+        for entity in entities:
+            get_mapping(entity)
+        with sync_lock(self.target_session):
+            recover_running(self.target_session)
+            batch = self._start_direct(entities, batch_no)
+            try:
+                return self.execute_claimed(batch, batch_size)
+            except BaseException:
+                self.target_session.rollback()
+                fail_batch(self.target_session, batch, "INTERRUPTED")
+                raise
+
+    def _start_direct(self, entities, batch_no):
+        batch = TPlusSyncBatch(
+            batch_no=batch_no or new_batch_no(),
+            sync_mode="FULL",
+            status="RUNNING",
+            source_database=get_settings().t_plus_database or None,
+            started_at=datetime.now(),
+            source_record_count=0,
+            staged_record_count=0,
+            failed_record_count=0,
+            skipped_record_count=0,
+            entity_types_json=list(entities),
+        )
+        self.target_session.add(batch)
+        self.target_session.commit()
+        return batch
+
+    def execute_claimed(self, batch: TPlusSyncBatch, batch_size: int = 500):
+        """处理已持锁且标记 RUNNING 的任务；仅由直接入口或 worker 调用。"""
+        if batch.status != "RUNNING" or not batch.entity_types_json:
+            raise ValueError("任务尚未领取或读取范围不完整")
+        mappings = [get_mapping(entity) for entity in batch.entity_types_json]
+        for mapping in mappings:
+            try:
+                self._read_entity(batch, mapping, batch_size)
+            except Exception as exc:
+                self.target_session.rollback()
+                self._record_error(batch, mapping, {}, "READ_ERROR", type(exc).__name__)
+                self.target_session.commit()
+        batch.status = "SUCCEEDED"
+        if batch.failed_record_count:
+            failed_rows = (
+                self.target_session.scalar(
+                    select(func.count())
+                    .select_from(TPlusSyncError)
+                    .where(
+                        TPlusSyncError.batch_no == batch.batch_no,
+                        TPlusSyncError.error_code == "STAGE_ERROR",
+                    )
+                )
+                or 0
+            )
+            batch.status = "PARTIAL" if batch.source_record_count > failed_rows else "FAILED"
+            batch.error_message = "部分数据未完成，请查询同步错误记录并重跑全量快照"
+        batch.finished_at = datetime.now()
+        batch.watermark_to = None
+        self.target_session.commit()
+        return batch
+
+    def _read_entity(self, batch: TPlusSyncBatch, mapping: TPlusMapping, size: int) -> None:
+        """批量写入一类数据；每页提交后即使进程中断也可幂等重跑。"""
+        source = self.reader.iter_entity(mapping.entity_type, batch_size=size)
+        for chunk in batched(source, size):
+            inserted, errors = stage_rows(self.target_session, batch.batch_no, mapping, list(chunk))
+            batch.source_record_count += len(chunk)
+            batch.staged_record_count += inserted
+            batch.skipped_record_count += len(chunk) - inserted - len(errors)
+            for row, reason in errors:
+                self._record_error(batch, mapping, row, "STAGE_ERROR", reason)
+            self.target_session.commit()
+
+    def _record_error(
+        self, batch: TPlusSyncBatch, mapping: TPlusMapping, row: dict, code: str, reason: str
+    ) -> None:
+        """只记录错误类别和源 ID，避免数据库异常携带个人信息或凭据。"""
+        self.target_session.add(
+            TPlusSyncError(
+                batch_no=batch.batch_no,
+                entity_type=mapping.entity_type,
+                source_table=mapping.source_table,
+                source_id=str(row["id"]) if row.get("id") is not None else None,
+                source_line_no=str(row["line_no"]) if row.get("line_no") is not None else None,
+                error_code=code,
+                error_message=f"源数据处理失败（{reason}），请核对连接和字段",
+            )
+        )
+        batch.failed_record_count += 1

backend/app/services/tplus/task_service.py

@@ -0,0 +1,184 @@
+"""持久化任务创建和重试，不在 HTTP 事务中读取源库。"""
+
+import secrets
+from contextlib import nullcontext
+from dataclasses import dataclass
+from datetime import datetime
+
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import Session
+
+from app.core.config import get_settings
+from app.core.exceptions import BadRequestError, NotFoundError
+from app.core.tplus_db import resolve_tplus_settings
+from app.models import TPlusSyncBatch, TPlusSyncError, User
+from app.repositories import tplus_repository as repository
+from app.schemas.tplus import TPlusCreateIn, TPlusRetryIn
+from app.services import audit_service
+from app.services.audit_service import AuditRecord, RequestMeta
+
+
+@dataclass(frozen=True)
+class TaskActor:
+    """HTTP 操作者与请求元信息，CLI 调用不传。"""
+
+    user: User
+    meta: RequestMeta | None = None
+
+
+def new_batch_no() -> str:
+    """生成最长 24 字符的可追溯批次号。"""
+    return f"TP{datetime.now():%Y%m%d%H%M%S}{secrets.token_hex(4)}"
+
+
+def _audit(session, actor, action, data):
+    if actor is None:
+        return nullcontext()
+    return audit_service.audit(
+        session,
+        operator=actor.user,
+        meta=actor.meta,
+        record=AuditRecord("T+数据同步", action, "T+ 数据任务" + action, data),
+    )
+
+
+def _source_database() -> str:
+    try:
+        return resolve_tplus_settings(get_settings()).t_plus_database
+    except RuntimeError as exc:
+        raise BadRequestError(str(exc)) from None
+
+
+def _same_intent(batch, mode, entities, parent):
+    if (
+        batch.sync_mode != mode
+        or sorted(batch.entity_types_json or []) != sorted(entities)
+        or batch.parent_batch_no != parent
+    ):
+        raise BadRequestError("幂等键已用于不同的任务请求")
+    return batch
+
+
+def _insert(session, values):
+    try:
+        with session.begin_nested():
+            batch = TPlusSyncBatch(**values)
+            session.add(batch)
+            session.flush()
+        return batch, True
+    except IntegrityError:
+        batch = repository.get_request(session, values["request_key"], for_update=True)
+        if batch is None:
+            raise
+        return _same_intent(
+            batch, values["sync_mode"], values["entity_types_json"], values["parent_batch_no"]
+        ), False
+
+
+def _values(payload, database, actor, parent=None):
+    return {
+        "batch_no": new_batch_no(),
+        "request_key": str(payload.request_key),
+        "status": "QUEUED",
+        "sync_mode": "AUDIT" if payload.task_type == "AUDIT" else "FULL",
+        "entity_types_json": sorted(payload.entity_types),
+        "source_database": database,
+        "requested_by": actor.user.id if actor else None,
+        "parent_batch_no": parent,
+        "started_at": None,
+        "source_record_count": 0,
+        "staged_record_count": 0,
+        "failed_record_count": 0,
+        "skipped_record_count": 0,
+    }
+
+
+def create_task(session: Session, payload: TPlusCreateIn, actor: TaskActor | None = None):
+    """原子创建任务；重放已存在的意图不依赖此刻的源连接状态。"""
+    with _audit(session, actor, "创建读取任务", payload.model_dump(mode="json")):
+        existing = repository.get_request(session, str(payload.request_key))
+        mode = "AUDIT" if payload.task_type == "AUDIT" else "FULL"
+        if existing:
+            batch = _same_intent(existing, mode, payload.entity_types, None)
+        else:
+            batch, _ = _insert(session, _values(payload, _source_database(), actor))
+        session.commit()
+    return batch
+
+
+def retry_task(
+    session: Session, parent_no: str, payload: TPlusRetryIn, actor: TaskActor | None = None
+):
+    """锁父批次后创建一个新任务；不覆盖原批次及异常历史。"""
+    with _audit(
+        session,
+        actor,
+        "重试读取任务",
+        {"parent_batch_no": parent_no, "request_key": str(payload.request_key)},
+    ):
+        existing = repository.get_request(session, str(payload.request_key))
+        if existing:
+            if existing.parent_batch_no != parent_no:
+                raise BadRequestError("幂等键已用于不同的重试请求")
+            session.commit()
+            batch = existing
+        else:
+            parent = repository.get_batch(session, parent_no, for_update=True)
+            if parent is None:
+                raise NotFoundError("同步批次不存在")
+            request = _retry_request(parent, payload)
+            active = repository.active_retry(session, parent_no)
+            if active:
+                if active.request_key != str(payload.request_key):
+                    raise BadRequestError("该批次已有未完成的重试任务，请刷新列表查看")
+                batch = active
+            else:
+                batch, inserted = _insert(
+                    session, _values(request, _source_database(), actor, parent_no)
+                )
+                if inserted:
+                    repository.increment_retry_errors(session, parent_no)
+            session.commit()
+    return batch
+
+
+def _retry_request(parent, payload):
+    if parent.status not in ("FAILED", "PARTIAL"):
+        raise BadRequestError("只能重试失败或部分失败的批次")
+    if parent.sync_mode not in ("FULL", "AUDIT"):
+        raise BadRequestError("该同步方式尚未启用，请重新选择读取任务")
+    if parent.sync_mode != "AUDIT" and parent.entity_types_json is None:
+        raise BadRequestError("历史批次未记录范围，请重新选择读取类型")
+    return TPlusCreateIn(
+        task_type="AUDIT" if parent.sync_mode == "AUDIT" else "READ",
+        entity_types=parent.entity_types_json or [],
+        request_key=payload.request_key,
+    )
+
+
+def fail_batch(session: Session, batch: TPlusSyncBatch, code: str) -> None:
+    """只保存固定错误摘要；即使原异常含连接密码也不直接落库。"""
+    messages = {
+        "INTERRUPTED": "执行进程中断，请重新读取；已暂存数据会自动去重",
+        "EXECUTION_ERROR": "任务执行失败，请检查只读连接和字段映射后重试",
+        "TYPE_UNRESOLVED": "核验所需单据类型缺失或不唯一，请核对源字典",
+    }
+    message = messages.get(code, "任务未完成，请核对运行配置")
+    batch.status, batch.finished_at, batch.error_message = "FAILED", datetime.now(), message
+    batch.failed_record_count = (batch.failed_record_count or 0) + 1
+    session.add(
+        TPlusSyncError(
+            batch_no=batch.batch_no,
+            entity_type="system",
+            source_table="system",
+            error_code=code,
+            error_message=message,
+        )
+    )
+    session.commit()
+
+
+def recover_running(session: Session) -> None:
+    """仅在持有执行锁时调用；锁证明之前的执行者已经退出。"""
+    for batch in repository.running_batches(session):
+        fail_batch(session, batch, "INTERRUPTED")

backend/app/services/tplus/worker.py

@@ -0,0 +1,61 @@
+"""持久化任务执行者；获取连接级锁后才恢复或领取任务。"""
+
+from datetime import datetime
+from pathlib import Path
+
+from app.core.tplus_db import get_tplus_session_factory, resolve_tplus_settings
+from app.core.tplus_lock import TPlusBusyError, sync_lock
+from app.repositories import tplus_repository as repository
+from app.services.tplus import task_service
+from app.services.tplus.relationship_audit import TPlusRelationshipAudit
+from app.services.tplus.sync_service import TPlusSyncService
+
+
+def run_once(session, source_factory=None):
+    """返回本轮处理的批次号；忙锁或队列为空时不假称已执行。"""
+    try:
+        with sync_lock(session):
+            task_service.recover_running(session)
+            batch = repository.next_queued(session)
+            if batch is None:
+                session.rollback()
+                return None
+            batch.status, batch.started_at = "RUNNING", datetime.now()
+            session.commit()
+            _execute(session, batch, source_factory)
+            return batch.batch_no
+    except TPlusBusyError:
+        session.rollback()
+        return None
+
+
+def _execute(session, batch, source_factory):
+    try:
+        settings = resolve_tplus_settings(task_service.get_settings())
+        if batch.source_database and batch.source_database != settings.t_plus_database:
+            raise RuntimeError("任务排队后的源账套配置已改变")
+        factory = source_factory or get_tplus_session_factory()
+        with factory() as source:
+            if batch.sync_mode == "AUDIT":
+                root = Path(settings.t_plus_query_dir) if settings.t_plus_query_dir else None
+                report = TPlusRelationshipAudit(source, root).run()
+                batch.audit_result_json = report
+                batch.source_record_count = sum(row["row_count"] for row in report["receipt_kinds"])
+                batch.finished_at, batch.status = datetime.now(), "SUCCEEDED"
+                if report["relationship_status"] != "OBSERVED":
+                    task_service.fail_batch(session, batch, "TYPE_UNRESOLVED")
+                else:
+                    session.commit()
+            else:
+                TPlusSyncService(source, session).execute_claimed(batch)
+    except BaseException as exc:
+        session.rollback()
+        task_service.fail_batch(
+            session,
+            batch,
+            "INTERRUPTED"
+            if isinstance(exc, (KeyboardInterrupt, SystemExit))
+            else "EXECUTION_ERROR",
+        )
+        if isinstance(exc, (KeyboardInterrupt, SystemExit)):
+            raise

backend/app/services/user_service.py

@@ -68,22 +68,8 @@ def _to_out(user: User) -> UserOut:
 
 def list_users(session: Session, query: UserQuery) -> PageResult[UserOut]:
     """分页查询用户列表。"""
-    users = user_repository.list_users(
-        session,
-        offset=query.offset,
-        limit=query.page_size,
-        username=query.username,
-        real_name=query.real_name,
-        status=query.status,
-        role_code=query.role_code,
-    )
-    total = user_repository.count_users(
-        session,
-        username=query.username,
-        real_name=query.real_name,
-        status=query.status,
-        role_code=query.role_code,
-    )
+    users = user_repository.list_users(session, query)
+    total = user_repository.count_users(session, query)
     return PageResult[UserOut](
         records=[_to_out(user) for user in users],
         total=total,

backend/pyproject.toml

@@ -38,6 +38,11 @@ dependencies = [
     "python-multipart>=0.0.32",
 ]
 
