feat: add node coverage and improve investigation presentation

.env.example

@@ -1,6 +1,7 @@
-# TJWater Server 环境变量模板
+# NextTJWater Server 环境变量模板
 # 复制为 .env 后按启用的功能填写；不要提交 .env 或任何真实凭据。
-# 生产配置由部署主机的受控 backend.env 注入，不要把完整配置保存为仓库密钥。
+# 生产配置由部署主机的受控 next-backend.env 注入，不随镜像或 CI/CD 同步本地 .env。
+# 进程环境变量优先；修改 Compose env_file 后需重新创建容器。
 
 # 基础运行
 PROJECT_NAME="TJWater Server"
@@ -69,9 +70,11 @@ KEYCLOAK_ALGORITHM="RS256"
 KEYCLOAK_AUDIENCE="account"
 KEYCLOAK_ACCESS_TOKEN_MAX_AGE_SECONDS=900
 
-# 工作台事件与跨域
-# 认证接口必须使用明确来源，逗号分隔；不要配置为 *。
-CORS_ALLOWED_ORIGINS="http://127.0.0.1:5173,http://localhost:5173,http://127.0.0.1:15173,http://localhost:15173"
+# 浏览器跨域访问（主 API）
+# 必须填写明确来源（协议、域名或 IP、端口），英文逗号分隔；不允许 * 或网段正则。
+CORS_ALLOWED_ORIGINS=http://127.0.0.1:5173,http://localhost:5173,http://127.0.0.1:15173,http://localhost:15173
+
+# 工作台事件
 WORKBENCH_EVENTS_ENABLED=true
 WORKBENCH_EVENT_RETENTION_DAYS=7
 
@@ -116,6 +119,17 @@ OPERATIONAL_ACTION_PERIPHERAL_URL=""
 OPERATIONAL_ACTION_COMMAND_GATEWAY_URL=""
 OPERATIONAL_ACTION_TIMEOUT_SECONDS=10
 
+# 外围样板（挂载于 /peripheral；故障注入默认关闭）
+# 独立启动时用于调用主 API 验证身份；容器内按网络填写主服务地址。
+PERIPHERAL_SAMPLE_NEXT_SERVER_URL="http://127.0.0.1:8000"
+# 以下两个默认库选择只供种子、导出等脚本使用，HTTP 请求仍按已授权项目路由。
+PERIPHERAL_SAMPLE_DB_NAME="tjwater_v2"
+PERIPHERAL_SAMPLE_PROJECT_CODE="tjwater_v2"
+# 子应用独立运行时的浏览器来源列表，使用 JSON 格式；挂载时还受主 API CORS 配置控制。
+PERIPHERAL_SAMPLE_CORS_ORIGINS='["http://127.0.0.1:5173","http://localhost:5173"]'
+PERIPHERAL_SAMPLE_ENABLE_FAULTS=false
+PERIPHERAL_SAMPLE_TIMEOUT_SECONDS=5
+
 # MCP 查询服务；OAuth 和所有项目业务库就绪前保持关闭。
 TJWATER_MCP_ENABLED=false
 TJWATER_MCP_ISSUER_URI=""
@@ -129,7 +143,20 @@ TJWATER_MCP_HOST_CLIENT_ID="tjwater-agent-mcp"
 TJWATER_MCP_ALLOWED_HOSTS='["127.0.0.1:*","localhost:*"]'
 TJWATER_MCP_ALLOWED_ORIGINS='[]'
 TJWATER_MCP_CONTEXT_TTL_SECONDS=600
-# 普通查询预算；计算工具由 Agent 通过 timeout_seconds 按次指定。
-TJWATER_MCP_QUERY_TIMEOUT_SECONDS=30
-# 计算等待上限为 120–86400 秒，实际上限通过 tools/list 发布。
+# 单次 HTTP 返回预算；由目录元数据公布给宿主，Agent 默认等待 180 秒。
+TJWATER_MCP_QUERY_TIMEOUT_SECONDS=120
+# HTTP 请求预算上限；Agent 加 60 秒余量并封顶 270 秒，外层 MCP 总时限 300 秒。
+TJWATER_MCP_MAX_REQUEST_TIMEOUT_SECONDS=240
+# 后台计算独立执行时限为 120–86400 秒，不受工具 timeout_seconds 控制。
 TJWATER_MCP_MAX_CALCULATION_TIMEOUT_SECONDS=3600
+
+# 独立计算队列消费者；显式项目 UUID，以逗号分隔。不会在 API lifespan 内启动。
+# 先在每个目标项目 BizDB 依次应用 026_calculation_jobs.sql 和 027_calculation_job_lifecycle.sql，再启动计算服务。
+CALCULATION_WORKER_PROJECT_IDS=""
+# 所有配置项目合计并发，独立于 Uvicorn --workers 4；单实例锁防止重复消费者扩容。
+CALCULATION_WORKER_MAX_CONCURRENCY=2
+CALCULATION_WORKER_POLL_SECONDS=2
+CALCULATION_WORKER_LEASE_SECONDS=30
+CALCULATION_WORKER_HEARTBEAT_SECONDS=5
+# 持久化本机恢复账本；容器模板已挂载独立 volume，不能当缓存删除。
+CALCULATION_WORKER_STATE_DIR="temp/calculation-jobs"

Dockerfile

@@ -23,7 +23,7 @@ COPY app ./app
 COPY contracts ./contracts
 COPY infra ./infra
 COPY resources ./resources
-COPY scripts/run_operational_worker.py scripts/sync_peripheral_work_orders.py ./scripts/
+COPY scripts/run_operational_worker.py scripts/run_calculation_worker.py scripts/sync_peripheral_work_orders.py ./scripts/
 COPY samples/peripheral ./samples/peripheral
 RUN python -c "from pathlib import Path; from zipfile import ZipFile; model_dir = Path('app/algorithms/pipe_health_prediction/model'); zip_path = model_dir / 'my_survival_forest_model_quxi.zip'; joblib_name = 'my_survival_forest_model_quxi.joblib'; joblib_path = model_dir / joblib_name; assert zip_path.exists(), f'Model archive not found: {zip_path}'; archive = ZipFile(zip_path); archive.extract(joblib_name, model_dir); archive.close(); assert joblib_path.exists(), f'Model file not extracted: {joblib_path}'" && \
     rm -f app/algorithms/pipe_health_prediction/model/my_survival_forest_model_quxi.zip

README.md

@@ -1,6 +1,6 @@
-# TJWaterServerBinary 内部后端
+# NextTJWater Server
 
-`TJWaterServerBinary` 是 TJWater 内部版 Python 后端，基于 FastAPI 提供认证、项目、管网、模拟、爆管、漏损、SCADA 和地图服务集成能力。该仓库用于内部开发和完整功能维护。
+`next-tjwater-server` 是 NextTJWater 供水系统的 FastAPI 后端，提供认证、项目、管网、模拟、爆管、漏损、SCADA、地图服务、工况任务和受控 MCP 能力。业务数据和时序数据按已认证项目路由，后台工作进程也必须使用明确的项目范围。
 
 ## 技术栈
 
@@ -30,6 +30,14 @@ infra/docker/         Docker Compose 编排
 
 ## 本地开发
 
+先从样板创建本地配置，再按启用的功能填写：
+
+```bash
+cp .env.example .env
+```
+
+`.env` 只用于本地或部署主机运行，不得提交。生产配置由部署主机的受控环境文件注入，不将完整应用 `.env` 保存为 Gitea Actions 密钥。
+
 推荐使用已有 conda 环境：
 
 ```bash
@@ -45,14 +53,50 @@ conda activate server
 
 `GET /health` 是无需鉴权的 HTTP 服务存活检查，返回 `200` 和 `{"status":"ok"}`，不查询数据库。
 `GET /ready` 检查应用启动状态、元数据库 `SELECT 1` 及已启用的 MCP 传输生命周期；
-全部就绪返回 `200`，否则返回 `503`，元数据库探测预算为2秒。MCP 关闭时标记为
+全部就绪返回 `200`，否则返回 `503`，元数据库探测预算为 2 秒。MCP 关闭时标记为
 `disabled`，不阻止就绪。返回结果不包含数据库地址、凭据或原始异常，禁止缓存。
 就绪状态不等同于 OAuth、用户权限或真实工具调用成功；指定项目的业务库和时序库仍通过
 带身份与 `X-Project-Id` 的 `GET /api/v1/projects/current/database-health` 检查。
 
+## 环境配置
+
+`.env.example` 是唯一应提交的环境配置样板，主要分为以下几组：
+
+- 元数据库、项目业务库、TimescaleDB 及项目数据库连接池。
+- GeoServer 项目供应、Keycloak JWT 校验、CORS 和工作台事件。
+- 天地图地理编码和地名搜索。
+- 调查任务 worker、外围工单同步、数据与动作适配器。
+- 可选的 MCP resource、OAuth issuer、宿主和超时预算。
+
+基础数据库变量用于启动和项目供应。运行时的项目请求仍以元数据库中的 `biz_data` 和 `iot_data` 路由为准，不会因为存在默认数据库变量而回退到固定项目。
+
+CORS 通过后端 `.env` 或部署进程环境变量中的 `CORS_ALLOWED_ORIGINS` 配置，进程环境变量优先。
+必须使用英文逗号分隔的明确页面来源（协议、域名或 IP、端口，不带路径），不能填写 `*`。
+未设置时默认只允许本机 `5173`、`15173` 开发来源。正式部署按实际浏览器地址填写，例如：
+
+```dotenv
+CORS_ALLOWED_ORIGINS=https://next-demo.waternetwork.cn,http://192.168.1.133,https://192.168.1.133,http://192.168.1.110,https://192.168.1.110
+```
+
+CORS 匹配浏览器页面来源，不是 Nginx 连接 IP 或域名解析结果；非默认端口须逐项列出。
+保留跨域凭据支持及 `Authorization`、`X-Project-Id` 请求头，接口认证和权限校验继续生效。
+修改后需重启进程；使用 Compose `env_file` 注入时需重新创建容器才能读取新值。
+
+## 业务 Skill 与闭环
+
+[爆管处置工程图](docs/burst-response-workflow.md)覆盖上报、候选定位、隔离阀门、影响分析、地图与数据展示、审批处置和反馈闭环。Server 通过任务 MCP 按 workflow 分发以下业务 Skill，当前版本均为 `2.0.0`：
+
+| workflow | Skill | 调查重点 |
+| --- | --- | --- |
+| burst_report | [爆管处置](resources/work-order-skills/burst-report/SKILL.md) | 候选定位、隔离阀门、影响分析、抢修及恢复闭环 |
+| water_quality_complaint | [水质投诉](resources/work-order-skills/water-quality-complaint/SKILL.md) | 投诉关联、监测与采样、候选影响、处置及复测回访 |
+| general | [预约跟进](resources/work-order-skills/appointment-follow-up/SKILL.md) | 审批与回执、现场履约、反馈复核及与原事件的闭环区别 |
+
+JSONL 整理遵守会话文件权限，后台图表和表格通过既有报告 `ui_blocks` 展示；Skill 不增加审批、执行或原工单关单权限。预约投影不自动启动 Agent 调查，普通协作与后台执行分别遵守其原有工具范围。
+
 ## 常用命令
 
-外围接入样板是独立进程，复用 PostgreSQL，不自动建表；启动、授权边界和契约见 [外围样板说明](docs/peripheral-sample.md)。
+外围接入样板作为 `/peripheral` 子应用随主 API 进程启动，复用 PostgreSQL 且不自动建表；授权边界和契约见 [外围样板说明](docs/peripheral-sample.md)。
 外围样板提供 synthetic 测试工单和测试数据的版本化 CRUD；工单经同步器投影为 `workbench.operational_tasks` 中的外围来源任务。固定演示时窗和造数方式见 [合成样本说明](docs/peripheral-generated-data.md)，运行态读取边界见 [工作台运行态说明](docs/workbench-runtime.md)。这些数据不代表现场事实，也不会触发设备控制。
 
 ```bash
@@ -84,8 +128,8 @@ docker compose -f infra/docker/docker-compose.yml config
 - 元数据、业务库和 TimescaleDB 可以部署在同一主机，也可以分别部署。
 
 在线重算会原子替换一个时刻的节点、管段基线。历史分块已经压缩时，写入事务使用
-`TIMESCALEDB_REALTIME_MAX_DECOMPRESSED_TUPLES` 设置解压预算，默认100万条；配置必须
-为正整数，不允许0。该设置仅在实时快照写入事务内生效，不改变数据库全局设置或普通读取。
+`TIMESCALEDB_REALTIME_MAX_DECOMPRESSED_TUPLES` 设置解压预算，默认 100 万条；配置必须
+为正整数，不允许 0。该设置仅在实时快照写入事务内生效，不改变数据库全局设置或普通读取。
 预算按需根据受影响压缩批次的规模调整，超过预算仍会失败并回滚当前快照；此前完成的其他
 时刻不会一起回滚。同步、异步及单表批量快照入口使用同一预算。
 
@@ -117,6 +161,19 @@ conda run -n server python scripts/publish_fengxian_map_layers.py --database tjw
 
 发布命令需要配置 `GEOSERVER_URL`、`GEOSERVER_USERNAME` 和 `GEOSERVER_PASSWORD`。其中 7 个基础管网图层仍由项目供应流程创建，奉贤发布脚本追加 9 个原始 GIS 图层和 1 个水质监测点图层。
 
+## 工况任务工作进程
+
+调查任务 worker 和外围工单同步器是独立进程，不随 API 进程自动启动：
+
+```bash
+conda run -n server python scripts/run_operational_worker.py
+conda run -n server python scripts/sync_peripheral_work_orders.py
+```
+
+`OPERATIONAL_WORKER_PROJECT_IDS` 是逗号分隔的显式项目 UUID 列表，非空时优先于单项目兼容变量 `OPERATIONAL_WORKER_PROJECT_ID`。如需覆盖所有已启用项目，应将这些项目的 UUID 全部明确写入列表。进程不会自动枚举项目，配置列表也不会自动授权；服务身份必须是每个目标项目的有效成员。
+
+每个项目保持独立的事件游标、会话、消息游标和 OAuth 缓存。`OPERATIONAL_WORKER_MAX_CONCURRENCY` 是每个项目的上限，而不是全局上限。更改项目列表后，需要同时重启 worker 和外围同步器。完整配置、授权和重试边界见 [工况工作进程说明](docs/operational-task-workers.md)。
+
 ## 测试与发布
 
 提交前根据改动范围运行最小有效测试：
@@ -131,7 +188,15 @@ conda run -n server python -m pytest tests/unit tests/auth -q
 docker build -t tjwater-server:local .
 ```
 
-Gitea 包工作流位于 `.gitea/workflows/package.yml`，通常由 tag 触发构建、推送镜像并通知部署 webhook。
+Dockerfile 的 `test` 阶段会执行 Python 编译检查、应用导入、OpenAPI 检查和完整 pytest。
+
+Gitea Actions 工作流位于 `.gitea/workflows/package.yml`，可手动触发，也会在推送 `v*` tag 时运行。工作流构建 `test` 阶段，推送不可变的 `sha-*` 镜像，然后通过受限 SSH 账号对 `next-backend` 执行候选验证。只有 tag 工作流会在验证通过后提升正式版本。仓库需要以下 Actions secrets：
+
+- `REGISTRY_USERNAME`
+- `REGISTRY_PASSWORD`
+- `DEV_DEPLOY_SSH_KEY`
+
+这些密钥只用于镜像仓库和受限开发机部署，不替代后端应用的运行环境配置。
 
 ## 安全规则
 
@@ -139,4 +204,19 @@ Gitea 包工作流位于 `.gitea/workflows/package.yml`，通常由 tag 触发
 
 ## 产品 MCP 查询
 
-Server 可选启用 MCP 2026-07-28 的25项项目查询及10项受控分析，默认关闭。它使用专用 OAuth resource Bearer、宿主绑定上下文及现有项目/任务权限，调查查询返回前持久化证据。启用需更新依赖、应用项目 SQL 迁移和配置 Keycloak token exchange；这些生产操作尚未执行。详见 [实施与启用](docs/mcp-query-implementation.md) 和 [协议/旧 CLI 能力对照](docs/mcp-query-design.md)。
+Server 可选启用 MCP 2026-07-28，默认关闭。当前目录包含 67 项类型化工具，其中 42 项查询、12 项计算、13 项预览式写入。MCP 使用专用 OAuth resource Bearer、宿主绑定上下文和现有项目或任务权限，不提供自由命令执行。
+
+写入工具默认只执行 `dry_run`。真实写入必须使用同一 `request_id`，带上预览返回的 `expected_digest`，并通过宿主原生许可和项目权限检查。超时或结果不明时先查询操作状态，不盲目重试。实施、启用和授权边界见 [MCP 实施与启用](docs/mcp-query-implementation.md) 和 [协议与旧 CLI 能力对照](docs/mcp-query-design.md)。
+
+## 后台计算队列
+
+长计算通过项目数据库队列和独立计算 Worker 执行，API 保持4个进程，HTTP超时不延长。接入需要新增数据库迁移和独立服务，详见[计算队列部署与MCP契约](docs/calculation-queue.md)。
+
+
+后台调查通过 `POST /api/v1/operational-tasks/{task_id}/questions` 保存结构化问题，并沿现有任务通知展示。当前 run 可在没有最终报告时正常结束，任务进入等待人工；具备 `operational_task.write` 权限的用户通过 `questions/{question_id}/answer` 回答，原任务排队新一轮。回答保存为交互记录，不自动转为现场证据或审批。问题支持最多三个单选、多选或自定义回答，版本校验和请求幂等保护重复提交。任务详情返回最近 100 组提问，完整历史由 MCP 的 `get_task_context(section="questions")` 分页读取。
+
+部署任务提问功能前，在各项目业务库执行 `resources/sql/028_task_questions.sql`，再同步 Server、Agent 与前端版本。迁移不修改既有调查记录；更新代码不会自动迁移数据库或重启服务。
+
+外围工单来源统一为系统数据。部署对应版本前，在各项目业务库执行 `resources/sql/029_peripheral_work_orders_system_data.sql`，再同步 Server、同步器、Agent 与前端版本。迁移会修正既有外围源工单、修订记录、任务快照和外围反馈证据，不修改测试数据或模拟监测数据。
+
+多 Agent 后台调查按实例归属运行。部署对应版本前，在每个启用调查 Worker 的项目 BizDB 执行 `resources/sql/030_operational_run_agent_ownership.sql`，再同步 Server、Agent 和 Worker，并为各 Agent 配置唯一且重启稳定的 `AGENT_INSTANCE_ID`。迁移只增加运行归属列和活跃运行索引；旧运行不猜测归属，待其结束或租约恢复后，新一代运行开始记录实例。

app/algorithms/_utils.py

@@ -3,6 +3,14 @@ import os
 import pandas as pd
 
 
+def calculation_temp_directory(kind: str) -> str:
+    """Keep native outputs inside the supervisor-owned job directory."""
+    root = os.environ.get("TJWATER_CALCULATION_JOB_DIR", "temp")
+    path = os.path.abspath(os.path.join(root, kind))
+    os.makedirs(path, exist_ok=True)
+    return path
+
+
 def fill_time_gaps(
     data: pd.DataFrame,
     time_col: str = "time",

app/algorithms/burst_localization/leak_signature.py

@@ -8,15 +8,14 @@ import sys
 import pandas as pd
 import wntr
 
-from app.algorithms._utils import _cleanup_temp_files
+from app.algorithms._utils import _cleanup_temp_files, calculation_temp_directory
 
 _PIPE2LEAKNODE = None
 _SIGNATURE_WORKER_DATA = {}
 
 
 def _make_temp_prefix(tag):
-    temp_dir = os.path.abspath(os.path.join("temp", "burst_location"))
-    os.makedirs(temp_dir, exist_ok=True)
+    temp_dir = calculation_temp_directory("burst_location")
     safe_tag = str(tag).replace(os.sep, "_").replace(" ", "_")
     return os.path.join(temp_dir, f"{safe_tag}_{os.getpid()}")
 

app/algorithms/dma_leakage_estimation/genetic_optimizer.py

@@ -15,7 +15,7 @@ from pymoo.optimize import minimize as pymoo_minimize
 from pymoo.termination.default import DefaultSingleObjectiveTermination
 
 
-from app.algorithms._utils import _cleanup_temp_files
+from app.algorithms._utils import _cleanup_temp_files, calculation_temp_directory
 
 _worker_data: dict[str, Any] = {}
 DEFAULT_N_WORKERS = max(1, min(cpu_count() - 1, 4))
@@ -92,7 +92,7 @@ def _worker_evaluate(raw_ratios: np.ndarray) -> float:
             demand_obj.base_value = original_val + per_node_leak
             modifications.append((demand_obj, original_val))
 
-    temp_dir = os.path.abspath(os.path.join("temp", "leakage"))
+    temp_dir = calculation_temp_directory("leakage")
     os.makedirs(temp_dir, exist_ok=True)
     prefix = os.path.join(temp_dir, f"temp_{os.getpid()}")
 
@@ -569,7 +569,7 @@ class LeakageProblem(Problem):
                 modifications.append((demand_obj, original_val))
 
         # 结果保存在根目录的temp/leakage文件夹中
-        temp_dir = os.path.abspath(os.path.join("temp", "leakage"))
+        temp_dir = calculation_temp_directory("leakage")
         os.makedirs(temp_dir, exist_ok=True)
         prefix = os.path.join(temp_dir, f"temp_{os.getpid()}")
 

app/algorithms/valve_isolation/topology_search.py

@@ -1,10 +1,11 @@
 """Topology-only valve isolation search."""
 
 from collections import defaultdict, deque
-from typing import Any, Iterable
+from typing import Any, Iterable, Literal
 
 
 VALVE_LINK_TYPE = "valve"
+AccidentElementType = Literal["auto", "link", "node"]
 
 
 def _parse_link_entry(link_entry: str) -> tuple[str, str, str, str]:
@@ -14,12 +15,37 @@ def _parse_link_entry(link_entry: str) -> tuple[str, str, str, str]:
     return parts[0], parts[1], parts[2], parts[3]
 
 
+def _reachable_nodes(
+    adjacency: dict[str, list[tuple[str, str]]],
+    start_nodes: Iterable[str],
+    blocked_links: set[str],
+) -> set[str]:
+    visited = set(start_nodes)
+    queue = deque(visited)
+    while queue:
+        for neighbor, link_id in adjacency.get(queue.popleft(), ()):
+            if link_id not in blocked_links and neighbor not in visited:
+                visited.add(neighbor)
+                queue.append(neighbor)
+    return visited
+
+
 def valve_isolation_analysis(
     link_entries: Iterable[str],
     accident_elements: str | list[str],
     disabled_valves: list[str] | None = None,
+    *,
+    accident_element_type: AccidentElementType = "auto",
+    water_source_nodes: Iterable[str] | None = None,
+    closed_links: Iterable[str] = (),
 ) -> dict[str, Any]:
-    """Determine boundary valves and affected nodes from a topology snapshot."""
+    """Find nearest valve boundaries, then check source connectivity.
+
+    Legacy untyped IDs prefer links; explicit node mode handles cross-type ID
+    collisions. Supply impact is a before/after connectivity difference, not
+    a hydraulic pressure or demand calculation. Without source metadata only
+    the provisional boundary search is available.
+    """
 
     disabled_valves_set = set(disabled_valves or [])
     target_elements = (
@@ -27,8 +53,13 @@ def valve_isolation_analysis(
         if isinstance(accident_elements, str)
         else accident_elements
     )
+    if not target_elements:
+        raise ValueError("At least one accident element is required")
+    if accident_element_type not in {"auto", "link", "node"}:
+        raise ValueError(f"Invalid accident element type: {accident_element_type}")
+    closed_links_set = set(closed_links)
 
-    pipe_adj: dict[str, set[str]] = defaultdict(set)
+    adjacency: dict[str, list[tuple[str, str]]] = defaultdict(list)
     all_valves: dict[str, tuple[str, str]] = {}
     link_lookup: dict[str, tuple[str, str, str]] = {}
     node_set: set[str] = set()
@@ -39,46 +70,31 @@ def valve_isolation_analysis(
         node_set.update((node1, node2))
         if link_type_name == VALVE_LINK_TYPE:
             all_valves[link_id] = (node1, node2)
-        else:
-            pipe_adj[node1].add(node2)
-            pipe_adj[node2].add(node1)
+        if link_id not in closed_links_set:
+            adjacency[node1].append((node2, link_id))
+            adjacency[node2].append((node1, link_id))
+
+    unknown_disabled = disabled_valves_set - all_valves.keys()
+    if unknown_disabled:
+        raise ValueError(f"Disabled valve was not found in topology: {sorted(unknown_disabled)}")
 
     start_nodes: set[str] = set()
     for element in target_elements:
-        if element in node_set:
-            start_nodes.add(element)
-        elif element in link_lookup:
+        if accident_element_type != "node" and element in link_lookup:
             node1, node2, _ = link_lookup[element]
             start_nodes.update((node1, node2))
+        elif accident_element_type != "link" and element in node_set:
+            start_nodes.add(element)
         else:
             raise ValueError(f"Accident element {element} was not found in topology")
 
-    extra_adj: dict[str, list[str]] = defaultdict(list)
-    boundary_valves: dict[str, tuple[str, str]] = {}
-    for valve_id, (node1, node2) in all_valves.items():
-        if valve_id in disabled_valves_set:
-            extra_adj[node1].append(node2)
-            extra_adj[node2].append(node1)
-        else:
-            boundary_valves[valve_id] = (node1, node2)
-
-    affected_nodes: set[str] = set()
-    queue = deque(start_nodes)
-    while queue:
-        node = queue.popleft()
-        if node in affected_nodes:
-            continue
-        affected_nodes.add(node)
-        queue.extend(pipe_adj.get(node, set()) - affected_nodes)
-        queue.extend(
-            neighbor
-            for neighbor in extra_adj.get(node, ())
-            if neighbor not in affected_nodes
-        )
+    closable_valves = set(all_valves) - disabled_valves_set - closed_links_set
+    affected_nodes = _reachable_nodes(adjacency, start_nodes, closable_valves)
 
     must_close_valves: list[str] = []
     optional_valves: list[str] = []
-    for valve_id, (node1, node2) in boundary_valves.items():
+    for valve_id in closable_valves:
+        node1, node2 = all_valves[valve_id]
         node1_affected = node1 in affected_nodes
         node2_affected = node2 in affected_nodes
         if node1_affected and node2_affected:
@@ -89,14 +105,37 @@ def valve_isolation_analysis(
     must_close_valves.sort()
     optional_valves.sort()
     isolatable = bool(must_close_valves)
+    failure_reason = None if isolatable else "no_boundary_valves"
+    sources = set(water_source_nodes or ())
+    source_check_performed = bool(sources)
+    supply_affected_nodes = None
+    if water_source_nodes is not None:
+        if not sources:
+            isolatable = False
+            failure_reason = "water_sources_unavailable"
+        else:
+            after_closure = _reachable_nodes(adjacency, sources, set(must_close_valves))
+            if affected_nodes & after_closure:
+                isolatable = False
+                failure_reason = "connected_to_water_source"
+            if isolatable:
+                before_closure = _reachable_nodes(adjacency, sources, set())
+                supply_affected_nodes = sorted(before_closure - after_closure)
     result: dict[str, Any] = {
         "accident_elements": target_elements,
+        "accident_element_type": accident_element_type,
         "disabled_valves": disabled_valves,
         "affected_nodes": sorted(affected_nodes) if isolatable else [],
         "affected_node_count": len(affected_nodes),
         "must_close_valves": must_close_valves,
         "optional_valves": optional_valves,
         "isolatable": isolatable,
+        "source_check_performed": source_check_performed,
+        "isolation_failure_reason": failure_reason,
+        "supply_affected_nodes": supply_affected_nodes,
+        "supply_affected_node_count": (
+            len(supply_affected_nodes) if supply_affected_nodes is not None else None
+        ),
     }
     if len(target_elements) == 1:
         result["accident_element"] = target_elements[0]

app/api/v1/endpoints/operational_tasks.py

@@ -26,6 +26,8 @@ from app.auth.project_dependencies import (
 from app.domain.domain_errors import DomainError
 from app.domain.schemas.workbench_runtime import WorkbenchDateTime
 from app.domain.schemas.operational_data import TaskDataQuery, TaskDataSnapshot
+from app.domain.schemas.task_question import TaskQuestionCreate, TaskQuestionReply, TaskQuestionRequest
+from app.services.task_questions import create_question, answer_question
 from app.domain.schemas.operational_task import (
     ActionDecisionCommand,
     ActionExecuteCommand,
@@ -74,6 +76,7 @@ from app.domain.schemas.operational_task import (
     TaskPresentationEventCommand,
     TaskPresentationEventResult,
     TaskPresentationCursor,
+    TaskPresentationHistoryPage,
 )
 from app.infra.db.postgresql.operational_tasks import OperationalTaskRepository
 from app.infra.integrations.operational_actions import (
@@ -521,6 +524,21 @@ async def get_operational_task_presentation_cursor(
         _raise_domain(exc)
 
 
+@router.get("/operational-tasks/{task_id}/presentation-events", response_model=TaskPresentationHistoryPage)
+async def get_operational_task_presentation_history(
+    task_id: UUID,
+    after: int = Query(default=0, ge=0),
+    through: int | None = Query(default=None, ge=0),
+    limit: int = Query(default=100, ge=1, le=200),
+    ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
+    conn: AsyncConnection = Depends(get_project_pg_connection),
+):
+    try:
+        return await _service(ctx, conn).presentation_history(task_id, after=after, through=through, limit=limit)
+    except DomainError as exc:
+        _raise_domain(exc)
+
+
 @router.post("/operational-tasks/{task_id}/presentation-events", response_model=TaskPresentationEventResult)
 async def publish_operational_task_presentation_event(
     task_id: UUID,
@@ -771,3 +789,27 @@ async def withdraw_operational_task_external_source(
         return await _service(ctx, conn).withdraw_external_source(task_id, command)
     except DomainError as exc:
         _raise_domain(exc)
+
+
+@router.post("/operational-tasks/{task_id}/questions", response_model=TaskQuestionRequest)
+async def create_operational_task_question(
+    task_id: UUID, command: TaskQuestionCreate,
+    ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
+    conn: AsyncConnection = Depends(get_project_pg_connection),
+):
+    try:
+        return await create_question(_service(ctx, conn), task_id, command)
+    except DomainError as exc:
+        _raise_domain(exc)
+
+
+@router.post("/operational-tasks/{task_id}/questions/{question_id}/answer", response_model=OperationalTaskDetail)
+async def answer_operational_task_question(
+    task_id: UUID, question_id: UUID, command: TaskQuestionReply,
+    ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
+    conn: AsyncConnection = Depends(get_project_pg_connection),
+):
+    try:
+        return await answer_question(_service(ctx, conn), task_id, question_id, command)
+    except DomainError as exc:
+        _raise_domain(exc)

app/api/v1/endpoints/simulation.py

@@ -161,6 +161,9 @@ def valve_isolation_endpoint(