+[project.optional-dependencies]
+tplus = [
+    "pyodbc>=5.2,<6",
+]
+
 [dependency-groups]
 dev = [
     "pytest>=8.3",

backend/tests/README.md

@@ -1,6 +1,30 @@
 # 后端测试
 
-> AI摘要：本项目后端的测试用例清单与运行记录。按「基础设施层随实现即时测、业务模块整模块完成时统一测」分层组织。当前 697 个用例全绿（2026-10-16，M3 增量批次后）。
+> AI 摘要：2026-10-28 的 P0.5 连接修复已通过 19 项定点测试和 1001 项完整回归；使用本轮独立 MySQL/Redis 容器，未连接客户 T+ 源库。真实备份源 10.50.1600 验收仍未执行。
+
+## 连接策略修复核验（2026-10-28）
+
+- Ruff（含兼容性脚本）通过；定点命令 `uv run pytest tests/unit/test_tplus_config.py tests/unit/test_tplus_connection_file.py tests/unit/test_tplus_compatibility_script.py -q`：19 passed。
+- 新增覆盖：数据库名/用户名/密码中的分号、花括号和中文；默认加密与显式回退；脚本复用连接文件校验；10.50.1600 与其他补丁、2022 版本区分；连接失败不自动降级且报告不带驱动原始异常。
+- 完整回归：1001 passed，1 warning，869.47 秒。使用已安装的 `mysql:8.0`、`redis:7-alpine` 镜像启动独立容器、随机宿主端口和命名卷，以 `db/init/init-dev.sql` 初始化；未连接现有开发业务库或 T+ 源库。
+- 完整测试入口为 `uv run python ../.ai/run_isolated_regression.py`，内部调用 `pytest.main(["-q", ...])`，不筛选或跳过用例。运行期配置指向临时隔离实例；单测的 `_env_file=None` 仍保持默认值测试语义。结束后已清理本轮容器、卷与临时凭据配置。
+- 兼容性脚本、worker 的 `--help` 可运行；保留现有 Starlette TestClient 弃用提醒。本轮未改前端，未重复前端构建。
+- 版本判断与失败分支是单元模拟结果，不能据此宣称真实 2008 R2 的驱动、TLS、断连和权限已验收。下方 2026-10-27 的结果保留为历史记录。
+
+## T+ 本轮核验（2026-10-27）
+
+- 完整回归在独立 MySQL `payroll_tplus_validation_20261027`、Redis 15 库运行：984 passed、2 failed。两项失败均为配置单测预期默认 Redis 0 库，而进程继承了集成用的 `REDIS_DB=15`；没有修改默认配置或削弱断言。
+- 后续将单元测试放在不设置上述集成覆盖值的独立进程中运行：602 passed。最终管理 API、真实并发和结构一致性定点测试：18 passed，包含新增加的配置关闭及创建/重试审计。
+- 真实并发：相同请求只生成一个批次；不同键同时重试同父批次只有一个成功，无死锁；执行锁跨 ORM commit 仍有效，退出后可再获取。
+- V15 在隔离库 downgrade 到 V14，再 upgrade 到 head 通过；当前 head `e5f607a8b9c0`。模型、Alembic、留存 SQL 和初始化脚本的结构检查通过；此前 V1–V14 留存 SQL 已从空库执行通过。
+- 人员重读 847 条，新增 0、重复跳过 847、失败 0；关联报告覆盖 103,318 条销货类收款明细，50,044 条有明确订单关联，六类业务表行数不变。独立证据位于 D 盘 reports/management_verification_2026-10-27.json。
+- Chrome 真实接口检查通过：排队、受控失败、异常查看、页面重试、人员去重、关联检查、筛选、普通用户拒绝访问，以及 320/768/1024/1440 宽度。无页面运行错误；临时账号和批次已清理。
+- 加载、空态、配置关闭、概况失败另外使用浏览器响应模拟验证，不能与真实源链路混记。证据位于 .ai/archive/ai-aux/preview/tplus/。
+- 320 像素主内容可操作，现有全局顶栏仍有文字挤压；未扩大本轮范围修改共享布局。宽表保留横向滚动，截图和限制见同步管理验收报告。
+- 前端类型检查、生产构建和 Ruff 通过；构建保留既有的大分块提醒，测试有现存 Starlette TestClient 弃用提醒。
+- 源查询保持参数化与旧版语法，时间增量和业务表写入均关闭。合法 Windows/SQL Server 2008 R2 安装介质仍未取得，不能把 2022 的结果标记为客户版本兼容。
+
+单元测试与集成测试建议分独立终端运行：单元测试不设置集成覆盖值；集成测试使用隔离 MySQL 和 Redis。通用集成夹具会清理数据，不能指向开发或客户业务库。Windows 可设置 `PYTEST_DEBUG_TEMPROOT` 到项目 D 盘的 ai-aux/pytest-tmp，避免不同权限令牌共享系统临时目录。
 
 ## 一、测试约定
 
@@ -98,9 +122,9 @@ uv run pytest tests/integration -q   # 仅集成
 | --- | --- | --- |
 | `tests/integration/test_personnel.py` | 未登录 401、新增默认在职、工资标准 str 直出、经理自引用/不存在或已离职/岗位停用 400、删除保护（团队成员与客户归属引用→引导改离职）、辐射区域与拓展计划全量替换（含清空与详情回显）、在职/离职切换、审计成功与失败双记录 | 20 |
 | `tests/integration/test_products.py` | 未登录 401、类别值域 400、名称+规格组合唯一（同名称不同规格放行）、单价负数/非数字 400、启停切换、改类别、审计留痕 | 14 |
-| `tests/integration/test_customers.py` | 未登录 401、客户分类值域 400、「公司名称+负责人」组合唯一（同公司不同负责人放行）、归属变更（主表更新+留痕字段断言）、归属人员离职 400、新旧相同含双方为空 400、留痕分页 join 姓名、编辑端点不触归属、审计留痕 | 14 |
+| `tests/integration/test_customers.py` | 未登录 401、客户分类值域 400、「公司名称+负责人」组合唯一（同公司不同负责人放行）、归属变更（主表更新+留痕字段断言）、归属人员离职 400、新旧相同含双方为空 400、留痕分页 join 姓名、编辑端点不触归属、审计留痕、评级保存（回显总分/时间、覆盖式保存）、评分 101→400、按评分排序未评分垫底（asc/desc） | 18 |
 
-运行：`uv run pytest tests/integration/test_personnel.py tests/integration/test_products.py tests/integration/test_customers.py -q` → **48 passed**（2026-10-13，连跑两次幂等）。
+运行：`uv run pytest tests/integration/test_personnel.py tests/integration/test_products.py tests/integration/test_customers.py -q` → **48 passed**（2026-10-13，连跑两次幂等）；2026-10-23 评级功能后客户单文件 **18 passed**（18.5s）。
 
 ### 业务数据管理（2026-10-13 完成）
 
@@ -159,6 +183,25 @@ uv run pytest tests/integration -q   # 仅集成
 
 **实施中发现并修复**：`alembic upgrade --sql` 在 Windows 控制台按 GBK 输出，重定向生成的 `V10__*.sql` 被 `test_schema_sync` 按 UTF-8 读取时抛 `UnicodeDecodeError`——导出命令须加 `PYTHONIOENCODING=utf-8 PYTHONUTF8=1`（已写入 backend/README）。
 
+### code-review 遗留 2 项（2026-10-20 完成：方案版本化 / 审批内容快照）
+
+| 文件 | 守什么 | 用例数 |
+| --- | --- | --- |
+| `tests/integration/test_approval_flow.py`（扩） | **提交时快照固化**（version_no=1、content_source=snapshot、方案比例/合计/余额与业务编号）；编辑方案为 80% 后新单 V2、旧单快照仍 50.00 且旧在途单撤回；**历史单无快照回退当前内容并标 live**（captured_at=null）；引用对象缺失且无快照 → missing；**方案版本化**：编辑后 history=[(1,0,50.00),(2,1,80.00)]、生效行仅 80.00；驳回后生效行为空但历史行保留（is_active=0） | +5 |
+| `tests/unit/test_approval_schema.py` | 对象类型筛选 1/2/3 均可、0/4 拒绝（修复原上界 `le=2` 导致调整单审批单无法筛选）；不传类型表示全部；驳回必须填意见、通过可不填 | 6 |
+| `tests/unit/test_approval_helpers.py` | **方案内容比较**（`business_service._plan_changed`）：完全相同/行序不同判未变、比例变/人员增减/同比例换人判已变、比例写法 `50` 与 `50.00` 等价、空方案对照 | 10 |
+| `tests/integration/test_secondary_allocation.py`（扩） | **同一会话内旧方案行随置历史同步刷新**（Core UPDATE 的会话同步策略退化为不刷新时，行对象会停留在 is_active=1 的过期内存值） | +1 |
+| `tests/integration/test_approval_flow.py`（扩） | **送审内容未变不重提**：方案与金额都没动 → 方案不新增版本、审批单不重提（旧单仍待财务复核）；**只改金额** → 方案版本不变但审批单 version_no+1 且新快照金额为改后值；**改方案比例** → 方案版本+1、审批单 version_no+1 且旧单撤回；**无在途单时保存仍提新单**（保守规则）；`pending_matches_snapshot` 四态（一致/不一致/无在途单/快照为 None） | +5 |
+
+### 审批单版本号并发防线（2026-10-21 完成：V12 唯一键 + 锁读取号）
+
+| 文件 | 守什么 | 用例数 |
+| --- | --- | --- |
+| `tests/integration/test_approval_version.py`（新） | **取号读最新已提交值**（两会话法：s1 锚定快照 + s2 提交 version_no=2 → s1 必须取到 3，守住「快照读 vs 当前读」根因）；**取号查询带 FOR UPDATE 且不含聚合**（事件钩子抓真实 SQL——MySQL 官方不允许聚合函数携带锁子句）；**唯一键生效**（重复 (order_type, ref_id, version_no) 直插撞 1062）；**编辑路径冲突收敛为 400** 且业务数据不写坏（monkeypatch `_next_version` 恒返回 1）；**正常路径版本号仍递增**（1→2→3） | +5 |
+| `tests/unit/test_models.py`（扩） | 审批单唯一键覆盖 (order_type, ref_id, version_no) 三列且约束名与迁移 DDL 一致 | +1 |
+
+运行（2026-10-22，容器恢复后）：`tests/integration/test_approval_version.py` + `test_dashboard.py` → **16 passed**；全量 `uv run pytest -q` → **897 passed（12:02）**；`uv run ruff check .` 绿、前端 `pnpm type-check` 绿。首次全量跑出两处**测试代码**适配（非产品缺陷）：① 工作台 `_insert_approval` 用同一 `(order_type=2, ref_id=0)` 连插两张单，撞 V12 新唯一键 → 每张单改给不同 `ref_id`；② 新用例跨模块引用 `switched_on` 夹具（pytest 夹具只在定义模块与 conftest 可见，报 fixture not found）→ 用 `_set_switch` 本地重建。
+
 ## 三、运行记录
 