     network: str = Query(..., description="管网名称（或数据库名称）"),
     accident_element: List[str] = Query(..., description="发生事故的管段/节点ID列表"),
     disabled_valves: List[str] = Query(None, description="已故障的阀门ID列表（可选）"),
+    accident_element_type: Literal["auto", "link", "node"] = Query(
+        "auto", description="事故要素类型；auto优先管段再节点；分析同名节点须指定node，纯管段分析可指定link",
+    ),
 ):
     """
     阀门隔离分析
@@ -168,13 +171,19 @@ def valve_isolation_endpoint(
     - **network**: 管网名称（或数据库名称）
     - **accident_element**: 发生事故的管段/节点ID列表
     - **disabled_valves**: 已故障的阀门ID列表（可选）
+    - **accident_element_type**: auto优先管段；link仅匹配管段；node仅匹配节点
     
     返回隔离方案，包括：
     - must_close_valves: 必须关闭的阀门列表
     - optional_valves: 可选关闭的阀门列表
-    - affected_nodes: 受影响的节点列表；不可隔离时为空列表
-    - affected_node_count: 受影响的节点总数
-    - isolatable: 是否可以有效隔离
+    - affected_nodes / affected_node_count: 最近阀门围成的直接隔离区；不可隔离时列表为空，计数保留
+    - supply_affected_nodes / supply_affected_node_count: 按建议关阀后新增失去水源连接的节点；未验证或不可隔离时为null
+    - source_check_performed: 是否已执行水源连通检查
+    - isolation_failure_reason: 不可隔离原因；可隔离时为null
+    - isolatable: 存在关阀边界，且直接隔离区不再连通任何水源
+
+    使用模型初始开关状态进行无向拓扑分析，水池按潜在水源处理。
+    不包含时变控制、水池蓄水量或压力流量计算，不是现场停水范围的水力证明。
     """
     # result = {
     #     "accident_element": "P461309",
@@ -193,7 +202,10 @@ def valve_isolation_endpoint(
     #     "must_close_valves": ["210521658", "V12974", "V12986", "V12993"],
     #     "optional_valves": [],
     # }
-    result = analyze_valve_isolation(network, accident_element, disabled_valves)
+    result = analyze_valve_isolation(
+        network, accident_element, disabled_valves,
+        accident_element_type=accident_element_type,
+    )
     return result
 
 

app/domain/schemas/answer_ui.py

@@ -7,7 +7,7 @@ from pathlib import Path
 from typing import Annotated, Any, Literal
 
 from jsonschema import Draft7Validator
-from pydantic import BaseModel, ConfigDict, Field, model_validator
+from pydantic import BaseModel, ConfigDict, Field, field_validator, model_serializer, model_validator
 
 _catalog = json.loads(
     (Path(__file__).resolve().parents[3] / "contracts/answer-ui-catalog.json").read_text()
@@ -34,9 +34,24 @@ class AnswerUiSpec(UiModel):
 class AnswerUiBlock(UiModel):
     version: Literal["answer-ui@1"]
     block_id: Key
+    phase: Literal["process", "final"] | None = None
     fallback_text: Annotated[str, Field(min_length=1, max_length=4000)]
     spec: AnswerUiSpec
 
+    @field_validator("phase", mode="before")
+    @classmethod
+    def validate_explicit_phase(cls, value):
+        if value is None:
+            raise ValueError("UI phase must be process or final when supplied")
+        return value
+
+    @model_serializer(mode="wrap")
+    def serialize_optional_phase(self, handler):
+        data = handler(self)
+        if self.phase is None:
+            data.pop("phase", None)
+        return data
+
     @model_validator(mode="after")
     def validate_components(self):
         spec = self.spec
@@ -83,6 +98,12 @@ class ReportUiBlock(UiModel):
     block: AnswerUiBlock
     evidence_refs: list[Annotated[str, Field(min_length=1, max_length=128)]] = Field(min_length=1, max_length=100)
 
+    @model_validator(mode="after")
+    def validate_report_phase(self):
+        if self.block.phase == "process":
+            raise ValueError("Report UI blocks must contain final results")
+        return self
+
 
 def validate_report_ui_blocks(blocks: list[ReportUiBlock], evidence_refs: list[str]) -> None:
     ids = [item.block.block_id for item in blocks]

app/domain/schemas/operational_data.py

@@ -58,6 +58,7 @@ McpSnapshotSource = Literal[
     'run_burst_detection',
     'run_burst_location',
     'analyze_valve_isolation',
+    'generate_node_coverage',
     'run_leakage_identification',
     'get_element_properties',
     'list_elements',

app/domain/schemas/operational_task.py

@@ -8,6 +8,7 @@ from uuid import UUID
 from pydantic import Field, JsonValue, model_validator
 from app.domain.schemas.answer_ui import ReportUiBlock, validate_report_ui_blocks
 
+from app.domain.schemas.task_question import TaskQuestionRequest
 from app.domain.schemas.shared import NonEmptyStr, PositiveInt, StrictBool, StrictModel
 from app.domain.schemas.operational_data import TaskDataSnapshot, TaskDataSource
 from app.domain.schemas.workbench_runtime import (
@@ -218,8 +219,12 @@ class InvestigationDiagramNodeRef(StrictModel):
     evidence_refs: list[NonEmptyStr] = Field(default_factory=list)
     action_refs: list[UUID] = Field(default_factory=list)
     status: Literal[
-        "pending", "running", "completed", "waiting_human", "failed", "skipped", "cancelled"
-    ] | None = None
+        "pending", "running", "completed", "waiting", "waiting_human", "failed", "skipped", "cancelled"
+    ] | None = Field(default=None, description=(
+        "Evidence-confirmed node state; waiting means prerequisites/external/system result, "
+        "waiting_human requires manual authorization/approval/feedback. "
+        "Unknown stays omitted/null; host controls colors."
+    ))
     map_targets: list[InvestigationMapTarget] = Field(default_factory=list, max_length=100)
 
 
@@ -240,6 +245,21 @@ class InvestigationDiagram(StrictModel):
         return self
 
 
+def _undeclared_diagram_node_refs(diagram: InvestigationDiagram | None) -> list[str]:
+    if diagram is None:
+        return []
+    missing: list[str] = []
+    for node in diagram.node_refs:
+        escaped = re.escape(node.node_id)
+        if not re.search(
+            rf"(?<![A-Za-z0-9_-]){escaped}(?=\s*(?:[\[({{@;]|:::|--|==|-\.|$))",
+            diagram.mermaid,
+            re.MULTILINE,
+        ):
+            missing.append(node.node_id)
+    return missing
+
+
 class InvestigationReport(StrictModel):
     """Read model: absent historical facts remain unknown, never inferred."""
 
@@ -341,9 +361,19 @@ class AgentArtifactsCommand(StrictModel):
     progress: InvestigationProgress | None = None
 
     @model_validator(mode="after")
-    def require_artifact(self):
+    def validate_artifacts(self):
         if not self.evidence and self.plan is None and self.report is None and self.progress is None:
             raise ValueError("at least one artifact must be supplied")
+        for diagram in (
+            self.report.diagram if self.report is not None else None,
+            self.progress.diagram if self.progress is not None else None,
+        ):
+            missing = _undeclared_diagram_node_refs(diagram)
+            if missing:
+                raise ValueError(
+                    "diagram node references must use ids declared in Mermaid source: "
+                    + ", ".join(missing)
+                )
         return self
 
 
@@ -502,7 +532,7 @@ class ExternalWorkOrderSnapshot(StrictModel):
     scenario_code: NonEmptyStr | None = None
     feedback_requirements: list[NonEmptyStr] = Field(default_factory=list)
     activity: list[WorkOrderActivity] = Field(default_factory=list)
-    origin: Literal["synthetic"]
+    origin: Literal["system_data"]
     updated_at: WorkbenchDateTime
 
     @model_validator(mode="before")
@@ -521,6 +551,7 @@ class OperationalTaskRun(StrictModel):
     status: OperationalTaskRunStatus
     revision: PositiveInt
     agent_session_id: NonEmptyStr | None = None
+    agent_instance_id: NonEmptyStr | None = None
     lease_expires_at: WorkbenchDateTime | None = None
     heartbeat_at: WorkbenchDateTime | None = None
     started_at: WorkbenchDateTime | None = None
@@ -570,6 +601,7 @@ class OperationalTaskPermissions(StrictModel):
     can_decide_commands: bool
     can_execute_commands: bool
     can_decide_access: bool = False
+    can_answer_questions: bool = False
 
 
 class OperationalTaskDetail(OperationalTaskSummary):
@@ -579,6 +611,7 @@ class OperationalTaskDetail(OperationalTaskSummary):
     evidence: list[EvidenceRecord]
     actions: list[HumanApprovalAction]
     access_requests: list[TaskAccessRequest] = Field(default_factory=list)
+    questions: list[TaskQuestionRequest] = Field(default_factory=list)
     permissions: OperationalTaskPermissions
     work_order_snapshot: ExternalWorkOrderSnapshot | None = None
 
@@ -605,6 +638,7 @@ class AgentRunCommand(StrictModel):
     request_id: UUID
     expected_revision: PositiveInt
     source_session_id: NonEmptyStr
+    agent_instance_id: NonEmptyStr | None = None
 
 
 class AgentRunFailureCommand(AgentRunCommand):
@@ -671,7 +705,59 @@ class TaskHighlightFeatures(StrictModel):
     action: Literal["highlight_features"]
     layer: TaskMapLayer
     feature_ids: Annotated[list[TaskFeatureId], Field(min_length=1, max_length=100)]
-    effect: Literal["steady", "pulse"] = "steady"
+    effect: Literal["steady", "pulse"]
+
+
+class TaskLocateFeatures(StrictModel):
+    action: Literal["locate_features"]
+    layer: TaskMapLayer
+    feature_ids: Annotated[list[TaskFeatureId], Field(min_length=1, max_length=100)]
+
+
+class TaskZoomToMap(StrictModel):
+    action: Literal["zoom_to_map"]
+    center: tuple[
+        Annotated[float, Field(strict=True, ge=-180, le=180, allow_inf_nan=False)],
+        Annotated[float, Field(strict=True, ge=-90, le=90, allow_inf_nan=False)],
+    ]
+    zoom: Annotated[float, Field(strict=True, ge=0, le=24, allow_inf_nan=False)]
+
+
+class TaskScadaAnalysisItem(StrictModel):
+    device_id: TaskFeatureId
+    level: Literal["high", "medium", "low", "unrated"]
+
+
+class TaskRenderScadaAnalysis(StrictModel):
+    action: Literal["render_scada_analysis"]
+    items: Annotated[list[TaskScadaAnalysisItem], Field(min_length=1, max_length=100)]
+
+    @model_validator(mode="after")
+    def validate_unique_devices(self):
+        if len({item.device_id for item in self.items}) != len(self.items):
+            raise ValueError("SCADA analysis device IDs must be unique")
+        return self
+
+
+class TaskClearScadaAnalysis(StrictModel):
+    action: Literal["clear_scada_analysis"]
+
+
+class TaskOverlayGeometry(StrictModel):
+    type: Literal["Point", "MultiPoint", "LineString", "MultiLineString", "Polygon", "MultiPolygon"]
+    coordinates: list[JsonValue]
+
+
+class TaskOverlayFeature(StrictModel):
+    type: Literal["Feature"]
+    geometry: TaskOverlayGeometry
+    properties: dict[str, JsonValue] | None
+    id: Annotated[str, Field(strict=True)] | Annotated[float, Field(strict=True, allow_inf_nan=False)] | None = None
+
+
+class TaskOverlayData(StrictModel):
+    type: Literal["FeatureCollection"]
+    features: Annotated[list[TaskOverlayFeature], Field(min_length=1, max_length=1000)]
 
 
 class TaskClearHighlights(StrictModel):
@@ -696,6 +782,7 @@ class TaskSetOverlay(StrictModel):
     title: Annotated[str, Field(min_length=1, max_length=200)]
     mode: Literal["geometry", "heatmap"]
     evidence_id: NonEmptyStr
+    data: TaskOverlayData | None = None
 
 
 class TaskClearOverlay(StrictModel):
@@ -704,7 +791,8 @@ class TaskClearOverlay(StrictModel):
 
 
 TaskPresentationCommand = Annotated[
-    TaskHighlightFeatures | TaskClearHighlights | TaskOpenChart | TaskCloseChart | TaskSetOverlay | TaskClearOverlay,
+    TaskHighlightFeatures | TaskLocateFeatures | TaskZoomToMap | TaskRenderScadaAnalysis
+    | TaskClearScadaAnalysis | TaskClearHighlights | TaskOpenChart | TaskCloseChart | TaskSetOverlay | TaskClearOverlay,
     Field(discriminator="action"),
 ]
 
@@ -736,6 +824,23 @@ class TaskPresentationCursor(StrictModel):
     event_cursor: int = Field(ge=0)
 
 
+class TaskPresentationHistoryItem(StrictModel):
+    event_seq: PositiveInt
+    run_id: UUID
+    execution_generation: PositiveInt
+    occurred_at: WorkbenchDateTime
+    # Stored records are validated against the current command contract by the
+    # reader; an obsolete record must not prevent reading the rest of the page.
+    command: dict[str, JsonValue]
+
+
+class TaskPresentationHistoryPage(StrictModel):
+    task_id: UUID
+    event_cursor: int = Field(ge=0)
+    items: list[TaskPresentationHistoryItem]
+    next_after: int | None = Field(default=None, ge=0)
+
+
 class TaskDataQueryResult(StrictModel):
     evidence: EvidenceRecord
     snapshot: TaskDataSnapshot
@@ -816,7 +921,7 @@ class OperationalTaskNotification(StrictModel):
     description: NonEmptyStr
     required_permission: NonEmptyStr | None = None
     lifecycle_status: Literal["active", "resolved"] = "active"
-    subject_type: Literal["task", "run", "access_request", "action"] = "task"
+    subject_type: Literal["task", "run", "access_request", "action", "question"] = "task"
     subject_id: UUID | None = None
     report_reference: InvestigationReportReference | None = None
     resolved_at: WorkbenchDateTime | None = None

app/domain/schemas/task_question.py

@@ -0,0 +1,75 @@
+"""Durable investigation questions; answers are user input, not field evidence."""
+from typing import Annotated, Literal
+from uuid import UUID
+
+from pydantic import Field, model_validator
+
+from app.domain.schemas.shared import StrictModel, PositiveInt
+from app.domain.schemas.workbench_runtime import WorkbenchDateTime
+
+Id = Annotated[str, Field(min_length=1, max_length=128)]
+
+
+class TaskQuestionOption(StrictModel):
+    id: Id
+    label: Annotated[str, Field(min_length=1, max_length=200)]
+
+
+class TaskQuestion(StrictModel):
+    question_id: Id
+    title: Annotated[str, Field(min_length=1, max_length=200)]
+    body: Annotated[str, Field(min_length=1, max_length=2000)]
+    options: list[TaskQuestionOption] = Field(max_length=12)
+    mode: Literal["single", "multiple"]
+    allow_custom: bool
+    required: bool
+
+    @model_validator(mode="after")
+    def valid_options(self):
+        if len({o.id for o in self.options}) != len(self.options):
+            raise ValueError("duplicate option ids")
+        if not self.options and not self.allow_custom:
+            raise ValueError("question must offer an answer input")
+        return self
+
+
+class TaskQuestionAnswer(StrictModel):
+    question_id: Id
+    selected_option_ids: list[Id] = Field(max_length=12)
+    custom_text: Annotated[str, Field(max_length=2000)]
+
+
+class TaskQuestionCreate(StrictModel):
+    request_id: UUID
+    source_session_id: Id
+    run_id: UUID
+    questions: list[TaskQuestion] = Field(min_length=1, max_length=3)
+
+    @model_validator(mode="after")
+    def unique_questions(self):
+        if len({q.question_id for q in self.questions}) != len(self.questions):
+            raise ValueError("duplicate question ids")
+        if len(self.model_dump_json().encode()) > 32768:
+            raise ValueError("question definition too large")
+        return self
+
+
+class TaskQuestionReply(StrictModel):