 | 日期 | 范围 | 用例数 | 结果 | 备注 |
@@ -184,9 +227,15 @@ uv run pytest tests/integration -q   # 仅集成
 | 2026-10-19 | **code-review 修复回归** | 全量 862 | ✅ 862 passed（9:36） | **M4 工作台 + 报表 3 必须 + 8 建议修复**：普通用户落地页/守卫（报表集成 27、工作台集成 11、报表单元 17，新增 10 例）；部门汇总 5% 按计入次数累计并与月结 `total_payable` 对拍、聚合行舍入差；工作台待办按角色过滤、逾期命中与阈值生效、终止业务排除、`warnings.total` 截断、趋势跨月归属。ruff 绿；type-check + build 绿 |
 | 2026-10-20 | e2e 全链路 | — | ✅ login 8 / system 13 / role_home 4 / report 12 / calc_settlement 8 | 重启后端加载新代码后实跑：普通用户落地页与受限页回退、报表四页/导出/打印、核算与月结链路全过。`report_flow` 回款统计改为展开目标桶后按本次业务编号定位明细，不依赖桶累计值，可重复运行 |
 | 2026-10-20 | 全量回归 | 862 | ✅ 862 passed（11:32） | **工作台/报表修复提交后的全量复核**；e2e 三脚本复跑：role_home 4/4、report 12/12、calc_settlement 8/8；calc 审批开关已还原为 1，无 `e2e_role_*` 残留 |
+| 2026-10-20 | code-review 遗留 2 项 | 全量 875 | ✅ 875 passed（12:13） | **方案版本化 + 审批内容快照**（V11 迁移）：审批单 `version_no`/`snapshot_json`，方案行 `is_active` 版本化；编辑/驳回不再物理删方案；历史单无快照时 live 回退并显式标注。新增集成 7 + 单元 6；定点审批/月结/核算/业务/报表全绿；ruff 与 type-check/build 绿 |
+| 2026-10-20 | 遗留项自评审收口 | 全量 891 | ✅ 891 passed（10:23） | **列表 defer 快照列**（审批列表/工作台待办/超时预警）+ **送审内容未变不重提、方案未变不新增版本**（两个 version_no 语义解耦）+ **月结提交加行锁**。新增单元 10 + 集成 5；单元 525、集成定点 96 全绿 |
+| 2026-10-22 | 清理批次（散参数 DTO 化） | 全量 900 | ✅ 900 passed（11:08） | **仓储层散参数收敛为查询 DTO**（`list_*(session, query)` 取代 keyword/status/offset/limit 四参数），同步调整服务层与测试调用；单元 526 绿、ruff 绿、前端 type-check 与 build 绿 |
 | 2026-10-19 | 单元全量 + 报表定点 | 484 | ✅ 484 passed（45.94s）；报表单元 11 例 0.70s | **M4 报表模块收尾**（前端四页/打印视图/导出与单元测试补齐）。ruff 绿；前端 `pnpm type-check` 绿。报表集成 7 例记录见上节 |
 | 2026-10-19 | 全部 | 806 | ✅ 806 passed（7:59） | **M4 ⑤ 报表统计全量回归**（报表新增 18 例：集成 7 + 单元 11；另含并行会话并入的 M3 前端修复与字体合规）。ruff 绿。e2e：`calc_settlement_flow` 8 步由并行会话跑通；`report_flow.py` 待 `E2E_ADMIN_PASSWORD` 运行 |
 
+| 2026-10-23 | 参数文案中文化 | 单元 526 + 参数集成 24 | ✅ 526 passed（45.15s）/ 24 passed（10.60s） | **计提参数页全面中文化**：`param_registry` 22 条 `description` 改写为业务语言（去掉「澄清 #10」「M3」「禁止按 0.011111 截断」等开发者语汇），`db/init/init-dev.sql` 种子同源同步；`param_service` 模板报错改中文模板名、产品类别报错去掉英文枚举；`dashboard_service` 概览口径说明重写。**这些文案无测试断言**（`rate_text` 断言 `11/1125`、`聚合:20%+5%` 未动），故用例数不变、全绿即口径未被破坏。ruff 绿；前端 `pnpm type-check` 绿 |
+| 2026-10-23 | 台账/预览的计提说明口语化 | 单元 526 + 集成 87（calc/settlements/param/reports） | ✅ 526 passed（47.29s）/ 87 passed（7:41） | **`rate_text`/`rate_note`/聚合行 `note` 改业务话**：`企业户固定计提比例`→「企业户：不按回款天数，统一按固定比例计提」、`业务员递减：10‰ 起，20 天（分界 360 天…）`→「回款 20 天，从起始 10‰ 按日递减（满 360 天后固定按 4‰ 计算）」、`信息员：本人销售额 × 计提比例`→「信息员：按本人销售额乘计提比例，不随回款天数递减」、聚合行 `#12 参数化：基数不含经理个人户业绩；9.8 方案公式`→「团队分成：基数不含经理本人名下业绩，按本团队与全公司两级比例合计计提」。**用例同步**：`test_经理聚合口径_预览拒绝` 原断言消息含「聚合」，改为断言含「营销经理」+「月度结算」（守「拒绝且给下一步」，不再绑定措辞）；`rate_text` 的 `11/1125`、`聚合:20%+5%` 断言未动（那是分数原文与百分比拼装，非说明文案） |
+
 ## 四、测试发现并修复的缺陷
 
 | 日期 | 缺陷 | 影响 | 处理 |
@@ -197,6 +246,15 @@ uv run pytest tests/integration -q   # 仅集成
 | 2026-10-19 | 经理个人户成员行取该人全部绩效 | 团队基数只含个人户行，成员行却含普通行——**基数≠成员合计，同一笔业绩在成员行与未分配重复** | 普通行与个人户分行记账；经理成员行只取个人户金额，普通行留在上级团队/未分配 |
 | 2026-10-19 | 导出取数与超限校验在审计上下文之外 | 超限 400 与取数异常**不留失败日志**，与 M4 计划 T12「导出 400/失败审计」不符 | `_export_payload` 移入 `with audit`；静态标题映射（AuditRecord frozen） |
 | 2026-10-19 | 月份只校验长度 | `2026-13` 传到 `calendar.monthrange` 抛 ValueError → **500**；导出手工构造模型抛裸 ValidationError 同样 500 | 新增 `core/month.py`（格式+01-12+年份）；四类 query、月结、调整单接入；导出 ValidationError 转 400 |
+| 2026-10-20 | 审批类型筛选上界写死 2 | `ApprovalQuery.order_type` 为 `le=2`：调整单（order_type=3）审批单在列表无法筛选，只能翻全量 | 上界改 3；单元用例锁定 1/2/3 与越界拒绝 |
+| 2026-10-20 | 编辑业务时送审快照取旧汇总 | `update_business` 中 `_recalc_totals` 在提单之后，快照里的销售额/成本额会固化上一版本金额 | 汇总重算移到提单之前；快照用例断言送审内容与本次编辑后方案一致 |
+| 2026-10-20 | 快照金额把「空」写成 0.00 | 业务尚无成本时 `total_cost_amount=None` 被文本化成 `0.00`，复核者会把「无成本」误读成「成本为零」 | `_amount_text(None)` 保留为空，前端显示「—」 |
+| 2026-10-20 | 每次保存带方案的业务都刷出版本与审批留痕 | 只改备注、方案一字未动，也会「置历史+新增版本+撤回旧单+提新单」，方案历史与审批留痕被噪音淹没，两个 `version_no` 语义混同 | 判据统一为**送审内容**：方案未变不新增版本行；在途单送审内容一致则不重提（`pending_matches_snapshot`，只比 `data`） |
+| 2026-10-20 | 三处列表查询整列拉快照 | 审批单列表、工作台待办、审批超时预警都用 `select(ApprovalOrder)` 取全列，快照 JSON 可达上百 KB，翻页与工作台负担随快照体积线性放大 | 三处加 `defer(ApprovalOrder.snapshot_json)`（沿用操作日志列表排除 `request_data` 的既有模式） |
+| 2026-10-20 | 月结提交未锁行 | 并发/双击提交同一月结时两次 `_next_version` 可能算出同一 `version_no`，送审次序失真 | `settlement_repository.get_by_id` 加 `for_update` 参数，提交路径锁行 |
+| 2026-10-21 | 审批单 `version_no` 无唯一约束 | 应用层取号即使正确，并发下 DB 层无兜底；且唯一键是「一个对象至多一张在途单」红线之外版本语义的最后防线 | V12 迁移加 `uk_approval_orders_order_type_ref_version (order_type, ref_id, version_no)`；加键前先重排存量重复分组（V11 把存量一律置 1 而 V11 前重提是新建行，直接加键必报 1062） |
+| 2026-10-21 | `_next_version` 是快照读 | 请求级鉴权与业务共用同一 DbSession，鉴权的普通 SELECT 已锚定 REPEATABLE READ 快照——父行锁只串行化写入时刻、**不刷新快照**，并发提单的 MAX 停在请求开始时刻 → 版本号重复 | 取号改锁读（`select(version_no)...with_for_update()`，当前读），最大值 Python 侧算；**不能用聚合 + FOR UPDATE**——MySQL 官方不允许聚合函数携带锁子句 |
+| 2026-10-21 | `act()` 驳回清空方案的注释与行为相反 | 注释写「清空该业务的二次分配行…如需保留历史版本，在 ④ 设计时演进」，而 V11 已实现 append-only（只置 `is_active=0` 行不删）——误导后来者按旧语义改代码 | 注释改为「置历史不删行，历史方案可回溯」 |
 | 2026-10-19 | 前端报表五处状态缺陷 | 打印超限判定恒假（静默截断）、部门下钻 query 未读取、回款多桶共用一个 drill 状态（面板串行）、summary 在请求序号守卫前写入（合计串号）、e2e 全页行数选择器必失败 | 分别修正判定/读取 query/按桶独立状态+数组参数/`onData` 回调/限定主桶表；补 e2e 多桶展开、下钻、截断、部门打印用例 |
 | 2026-10-19 | 工作台权限收紧后守卫仍回退 `/dashboard` | 普通用户登录/直连受限页被反复重定向到同一受限页 → **无法进入系统**（导航被判定无限重定向） | `router/access.ts` 统一准入与落地页解析（工作台 → 业务数据 → 回款）；登录回跳经 `canAccessPath` 校验；菜单/面包屑复用同一判定 |
 | 2026-10-19 | 部门汇总应发只计一次全公司 5% | 月结每位有聚合行的经理各计一次 5%，签字稿少计 N-1 份；**导出/打印金额小于月结应发** | 团队行增 `has_aggregate/aggregate_amount`（优先冻结聚合行）；对账改按聚合额求和，新增「聚合行舍入差」吸收分列 q2 尾差；补「与 `total_payable` 对拍」用例 |

backend/tests/conftest.py

@@ -4,6 +4,36 @@ import pytest
 from fastapi.testclient import TestClient
 
 
+@pytest.fixture
+def tplus_session(monkeypatch):
+    """T+ 管理逻辑的真实内存事务；MySQL 并发锁另做集成验证。"""
+    from sqlalchemy import DefaultClause, Integer, MetaData, create_engine, text
+    from sqlalchemy.orm import Session
+
+    from app.core.config import Settings
+    from app.models import TPlusSyncBatch, TPlusSyncError, TPlusSyncRecord
+
+    engine = create_engine("sqlite:///:memory:")
+    metadata = MetaData()
+    for model in (TPlusSyncBatch, TPlusSyncError, TPlusSyncRecord):
+        table = model.__table__.to_metadata(metadata)
+        table.c.id.type = Integer()
+        table.c.update_time.server_default = DefaultClause(text("CURRENT_TIMESTAMP"))
+    metadata.create_all(engine)
+    settings = Settings(
+        _env_file=None,
+        app_secret_key="x" * 32,
+        t_plus_enabled=True,
+        t_plus_database="unit_tplus",
+        t_plus_user="reader",
+        t_plus_odbc_driver="unit",
+    )
+    monkeypatch.setattr("app.services.tplus.task_service.get_settings", lambda: settings)
+    with Session(engine, expire_on_commit=False) as session:
+        yield session
+    engine.dispose()
+
+
 @pytest.fixture(scope="session")
 def client() -> TestClient:
     """构造应用测试客户端（会话级复用，避免重复构建应用）。

backend/tests/integration/conftest.py

@@ -324,3 +324,36 @@ def normal(make_user, auth_headers) -> dict:
     user = make_user([ROLE_USER])
     user["headers"] = auth_headers(user)
     return user
+
+
+@pytest.fixture
+def tplus_tasks(monkeypatch):
+    """隔离测试任务，仅清理此夹具创建的批次和异常。"""
+    from app.core.config import Settings
+    from app.models import TPlusSyncBatch, TPlusSyncError
+
+    settings = Settings(
+        _env_file=None,
+        app_secret_key="x" * 32,
+        t_plus_enabled=True,
+        t_plus_database="unit_tplus",
+        t_plus_user="reader",
+        t_plus_odbc_driver="unit",
+    )
+    monkeypatch.setattr("app.services.tplus.task_service.get_settings", lambda: settings)
+    keys = []
+
+    def payload():
+        key = str(uuid.uuid4())
+        keys.append(key)
+        return {"task_type": "READ", "entity_types": ["personnel"], "request_key": key}
+
+    yield payload
+    with get_session_factory()() as db:
+        numbers = list(
+            db.scalars(select(TPlusSyncBatch.batch_no).where(TPlusSyncBatch.request_key.in_(keys)))
+        )
+        if numbers:
+            db.execute(delete(TPlusSyncError).where(TPlusSyncError.batch_no.in_(numbers)))
+            db.execute(delete(TPlusSyncBatch).where(TPlusSyncBatch.batch_no.in_(numbers)))
+            db.commit()

backend/tests/integration/test_approval_flow.py

@@ -388,3 +388,336 @@ class TestModuleAccess:
             user = make_user([code])
             user["headers"] = auth_headers(user)
             assert client.get(_APPROVALS, headers=user["headers"]).status_code == 200
+
+
+class TestContentSnapshot:
+    """审批单内容快照与版本号（code-review 遗留②）。
+
+    守的是：复核者看到的是**送审时**的方案，而不是审批期间被改过的当前
+    方案；历史单（升级前无快照）必须显式标注「当前内容回退」，不能把
+    当前内容冒充送审内容。
+    """
+
+    def test_提交时快照固化_再改方案不回写旧单(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        first = _latest_order(client, finance["headers"], data["business_no"])
+
+        assert first["version_no"] == 1
+        assert first["content_source"] == "snapshot"
+        snapshot = first["snapshot"]
+        assert snapshot["schema_version"] == 1
+        assert snapshot["order_type"] == 1
+        assert snapshot["captured_at"], "快照须固化采集时间"
+        assert snapshot["data"]["business_no"] == data["business_no"]
+        assert snapshot["data"]["allocations"][0]["ratio"] == "50.00"
+        assert snapshot["data"]["allocation_total"] == "50.00"
+        assert snapshot["data"]["allocation_balance"] == "50.00"
+
+        # 编辑方案为 80%：旧在途单撤回，新单 V2 快照取新值；旧单快照不变
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        payload = business_payload(
+            detail["customer_id"],
+            secondary_allocations=[{"personnel_id": p["id"], "ratio": "80.00"}],
+        )
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 200, resp.text
+
+        second = _latest_order(client, finance["headers"], data["business_no"])
+        assert second["version_no"] == 2
+        assert second["snapshot"]["data"]["allocations"][0]["ratio"] == "80.00"
+
+        old = client.get(f"{_APPROVALS}/{first['id']}", headers=finance["headers"]).json()["data"]
+        assert old["status"] == 5, "重新提交须撤回旧在途单"
+        assert old["snapshot"]["data"]["allocations"][0]["ratio"] == "50.00"
+
+    def test_历史单无快照_回退当前内容并标注(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        """升级前的审批单没有快照列：回退读当前内容，但来源必须标 live。"""
+        from app.core.db import get_session_factory
+        from app.models import ApprovalOrder
+
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        order = _latest_order(client, finance["headers"], data["business_no"])
+
+        factory = get_session_factory()
+        with factory() as session:
+            row = session.get(ApprovalOrder, order["id"])
+            assert row is not None
+            row.snapshot_json = None
+            session.commit()
+
+        detail = client.get(f"{_APPROVALS}/{order['id']}", headers=finance["headers"]).json()[
+            "data"
+        ]
+        assert detail["content_source"] == "live"
+        assert detail["snapshot"]["captured_at"] is None
+        assert detail["snapshot"]["data"]["business_no"] == data["business_no"]
+
+    def test_引用对象缺失且无快照_内容缺失(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        from app.core.db import get_session_factory
+        from app.models import ApprovalOrder
+
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        order = _latest_order(client, finance["headers"], data["business_no"])
+
+        factory = get_session_factory()
+        with factory() as session:
+            row = session.get(ApprovalOrder, order["id"])
+            assert row is not None
+            row.snapshot_json = None
+            row.ref_id = 999_999_999
+            session.commit()
+
+        detail = client.get(f"{_APPROVALS}/{order['id']}", headers=finance["headers"]).json()[
+            "data"
+        ]
+        assert detail["content_source"] == "missing"
+        assert detail["snapshot"] is None
+
+
+def _payload_from_detail(detail: dict, **overrides: object) -> dict:
+    """按业务详情原样重建提交载荷（明细行照抄，用于「内容未变」的场景）。"""
+    items = [
+        {
+            "product_name": it["product_name"],
+            "product_category": it["product_category"],
+            "quantity": it["quantity"],
+            "sale_price": it["sale_price"],
+        }
+        for it in detail["items"]
+    ]
+    body = business_payload(
+        detail["customer_id"],
+        items=items,
+        secondary_allocations=detail["secondary_allocations"],
+    )
+    body.update(overrides)
+    return body
+
+
+class TestNoRedundantResubmit:
+    """送审内容未变不重提、方案未变不新增版本（版本与留痕降噪）。
+
+    守的是：反复保存同一份业务（只改备注）不应刷出一串方案版本与
+    「撤回+新建」的审批留痕；而内容真变了（金额或方案）必须重提，
+    否则审批人看到的是过期内容。
+    """
+
+    def test_方案与金额均未变_不新增版本也不重提(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        first = _latest_order(client, finance["headers"], data["business_no"])
+        assert first["status"] == 1
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        payload = _payload_from_detail(detail)
+        payload["remark"] = "只改了备注"
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 200, resp.text
+
+        # 方案版本不增
+        after = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        assert [(h["version_no"], h["is_active"]) for h in after["allocation_history"]] == [
+            (1, 1)
+        ], "方案内容未变不应新增版本行"
+
+        # 审批单不重提：仍是原来那一张、仍待财务复核
+        rows = client.get(
+            _APPROVALS,
+            params={"keyword": data["business_no"], "pageSize": 10},
+            headers=finance["headers"],
+        ).json()["data"]["list"]
+        assert len(rows) == 1, f"送审内容未变不应重提，实得 {len(rows)} 张单"
+        assert rows[0]["id"] == first["id"]
+        assert rows[0]["status"] == 1, "在途单不应被撤回"
+        assert rows[0]["version_no"] == 1
+
+    def test_只改金额不改方案_方案版本不变但审批单重提(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        """金额变了 → 送审内容变了 → 必须重提；但方案本身没变 → 不新增版本。"""
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        first = _latest_order(client, finance["headers"], data["business_no"])
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        payload = _payload_from_detail(detail)
+        payload["items"][0]["sale_price"] = "200.00"
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 200, resp.text
+
+        after = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        assert [(h["version_no"], h["is_active"]) for h in after["allocation_history"]] == [
+            (1, 1)
+        ], "方案未变不应新增版本行"
+
+        second = _latest_order(client, finance["headers"], data["business_no"])
+        assert second["version_no"] == 2, "金额变了必须重提"
+        assert second["snapshot"]["data"]["sale_amount"] == "20000.00"
+        old = client.get(f"{_APPROVALS}/{first['id']}", headers=finance["headers"]).json()["data"]
+        assert old["status"] == 5
+
+    def test_改方案比例_方案版本与审批单版本同时递增(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        first = _latest_order(client, finance["headers"], data["business_no"])
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        payload = _payload_from_detail(
+            detail, secondary_allocations=[{"personnel_id": p["id"], "ratio": "80.00"}]
+        )
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 200, resp.text
+
+        after = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        assert [(h["version_no"], h["is_active"]) for h in after["allocation_history"]] == [
+            (1, 0),
+            (2, 1),
+        ], "方案变了应新增版本并把旧行置历史"
+
+        second = _latest_order(client, finance["headers"], data["business_no"])
+        assert second["version_no"] == 2
+        old = client.get(f"{_APPROVALS}/{first['id']}", headers=finance["headers"]).json()["data"]
+        assert old["status"] == 5
+
+    def test_无在途单时保存仍提新单(
+        self, client: TestClient, admin: dict, finance: dict, leader: dict, switched_on
+    ) -> None:
+        """保守规则：只有「存在在途单且内容一致」才不重提。
+        方案已审批通过后再保存，仍按原语义重新送审。"""
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        order = _latest_order(client, finance["headers"], data["business_no"])
+        client.post(
+            f"{_APPROVALS}/{order['id']}/action", json={"action": 1}, headers=finance["headers"]
+        )
+        client.post(
+            f"{_APPROVALS}/{order['id']}/action", json={"action": 1}, headers=leader["headers"]
+        )
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        payload = _payload_from_detail(detail)
+        payload["remark"] = "已通过后再保存"
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 200, resp.text
+
+        rows = client.get(
+            _APPROVALS,
+            params={"keyword": data["business_no"], "pageSize": 10},
+            headers=finance["headers"],
+        ).json()["data"]["list"]
+        assert len(rows) == 2, "无在途单时保存应重新送审"
+        assert max(r["version_no"] for r in rows) == 2
+
+
+class TestPendingSnapshotMatch:
+    """在途单送审内容一致性判定（approval_service.pending_matches_snapshot）。"""
+
+    def test_三种情形(self, client: TestClient, admin: dict, finance: dict, switched_on) -> None:
+        from app.core.db import get_session_factory
+        from app.services import approval_service
+
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        order = _latest_order(client, finance["headers"], data["business_no"])
+        snapshot = order["snapshot"]["data"]
+
+        factory = get_session_factory()
+        with factory() as session:
+            # 1) 内容一致 → True
+            assert (
+                approval_service.pending_matches_snapshot(
+                    session,
+                    order_type=1,
+                    ref_id=data["id"],
+                    snapshot=snapshot,
+                )
+                is True
+            )
+            # 2) 内容不同 → False
+            assert (
+                approval_service.pending_matches_snapshot(
+                    session, order_type=1, ref_id=data["id"], snapshot={"allocations": []}
+                )
+                is False
+            )
+            # 3) 无在途单 → False（引用对象换成别的）
+            assert (
+                approval_service.pending_matches_snapshot(
+                    session, order_type=1, ref_id=999_999_999, snapshot=snapshot
+                )
+                is False
+            )
+            # 4) 快照为 None → False
+            assert (
+                approval_service.pending_matches_snapshot(
+                    session, order_type=1, ref_id=data["id"], snapshot=None
+                )
+                is False
+            )
+
+
+class TestAllocationVersions:
+    """二次分配方案版本化（code-review 遗留①）。
+
+    守的是：编辑替换与驳回清空只把旧行置历史（is_active=0），不物理删除，
+    方案参与人与比例可回溯；生效行始终唯一，核算取数不受历史行干扰。
+    """
+
+    def test_编辑方案产生新版本_旧版本保留(
+        self, client: TestClient, admin: dict, switched_on
+    ) -> None:
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        assert [(h["version_no"], h["is_active"]) for h in detail["allocation_history"]] == [(1, 1)]
+
+        payload = business_payload(
+            detail["customer_id"],
+            secondary_allocations=[{"personnel_id": p["id"], "ratio": "80.00"}],
+        )
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 200, resp.text
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        history = [
+            (h["version_no"], h["is_active"], h["ratio"]) for h in detail["allocation_history"]
+        ]
+        assert history == [(1, 0, "50.00"), (2, 1, "80.00")]
+        assert [a["ratio"] for a in detail["secondary_allocations"]] == ["80.00"]
+
+    def test_驳回后生效行为空_历史行保留(
+        self, client: TestClient, admin: dict, finance: dict, switched_on
+    ) -> None:
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        order = _latest_order(client, finance["headers"], data["business_no"])
+
+        resp = client.post(
+            f"{_APPROVALS}/{order['id']}/action",
+            json={"action": 2, "opinion": "方案不合理"},
+            headers=finance["headers"],
+        )
+        assert resp.status_code == 200, resp.text
+
+        detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        assert detail["secondary_allocations"] == [], "被驳回方案不得留在生效行"
+        assert detail["allocation_hint"] is None
+        history = detail["allocation_history"]
+        assert len(history) == 1
+        assert history[0]["version_no"] == 1
+        assert history[0]["is_active"] == 0
+        assert history[0]["ratio"] == "50.00"