+    request_id: UUID
+    expected_version: PositiveInt
+    answers: list[TaskQuestionAnswer] = Field(max_length=3)
+
+
+class TaskQuestionRequest(StrictModel):
+    question_request_id: UUID
+    task_id: UUID
+    run_id: UUID
+    execution_generation: PositiveInt
+    source_session_id: Id
+    questions: list[TaskQuestion]
+    status: Literal["pending", "answered", "cancelled"]
+    version: PositiveInt
+    created_at: WorkbenchDateTime
+    answered_at: WorkbenchDateTime | None = None
+    answered_by: str | None = None
+    answers: list[TaskQuestionAnswer] = Field(default_factory=list)

app/infra/db/calculation_worker.py

@@ -0,0 +1,74 @@
+"""Database resources owned by the standalone calculation worker."""
+from __future__ import annotations
+
+from contextlib import asynccontextmanager
+from uuid import UUID
+
+from sqlalchemy import text
+
+
+@asynccontextmanager
+async def queue_repository(project_id):
+    """Resolve the current project route and open its calculation queue."""
+    from app.core.config import settings
+    from app.infra.db.dynamic_manager import project_connection_manager
+    from app.infra.db.metadb.database import SessionLocal
+    from app.infra.db.metadb.repositories.metadata_repository import MetadataRepository
+    from app.infra.db.postgresql.calculation_jobs import CalculationJobRepository
+
+    project_uuid = UUID(str(project_id))
+    async with SessionLocal() as session:
+        metadata = MetadataRepository(session)
+        project = await metadata.get_project_by_id(project_uuid)
+        if not project or project.status != "active":
+            raise ValueError("Calculation project is not active")
+        routing = await metadata.get_project_db_routing(project_uuid, "biz_data")
+        if not routing or routing.db_type != "postgresql":
+            raise ValueError("Project business PostgreSQL routing is unavailable")
+        await session.commit()
+        pool_min_size = (
+            routing.pool_min_size
+            if routing.pool_min_size is not None
+            else settings.PROJECT_PG_POOL_MIN_SIZE
+        )
+        pool_max_size = (
+            routing.pool_max_size
+            if routing.pool_max_size is not None
+            else settings.PROJECT_PG_POOL_SIZE
+        )
+        async with project_connection_manager.pg_connection(
+            project_uuid,
+            "biz_data",
+            routing.dsn,
+            pool_min_size,
+            pool_max_size,
+        ) as connection:
+            yield CalculationJobRepository(connection, project_id)
+
+
+class CalculationSupervisorLease:
+    """Keep the singleton advisory lock alive on one metadata connection."""
+
+    def __init__(self, connection):
+        self.connection = connection
+
+    async def ping(self) -> None:
+        await self.connection.execute(text("SELECT 1"))
+        await self.connection.commit()
+
+
+@asynccontextmanager
+async def calculation_supervisor_lease():
+    """Acquire and release the session-level calculation supervisor lock."""
+    from app.infra.db.metadb.database import engine
+
+    async with engine.connect() as connection:
+        acquired = await connection.scalar(text("SELECT pg_try_advisory_lock(1748531781, 2)"))
+        await connection.commit()
+        if not acquired:
+            raise RuntimeError("Another calculation supervisor is already active")
+        try:
+            yield CalculationSupervisorLease(connection)
+        finally:
+            await connection.execute(text("SELECT pg_advisory_unlock(1748531781, 2)"))
+            await connection.commit()

app/infra/db/postgresql/calculation_jobs.py

@@ -0,0 +1,150 @@
+"""Durable project calculation queue. Every worker update is generation fenced."""
+from uuid import uuid4
+
+from fastapi import HTTPException
+from psycopg.types.json import Jsonb
+from app.mcp.errors import CalculationCapacityExceeded
+
+
+# Explicit metadata projection prevents status polling from decoding large JSON results.
+JOB_COLUMNS = "job_id,project_id,owner_user_id,request_key,tool,parameters,binding,timeout_seconds,status,error,lease_generation,lease_owner,lease_expires_at,created_at,updated_at,started_at,finished_at,queue_expires_at,result_expires_at"
+
+
+PUBLIC_JOB_COLUMNS = "job_id,status,tool,error,created_at,updated_at,started_at,finished_at,queue_expires_at,result_expires_at"
+
+
+class CalculationJobRepository:
+    def __init__(self, connection, project_id):
+        self.connection = connection
+        self.project_id = str(project_id)
+
+    async def maintain(self):
+        await self.connection.execute(
+            """UPDATE workbench.calculation_jobs SET status='timed_out',
+            error='{"code":"QUEUE_TIMEOUT"}'::jsonb,finished_at=clock_timestamp(),updated_at=clock_timestamp()
+            WHERE project_id=%s AND status='queued' AND queue_expires_at<=clock_timestamp()""",
+            (self.project_id,),
+        )
+        await self.connection.execute(
+            """UPDATE workbench.calculation_jobs SET result=NULL
+            WHERE project_id=%s AND result IS NOT NULL AND result_expires_at<=clock_timestamp()""",
+            (self.project_id,),
+        )
+
+    async def enqueue(self, *, owner_user_id, request_key, tool, parameters, binding, timeout_seconds,
+                      queue_wait_seconds=3600, max_active_jobs=100):
+        async with self.connection.transaction():
+            return await self._enqueue(owner_user_id=owner_user_id, request_key=request_key,
+                tool=tool, parameters=parameters, binding=binding, timeout_seconds=timeout_seconds,
+                queue_wait_seconds=queue_wait_seconds, max_active_jobs=max_active_jobs)
+
+    async def _enqueue(self, *, owner_user_id, request_key, tool, parameters, binding,
+                       timeout_seconds, queue_wait_seconds, max_active_jobs):
+        # Serialize admission within the caller's transaction, across every API process.
+        await self.connection.execute(
+            "SELECT pg_advisory_xact_lock(hashtextextended(%s, 1748531781))", (self.project_id,))
+        await self.maintain()
+        cursor = await self.connection.execute(
+            f"SELECT {JOB_COLUMNS} FROM workbench.calculation_jobs WHERE project_id=%s AND owner_user_id=%s AND request_key=%s",
+            (self.project_id, owner_user_id, request_key),
+        )
+        existing = await cursor.fetchone()
+        if existing:
+            row = dict(existing)
+            if (row['tool'] != tool
+                    or {key: value for key, value in row['parameters'].items() if key != 'timeout_seconds'}
+                    != {key: value for key, value in parameters.items() if key != 'timeout_seconds'}
+                    or row['binding'] != binding):
+                raise HTTPException(409, 'Calculation request_key was used with different inputs')
+            return row
+        cursor = await self.connection.execute(
+            """SELECT count(*) AS active FROM workbench.calculation_jobs WHERE project_id=%s
+            AND status IN ('queued','running','cancel_requested')""", (self.project_id,))
+        if (await cursor.fetchone())['active'] >= max_active_jobs:
+            raise CalculationCapacityExceeded()
+        cursor = await self.connection.execute(
+            f"""INSERT INTO workbench.calculation_jobs
+            (job_id,project_id,owner_user_id,request_key,tool,parameters,binding,timeout_seconds,queue_expires_at)
+            VALUES (%s,%s,%s,%s,%s,%s,%s,%s,clock_timestamp()+%s*interval '1 second')
+            RETURNING {JOB_COLUMNS}""",
+            (uuid4(), self.project_id, owner_user_id, request_key, tool,
+             Jsonb(parameters), Jsonb(binding), timeout_seconds, queue_wait_seconds),
+        )
+        return dict(await cursor.fetchone())
+
+    async def get(self, job_id, owner_user_id=None, *, include_result=False):
+        await self.maintain()
+        columns = JOB_COLUMNS + (",result" if include_result else "")
+        cursor = await self.connection.execute(
+            f"""SELECT {columns} FROM workbench.calculation_jobs WHERE project_id=%s AND job_id=%s
+            AND (%s::uuid IS NULL OR owner_user_id=%s)""",
+            (self.project_id, job_id, owner_user_id, owner_user_id),
+        )
+        row = await cursor.fetchone()
+        return dict(row) if row else None
+
+    async def list(self, *, owner_user_id, binding, limit, offset, status=None):
+        await self.maintain()
+        cursor = await self.connection.execute(
+            f"""SELECT {PUBLIC_JOB_COLUMNS} FROM workbench.calculation_jobs WHERE project_id=%s AND owner_user_id=%s
+            AND binding @> %s AND (%s::text IS NULL OR status=%s) ORDER BY created_at DESC,job_id LIMIT %s OFFSET %s""",
+            (self.project_id, owner_user_id, Jsonb(binding), status, status, limit, offset),
+        )
+        return [dict(row) for row in await cursor.fetchall()]
+
+    async def cancel(self, job_id, owner_user_id):
+        await self.connection.execute(
+            """UPDATE workbench.calculation_jobs SET status=CASE WHEN status='queued' THEN 'cancelled'
+            ELSE 'cancel_requested' END,finished_at=CASE WHEN status='queued' THEN clock_timestamp() ELSE finished_at END,updated_at=clock_timestamp()
+            WHERE project_id=%s AND job_id=%s AND owner_user_id=%s AND status IN ('queued','running')""",
+            (self.project_id, job_id, owner_user_id),
+        )
+        return await self.get(job_id, owner_user_id)
+
+    async def recover(self):
+        await self.maintain()
+        # A crashed write is never automatically replayed.
+        await self.connection.execute(
+            """UPDATE workbench.calculation_jobs SET status='interrupted',
+            error='{"code":"WORKER_INTERRUPTED"}'::jsonb,finished_at=clock_timestamp(),updated_at=clock_timestamp()
+            WHERE project_id=%s AND status IN ('running','cancel_requested')
+            AND lease_expires_at < clock_timestamp()""", (self.project_id,),
+        )
+
+    async def claim(self, owner, lease_seconds):
+        await self.maintain()
+        cursor = await self.connection.execute(
+            """UPDATE workbench.calculation_jobs j SET status='running',started_at=clock_timestamp(),lease_owner=%s,
+            lease_generation=j.lease_generation+1,
+            lease_expires_at=clock_timestamp()+%s*interval '1 second',updated_at=clock_timestamp()
+            WHERE j.job_id=(SELECT job_id FROM workbench.calculation_jobs
+              WHERE project_id=%s AND status='queued' AND queue_expires_at>clock_timestamp() ORDER BY created_at,job_id
+              FOR UPDATE SKIP LOCKED LIMIT 1) RETURNING j.*""",
+            (owner, lease_seconds, self.project_id),
+        )
+        row = await cursor.fetchone()
+        return dict(row) if row else None
+
+    async def heartbeat(self, row, owner, lease_seconds):
+        cursor = await self.connection.execute(
+            """UPDATE workbench.calculation_jobs SET lease_expires_at=clock_timestamp()+%s*interval '1 second',
+            updated_at=clock_timestamp() WHERE project_id=%s AND job_id=%s AND lease_owner=%s
+            AND lease_generation=%s AND lease_expires_at>clock_timestamp()
+            AND status IN ('running','cancel_requested') RETURNING status""",
+            (lease_seconds, self.project_id, row['job_id'], owner, row['lease_generation']),
+        )
+        result = await cursor.fetchone()
+        return result['status'] if result else None
+
+    async def finish(self, row, owner, status, *, result=None, error=None, result_retention_seconds=604800):
+        cursor = await self.connection.execute(
+            """UPDATE workbench.calculation_jobs SET status=%s,result=%s,error=%s,finished_at=clock_timestamp(),
+            result_expires_at=clock_timestamp()+%s*interval '1 second',updated_at=clock_timestamp()
+            WHERE project_id=%s AND job_id=%s AND lease_owner=%s AND lease_generation=%s
+            AND lease_expires_at>clock_timestamp() AND status IN ('running','cancel_requested')
+            AND (status!='cancel_requested' OR %s IN ('cancelled','interrupted')
+                 OR (%s='failed' AND %s='CLEANUP_FAILED')) RETURNING job_id""",
+            (status, Jsonb(result) if result is not None else None, Jsonb(error) if error else None, result_retention_seconds,
+             self.project_id, row['job_id'], owner, row['lease_generation'], status, status, (error or {}).get('code')),
+        )
+        return await cursor.fetchone() is not None

app/infra/db/postgresql/operational_tasks.py

@@ -19,6 +19,7 @@ class OperationalTaskRepository:
     def __init__(self, connection: AsyncConnection, project_id):
         self.connection = connection
         self.project_id = str(project_id)
+        self._agent_ownership_available: bool | None = None
 
     @asynccontextmanager
     async def transaction(self):
@@ -75,6 +76,48 @@ class OperationalTaskRepository:
                 status=409,
             )
 
+    async def pending_calculation_jobs(
+        self, task_id: str, run_id: str, execution_request_id: str,
+    ) -> list[str]:
+        """Only the exact original run blocks final artifacts and completion."""
+        cursor = await self.connection.execute(
+            """SELECT job_id FROM workbench.calculation_jobs
+               WHERE project_id=%s AND binding->>'task_id'=%s
+                 AND binding->>'run_id'=%s AND binding->>'execution_request_id'=%s
+                 AND status IN ('queued','running','cancel_requested')
+               ORDER BY created_at,job_id LIMIT 100""",
+            (self.project_id, task_id, run_id, execution_request_id),
+        )
+        return [str(row["job_id"]) for row in await cursor.fetchall()]
+
+    async def list_questions(self, task_id: str) -> list[dict]:
+        cursor = await self.connection.execute(
+            """SELECT document FROM workbench.operational_task_questions
+               WHERE project_id=%s AND task_id=%s
+               ORDER BY document->>'created_at' DESC, question_request_id DESC LIMIT 100""",
+            (self.project_id, task_id),
+        )
+        return [row["document"] for row in await cursor.fetchall()]
+
+    async def get_question(self, task_id: str, question_id: str) -> dict | None:
+        cursor = await self.connection.execute(
+            """SELECT document FROM workbench.operational_task_questions
+               WHERE project_id=%s AND task_id=%s AND question_request_id=%s""",
+            (self.project_id, task_id, question_id),
+        )
+        row = await cursor.fetchone()
+        return row["document"] if row else None