backend/tests/integration/test_approval_version.py

@@ -0,0 +1,183 @@
+"""审批单版本号并发防线（V12 唯一键 + 锁读取号）。
+
+守的是：
+- **取号必须读最新已提交值**：请求级鉴权与业务共用同一 DbSession，鉴权
+  的普通 SELECT 已把 REPEATABLE READ 快照锚定在请求开始时刻；父行锁只
+  串行化写入时刻、不刷新快照——取号若是快照读，并发提单会算出同一个
+  version_no（两会话法做行为级回归锁，不依赖真多线程）
+- **取号查询带 FOR UPDATE 且不用聚合**：MySQL 官方不允许聚合函数携带
+  锁子句（Locking clauses are not permitted in queries with aggregate
+  functions or GROUP BY），聚合上的 FOR UPDATE 语义无保证
+- **DB 层唯一键生效**：即使应用层取号缺陷被绕过，重复
+  (order_type, ref_id, version_no) 也必须在库上被拒（1062）
+- **冲突收敛为 400**：编辑路径撞唯一键 → 审计上下文回滚 → 业务提示
+  可重试，业务数据不得写坏
+- **正常路径版本号仍递增**：锁读改写不破坏既有 1→2→3 行为
+"""
+
+import uuid
+
+import pytest
+from fastapi.testclient import TestClient
+from sqlalchemy import event, text
+
+from app.core.db import get_engine, get_session_factory
+from app.core.db_errors import is_duplicate_entry
+from app.models import ApprovalOrder
+from tests.integration.factories import business_payload
+from tests.integration.test_approval_flow import (
+    _BUSINESSES,
+    _create_with_plan,
+    _latest_order,
+    _person,
+    _set_switch,
+)
+
+pytestmark = pytest.mark.integration
+
+
+@pytest.fixture
+def switched_on(client: TestClient, admin: dict) -> None:
+    """把二次分配审批开关置 1（用例结束由 conftest 恢复种子基线）。
+
+    pytest 夹具只在定义它的模块与 conftest 中可见，跨模块直接引用会报
+    ``fixture 'switched_on' not found``；这里用 ``_set_switch`` 本地重建。
+    """
+    _set_switch(client, admin["headers"], "1")
+
+
+class TestNextVersionLockRead:
+    """取号查询的锁读口径（真实 SQL 形态 + 行为级快照回归）。"""
+
+    def test_取号查询带FOR_UPDATE_不含聚合(self, client: TestClient, admin: dict) -> None:
+        """抓 _next_version 真实发出的 SQL（比本地重造语句再编译更贴真）：
+        必须含 FOR UPDATE、不含 MAX( —— 聚合上的锁子句 MySQL 不保证语义。"""
+
+        captured: list[str] = []
+
+        def _spy(_conn, _cursor, statement, *_rest):
+            if "approval_orders" in statement and "version_no" in statement:
+                captured.append(statement)
+
+        engine = get_engine()
+        event.listen(engine, "before_cursor_execute", _spy)
+        try:
+            p = _person(client, admin["headers"])
+            _create_with_plan(client, admin["headers"], p["id"])  # 触发一次真实提单
+        finally:
+            event.remove(engine, "before_cursor_execute", _spy)
+
+        assert captured, "未捕获到取号查询"
+        assert any("FOR UPDATE" in sql.upper() for sql in captured), captured
+        assert not any("MAX(" in sql.upper() for sql in captured), captured
+
+    def test_版本号读最新已提交值_不受事务旧快照影响(self) -> None:
+        """两会话法（无需线程）：s1 先锚定快照 → s2 提交新版本 →
+        s1 取号必须读到 3 而非旧快照里的 2——守住「快照读 vs 当前读」根因。"""
+        from app.services import approval_service
+
+        factory = get_session_factory()
+        with factory() as s1:
+            # 锚定快照（模拟请求级鉴权读先于业务写发生）
+            s1.execute(text("SELECT order_no FROM approval_orders LIMIT 1")).all()
+
+            with factory() as s2:
+                order = ApprovalOrder(
+                    order_no=f"AP-{uuid.uuid4().hex[:12]}",
+                    order_type=1,
+                    ref_id=987654321,
+                    refer_no="YW-SNAPSHOT-TEST",
+                    version_no=2,
+                    status=5,
+                    submitter_name="t_snapshot_probe",
+                )
+                s2.add(order)
+                s2.commit()
+
+            # s1 的普通读仍停在旧快照（该分组在快照里不存在）
+            stale = s1.execute(
+                text(
+                    "SELECT COALESCE(MAX(version_no),0) FROM approval_orders "
+                    "WHERE order_type=1 AND ref_id=987654321"
+                )
+            ).scalar_one()
+            assert stale == 0, "前置失效：快照读应停在锚定时刻，先修隔离级别再跑本用例"
+
+            assert approval_service._next_version(s1, 1, 987654321) == 3, (
+                "取号必须读到最新已提交值（2+1=3），读到 2 说明退化为快照读"
+            )
+            s1.rollback()
+
+    def test_唯一键生效_重复版本号被拒(self) -> None:
+        """重复 (order_type, ref_id, version_no) 直插必须撞 1062——
+        应用层取号即使被绕过，DB 也是最后一道防线。"""
+        from sqlalchemy.exc import IntegrityError
+
+        factory = get_session_factory()
+        with factory() as s:
+            common = {
+                "order_type": 1,
+                "ref_id": 987654322,
+                "refer_no": "YW-UK-TEST",
+                "status": 5,
+                "submitter_name": "t_uk_probe",
+            }
+            s.add(ApprovalOrder(order_no=f"AP-{uuid.uuid4().hex[:12]}", version_no=1, **common))
+            s.flush()
+            s.add(ApprovalOrder(order_no=f"AP-{uuid.uuid4().hex[:12]}", version_no=1, **common))
+            with pytest.raises(IntegrityError) as exc_info:
+                s.flush()
+            assert is_duplicate_entry(exc_info.value), "应为唯一键冲突（1062）而非其他完整性错误"
+            s.rollback()
+
+    def test_正常路径版本号仍递增(self, client: TestClient, admin: dict, switched_on) -> None:
+        """常规编辑两次：送审版本号 1→2→3——锁读改写不得破坏既有行为。"""
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+
+        first = _latest_order(client, admin["headers"], data["business_no"])
+        assert first["version_no"] == 1
+
+        for ratio in ("60.00", "70.00"):
+            detail = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()[
+                "data"
+            ]
+            payload = business_payload(
+                detail["customer_id"],
+                secondary_allocations=[{"personnel_id": p["id"], "ratio": ratio}],
+            )
+            resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+            assert resp.status_code == 200, resp.text
+
+        third = _latest_order(client, admin["headers"], data["business_no"])
+        assert third["version_no"] == 3
+
+
+class TestConflictConvergence:
+    """编辑路径撞版本号唯一键 → 收敛为可重试的 400，业务数据不写坏。"""
+
+    def test_编辑路径版本号冲突_收敛为400(
+        self, client: TestClient, admin: dict, switched_on, monkeypatch
+    ) -> None:
+        from app.services import approval_service
+
+        p = _person(client, admin["headers"])
+        data = _create_with_plan(client, admin["headers"], p["id"])
+        before = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+
+        # 模拟取号读到过期快照：恒返回 1（库中 V1 已存在）→ 撞唯一键
+        monkeypatch.setattr(approval_service, "_next_version", lambda *a, **k: 1)
+
+        payload = business_payload(
+            before["customer_id"],
+            secondary_allocations=[{"personnel_id": p["id"], "ratio": "80.00"}],
+        )
+        resp = client.put(f"{_BUSINESSES}/{data['id']}", json=payload, headers=admin["headers"])
+        assert resp.status_code == 400, resp.text
+        assert "版本号生成重复" in resp.json()["message"]
+
+        monkeypatch.undo()
+        # 业务数据不得写坏：整个事务回滚，方案仍是 50.00（V1 生效行）
+        after = client.get(f"{_BUSINESSES}/{data['id']}", headers=admin["headers"]).json()["data"]
+        assert [a["ratio"] for a in after["secondary_allocations"]] == ["50.00"]
+        assert [h["version_no"] for h in after["allocation_history"]] == [1]

backend/tests/integration/test_audit_retry.py

@@ -13,21 +13,24 @@ audit_service.py「上下文内不得吞掉异常后正常退出」的约定相
 """
 
 import uuid
+from datetime import date
 
 import pytest
 from fastapi.testclient import TestClient
 from sqlalchemy import func, select
 
-from app.core.constants import ACTION_CREATE, MODULE_BUSINESS
+from app.core.constants import ACTION_CREATE, MODULE_BUSINESS, MODULE_RECEIPT
 from app.core.db import get_session_factory
 from app.main import app
 from app.models import OperationLog
-from app.services import business_service
+from app.services import business_service, receipt_service
+from tests.integration.factories import create_business
 
 pytestmark = pytest.mark.integration
 
 _BUSINESSES = "/api/v1/business/businesses"
 _CUSTOMERS = "/api/v1/base/customers"
+_RECEIPTS = "/api/v1/business/receipts"
 
 
 @pytest.fixture
@@ -160,3 +163,94 @@ class TestAuditRetry:
         # 每层各写一条记录；修复后重试有上限,最终失败仅一条失败日志
         assert _log_count(normal["username"], success=False) == base_fail + 1
         assert _log_count(normal["username"], success=True) == base_ok
+
+
+def _receipt_log_count(username: str, *, success: bool) -> int:
+    """按操作人统计「新增回款」操作日志条数。"""
+    factory = get_session_factory()
+    with factory() as session:
+        stmt = (
+            select(func.count())
+            .select_from(OperationLog)
+            .where(
+                OperationLog.username == username,
+                OperationLog.module == MODULE_RECEIPT,
+                OperationLog.action == ACTION_CREATE,
+                OperationLog.summary.like("新增回款%"),
+                OperationLog.success == int(success),
+            )
+        )
+        return int(session.execute(stmt).scalar_one())
+
+
+def _receipt_payload(business_id: int) -> dict:
+    return {
+        "business_id": business_id,
+        "receipt_date": str(date.today()),
+        "amount": "1200.00",
+    }
+
+
+class TestReceiptAuditRetry:
+    """回款新增的 1062 重试路径（回款与业务同构，但走独立实现）。
+
+    回款侧的重试循环本次从 create 函数体内移入 helper 调用，重试边界、
+    审计条数、批次号重取都必须保持原样 —— 用例会盯住这三条。
+    """
+
+    def test_首次撞号重试成功_审计恰一条真成功(
+        self, quiet_client: TestClient, normal: dict, monkeypatch
+    ) -> None:
+        client = quiet_client
+        business = create_business(client, normal["headers"])
+        first = client.post(
+            _RECEIPTS, json=_receipt_payload(business["id"]), headers=normal["headers"]
+        )
+        assert first.status_code == 200, first.text
+        existing_no = first.json()["data"]["receipt_batch_no"]
+
+        real_generate = receipt_service._generate_no
+        calls = {"n": 0}
+
+        def flaky_generate(session):
+            calls["n"] += 1
+            if calls["n"] == 1:
+                return existing_no  # 首次返回库内已存在批次号 → flush 撞 1062
+            return real_generate(session)
+
+        monkeypatch.setattr(receipt_service, "_generate_no", flaky_generate)
+        base_ok = _receipt_log_count(normal["username"], success=True)
+        base_fail = _receipt_log_count(normal["username"], success=False)
+
+        second = client.post(
+            _RECEIPTS, json=_receipt_payload(business["id"]), headers=normal["headers"]
+        )
+        assert second.status_code == 200, second.text
+        assert calls["n"] == 2, "重试应恰好再取一次号"
+        assert _receipt_log_count(normal["username"], success=True) == base_ok + 1
+        assert _receipt_log_count(normal["username"], success=False) == base_fail
+
+    def test_持续冲突_最终失败且审计恰一条失败(
+        self, quiet_client: TestClient, normal: dict, monkeypatch
+    ) -> None:
+        client = quiet_client
+        business = create_business(client, normal["headers"])
+        first = client.post(
+            _RECEIPTS, json=_receipt_payload(business["id"]), headers=normal["headers"]
+        )
+        assert first.status_code == 200, first.text
+        existing_no = first.json()["data"]["receipt_batch_no"]
+
+        def always_conflict(session):
+            return existing_no
+
+        monkeypatch.setattr(receipt_service, "_generate_no", always_conflict)
+        base_ok = _receipt_log_count(normal["username"], success=True)
+        base_fail = _receipt_log_count(normal["username"], success=False)
+
+        second = client.post(
+            _RECEIPTS, json=_receipt_payload(business["id"]), headers=normal["headers"]
+        )
+        assert second.status_code == 400, second.text
+        assert _receipt_log_count(normal["username"], success=False) == base_fail + 1
+        assert _receipt_log_count(normal["username"], success=True) == base_ok

backend/tests/integration/test_businesses.py

@@ -452,9 +452,7 @@ class TestEdit:
         assert resp.status_code == 400, resp.text
         assert "含合同负责营销人员" in resp.json()["message"]
 
-    def test_二次分配方案合法比例_保存成功且回显(
-        self, client: TestClient, normal: dict
-    ) -> None:
+    def test_二次分配方案合法比例_保存成功且回显(self, client: TestClient, normal: dict) -> None:
         headers = normal["headers"]
         customer = _customer(client, headers)
         owner = _personnel(client, headers)

backend/tests/integration/test_customers.py

@@ -376,3 +376,73 @@ class TestAudit:
         latest = logs[0]
         assert customer["company_name"] in latest.summary, "摘要应含客户名"
         assert "→" in latest.summary, "摘要应体现新旧归属走向"
+
+
+class TestRating:
+    """评级保存与列表排序 —— 评分是档案属性，只存总分与评级时间。
+
+    守的是：评分只走专属端点（编辑端点不含评分，旁路写评分会绕开口径）；
+    分值红线 0-100 在入口拦截；按评分排序时未评分客户必须垫底，不能因
+    MySQL 把 NULL 视为最小值而在升序里压过打了低分的客户。
+    """
+
+    def test_保存评级_回显总分与时间_且编辑端点不含评分(
+        self, client: TestClient, normal: dict
+    ) -> None:
+        headers = normal["headers"]
+        customer = _create_customer(client, headers)
+        assert customer["rating_score"] is None, "新客户应无评分"
+
+        resp = client.put(
+            f"{_CUSTOMERS}/{customer['id']}/rating", json={"score": 88}, headers=headers
+        )
+        assert resp.status_code == 200, resp.text
+        body = resp.json()["data"]
+        assert body["rating_score"] == 88
+        assert body["rating_time"] is not None, "保存评分必须刷新评级时间"
+
+        # 覆盖式保存：新分数替换旧分数
+        resp = client.put(
+            f"{_CUSTOMERS}/{customer['id']}/rating", json={"score": 60}, headers=headers
+        )
+        assert resp.status_code == 200, resp.text
+        assert resp.json()["data"]["rating_score"] == 60
+
+    def test_评分越界_400(self, client: TestClient, normal: dict) -> None:
+        headers = normal["headers"]
+        customer = _create_customer(client, headers)
+        resp = client.put(
+            f"{_CUSTOMERS}/{customer['id']}/rating", json={"score": 101}, headers=headers
+        )
+        assert resp.status_code == 400, resp.text
+
+    def test_按评分排序_未评分垫底_升序降序都不压过低分(
+        self, client: TestClient, normal: dict
+    ) -> None:
+        headers = normal["headers"]
+        high = _create_customer(client, headers)
+        low = _create_customer(client, headers)
+        unrated = _create_customer(client, headers)
+        for target, score in ((high, 80), (low, 60)):
+            resp = client.put(
+                f"{_CUSTOMERS}/{target['id']}/rating", json={"score": score}, headers=headers
+            )
+            assert resp.status_code == 200, resp.text
+
+        def _order_names(order: str) -> list[str]:
+            resp = client.get(
+                _CUSTOMERS,
+                params={"pageNum": 1, "pageSize": 50, "order_by": "rating_score", "order": order},
+                headers=headers,
+            )
+            assert resp.status_code == 200, resp.text
+            names = [row["company_name"] for row in resp.json()["data"]["list"]]
+            return names
+
+        # 降序：高分在前，未评分最后；升序：低分在前，未评分仍垫底
+        desc_names = _order_names("desc")
+        assert desc_names.index(high["company_name"]) < desc_names.index(low["company_name"])
+        assert desc_names[-1] == unrated["company_name"]
+        asc_names = _order_names("asc")
+        assert asc_names.index(low["company_name"]) < asc_names.index(high["company_name"])
+        assert asc_names[-1] == unrated["company_name"], "未评分必须垫底"

backend/tests/integration/test_dashboard.py

@@ -118,10 +118,20 @@ def _dashboard(client: TestClient, headers: dict, month: str | None = None) -> d
     return resp.json()["data"]
 
 
+def _activities(client: TestClient, headers: dict, **params) -> dict:
+    """最近动态分页端点。"""
+    resp = client.get(f"{_DASHBOARD}/activities", params=params, headers=headers)
+    assert resp.status_code == 200, resp.text
+    return resp.json()["data"]
+
+
 def test_权限_普通用户403文案(client: TestClient, normal: dict):
     resp = client.get(_DASHBOARD, headers=normal["headers"])
     assert resp.status_code == 403
     assert resp.json()["message"] == "无权访问工作台"
+    # 最近动态分页端点同属工作台，权限一致
+    resp = client.get(f"{_DASHBOARD}/activities", headers=normal["headers"])
+    assert resp.status_code == 403
 
 
 def test_总览_绩效合计与台账一致且非法月份400(client: TestClient, admin: dict):
@@ -138,9 +148,7 @@ def test_总览_绩效合计与台账一致且非法月份400(client: TestClient
         expected = Decimal(
             str(
                 session.execute(
-                    select(
-                        func.coalesce(func.sum(PerformanceLedger.performance_amount), 0)
-                    ).where(
+                    select(func.coalesce(func.sum(PerformanceLedger.performance_amount), 0)).where(
                         PerformanceLedger.settlement_month == _month(),
                         PerformanceLedger.calc_type == 1,
                         PerformanceLedger.is_active == 1,
@@ -166,16 +174,12 @@ def test_待办_待审批与未指定批次与同步降级(client: TestClient, a
     )
     assert resp.status_code == 200, resp.text
     settlement_id = resp.json()["data"]["id"]
-    resp = client.post(
-        f"{_SETTLEMENTS}/{settlement_id}/submit", headers=admin["headers"]
-    )
+    resp = client.post(f"{_SETTLEMENTS}/{settlement_id}/submit", headers=admin["headers"])
     assert resp.status_code == 200, resp.text
 
     data = _dashboard(client, admin["headers"])
     assert data["todos"]["pending_approval_total"] >= 1
-    assert any(
-        item["order_type"] == 2 for item in data["todos"]["pending_approvals"]
-    )
+    assert any(item["order_type"] == 2 for item in data["todos"]["pending_approvals"])
     assert data["todos"]["unassigned_receipt_total"] >= 1
     assert data["todos"]["unassigned_receipts"][0]["receipt_id"] == ctx["receipt"]["id"]
     # ⑥ 未建：显式降级
@@ -227,9 +231,40 @@ def test_预警_审批超时命中且参数关闭后不产生(client: TestClient
 
 def test_动态_来源操作日志(client: TestClient, admin: dict):
     _chain(client, admin["headers"])
-    data = _dashboard(client, admin["headers"])
-    assert data["activities"], "工作台动态为空：写操作应产生操作日志"
-    assert {"module", "action", "summary", "create_time"} <= set(data["activities"][0])
+    data = _activities(client, admin["headers"])
+    assert data["list"], "工作台动态为空：写操作应产生操作日志"
+    assert {"module", "action", "summary", "create_time"} <= set(data["list"][0])
+
+
+def test_动态_分页翻页不重不漏(client: TestClient, admin: dict):
+    _chain(client, admin["headers"])
+    first = _activities(client, admin["headers"], pageNum=1, pageSize=2)
+    assert first["pageNum"] == 1
+    assert first["pageSize"] == 2
+    assert len(first["list"]) == 2
+    assert first["total"] >= 2
+
+    second = _activities(client, admin["headers"], pageNum=2, pageSize=2)
+    assert len(second["list"]) == 2
+    # 两页不重不漏（倒序取，按 id 倒排）
+    first_ids = {item["id"] for item in first["list"]}
+    second_ids = {item["id"] for item in second["list"]}
+    assert first_ids.isdisjoint(second_ids)
+    assert first["list"][0]["id"] > second["list"][0]["id"]
+
+
+def test_动态_模块筛选与筛选项(client: TestClient, admin: dict):
+    _chain(client, admin["headers"])
+    page = _activities(client, admin["headers"], pageSize=50)
+    assert page["modules"], "模块筛选项不应为空"
+    module = page["modules"][0]
+
+    filtered = _activities(client, admin["headers"], module=module, pageSize=50)
+    assert filtered["total"] >= 1
+    assert filtered["list"], "按模块筛选后应有数据"
+    assert all(item["module"] == module for item in filtered["list"])
+    # 筛选项不随筛选条件变化（下拉里始终是全量模块）
+    assert filtered["modules"] == page["modules"]
 
 
 # ==== 角色待办 / 预警阈值 / 趋势（2026-10-19 code-review 修复回归）====
@@ -250,7 +285,13 @@ def leader(make_user, auth_headers) -> dict:
 
 
 def _insert_approval(status: int) -> None:
-    """直插在途审批单（避免依赖两次流转，专注工作台过滤口径）。"""
+    """直插在途审批单（避免依赖两次流转，专注工作台过滤口径）。
+
+    ``ref_id`` 仅作对象分组锚点、本用例不校验：**每张单必须给不同的
+    ref_id**——V12 的 ``uk_approval_orders_order_type_ref_version`` 要求
+    (order_type, ref_id, version_no) 唯一，同一假对象连插两张 version_no=1
+    会撞 1062。
+    """
     from app.core.db import get_session_factory
     from app.models import ApprovalOrder
 
@@ -260,7 +301,7 @@ def _insert_approval(status: int) -> None:
             ApprovalOrder(
                 order_no=f"AP{uuid.uuid4().hex[:12]}",
                 order_type=2,
-                ref_id=0,
+                ref_id=status,
                 refer_no="t_工作台",
                 status=status,
                 submitter_name="t_admin",
@@ -329,17 +370,17 @@ def test_待办_按角色仅显示可操作节点(
     assert _dashboard(client, both["headers"])["todos"]["pending_approval_total"] == 2
 
     finance_orders = {
-        item["order_no"] for item in _dashboard(client, finance["headers"])["todos"]["pending_approvals"]
+        item["order_no"]
+        for item in _dashboard(client, finance["headers"])["todos"]["pending_approvals"]
     }
     leader_orders = {
-        item["order_no"] for item in _dashboard(client, leader["headers"])["todos"]["pending_approvals"]
+        item["order_no"]
+        for item in _dashboard(client, leader["headers"])["todos"]["pending_approvals"]
     }
     assert finance_orders.isdisjoint(leader_orders)
 
 
-def test_待办_财务通过后领导出现待办(
-    client: TestClient, admin: dict, finance: dict, leader: dict
-):
+def test_待办_财务通过后领导出现待办(client: TestClient, admin: dict, finance: dict, leader: dict):
     """真实流转：财务节点通过后，条目从财务待办消失并进入领导待办。"""
     _chain(client, admin["headers"])
     resp = client.post(
@@ -347,9 +388,7 @@ def test_待办_财务通过后领导出现待办(
     )
     assert resp.status_code == 200, resp.text
     settlement_id = resp.json()["data"]["id"]
-    resp = client.post(
-        f"{_SETTLEMENTS}/{settlement_id}/submit", headers=admin["headers"]
-    )
+    resp = client.post(f"{_SETTLEMENTS}/{settlement_id}/submit", headers=admin["headers"])
     assert resp.status_code == 200, resp.text
 
     order = _dashboard(client, admin["headers"])["todos"]["pending_approvals"][0]

backend/tests/integration/test_param_management.py

@@ -326,6 +326,38 @@ class TestPreview:
         # Fraction 自动约分：4/1000 存储为 1/250，按数值断言
         assert Fraction(data["rate"]) == Fraction(4, 1000)
 
+    def test_预览参数使用顺序稳定(self, client: TestClient, admin: dict) -> None:
+        """params_used 的顺序即前端「用了哪些参数」的展示顺序。
+
+        预览实现拆成多段后，各段的调用先后决定参数读取先后；顺序漂移会让
+        复核者看到的参数清单与核算口径对不上（不报错，但会误导判读）。
+        """
+        payload = {
+            "business_role": "salesperson",
+            "customer_type": "personal",
+            "product_category": "municipal_water",
+            "ship_date": "2025-12-15",
+            "receipt_date": "2026-01-10",
+            "revenue": "100000",
+            "profit": "45000",
+            "received_amount": "60000",
+            "barter": True,
+            "probation": True,
+        }
+        data = client.post(f"{_PARAMS}/preview", json=payload, headers=admin["headers"]).json()[
+            "data"
+        ]
+        assert [p["param_code"] for p in data["params_used"]] == [
+            "salesperson.rate.start",
+            "salesperson.rate.slope",
+            "salesperson.rate.floor",
+            "salesperson.days.boundary",
+            "product.threshold",
+            "barter.discount",
+            "payment.start.ratio",
+            "probation.annual.threshold",
+        ]
+
     def test_经理聚合口径_预览拒绝(self, client: TestClient, admin: dict) -> None:
         payload = {
             "business_role": "sales_manager",
@@ -339,7 +371,11 @@ class TestPreview:
         }
         resp = client.post(f"{_PARAMS}/preview", json=payload, headers=admin["headers"])
         assert resp.status_code == 400
-        assert "聚合" in resp.json()["message"]
+        # 守两件事：① 聚合口径不允许单笔试算（拒绝而不是算出个错数）；
+        # ② 提示要能指导下一步（点名是谁的口径 + 去哪儿看），不是一句「不支持」。
+        message = resp.json()["message"]
+        assert "营销经理" in message
+        assert "月度结算" in message
 
     def test_回款日早于发货日_400(self, client: TestClient, admin: dict) -> None:
         payload = {

backend/tests/integration/test_reports.py

@@ -44,7 +44,9 @@ def _role_id(code: str) -> int:
         )
 