+
+    async def save_question(self, task_id: str, document: dict) -> None:
+        # All writers hold the owning task advisory lock.
+        await self.connection.execute(
+            """INSERT INTO workbench.operational_task_questions
+               (project_id,task_id,question_request_id,document) VALUES (%s,%s,%s,%s)
+               ON CONFLICT (project_id,question_request_id) DO UPDATE SET document=excluded.document
+               WHERE operational_task_questions.task_id=excluded.task_id""",
+            (self.project_id, task_id, document["question_request_id"], Jsonb(document)),
+        )
+
     async def get_task(self, task_id: str) -> dict | None:
         cursor = await self.connection.execute(
             """SELECT document FROM workbench.operational_tasks
@@ -87,7 +130,8 @@ class OperationalTaskRepository:
     async def get_active_run(self, task_id: str) -> dict | None:
         cursor = await self.connection.execute(
             """SELECT run_id,task_id,generation,execution_request_id,status,revision,
-                      agent_session_id,lease_expires_at,heartbeat_at,started_at,finished_at,
+                      agent_session_id,to_jsonb(operational_task_runs)->>'agent_instance_id' AS agent_instance_id,
+                      lease_expires_at,heartbeat_at,started_at,finished_at,
                       failure_code,failure_detail,dispatch_attempts,next_dispatch_at,
                       created_at,updated_at
                FROM workbench.operational_task_runs
@@ -102,7 +146,8 @@ class OperationalTaskRepository:
     async def get_run(self, run_id: str) -> dict | None:
         cursor = await self.connection.execute(
             """SELECT run_id,task_id,generation,execution_request_id,status,revision,
-                      agent_session_id,lease_expires_at,heartbeat_at,started_at,finished_at,
+                      agent_session_id,to_jsonb(operational_task_runs)->>'agent_instance_id' AS agent_instance_id,
+                      lease_expires_at,heartbeat_at,started_at,finished_at,
                       failure_code,failure_detail,dispatch_attempts,next_dispatch_at,
                       created_at,updated_at
                FROM workbench.operational_task_runs
@@ -129,13 +174,23 @@ class OperationalTaskRepository:
     async def require_execution_storage(self) -> None:
         """Reject claims before changing state if the inbox cannot be used."""
         cursor = await self.connection.execute(
-            "SELECT to_regclass('workbench.agent_messages') IS NOT NULL AS available"
+            """SELECT to_regclass('workbench.agent_messages') IS NOT NULL
+                      AND to_regclass('workbench.operational_task_questions') IS NOT NULL AS available,
+                      EXISTS (
+                        SELECT 1 FROM pg_attribute
+                        WHERE attrelid=to_regclass('workbench.operational_task_runs')
+                          AND attname='agent_instance_id' AND NOT attisdropped
+                      ) AS agent_ownership_available"""
         )
-        if not (await cursor.fetchone())["available"]:
+        readiness = await cursor.fetchone()
+        self._agent_ownership_available = bool(readiness.get("agent_ownership_available"))
+        if not readiness["available"] or not self._agent_ownership_available:
             raise DomainError(
                 "DEPENDENCY_UNAVAILABLE",
-                "Agent message storage is not provisioned; apply "
-                "resources/sql/024_agent_messages.sql to this project database before starting runs",
+                "Agent execution storage is not provisioned; apply "
+                "resources/sql/024_agent_messages.sql, resources/sql/028_task_questions.sql, "
+                "and resources/sql/030_operational_run_agent_ownership.sql "
+                "to this project database before starting runs",
                 status=503,
             )
 
@@ -176,14 +231,26 @@ class OperationalTaskRepository:
         )
 
     async def update_run(self, run: dict, expected_revision: int) -> None:
+        if self._agent_ownership_available is None:
+            cursor = await self.connection.execute(
+                """SELECT EXISTS (
+                     SELECT 1 FROM pg_attribute
+                     WHERE attrelid=to_regclass('workbench.operational_task_runs')
+                       AND attname='agent_instance_id' AND NOT attisdropped
+                   ) AS available"""
+            )
+            self._agent_ownership_available = bool((await cursor.fetchone())["available"])
+        owner_assignment = "agent_instance_id=%s," if self._agent_ownership_available else ""
+        owner_values = (run.get("agent_instance_id"),) if self._agent_ownership_available else ()
         cursor = await self.connection.execute(
-            """UPDATE workbench.operational_task_runs SET
-                 status=%s,revision=%s,agent_session_id=%s,lease_expires_at=%s,
+            f"""UPDATE workbench.operational_task_runs SET
+                 status=%s,revision=%s,agent_session_id=%s,{owner_assignment}lease_expires_at=%s,
                  heartbeat_at=%s,started_at=%s,finished_at=%s,failure_code=%s,
                  failure_detail=%s,dispatch_attempts=%s,next_dispatch_at=%s,updated_at=%s
                WHERE project_id=%s AND run_id=%s AND revision=%s""",
             (
                 run["status"], run["revision"], run.get("agent_session_id"),
+                *owner_values,
                 run.get("lease_expires_at"), run.get("heartbeat_at"),
                 run.get("started_at"), run.get("finished_at"), run.get("failure_code"),
                 run.get("failure_detail"), run.get("dispatch_attempts", 0),
@@ -200,7 +267,8 @@ class OperationalTaskRepository:
     ) -> list[dict]:
         result = await self.connection.execute(
             """SELECT run_id,task_id,generation,execution_request_id,status,revision,
-                      agent_session_id,lease_expires_at,heartbeat_at,started_at,finished_at,
+                      agent_session_id,to_jsonb(operational_task_runs)->>'agent_instance_id' AS agent_instance_id,
+                      lease_expires_at,heartbeat_at,started_at,finished_at,
                       failure_code,failure_detail,dispatch_attempts,next_dispatch_at,
                       created_at,updated_at
                FROM workbench.operational_task_runs
@@ -229,7 +297,8 @@ class OperationalTaskRepository:
     ) -> list[dict]:
         cursor = await self.connection.execute(
             """SELECT run_id,task_id,generation,execution_request_id,status,revision,
-                      agent_session_id,lease_expires_at,heartbeat_at,started_at,finished_at,
+                      agent_session_id,to_jsonb(operational_task_runs)->>'agent_instance_id' AS agent_instance_id,
+                      lease_expires_at,heartbeat_at,started_at,finished_at,
                       failure_code,failure_detail,dispatch_attempts,next_dispatch_at,
                       created_at,updated_at
                FROM workbench.operational_task_runs
@@ -286,6 +355,18 @@ class OperationalTaskRepository:
     async def current_event_cursor(self) -> int:
         return await WorkbenchEventRepository(self.connection, self.project_id).current_cursor()
 
+    async def list_presentation_events(self, task_id: str, after: int, through: int, limit: int) -> list[dict]:
+        cursor = await self.connection.execute(
+            """SELECT event_seq,occurred_at,metadata
+               FROM workbench.domain_event_log
+               WHERE project_id=%s AND topic='operational.presentation.requested'
+                 AND aggregate_type='operational_task_presentation' AND aggregate_id=%s
+                 AND event_seq>%s AND event_seq<=%s
+               ORDER BY event_seq LIMIT %s""",
+            (self.project_id, task_id, after, through, limit),
+        )
+        return [dict(row) for row in await cursor.fetchall()]
+
     async def get_task_by_create_request(self, request_id: str) -> dict | None:
         cursor = await self.connection.execute(
             """SELECT document FROM workbench.operational_tasks

app/infra/db/postgresql/read_queries.py

@@ -337,6 +337,7 @@ async def task_context_page(
         "evidence": ("operational_task_evidence", "evidence_id"),
         "actions": ("operational_task_actions", "action_id"),
         "access_requests": ("operational_task_access_requests", "access_request_id"),
+        "questions": ("operational_task_questions", "question_request_id"),
     }
     documents = {"task": "operational_tasks", "report": "operational_task_reports", "plan": "operational_task_plans"}
     if section in {"access_requests", "progress"}:

app/infra/integrations/operational_actions.py

@@ -52,7 +52,7 @@ class SampleWorkOrderReceipt(BaseModel):
     work_order_id: UUID
     project_id: UUID
     created_at: datetime
-    origin: Literal["synthetic"]
+    origin: Literal["system_data"]
     status: Literal[
         "draft", "open", "in_progress", "awaiting_reply", "completed", "cancelled"
     ]
@@ -195,7 +195,8 @@ class HttpOperationalActionDispatcher:
             receipt = ActionReceipt(
                 adapter="peripheral_sample", receipt_id=str(order.work_order_id),
                 submitted_at=order.created_at, completed_at=order.created_at,
-                external_status=order.status, detail="合成外围工单已创建，现场工作进度由外围系统继续同步",
+                external_status=order.status,
+                detail="外围工单已创建，现场工作进度由外围系统继续同步",
             )
             return ActionDispatchOutcome("succeeded", receipt.detail, receipt)
         result = CommandGatewayReceipt.model_validate(value)

app/infra/integrations/operational_data.py

@@ -5,6 +5,7 @@ import hashlib
 import json
 from datetime import datetime, timezone
 from pathlib import Path
+from typing import Literal
 
 import httpx
 from pydantic import Field
@@ -33,6 +34,23 @@ class TaskDataError(Exception):
         self.detail = detail
 
 
+def peripheral_records_origin(items: object) -> Literal["system_data", "synthetic"]:
+    """Classify a record collection; mixed collections retain per-item provenance."""
+    if not isinstance(items, list) or any(
+        not isinstance(item, dict) or item.get("origin") not in {"system_data", "synthetic"}
+        for item in items
+    ):
+        raise TaskDataError(502, "外围记录缺少明确来源，不能作为已核实证据")
+    return "synthetic" if items and all(item["origin"] == "synthetic" for item in items) else "system_data"
+
+
+def normalize_peripheral_work_orders(items: object) -> list[dict[str, object]]:
+    """Expose peripheral work orders only as project-owned system data."""
+    if not isinstance(items, list) or any(not isinstance(item, dict) for item in items):
+        raise TaskDataError(502, "外围工单格式无效，不能作为已核实证据")
+    return [{**item, "origin": "system_data"} for item in items]
+
+
 def _matches_project(value: object, project_id: str) -> bool:
     if isinstance(value, list):
         return all(_matches_project(item, project_id) for item in value)
@@ -147,7 +165,14 @@ async def read_task_data(
         data = _select_demo_observations(data, query, str(context.project_id))
     elif not _matches_project(data, str(context.project_id)):
         raise TaskDataError(502, "数据源返回了其他项目的数据")
-    origin = "synthetic" if peripheral else "system_data"
+    origin = "system_data"
+    if query.source == "peripheral_work_orders":
+        items = data.get("items") if isinstance(data, dict) else None
+        data = {**data, "items": normalize_peripheral_work_orders(items)}
+    elif query.source == "peripheral_test_data":
+        items = data.get("items") if isinstance(data, dict) else None
+        origin = peripheral_records_origin(items)
+        data = {**data, "provenance_scope": "per_record", "source_origins": sorted({item["origin"] for item in items})}
     historical_replay = False
     if query.source in {"peripheral_points", "peripheral_observations"}:
         source_origin = data.get("origin") if isinstance(data, dict) else None

app/main.py

@@ -14,6 +14,7 @@ from app.infra.db.metadb.database import close_metadata_engine
 from app.infra.db.timescaledb.sync_pool import close_all_timescale_pools
 from app.native.wndb.core.connection import close_all_project_pools
 from app.core.config import settings
+from samples.peripheral.api import create_app as create_peripheral_app
 
 # 导入审计中间件
 from app.infra.audit.middleware import AuditMiddleware
@@ -80,6 +81,7 @@ install_problem_details_handlers(app)
 from app.api.v1.endpoints.mcp_queries import router as mcp_context_router
 app.state.mcp_runtime = mcp_runtime
 app.include_router(mcp_context_router, prefix="/api/v1")
+app.mount("/peripheral", create_peripheral_app(), name="peripheral")
 # Legacy routers without version prefix
 # app.include_router(api_router)
 

app/mcp/analysis_models.py

@@ -4,21 +4,51 @@ from datetime import datetime, timedelta
 from typing import Annotated, Literal
 from uuid import UUID
 
-from pydantic import Field, model_validator
+from pydantic import BaseModel, ConfigDict, Field, model_validator
 
 from app.mcp.models import Id, StrictInput
 
 
-MAX_CALCULATION_TIMEOUT_SECONDS = 86400
+MAX_REQUEST_TIMEOUT_SECONDS = 240
 
 
 class CalculationInput(StrictInput):
     timeout_seconds: int = Field(
-        default=120, ge=1, le=MAX_CALCULATION_TIMEOUT_SECONDS, strict=True,
-        description="本次计算等待预算，秒；由Agent按模型规模和任务时长选择，受工具目录中的服务端上限约束；与模拟duration不同，超时不保证已启动的计算停止或回滚",
+        default=120, ge=1, le=MAX_REQUEST_TIMEOUT_SECONDS, strict=True,
+        description="本次HTTP请求响应等待预算，秒，默认120秒；不是后台计算执行时限，受工具目录中的请求上限约束；后台任务通过状态和结果工具查询",
     )
 
 
+class NodeCoverageGroup(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+    group_id: Id = Field(max_length=80)
+    title: str = Field(min_length=1, max_length=200)
+    node_ids: list[Id] = Field(min_length=1, max_length=10000)
+
+
+class NodeCoverageInput(CalculationInput):
+    groups: list[NodeCoverageGroup] = Field(min_length=1, max_length=8,
+        description="已核实的模型节点集合；各组为需要比较的业务或空间集合，可重叠，不接受任意坐标")
+    concave_ratio: float = Field(default=0.25, ge=0, le=1, allow_inf_nan=False,
+        description="凹包比例；越接近1越平滑接近凸包，参数只决定示意形状")
+    buffer_m: float = Field(default=25, gt=0, le=1000, allow_inf_nan=False,
+        description="展示缓冲距离，米；不是服务半径或影响距离")
+    split_by_connectivity: bool = Field(default=False, strict=True,
+        description="需要按模型连接分组时设true；地理分布比较默认false，不能据此认定实际供水关系")
+    excluded_link_ids: list[Id] = Field(default_factory=list, max_length=200,
+        description="连通分组时额外视为断开的管段/阀门，例如已有隔离建议中的关阀ID；仅影响绘图分组")
+
+    @model_validator(mode="after")
+    def bounded_groups(self):