 
-def _person(client: TestClient, headers: dict, *, role: str = "salesperson", manager_id=None) -> dict:
+def _person(
+    client: TestClient, headers: dict, *, role: str = "salesperson", manager_id=None
+) -> dict:
     body = {
         "personnel_name": f"t_报表_{uuid.uuid4().hex[:8]}",
         "business_role_id": _role_id(role),
@@ -194,9 +196,7 @@ def test_部门汇总_无成员经理出零值行(admin, client: TestClient):
         headers=admin["headers"],
     )
     assert resp.status_code == 200, resp.text
-    team = next(
-        (t for t in resp.json()["data"]["teams"] if t["manager_id"] == manager["id"]), None
-    )
+    team = next((t for t in resp.json()["data"]["teams"] if t["manager_id"] == manager["id"]), None)
     assert team is not None
     assert Decimal(str(team["team_base"])) == Decimal("0")
     assert team["member_count"] == 0
@@ -224,7 +224,14 @@ def test_回款统计_空桶恒在且分摊拆行金额只计一次(admin, clien
 
     assert len(data["buckets"]) == 8
     assert [b["code"] for b in data["buckets"]] == [
-        "le0", "1-30", "31-60", "61-90", "91-180", "181-360", "361+", "none",
+        "le0",
+        "1-30",
+        "31-60",
+        "61-90",
+        "91-180",
+        "181-360",
+        "361+",
+        "none",
     ]
     # 60 天前发货 → 落在 61-90 桶（发货日首日计第 1 天）
     bucket = next(b for b in data["buckets"] if b["code"] == "61-90")
@@ -293,11 +300,15 @@ def test_导出_excel含签字栏且入操作日志(admin, client: TestClient):
 
     factory = get_session_factory()
     with factory() as session:
-        log = session.execute(
-            select(OperationLog)
-            .where(OperationLog.module == "报表统计", OperationLog.action == "导出")
-            .order_by(OperationLog.id.desc())
-        ).scalars().first()
+        log = (
+            session.execute(
+                select(OperationLog)
+                .where(OperationLog.module == "报表统计", OperationLog.action == "导出")
+                .order_by(OperationLog.id.desc())
+            )
+            .scalars()
+            .first()
+        )
     assert log is not None
 
 
@@ -389,19 +400,21 @@ def _latest_export_log():
 
     factory = get_session_factory()
     with factory() as session:
-        return session.execute(
-            select(OperationLog)
-            .where(OperationLog.module == "报表统计", OperationLog.action == "导出")
-            .order_by(OperationLog.id.desc())
-        ).scalars().first()
+        return (
+            session.execute(
+                select(OperationLog)
+                .where(OperationLog.module == "报表统计", OperationLog.action == "导出")
+                .order_by(OperationLog.id.desc())
+            )
+            .scalars()
+            .first()
+        )
 
 
 def test_部门汇总_无台账经理挂上级不报500(admin, client: TestClient):
     """经理 A 无台账且上级是经理 B：不能因 member_map 直接索引而 KeyError。"""
     manager_b = _person(client, admin["headers"], role="sales_manager")
-    manager_a = _person(
-        client, admin["headers"], role="sales_manager", manager_id=manager_b["id"]
-    )
+    manager_a = _person(client, admin["headers"], role="sales_manager", manager_id=manager_b["id"])
 
     resp = client.get(
         f"{_REPORTS}/department-summary",
@@ -467,9 +480,7 @@ def test_部门汇总_经理个人户计入时基数等于成员合计(admin, cl
             headers=admin["headers"],
         )
         assert resp.status_code == 200, resp.text
-        team = next(
-            t for t in resp.json()["data"]["teams"] if t["manager_id"] == manager["id"]
-        )
+        team = next(t for t in resp.json()["data"]["teams"] if t["manager_id"] == manager["id"])
         assert Decimal(str(team["team_base"])) == Decimal("100.00")
         assert sum(
             (Decimal(str(m["performance_sum"])) for m in team["members"]), Decimal("0")
@@ -516,9 +527,7 @@ def test_部门汇总_按经理筛选不误判其他团队下属(admin, client:
     assert [t["manager_id"] for t in data["teams"]] == [manager_1["id"]]
     assert data["unassigned"] == []
     assert all(
-        m["personnel_id"] != salesperson_2["id"]
-        for team in data["teams"]
-        for m in team["members"]
+        m["personnel_id"] != salesperson_2["id"] for team in data["teams"] for m in team["members"]
     )
     assert Decimal(str(data["reconciliation"]["other_team_base"])) > 0
 
@@ -557,9 +566,7 @@ def test_部门汇总_对账恒等式(admin, client: TestClient):
     )
 
 
-def _two_manager_scene(
-    client: TestClient, headers: dict
-) -> tuple[dict, dict, dict, dict, dict]:
+def _two_manager_scene(client: TestClient, headers: dict) -> tuple[dict, dict, dict, dict, dict]:
     """两位经理各有下属与业务 + 一位零成员经理（无聚合行）。"""
     manager_1 = _person(client, headers, role="sales_manager")
     manager_2 = _person(client, headers, role="sales_manager")
@@ -777,9 +784,7 @@ def test_导出取数异常_失败审计留痕(admin, client: TestClient, monkey
     assert "导出个人绩效工资明细表" in (log.summary or "")
 
 
-@pytest.mark.parametrize(
-    "path", ["/personal-detail", "/department-summary", "/payment-stats"]
-)
+@pytest.mark.parametrize("path", ["/personal-detail", "/department-summary", "/payment-stats"])
 def test_非法月份_查询返回400(admin, client: TestClient, path: str):
     """非法月份必须是参数错误 400，不能落 500。"""
     resp = client.get(
@@ -790,9 +795,7 @@ def test_非法月份_查询返回400(admin, client: TestClient, path: str):
     assert resp.status_code == 400, resp.text
 
 
-@pytest.mark.parametrize(
-    "code", ["personal-detail", "department-summary", "payment-stats"]
-)
+@pytest.mark.parametrize("code", ["personal-detail", "department-summary", "payment-stats"])
 def test_非法月份_导出返回400(admin, client: TestClient, code: str):
     """导出手工构造查询模型同样要转 400。"""
     resp = client.get(

backend/tests/integration/test_secondary_allocation.py

@@ -247,6 +247,36 @@ class TestGuards:
 class TestDeleteProtection:
     """RESTRICT 外键：被方案引用的人员禁删，方案随业务删除而清。"""
 
+    def test_同一会话内旧方案行同步为历史态(self, client: TestClient, admin: dict) -> None:
+        """守「置历史后同一会话内的旧行必须同步刷新」这一行为：Core UPDATE
+        走的是 SQLAlchemy 的会话同步策略，条件复杂化时可能退化为不刷新，
+        届时行对象会停留在 is_active=1 的过期内存值（库中实际已置历史），
+        被序列化或参与判断就会呈现「生效态」的假象。"""
+        from app.core.db import get_session_factory
+        from app.models import SecondaryAllocation
+        from app.repositories import business_repository
+
+        p = _person(client, admin["headers"])
+        data = _create_with_allocations(
+            client, admin["headers"], [{"personnel_id": p["id"], "ratio": "50.00"}]
+        )
+
+        factory = get_session_factory()
+        with factory() as session:
+            loaded = business_repository.list_allocations(session, data["id"])
+            assert loaded and all(a.is_active == 1 for a in loaded)
+            business_repository.deactivate_allocations(session, data["id"])
+            # 同一 session 内的行对象必须已是历史态，不能停留在过期内存值
+            assert all(a.is_active == 0 for a in loaded), "旧行未随置历史同步刷新"
+            # 库中同样已置历史
+            rows = (
+                session.query(SecondaryAllocation)
+                .filter(SecondaryAllocation.business_id == data["id"])
+                .all()
+            )
+            assert rows and all(r.is_active == 0 for r in rows)
+            session.rollback()
+
     def test_被方案引用的人员禁删(self, client: TestClient, admin: dict) -> None:
         p = _person(client, admin["headers"])
         data = _create_with_allocations(

backend/tests/integration/test_settlements.py

@@ -229,6 +229,19 @@ class TestSubmitAndLock:
         assert out["approval_order_no"] and out["approval_order_no"].startswith("AP")
         assert out["locked_at"] is not None
 
+        # 自动通过路径同样固化送审快照（月结头 + 按人明细）
+        order = client.get(
+            "/api/v1/system/approvals",
+            params={"keyword": out["approval_order_no"], "pageSize": 1},
+            headers=admin["headers"],
+        ).json()["data"]["list"][0]
+        detail = client.get(
+            f"/api/v1/system/approvals/{order['id']}", headers=admin["headers"]
+        ).json()["data"]
+        assert detail["content_source"] == "snapshot"
+        assert detail["snapshot"]["data"]["settlement_month"] == "2026-10"
+        assert detail["snapshot"]["data"]["items"], "月结快照须含按人明细"
+
     def test_锁定后回款禁改(self, client: TestClient, admin: dict) -> None:
         from sqlalchemy import select
 
@@ -474,8 +487,18 @@ class TestManagerAggregate:
         assert len([i for i in out["items"] if i["calc_type"] == 2]) == 1
         assert len([i for i in out["items"] if i["calc_type"] == 1]) == 1
 
-    def test_月末取参_预约变更影响本月聚合(self, client: TestClient, admin: dict) -> None:
+    def test_月末取参_预约变更影响本月聚合(
+        self, client: TestClient, admin: dict, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
         """取参锚点=所属月月末：10-25 生效的预约变更应影响 2026-10 月结。"""
+
+        class ScenarioDate(date):
+            @classmethod
+            def today(cls):
+                return cls(2026, 10, 20)
+
+        # 固定预约场景的申请日，避免现实日期超过 10-25 后误测成“禁止追溯修改”。
+        monkeypatch.setattr("app.services.param_service.date", ScenarioDate)
         resp = client.post(
             "/api/v1/system/params/changes",
             json={
@@ -603,6 +626,21 @@ class TestAdjustment:
         out = resp.json()["data"]
         assert out["status"] == 2  # 已生效
 
+        # 调整单送审快照：复核者能看到金额/方向/说明，而非只有调整单号
+        order = client.get(
+            "/api/v1/system/approvals",
+            params={"keyword": out["order_no"], "pageSize": 1},
+            headers=admin["headers"],
+        ).json()["data"]["list"][0]
+        assert order["order_type"] == 3
+        detail = client.get(
+            f"/api/v1/system/approvals/{order['id']}", headers=admin["headers"]
+        ).json()["data"]
+        assert detail["content_source"] == "snapshot"
+        assert detail["snapshot"]["data"]["amount"] == "500.00"
+        assert detail["snapshot"]["data"]["direction"] == 1
+        assert detail["snapshot"]["data"]["reason"] == "补发差额"
+
     def test_金额恒正校验(self, client: TestClient, admin: dict) -> None:
         """守金额无校验缺陷：负数会反转补发/扣减语义，非数值直落库 500。"""
         _set_switch(client, admin["headers"], "approval.adjustment.enabled", "0")

backend/tests/integration/test_tplus_concurrency.py

@@ -0,0 +1,90 @@
+"""在隔离 MySQL 中验证幂等竞争、父批次串行化和连接级锁。"""
+
+from concurrent.futures import ThreadPoolExecutor
+from threading import Barrier
+from uuid import uuid4
+
+import pytest
+from sqlalchemy import delete, func, select, text
+
+from app.core.db import get_session_factory
+from app.core.exceptions import BadRequestError
+from app.core.tplus_lock import TPlusBusyError, sync_lock
+from app.models import TPlusSyncBatch
+from app.schemas.tplus import TPlusCreateIn, TPlusRetryIn
+from app.services.tplus import task_service
+
+pytestmark = pytest.mark.integration
+
+
+def _parallel(action):
+    barrier = Barrier(2)
+
+    def run():
+        with get_session_factory()() as session:
+            barrier.wait(timeout=10)
+            return action(session)
+
+    with ThreadPoolExecutor(max_workers=2) as pool:
+        futures = [pool.submit(run) for _ in range(2)]
+        return [future.result(timeout=20) for future in futures]
+
+
+def test_并发相同请求只创建一个批次(tplus_tasks):
+    payload = TPlusCreateIn(**tplus_tasks())
+    numbers = _parallel(lambda session: task_service.create_task(session, payload).batch_no)
+    assert numbers[0] == numbers[1]
+    with get_session_factory()() as session:
+        assert (
+            session.scalar(
+                select(func.count())
+                .select_from(TPlusSyncBatch)
+                .where(TPlusSyncBatch.request_key == str(payload.request_key))
+            )
+            == 1
+        )
+
+
+def test_并发不同请求重试同父批次只允许一个(tplus_tasks):
+    with get_session_factory()() as session:
+        parent = task_service.create_task(session, TPlusCreateIn(**tplus_tasks()))
+        parent.status = "FAILED"
+        session.commit()
+        number = parent.batch_no
+
+    def retry(session):
+        try:
+            return task_service.retry_task(
+                session, number, TPlusRetryIn(request_key=uuid4())
+            ).batch_no
+        except BadRequestError:
+            session.rollback()
+            return "REJECTED"
+
+    try:
+        numbers = _parallel(retry)
+        assert numbers.count("REJECTED") == 1
+        with get_session_factory()() as session:
+            assert (
+                session.scalar(
+                    select(func.count())
+                    .select_from(TPlusSyncBatch)
+                    .where(TPlusSyncBatch.parent_batch_no == number)
+                )
+                == 1
+            )
+    finally:
+        with get_session_factory()() as session:
+            session.execute(delete(TPlusSyncBatch).where(TPlusSyncBatch.parent_batch_no == number))
+            session.commit()
+
+
+def test_执行锁跨提交保持且释放后可获取():
+    with get_session_factory()() as owner, get_session_factory()() as other:
+        with sync_lock(owner):
+            owner.execute(text("SELECT 1"))
+            owner.commit()
+            with pytest.raises(TPlusBusyError), sync_lock(other):
+                pytest.fail("另一执行者不应取得锁")
+        with sync_lock(other):
+            assert other.scalar(text("SELECT 1")) == 1

backend/tests/integration/test_tplus_management.py

@@ -0,0 +1,129 @@
+"""管理员可管理原始读取任务；权限、幂等和异常脱敏由真实接口验证。"""
+
+import pytest
+from sqlalchemy import select
+
+from app.core.db import get_session_factory
+from app.models import TPlusSyncBatch, TPlusSyncError
+
+pytestmark = pytest.mark.integration
+BASE = "/api/v1/system/tplus"
+
+
+def test_未登录和普通用户不能访问管理(client, admin, normal, tplus_tasks):
+    assert client.get(BASE + "/overview").status_code == 401
+    assert client.get(BASE + "/overview", headers=normal["headers"]).status_code == 403
+    assert (
+        client.post(BASE + "/batches", json=tplus_tasks(), headers=normal["headers"]).status_code
+        == 403
+    )
+
+
+def test_创建重复请求返回同一批次并可分页查看(client, admin, tplus_tasks):
+    payload = tplus_tasks()
+    first = client.post(BASE + "/batches", json=payload, headers=admin["headers"])
+    assert first.status_code == 200
+    second = client.post(BASE + "/batches", json=payload, headers=admin["headers"])
+    assert second.status_code == 200
+    number = first.json()["data"]["batch_no"]
+    assert second.json()["data"]["batch_no"] == number
+    page = client.get(
+        BASE + "/batches", params={"pageSize": 1, "status": "QUEUED"}, headers=admin["headers"]
+    )
+    assert page.status_code == 200 and page.json()["data"]["pageSize"] == 1
+    detail = client.get(BASE + "/batches/" + number, headers=admin["headers"])
+    assert detail.status_code == 200
+    assert detail.json()["data"]["status"] == "QUEUED"
+
+
+def test_异常返回固定说明不泄漏数据库参数(client, admin, tplus_tasks):
+    response = client.post(BASE + "/batches", json=tplus_tasks(), headers=admin["headers"])
+    assert response.status_code == 200
+    number = response.json()["data"]["batch_no"]
+    with get_session_factory()() as db:
+        batch = db.scalar(select(TPlusSyncBatch).where(TPlusSyncBatch.batch_no == number))
+        batch.status, batch.error_message = "FAILED", "password=private-value phone=13800000000"
+        db.add(
+            TPlusSyncError(
+                batch_no=number,
+                entity_type="personnel",
+                source_table="AA_Person",
+                source_id="12",
+                error_code="READ_ERROR",
+                error_message=batch.error_message,
+            )
+        )
+        db.commit()
+    for path in (f"/batches/{number}", f"/batches/{number}/errors"):
+        result = client.get(BASE + path, headers=admin["headers"])
+        assert result.status_code == 200
+        assert "private-value" not in result.text and "13800000000" not in result.text
+
+
+def test_不支持的实体和不存在的批次有明确错误(client, admin, tplus_tasks):
+    payload = tplus_tasks()
+    payload["entity_types"] = ["users;drop"]
+    assert client.post(BASE + "/batches", json=payload, headers=admin["headers"]).status_code == 400
+    assert client.get(BASE + "/batches/TPmissing", headers=admin["headers"]).status_code == 404
+
+
+def test_配置关闭时可查询历史但不能排队(client, admin, tplus_tasks, monkeypatch):
+    from app.services.tplus import task_service
+
+    settings = task_service.get_settings().model_copy(update={"t_plus_enabled": False})
+    monkeypatch.setattr(task_service, "get_settings", lambda: settings)
+    response = client.get(BASE + "/overview", headers=admin["headers"])
+    assert response.status_code == 200
+    assert response.json()["data"]["is_configured"] is False
+    assert response.json()["data"]["business_import_enabled"] is False
+    assert response.json()["data"]["compatibility_status"] == "PENDING"
+    assert (
+        client.post(BASE + "/batches", json=tplus_tasks(), headers=admin["headers"]).status_code
+        == 400
+    )
+
+
+def test_失败批次重试保留异常并返回新批次(client, admin, tplus_tasks):
+    from app.models import OperationLog
+
+    parent = client.post(BASE + "/batches", json=tplus_tasks(), headers=admin["headers"]).json()[
+        "data"
+    ]
+    number = parent["batch_no"]
+    with get_session_factory()() as db:
+        batch = db.scalar(select(TPlusSyncBatch).where(TPlusSyncBatch.batch_no == number))
+        batch.status, batch.failed_record_count = "FAILED", 1
+        db.add(
+            TPlusSyncError(
+                batch_no=number,
+                entity_type="personnel",
+                source_table="AA_Person",
+                source_id="1",
+                error_code="READ_ERROR",
+                error_message="读取失败",
+            )
+        )
+        db.commit()
+    request = {"request_key": tplus_tasks()["request_key"]}
+    first = client.post(BASE + f"/batches/{number}/retry", json=request, headers=admin["headers"])
+    replay = client.post(BASE + f"/batches/{number}/retry", json=request, headers=admin["headers"])
+    assert first.status_code == replay.status_code == 200
+    child = first.json()["data"]
+    assert child["parent_batch_no"] == number and child["batch_no"] != number
+    assert replay.json()["data"]["batch_no"] == child["batch_no"]
+    detail = client.get(BASE + f"/batches/{number}", headers=admin["headers"]).json()["data"]
+    assert detail["can_retry"] is False
+    with get_session_factory()() as db:
+        assert (
+            db.scalar(select(TPlusSyncError.retry_count).where(TPlusSyncError.batch_no == number))
+            == 1
+        )
+        assert db.scalar(
+            select(OperationLog.id)
+            .where(
+                OperationLog.user_id == admin["id"],
+                OperationLog.module == "T+数据同步",
+                OperationLog.action == "重试读取任务",
+            )
+            .limit(1)
+        )

backend/tests/unit/test_approval_helpers.py

@@ -0,0 +1,72 @@
+"""审批与方案版本化的纯函数单元测试（秒级，无需容器）。
+
+守的是「方案内容是否真的变了」这一判定（``business_service._plan_changed``）：
+它决定要不要新增一个方案版本行、要不要重提审批单。判错的后果是双向的——
+判成「变了」会刷出版本与审批留痕噪音，判成「没变」会让改过的方案
+沿用旧版本号，版本语义失真。
+
+比较口径是**集合**（人员 ID + 其比例），与提交顺序无关：用户只是把行
+拖了个序，不该被算成方案变了。
+"""
+
+from decimal import Decimal
+
+import pytest
+
+from app.models.business import SecondaryAllocation
+from app.services.business_service import _plan_changed
+
+
+def _row(personnel_id: int, ratio: str) -> SecondaryAllocation:
+    """构造一个未入库的方案行（ORM 实例化不触数据库）。"""
+    return SecondaryAllocation(
+        business_id=1,
+        version_no=1,
+        personnel_id=personnel_id,
+        personnel_name=f"t_人员{personnel_id}",
+        ratio=Decimal(ratio),
+        is_active=1,
+    )
+
+
+class TestPlanChanged:
+    """方案内容比较。"""
+
+    def test_完全相同_判未变(self) -> None:
+        old = [_row(1, "50.00"), _row(2, "30.00")]
+        new = [_row(1, "50.00"), _row(2, "30.00")]
+        assert _plan_changed(old, new) is False
+
+    def test_顺序不同_判未变(self) -> None:
+        """用户只是把行拖了个序 —— 送审快照按库内生效行顺序生成，
+        按顺序比较会把这种编辑误判成「方案变了」。"""
+        old = [_row(1, "50.00"), _row(2, "30.00")]
+        new = [_row(2, "30.00"), _row(1, "50.00")]
+        assert _plan_changed(old, new) is False
+
+    def test_比例变化_判已变(self) -> None:
+        old = [_row(1, "50.00")]
+        new = [_row(1, "80.00")]
+        assert _plan_changed(old, new) is True
+
+    def test_参与人增减_判已变(self) -> None:
+        old = [_row(1, "50.00")]
+        assert _plan_changed(old, [_row(1, "50.00"), _row(2, "30.00")]) is True
+        assert _plan_changed([_row(1, "50.00"), _row(2, "30.00")], [_row(1, "50.00")]) is True
+
+    def test_同一人的比例从一人换成另一人_判已变(self) -> None:
+        """参与人不同但比例相同：人员 ID 是比较键的一部分，不能只看比例。"""
+        old = [_row(1, "50.00")]
+        new = [_row(2, "50.00")]
+        assert _plan_changed(old, new) is True
+
+    @pytest.mark.parametrize("written", ["50", "50.0", "50.00"])
+    def test_比例写法不同_判未变(self, written: str) -> None:
+        """比例入参是字符串（'50' 与 '50.00' 都合法），数值相同即未变。"""
+        assert _plan_changed([_row(1, "50.00")], [_row(1, written)]) is False
+
+    def test_空方案与空方案_判未变(self) -> None:
+        assert _plan_changed([], []) is False
+
+    def test_空方案与有方案_判已变(self) -> None:
+        assert _plan_changed([], [_row(1, "50.00")]) is True

backend/tests/unit/test_approval_schema.py

@@ -0,0 +1,38 @@
+"""审批管理入参校验（schema 层，秒级）。
+
+守的是：对象类型筛选必须覆盖三类审批对象（1 二次分配方案 / 2 月度结算 /
+3 调整单）——早期上界写死在 2，调整单审批单在列表里无法筛选；越界值在
+参数绑定阶段即拒绝。驳回必须带意见（空驳无从追溯）。
+"""
+
+import pytest
+from pydantic import ValidationError
+
+from app.schemas.approval import ApprovalActionIn, ApprovalQuery
+
+
+class TestQueryOrderType:
+    """对象类型筛选边界。"""
+
+    @pytest.mark.parametrize("value", [1, 2, 3])
+    def test_三类对象均可筛选(self, value: int) -> None:
+        assert ApprovalQuery(order_type=value).order_type == value
+
+    @pytest.mark.parametrize("value", [0, 4])
+    def test_越界类型被拒(self, value: int) -> None:
+        with pytest.raises(ValidationError):
+            ApprovalQuery(order_type=value)
+
+    def test_不传类型表示全部(self) -> None:
+        assert ApprovalQuery().order_type is None
+
+
+class TestActionOpinion:
+    """节点动作的驳回意见要求。"""
+
+    def test_驳回必须填意见(self) -> None:
+        with pytest.raises(ValidationError):
+            ApprovalActionIn(action=2)
+
+    def test_通过可不填意见(self) -> None:
+        assert ApprovalActionIn(action=1).opinion is None

backend/tests/unit/test_config.py

@@ -13,7 +13,12 @@ VALID_SECRET = "x" * 32
 
 def _settings(**overrides: object) -> Settings:
     """构造一个不读取 .env 的配置实例。"""
-    params: dict[str, object] = {"_env_file": None, "app_secret_key": VALID_SECRET}
+    # _env_file=None 只禁用文件读取；显式值避免隔离测试库的环境变量影响默认值断言。
+    params: dict[str, object] = {
+        "_env_file": None,
+        "app_secret_key": VALID_SECRET,
+        "mysql_database": "payroll",
+    }
     params.update(overrides)
     return Settings(**params)  # type: ignore[arg-type]
 

backend/tests/unit/test_models.py

@@ -44,6 +44,10 @@ EXPECTED_TABLES = {
     "monthly_settlements",
     "settlement_items",
     "adjustment_orders",
+    "tplus_sync_batches",
+    "tplus_sync_errors",
+    "tplus_sync_records",
+    "tplus_sync_watermarks",
 }
 
 # 单复数同形的集合名词：本身即复数语义，不受「以 s 结尾」检查约束
@@ -96,6 +100,20 @@ class TestNamingConvention:
             for fk in table.foreign_keys:
                 assert fk.constraint.name.startswith(f"fk_{table.name}_"), fk.constraint.name
 
+    def test_审批单版本号唯一键覆盖三列(self) -> None:
+        """并发提单算重版本号时由 DB 兜底（应用层取号走锁读，见
+        approval_service._next_version）；约束名必须与迁移 DDL 严格一致，
+        否则 alembic autogenerate 会误判成「约束缺失/多余」。"""
+        table = Base.metadata.tables["approval_orders"]
+        uks = [c for c in table.constraints if c.__class__.__name__ == "UniqueConstraint"]
+        hit = [
+            c
+            for c in uks
+            if {col.name for col in c.columns} == {"order_type", "ref_id", "version_no"}
+        ]
+        assert hit, "审批单缺 (order_type, ref_id, version_no) 唯一键"
+        assert hit[0].name == "uk_approval_orders_order_type_ref_version"
+
 
 class TestComments:
     """规范红线要求中文 COMMENT。"""