+        if len({group.group_id for group in self.groups}) != len(self.groups):
+            raise ValueError("coverage group IDs must be distinct")
+        if len({node for group in self.groups for node in group.node_ids}) > 10000:
+            raise ValueError("coverage supports at most 10000 distinct nodes")
+        if self.excluded_link_ids and not self.split_by_connectivity:
+            raise ValueError("excluded_link_ids requires split_by_connectivity")
+        return self
+
+
 class BurstDetectionInput(CalculationInput):
     target_time: datetime = Field(description="带时区的侦测时刻；读取此前15天同周期压力数据，不是单小时定位")
     sensor_nodes: list[Id] = Field(min_length=5, max_length=20, description="已核实的压力监测节点ID，至少5个完整传感器")
@@ -38,7 +68,7 @@ class BurstDetectionInput(CalculationInput):
 class BurstLocationInput(CalculationInput):
     data_source: Literal["monitoring", "simulation"] = "monitoring"
     simulation_run_id: UUID | None = Field(default=None, description="已有分析运行UUID，不是方案名称；simulation模式只读取已有异常结果及模型基线")
-    burst_leakage: float = Field(gt=0, allow_inf_nan=False, description="有依据的假定漏水流量，WNTR SI单位m³/s；不是已知爆管管段，不能随意猜测")
+    burst_leakage: float = Field(gt=0, allow_inf_nan=False, description="当前爆管定位算法必需的假定漏水流量，WNTR SI单位m³/s；可用有数据和方法依据的估计或明确给定的情景假设，说明来源与不确定性，不要求现场实测确认；不能任填数值")
     pressure_scada_ids: list[Id] | None = Field(default=None, min_length=1, max_length=20, description="压力监测设备关联的模型节点ID，不是设备查询编号；省略时自动读取当前项目SCADA配置中的压力测点，不需要猜测节点ID")
     flow_scada_ids: list[Id] | None = Field(default=None, min_length=1, max_length=20, description="流量监测设备关联的模型要素ID，不是设备查询编号")
     scada_burst_start: datetime
@@ -69,6 +99,7 @@ class BurstLocationInput(CalculationInput):
 
 class ValveIsolationInput(CalculationInput):
     accident_element: list[Id] = Field(min_length=1, max_length=20, description="经核实或明确作为候选的事故管段/节点ID")
+    accident_element_type: Literal["auto", "link", "node"] = Field(default="auto", description="事故要素类型；auto优先识别管段，再识别节点；节点与管段重号时，节点分析须指定node，纯管段分析可指定link")
     disabled_valves: list[Id] = Field(default_factory=list, max_length=200, description="已知无法关闭的阀门；未提供不代表现场确认全部可用")
 
 

app/mcp/analysis_service.py

@@ -71,6 +71,10 @@ def _calculate(name: str, parameters: dict[str, Any], network: str, username: st
             from app.services.valve_isolation import analyze_valve_isolation
 
             return analyze_valve_isolation(network=network, **parameters)
+        if name == "generate_node_coverage":
+            from app.services.node_coverage import generate_node_coverage
+
+            return generate_node_coverage(network=network, **parameters)
         raise ValueError("Calculation is not registered")
     finally:
         _slots.release()

app/mcp/application.py

@@ -203,7 +203,11 @@ class McpQueryApplication:
             return None
         schema = tool.model.model_json_schema()
         if tool.computation or tool.mutating:
-            schema["properties"]["timeout_seconds"]["maximum"] = self.settings.max_calculation_timeout_seconds
+            schema["properties"]["timeout_seconds"]["maximum"] = self.settings.max_request_timeout_seconds
+            schema["properties"]["timeout_seconds"]["default"] = min(
+                self.settings.query_timeout_seconds,
+                self.settings.max_request_timeout_seconds,
+            )
         return schema
 
     async def list_tools(self, ctx, params):
@@ -213,6 +217,10 @@ class McpQueryApplication:
         for name in sorted(TOOLS):
             tool = TOOLS[name].protocol_tool()
             tool.input_schema.update(self.tool_input_schema(name))
+            tool.meta = {**(tool.meta or {}), "io.tjwater/request-timeout": {
+                "defaultSeconds": self.settings.query_timeout_seconds,
+                "maxSeconds": self.settings.max_request_timeout_seconds,
+            }}
             tools.append(tool)
         return types.ListToolsResult(
             tools=tools,
@@ -291,11 +299,13 @@ class McpQueryApplication:
             return tool_error(
                 "INVALID_QUERY", "Query parameters do not match the tool schema"
             )
+        if (tool.computation or tool.mutating) and "timeout_seconds" not in args.model_fields_set:
+            args.timeout_seconds = min(self.settings.query_timeout_seconds, self.settings.max_request_timeout_seconds)
         timeout_seconds = (
             args.timeout_seconds if tool.computation or tool.mutating else self.settings.query_timeout_seconds
         )
-        if (tool.computation or tool.mutating) and timeout_seconds > self.settings.max_calculation_timeout_seconds:
-            return tool_error("INVALID_QUERY", "Calculation timeout exceeds the server limit")
+        if (tool.computation or tool.mutating) and timeout_seconds > self.settings.max_request_timeout_seconds:
+            return tool_error("INVALID_QUERY", "HTTP request timeout exceeds the server limit")
         request = ctx.request
         project = request.headers.get("x-project-id", "") if request else ""
         meta = ctx.meta or {}
@@ -347,20 +357,28 @@ class McpQueryApplication:
                         )
                     result = QueryResult.model_validate(previous["result"])
                 else:
-                    result = (
-                        await self.executor(
-                            scope, binding, tool, args, full_snapshot=True
+                    from app.mcp.calculation_jobs import should_queue, submit_calculation
+                    if should_queue(tool, args):
+                        result = await submit_calculation(
+                            scope, binding, tool, args, call_id,
+                            job_timeout_seconds=self.settings.max_calculation_timeout_seconds,
+                        )
+                    else:
+                        result = (
+                            await self.executor(
+                                scope, binding, tool, args, full_snapshot=True
+                            )
+                            if include_full
+                            else await self.executor(scope, binding, tool, args)
                         )
-                        if include_full
-                        else await self.executor(scope, binding, tool, args)
-                    )
                     # Never serialize an unbounded response into a model message or database receipt.
                     if len(result.model_dump_json().encode()) > 8 * 1024 * 1024:
                         return tool_error(
                             "RESULT_TOO_LARGE",
                             "Narrow the query or request a smaller page",
                         )
-                    if binding.task_id and name not in {
+                    if binding.task_id and not should_queue(tool, args) and name not in {
+                        "get_calculation_job", "get_calculation_result", "cancel_calculation_job", "list_calculation_jobs",
                         "get_task_context",
                         "get_task_evidence",
                     }:
@@ -423,6 +441,10 @@ class McpQueryApplication:
                             "resource_uri": uri,
                             "preview": "The complete query snapshot is saved in task evidence",
                         }
+                        from app.mcp.presentation_preview import saved_presentation_preview
+                        presentation = saved_presentation_preview(result)
+                        if presentation is not None:
+                            output["data"]["presentation"] = presentation
                         output["truncated"] = True
                         output["data_gaps"] = [
                             *result.data_gaps,
@@ -458,6 +480,7 @@ class McpQueryApplication:
                 403: "QUERY_CONTEXT_DENIED",
                 404: "NOT_FOUND",
                 409: "QUERY_CONTEXT_STALE",
+                410: "RESULT_EXPIRED",
                 413: "RESULT_TOO_LARGE",
                 422: "INVALID_QUERY",
                 503: "SOURCE_UNAVAILABLE",
@@ -466,7 +489,7 @@ class McpQueryApplication:
                 codes.get(exc.status_code, "QUERY_FAILED"),
                 str(exc.detail)
                 if isinstance(exc.detail, str)
-                and exc.status_code in {403, 404, 409, 413, 422}
+                and exc.status_code in {403, 404, 409, 410, 413, 422}
                 else "Query is unavailable",
             )
         except (UndefinedTable, InvalidSchemaName):
@@ -475,7 +498,7 @@ class McpQueryApplication:
                 "MCP project storage has not been provisioned",
             )
         except TimeoutError:
-            return tool_error("QUERY_TIMEOUT", "Query execution timed out")
+            return tool_error("QUERY_TIMEOUT", "HTTP request timed out")
         except ValueError:
             return tool_error(
                 "INVALID_QUERY", "Invalid query value or incompatible source data"

app/mcp/calculation_jobs.py

@@ -0,0 +1,205 @@
+"""Durable MCP calculation submission and owner-scoped result retrieval.
+
+Only safe server-owned binding fields are persisted. OAuth tokens and context
+handles are never used to resume a calculation.
+"""
+from datetime import datetime, timezone
+from uuid import uuid5, NAMESPACE_URL
+from fastapi import HTTPException
+from app.mcp.models import BoundContext, QueryResult, PageInfo
+from app.infra.db.postgresql.calculation_jobs import CalculationJobRepository
+
+QUEUED_TOOLS = frozenset({
+    "run_burst_location", "run_leakage_identification", "run_burst_detection",
+    "run_simulation", "run_burst_simulation", "run_contaminant_simulation",
+    "analyze_valve_closure", "analyze_flushing", "analyze_pressure_regulation",
+    "optimize_sensor_placement", "save_simulation_scenario", "create_sensor_placement",
+})
+CONTROL_TOOLS = frozenset({"get_calculation_job", "get_calculation_result", "cancel_calculation_job", "list_calculation_jobs"})
+BINDING_FIELDS = {"project_id", "subject", "client_id", "native_session_id", "purpose", "parent_session_id", "task_id", "run_id", "execution_request_id", "tool_cap", "permission_cap", "task_scope_cap"}
+
+
+def durable_binding(binding: BoundContext) -> dict:
+    return binding.model_dump(mode="json", include=BINDING_FIELDS)
+
+
+def restore_binding(document: dict) -> BoundContext:
+    # Expiration governs admission via a context handle; durable work has its own
+    # lifecycle and must revalidate identity, membership and run on every use.
+    return BoundContext.model_validate({**document, "context_handle": "durable-calculation", "expires_at": datetime.max.replace(tzinfo=timezone.utc)})
+
+
+def should_queue(tool, args) -> bool:
+    return tool.name in QUEUED_TOOLS and (not tool.mutating or not args.dry_run)
+
+
+def envelope(source: str, data, *, job_id="list", page=None) -> QueryResult:
+    return QueryResult(source=source, source_ref="calculation:" + str(job_id), queried_at=datetime.now(timezone.utc), semantics="task", data=data, page=page, truncated=bool(page and (page.offset or page.has_more)),
+        data_gaps=["Calculation status and queue receipts are not analytical evidence."])
+
+
+def public_job(row: dict) -> dict:
+    fields = {"job_id", "status", "tool", "created_at", "started_at", "finished_at", "updated_at", "error", "queue_expires_at", "result_expires_at"}
+    from fastapi.encoders import jsonable_encoder
+    expiry = row.get("result_expires_at")
+    available = row.get("status") == "succeeded" and (expiry is None or expiry > datetime.now(timezone.utc))
+    return {**jsonable_encoder({key: row[key] for key in fields if key in row}),
+            "result_available": available, "poll_after_ms": 1500}
+
+
+async def submit_calculation(scope, binding, tool, args, call_id: str, *, job_timeout_seconds: int = 3600) -> QueryResult:
+    from app.mcp.query_service import authorize_query
+    await authorize_query(scope, binding, tool, args)
+    if binding.purpose == "task_reader":
+        raise HTTPException(403, "Read-only task agents cannot start calculations")
+    if tool.mutating:
+        from app.mcp.mutation_service import (
+            preview_change, check_digest, mutation_arguments, retryable_capacity_rejection,
+        )
+        from app.infra.db.postgresql.mcp_queries import McpQueryRepository
+        mutation_repo = McpQueryRepository(scope.connection, scope.project.project_id)
+        mutation_key = "mutation:" + binding.subject + ":" + binding.native_session_id
+        approved_parameters = {"tool": tool.name, "arguments": mutation_arguments(args, preview=False)}
+        async with scope.connection.transaction():
+            await mutation_repo.lock_receipt(mutation_key, str(args.request_id))
+            previous = await mutation_repo.receipt(mutation_key, str(args.request_id))
+            if previous:
+                if previous["parameters"] != approved_parameters:
+                    raise HTTPException(409, "request_id already has different parameters")
+                if previous["result"].get("status") == "completed":
+                    return QueryResult.model_validate(previous["result"]["output"])
+                if not retryable_capacity_rejection(previous["result"]):
+                    raise HTTPException(409, "Operation already started; inspect the existing calculation or mutation status")
+        check_digest(args, await preview_change(scope, binding, tool.name, args))
+    parameters = args.model_dump(mode="json", exclude={"context_handle", "timeout_seconds"})
+    request_id = str(args.request_id) if tool.mutating else call_id
+    key = str(uuid5(NAMESPACE_URL, ":".join([binding.subject, binding.client_id, binding.native_session_id, str(binding.task_id), str(binding.run_id), request_id])))
+    repo = CalculationJobRepository(scope.connection, scope.project.project_id)
+    if binding.task_id:
+        from app.mcp.query_service import task_service
+        from app.mcp.contexts import verify_task
+        task_repo = task_service(scope).repo
+        # Share the operational transaction lock with finalization. Admission
+        # cannot race a pending-job check and then publish work into a closed run.
+        async with task_repo.transaction():
+            await task_repo.lock(f"task:{binding.task_id}")
+            await task_repo.lock(f"run:{binding.run_id}")
+            await verify_task(scope, binding)
+            row = await repo.enqueue(
+                owner_user_id=scope.project.user_id, request_key=key, tool=tool.name,
+                parameters=parameters, binding=durable_binding(binding), timeout_seconds=job_timeout_seconds,
+            )
+    else:
+        row = await repo.enqueue(
+            owner_user_id=scope.project.user_id, request_key=key, tool=tool.name,
+            parameters=parameters, binding=durable_binding(binding), timeout_seconds=job_timeout_seconds,
+        )
+    await scope.connection.commit()
+    return envelope(tool.name, public_job(row), job_id=row["job_id"])
+
+
+def check_binding(binding, row):
+    original = row["binding"]
+    if any(str(original.get(key)) != str(getattr(binding, key)) for key in ("subject", "client_id", "native_session_id", "task_id", "run_id", "execution_request_id")):
+        raise HTTPException(403, "Calculation belongs to a different session or task run")
+
+
+async def authorize_original(scope, row):
+    from app.mcp.catalog import TOOLS
+    from app.mcp.query_service import authorize_query
+    from app.mcp.contexts import verify_task
+    original = restore_binding(row["binding"])
+    tool = TOOLS[row["tool"]]
+    # Legacy queue rows stored the old HTTP timeout. It does not govern durable execution.
+    parameters = {key: value for key, value in row["parameters"].items() if key != "timeout_seconds"}
+    args = tool.model.model_validate({**parameters, "context_handle": "durable-calculation-context-handle"})
+    if (str(scope.project.user_id) != str(row["owner_user_id"])
+            or str(scope.project.project_id) != str(original.project_id)):
+        raise HTTPException(403, "Calculation owner is no longer authorized")
+    if original.task_id:
+        await verify_task(scope, original)
+    await authorize_query(scope, original, tool, args)
+    return original, tool, args
+
+
+async def read_calculation_job(scope, binding, tool, args) -> QueryResult:
+    repo = CalculationJobRepository(scope.connection, scope.project.project_id)
+    if tool.name == "list_calculation_jobs":
+        identity_scope = binding.model_dump(mode="json", include={
+            "subject", "client_id", "native_session_id", "task_id", "run_id", "execution_request_id",
+        })
+        rows = await repo.list(owner_user_id=scope.project.user_id, binding=identity_scope, limit=args.limit + 1, offset=args.offset, status=args.status)
+        return envelope(tool.name, [public_job(row) for row in rows[:args.limit]], page=PageInfo(limit=args.limit, offset=args.offset, has_more=len(rows) > args.limit))
+    row = await repo.get(args.job_id, owner_user_id=scope.project.user_id,
+                         include_result=tool.name == "get_calculation_result")
+    if row is None:
+        raise HTTPException(404, "Calculation was not found")
+    check_binding(binding, row)
+    await authorize_original(scope, row)
+    if tool.name == "cancel_calculation_job":
+        if binding.purpose == "task_reader":
+            raise HTTPException(403, "Read-only task agents cannot cancel calculations")
+        row = await repo.cancel(args.job_id, owner_user_id=scope.project.user_id)
+        await scope.connection.commit()
+    if tool.name != "get_calculation_result" or row["status"] != "succeeded":
+        return envelope(tool.name, public_job(row), job_id=row["job_id"])
+    expiry = row.get("result_expires_at")
+    if row.get("result") is None or (expiry is not None and expiry <= datetime.now(timezone.utc)):
+        raise HTTPException(410, "Calculation result retention period has expired")
+    result = QueryResult.model_validate(row["result"])
+    result = await attach_calculation_evidence(scope, row, result)
+    value = result.data
+    for part in args.data_path:
+        if isinstance(value, dict) and part in value:
+            value = value[part]
+        else:
+            raise HTTPException(422, "data_path must select an existing object field")
+    if isinstance(value, list):
+        result.data = value[args.offset:args.offset + args.limit]
+        result.page = PageInfo(limit=args.limit, offset=args.offset, total=len(value), has_more=args.offset + args.limit < len(value))
+        result.truncated = result.truncated or args.offset > 0 or result.page.has_more
+    elif args.offset:
+        raise HTTPException(422, "offset requires a collection data_path")
+    else:
+        result.data = value
+    if len(result.model_dump_json().encode()) > 11000:
+        raise HTTPException(413, "Select a narrower data_path or smaller result page")
+    return result
+
+
+async def execute_calculation_job(scope, row) -> QueryResult:
+    """Worker entry point; success is persisted by the fenced supervisor."""
+    from app.mcp.query_service import execute_query
+    original, tool, args = await authorize_original(scope, row)
+    result = await execute_query(scope, original, tool, args)
+    return result
+
+
+async def attach_calculation_evidence(scope, row, result):
+    from app.mcp.query_service import task_service
+    from app.mcp.contexts import verify_task
+    original, tool, _ = await authorize_original(scope, row)
+    if original.task_id:
+        from app.domain.schemas.operational_task import TaskDataSnapshot
+        from app.infra.db.postgresql.mcp_queries import McpQueryRepository
+        repo = McpQueryRepository(scope.connection, scope.project.project_id)
+        key, call_id = "calculation:" + str(row["job_id"]), "result"
+        # Result lookup can leave an implicit SELECT transaction open. End it so
+        # evidence and its receipt commit together before returning a response.
+        await scope.connection.commit()
+        task_repo = task_service(scope).repo
+        # Match operational finalization's lock order: event lock, then task.
+        async with task_repo.transaction():
+            await repo.lock_receipt(key, call_id)
+            previous = await repo.receipt(key, call_id)
+            if previous:
+                return QueryResult.model_validate(previous["result"])
+            await task_repo.lock(f"task:{original.task_id}")
+            await verify_task(scope, original)
+            result.source_ref = key
+            evidence, _ = await task_service(scope).record_data_snapshot(original.task_id, TaskDataSnapshot(
+                source=tool.name, source_ref=key, queried_at=result.queried_at, origin=result.origin,
+                data=result.model_dump(mode="json"), truncated=result.truncated))
+            result.evidence_id = evidence.evidence_id
+            await repo.save_receipt(key, call_id, {"job_id": str(row["job_id"])}, result.model_dump(mode="json"))
+    return result

app/mcp/calculation_models.py

@@ -0,0 +1,21 @@
+"""Bounded inputs for durable computation control."""
+from typing import Literal
+from uuid import UUID
+from pydantic import Field
+from app.mcp.models import StrictInput, PageInput, Id
+
+CalculationStatus = Literal["queued", "running", "cancel_requested", "succeeded", "failed", "cancelled", "interrupted", "timed_out"]
+
+
+class CalculationJobInput(StrictInput):
+    job_id: UUID
+
+
+class CalculationResultInput(CalculationJobInput):
+    data_path: list[Id] = Field(default_factory=list, max_length=8)
+    limit: int = Field(default=20, ge=1, le=500)
+    offset: int = Field(default=0, ge=0, le=1000000)
+
+
+class CalculationJobsInput(PageInput):
+    status: CalculationStatus | None = None

app/mcp/catalog.py

@@ -4,8 +4,9 @@ from dataclasses import dataclass
 from pydantic import BaseModel
 from mcp import types
 from app.mcp import models as m
+from app.mcp import calculation_models as jobs
 from app.mcp import extended_models as e
-from app.mcp.analysis_models import BurstDetectionInput, BurstLocationInput, ValveIsolationInput, LeakageIdentificationInput
+from app.mcp.analysis_models import BurstDetectionInput, BurstLocationInput, ValveIsolationInput, LeakageIdentificationInput, NodeCoverageInput
 from app.mcp.analysis_models import ScenarioInput, ContaminantSimulationInput, ValveClosureInput, FlushingInput, PressureRegulationInput, SensorPlacementInput
 
 
@@ -36,11 +37,15 @@ class QueryTool:
             )
         return types.Tool(
             name=self.name,
-            description=self.description + suffix,
+            description=self.description + suffix + (
+                "。执行返回持久job_id；排队回执不是分析证据，使用get_calculation_job查询状态、get_calculation_result分页读取成功结果，不重复提交。"
+                if self.name in {"run_burst_location", "run_leakage_identification", "run_burst_detection", "run_simulation", "run_burst_simulation", "run_contaminant_simulation", "analyze_valve_closure", "analyze_flushing", "analyze_pressure_regulation", "optimize_sensor_placement", "save_simulation_scenario", "create_sensor_placement"}
+                else ""
+            ),
             inputSchema=self.model.model_json_schema(),
             outputSchema=m.QueryResult.model_json_schema(),
             annotations=types.ToolAnnotations(
-                readOnlyHint=not (self.computation or self.mutating), destructiveHint=self.mutating or self.name == "run_simulation", openWorldHint=self.name in {"search_project_places", "reverse_geocode"}
+                readOnlyHint=not (self.computation or self.mutating or self.name == "cancel_calculation_job"), destructiveHint=self.mutating or self.name == "run_simulation", openWorldHint=self.name in {"search_project_places", "reverse_geocode"}
             ),
         )
 
@@ -54,8 +59,9 @@ _specs = [
     ("optimize_sensor_placement", "使用灵敏度或K-means算法推荐压力监测点；仅返回候选节点，不保存正式方案，不保证污染溯源可观测性", SensorPlacementInput, "optimization.run", "analysis", False, True),
     ("run_leakage_identification", "基于SCADA压力与模型进行DMA漏损分配识别，需有依据的总漏损流量和时间对齐；返回区域估计，不是已确认漏点。内部执行遗传算法与水力计算", LeakageIdentificationInput, "burst.run", "analysis", False, True),
     ("run_burst_detection", "爆管异常侦测：使用至少5个完整压力传感器此前15天数据；只提示压力异常，不定位、不证明爆管", BurstDetectionInput, "burst.run", "analysis", False, True),
-    ("run_burst_location", "爆管候选定位：对比异常与正常压力，可辅助流量；需有依据的漏水量假设。压力节点可省略，自动读取项目SCADA压力测点；monitoring正常窗可成对省略，默认前一天同时间段，须说明该对照假设。simulation须有已有分析运行UUID并读取同窗模型基线，不是实测；定位内部会执行水力计算，不创建新的事件方案", BurstLocationInput, "burst.run", "analysis", False, True),
-    ("analyze_valve_isolation", "按事故候选要素分析拓扑隔离边界、建议关闭阀门和受影响节点；不是污染扩散或水力影响范围，不执行关阀，须现场核实", ValveIsolationInput, "simulation.run", "analysis", False, True),
+    ("run_burst_location", "爆管候选定位：对比异常与正常压力，可辅助流量；burst_leakage必需，可使用有依据的估计或明确给定的情景假设，不要求现场确诊；本工具不反演总漏水量。压力节点可省略，自动读取项目SCADA压力测点；monitoring正常窗可成对省略，默认前一天同时间段，须说明该对照假设。simulation须有已有分析运行UUID并读取同窗模型基线，不是实测；定位内部会执行水力计算，不创建新的事件方案", BurstLocationInput, "burst.run", "analysis", False, True),
+    ("analyze_valve_isolation", "按事故候选要素分析拓扑隔离边界、建议关闭阀门和受影响节点；候选分析无需等待现场确诊。输出不是污染扩散或水力影响范围；现场操作前须核实阀门映射及可用性，本工具不执行关阀", ValveIsolationInput, "simulation.run", "analysis", False, True),
+    ("generate_node_coverage", "为当前项目已核实的模型节点集合生成空间覆盖示意，适用于候选分布、监测覆盖比较、区域成员、隔离或模拟结果节点等，不限调查任务。支持最多8组、合计10000个不同节点，从真实GIS查询坐标；监测设备、投诉或地址须先关联模型节点。可按连接分组，记录凹包/缓冲参数、缺失节点和display_only；不计算故障、污染、实际停水或服务边界。返回result.map_payload及coverage，任务中用set_overlay引用本次evidence_id、省略data展示；生成不自动操作地图", NodeCoverageInput, "webgis.view", "analysis", False, True),
     (
         "get_element_properties",
         "读取当前项目的单个管网元素属性",
@@ -208,7 +214,7 @@ _specs = [
     ),
     (
         "query_peripheral_data",
-        "查询有限外围数据源；synthetic是合成证据",
+        "查询外围业务工单及监测数据；外围工单属于系统数据，监测数据按实际来源保留来源信息",
         m.PeripheralInput,
         "simulation.view",
         "peripheral",
@@ -256,4 +262,10 @@ _specs += [
     ("retry_operational_task", "预览或重新入队已终止任务；不自动启动新Agent会话", e.TaskRetryInput, "operational_task.write", "task", False, False, True),
     ("cancel_task_run", "预览或取消指定任务运行，校验运行版本", e.TaskCancelInput, "operational_task.write", "task", False, False, True),
 ]
+_specs += [
+    ("get_calculation_job", "读取当前身份与会话的持久计算状态；排队回执不是分析证据", jobs.CalculationJobInput, "webgis.view", "task"),
+    ("get_calculation_result", "读取成功计算的原始结果；data_path选择嵌套集合，limit/offset按需分页", jobs.CalculationResultInput, "webgis.view", "task"),
+    ("cancel_calculation_job", "请求取消本人当前会话的计算；运行中取消由独立计算进程处理，须继续核实最终状态", jobs.CalculationJobInput, "webgis.view", "task"),
+    ("list_calculation_jobs", "分页读取本人当前会话及任务范围的计算状态", jobs.CalculationJobsInput, "webgis.view", "task"),
+]
 TOOLS = {row[0]: QueryTool(*row) for row in _specs}

app/mcp/config.py

@@ -21,7 +21,8 @@ class McpSettings(BaseSettings):
     )
     allowed_origins: list[str] = Field(default_factory=list)
     context_ttl_seconds: int = Field(default=600, ge=30, le=600)
-    query_timeout_seconds: int = Field(default=30, ge=1, le=60)
+    query_timeout_seconds: int = Field(default=120, ge=1, le=240)
+    max_request_timeout_seconds: int = Field(default=240, ge=1, le=240)
     max_calculation_timeout_seconds: int = Field(default=3600, ge=120, le=86400)
     model_config = SettingsConfigDict(
         env_prefix="TJWATER_MCP_",
@@ -32,6 +33,8 @@ class McpSettings(BaseSettings):
 
     @model_validator(mode="after")
     def validate_resource(self):
+        if self.query_timeout_seconds > self.max_request_timeout_seconds:
+            raise ValueError("query timeout exceeds the HTTP request timeout limit")
         if self.dev_http_issuer_uri:
             dev_url = urlsplit(self.dev_http_issuer_uri)
             try:

app/mcp/contexts.py

@@ -70,7 +70,7 @@ async def bind_context(
         subject=principal.subject,
         client_id=principal.client_id,
         tool_cap=sorted(name for name, tool in TOOLS.items()
-                        if (command.purpose != "task_reader" or not (tool.computation or tool.mutating))
+                        if (command.purpose != "task_reader" or not (tool.computation or tool.mutating or name == "cancel_calculation_job"))
                         and (not tool.mutating or command.purpose == "ordinary_chat")),
         permission_cap=sorted(permissions_for_context(scope.project)),
         native_session_id=command.native_session_id,

app/mcp/models.py

@@ -22,7 +22,7 @@ class PageInput(StrictInput):
 
 
 class TaskContextInput(PageInput):
-    section: Literal["summary", "task", "report", "evidence", "actions", "progress", "access_requests", "plan", "run", "skill"] = "summary"
+    section: Literal["summary", "task", "report", "evidence", "actions", "progress", "access_requests", "plan", "run", "skill", "questions"] = "summary"
     report_revision: int | None = Field(default=None, ge=1)
     data_path: list[Id] = Field(default_factory=list, max_length=8)
 

app/mcp/peripheral.py

@@ -7,7 +7,12 @@ from app.infra.db.postgresql.read_queries import MAX_QUERY_BYTES, read_rows, sca
 from app.domain.schemas.operational_data import TaskDataSnapshot, TaskDataQuery
 from app.services.content_digest import digest
 from samples.peripheral.store import PostgresStore
-from app.infra.integrations.operational_data import _select_demo_observations
+from app.infra.integrations.operational_data import (
+    TaskDataError,
+    _select_demo_observations,
+    normalize_peripheral_work_orders,
+    peripheral_records_origin,
+)
 
 
 async def read_peripheral_query(scope, args):
@@ -37,17 +42,25 @@ async def read_peripheral_query(scope, args):
         truncated = len(rows) > args.limit
         if any(item.get("project_id") != project for item in items):
             raise HTTPException(502, "Peripheral project mismatch")
+        try:
+            if args.source == "peripheral_work_orders":
+                items = normalize_peripheral_work_orders(items)
+                origin = "system_data"
+            else:
+                origin = peripheral_records_origin(items)
+        except TaskDataError as exc:
+            raise HTTPException(exc.status, exc.detail) from exc
         data = {
             "items": items,
             "truncated": truncated,
             "limit": args.limit,
             "offset": args.offset,
         }
-        origin = (
-            "synthetic"
-            if items and any(item.get("origin") == "synthetic" for item in items)
-            else "system_data"
-        )
+        if args.source == "peripheral_test_data":
+            data.update(
+                provenance_scope="per_record",
+                source_origins=sorted({item["origin"] for item in items}),
+            )
     elif args.source == "peripheral_points":
         rows = await scada_devices(conn, limit=args.limit + 1, offset=args.offset)
         data = {

app/mcp/presentation_preview.py

@@ -0,0 +1,31 @@
+"""Keep a saved geometry discoverable even when its inline response is shortened."""
+
+from uuid import NAMESPACE_URL, uuid5
+
+from app.mcp.models import QueryResult
+
+
+def saved_presentation_preview(result: QueryResult) -> dict | None:
+    if result.source != "generate_node_coverage" or not result.evidence_id:
+        return None
+    analysis = result.data.get("result") if isinstance(result.data, dict) else None
+    if not isinstance(analysis, dict):
+        return None
+    payload = analysis.get("map_payload")
+    if not isinstance(payload, dict) or not payload.get("features"):
+        return None
+    coverage = analysis.get("coverage", {})
+    return {
+        "command": {
+            "action": "set_overlay", "mode": "geometry", "evidence_id": result.evidence_id,
+            "overlay_id": f"node-coverage-{uuid5(NAMESPACE_URL, result.evidence_id)}",
+            "title": "节点空间覆盖示意",
+        },
+        "display_only": True,
+        "coverage_status": coverage.get("status"),
+        "groups": {role: {
+            "node_count": len(group.get("node_ids", [])),
+            "mapped_node_count": group.get("mapped_node_count", 0),
+            "missing_node_count": len(group.get("missing_node_ids", [])),
+        } for role, group in coverage.get("groups", {}).items()},
+    }

app/mcp/query_service.py

@@ -131,6 +131,9 @@ async def execute_query(
     full_snapshot: bool = False,
 ) -> QueryResult:
     await authorize_query(scope, binding, tool, args)
+    if tool.name in {"get_calculation_job", "get_calculation_result", "cancel_calculation_job", "list_calculation_jobs"}:
+        from app.mcp.calculation_jobs import read_calculation_job
+        return await read_calculation_job(scope, binding, tool, args)
     if tool.mutating:
         from app.mcp.mutation_service import execute_mutation
         return await execute_mutation(scope, binding, tool, args)
@@ -176,6 +179,7 @@ async def execute_query(
             "run_burst_detection": "Pressure anomalies alone do not confirm a burst or locate a pipe.",
             "run_burst_location": "Candidate ranking depends on the leakage assumption, model and sensor coverage; field confirmation is required.",
             "analyze_valve_isolation": "Topology-only isolation advice; valve operability, hydraulic impact and field actions are not verified or executed.",
+            "generate_node_coverage": "Derived node coverage for display only; geometry does not establish a fault, contamination, outage or service boundary. Missing GIS positions are reported separately.",
             "calculate_demand_allocation": "Pipe-length-weighted allocation only; model demands were not modified. Units follow the current model.",
             "run_burst_simulation": "Forward hydraulic simulation for assumed orifice areas, not detection or confirmed burst location; only requested elements returned.",
         }[name]]
@@ -409,14 +413,14 @@ async def execute_query(
                 data["display_time_zone"] = "Asia/Shanghai"
                 data["approved_data_scopes"] = binding.task_scope_cap
                 data["bound_run_id"] = str(binding.run_id) if binding.run_id else None
-                data["available_sections"] = ["task", "report", "evidence", "actions", "progress", "access_requests", "plan", "run", "skill"]
+                data["available_sections"] = ["task", "report", "evidence", "actions", "progress", "access_requests", "plan", "run", "skill", "questions"]
                 truncated = True
                 gaps.append("Task context is a bounded summary; use section, data_path, limit and offset for detail. Previews are not complete reports.")
         else:
             projection = await reads.task_context_page(conn, project_id=binding.project_id, task_id=binding.task_id,
                 section=args.section, report_revision=args.report_revision, data_path=args.data_path, limit=args.limit, offset=args.offset)
             total = projection["total"]
-            if args.section in {"evidence", "actions", "access_requests"} and not args.data_path:
+            if args.section in {"evidence", "actions", "access_requests", "questions"} and not args.data_path:
                 total = projection["snapshot_metadata"].get("total", total)
             if projection["value_type"] in {"array", "object", "string"}:
                 page = PageInfo(limit=args.limit, offset=args.offset, total=total, has_more=args.offset + args.limit < total)

app/native/wndb/core/projects.py

@@ -201,6 +201,8 @@ def temporary_project_name(project: str, purpose: str) -> str:
 def temporary_project_database(project: str, purpose: str):
     """Clone one project's runnable model into an isolated temporary database."""
     temporary_name = temporary_project_name(project, purpose)
+    from app.services.calculation_cleanup import record_calculation_database
+    record_calculation_database(temporary_name)
     try:
         copy_project(
             get_project_template_database_name(project),
@@ -211,12 +213,15 @@ def temporary_project_database(project: str, purpose: str):
     finally:
         if have_project(temporary_name):
             delete_project(temporary_name)
+        record_calculation_database(temporary_name, removed=True)
 
 
 @contextmanager
 def temporary_template_database(name_hint: str, purpose: str):
     """Create an empty schema-only temporary database from the fixed template."""
     temporary_name = temporary_project_name(name_hint, purpose)
+    from app.services.calculation_cleanup import record_calculation_database
+    record_calculation_database(temporary_name)
     try:
         copy_project(
             get_schema_template_database_name(),
@@ -227,6 +232,7 @@ def temporary_template_database(name_hint: str, purpose: str):
     finally:
         if have_project(temporary_name):
             delete_project(temporary_name)
+        record_calculation_database(temporary_name, removed=True)
 
 
 def have_project(name: str) -> bool:

app/native/wndb/gis/network_views.py

@@ -1,10 +1,39 @@
-"""Read-only projections backed by the GIS materialized-view query layer."""
+"""Read-only GIS projections and authoritative network topology snapshots."""
 
 from typing import Any
 
 from ..core.database import read, read_all
 
 
+def get_node_geographic_positions(name: str, node_ids: list[str]) -> list[Any]:
+    """Resolve only the selected project's stored node geometries to WGS84."""
+    if not node_ids:
+        return []
+    return read_all(name, """
+        SELECT node_id AS id,
+               ST_X(ST_Transform(geom,4326)) AS longitude,
+               ST_Y(ST_Transform(geom,4326)) AS latitude
+        FROM gis.node_geometries
+        WHERE node_id = ANY(%s) AND geom IS NOT NULL AND NOT ST_IsEmpty(geom)
+        ORDER BY node_id
+    """, (node_ids,))
+
+
+def get_node_coverage_links(name: str, node_ids: list[str]) -> list[Any]:
+    """Read the selected subgraph and initial statuses without changing the model."""
+    if not node_ids:
+        return []
+    return read_all(name, """
+        SELECT l.id, l.start_node_id, l.end_node_id,
+               COALESCE(s.status::text, p.status::text) AS status
+        FROM network.links l
+        LEFT JOIN network.pipes p ON p.link_id = l.id
+        LEFT JOIN network.link_initial_settings s ON s.link_id = l.id
+        WHERE l.start_node_id = ANY(%s) AND l.end_node_id = ANY(%s)
+        ORDER BY l.id
+    """, (node_ids, node_ids))
+
+
 def _node_coord_rows(
     rows: list[Any],
 ) -> dict[str, dict[str, Any]]:
@@ -69,6 +98,46 @@ def get_network_link_nodes(name: str) -> list[str]:
     ]
 
 
+def get_valve_isolation_topology(name: str) -> tuple[list[str], list[str], list[str]]:
+    """Read links, closed statuses and sources from one consistent SQL snapshot.
+
+    Isolation must include model links without published geometries, and use
+    initial statuses rather than treating every link as open. UNION ALL keeps
+    source nodes in the same statement snapshot without repeating them for
+    every link or relying on materialized view refresh timing.
+    """
+    rows = read_all(
+        name,
+        """
+        SELECT l.id, l.link_type::text AS element_type,
+               l.start_node_id, l.end_node_id,
+               COALESCE(s.status::text, p.status::text) AS status
+        FROM network.links AS l
+        LEFT JOIN network.pipes AS p ON p.link_id = l.id
+        LEFT JOIN network.link_initial_settings AS s ON s.link_id = l.id
+        UNION ALL
+        SELECT n.id, n.node_type::text AS element_type,
+               NULL, NULL, NULL
+        FROM network.nodes AS n
+        WHERE n.node_type::text IN ('reservoir', 'tank')
+        """,
+    )
+    link_entries: list[str] = []
+    source_nodes: list[str] = []
+    closed_links: list[str] = []
+    for row in rows:
+        element_id = str(row["id"])
+        if row["element_type"] in {"reservoir", "tank"}:
+            source_nodes.append(element_id)
+        else:
+            link_entries.append(
+                f"{element_id}:{row['element_type']}:{row['start_node_id']}:{row['end_node_id']}"
+            )
+            if row["status"] == "CLOSED":
+                closed_links.append(element_id)
+    return link_entries, source_nodes, closed_links
+
+
 def get_major_pipe_nodes(name: str, diameter: int) -> list[str]:
     """Return large pipes in the established API wire format."""
     rows = read_all(

app/services/calculation_cleanup.py

@@ -0,0 +1,74 @@
+"""Record only job-owned native databases so abrupt cancellation is recoverable."""
+import asyncio
+import json
+import os
+from pathlib import Path
+import re
+from uuid import UUID
+
+
+def _names(directory: Path) -> list[str]:
+    path = directory / "databases.json"
+    if not path.exists():
+        return []
+    if path.is_symlink() or path.stat().st_size > 65536:
+        raise ValueError("Invalid calculation database ledger")
+    names = json.loads(path.read_text())
+    if not isinstance(names, list) or len(names) > 100 or any(
+        not isinstance(name, str) or not re.fullmatch(r"tjw_tmp_[a-z0-9_]+_[a-f0-9]{32}", name)
+        or len(name) > 63 for name in names
+    ):
+        raise ValueError("Invalid calculation database ledger")
+    return names
+
+
+def _write(directory: Path, names: list[str]) -> None:
+    path = directory / "databases.json"
+    staging = directory / "databases.json.new"
+    with staging.open("w") as output:
+        json.dump(names, output)
+        output.flush()
+        os.fsync(output.fileno())
+    staging.replace(path)
+
+
+def record_calculation_database(name: str, *, removed: bool = False) -> None:
+    """Called before clone creation and after successful normal cleanup."""
+    root = os.environ.get("TJWATER_CALCULATION_JOB_DIR")
+    if not root:
+        return
+    directory = Path(root)
+    names = _names(directory)
+    if removed:
+        names = [value for value in names if value != name]
+    elif name not in names:
+        names.append(name)
+    _write(directory, names)
+
+
+async def cleanup_calculation_databases(project_id: str, job_directory: str) -> None:
+    directory = Path(job_directory)
+    names = _names(directory)
+    if not names:
+        return
+    from app.infra.db.metadb.database import SessionLocal
+    from app.infra.db.metadb.repositories.metadata_repository import MetadataRepository
+    from app.infra.db.project_routing import ActiveProjectRouting, activate_project_routing
+    from app.native.wndb.core.projects import have_project, delete_project
+    async with SessionLocal() as session:
+        metadata = MetadataRepository(session)
+        project = await metadata.get_project_by_id(UUID(str(project_id)))
+        route = await metadata.get_project_db_routing(UUID(str(project_id)), "biz_data")
+        if not project or not route or route.db_type != "postgresql":
+            raise ValueError("Calculation cleanup routing unavailable")
+        routing = ActiveProjectRouting(project.code, route.dsn, project_id=project.id)
+        await session.commit()
+
+    def cleanup():
+        with activate_project_routing(routing):
+            for name in names[:]:
+                if have_project(name):
+                    delete_project(name)
+                names.remove(name)
+                _write(directory, names)
+    await asyncio.to_thread(cleanup)

app/services/calculation_worker.py

@@ -0,0 +1,299 @@
+"""Independent, singleton calculation supervisor and isolated job subprocesses.
+
+Never imported by API lifespan. Child output contains the original QueryResult;
+algorithm stdout/stderr is discarded rather than interpreted as a protocol.
+"""
+from __future__ import annotations
+
+import asyncio
+from contextlib import asynccontextmanager, contextmanager
+from dataclasses import dataclass
+from datetime import datetime, timezone
+import json
+import os
+from pathlib import Path
+import signal
+import shutil
+import sys
+import tempfile
+import time
+from uuid import UUID, uuid4
+
+
+@dataclass(frozen=True)
+class WorkerSettings:
+    project_ids: tuple[str, ...]
+    max_concurrency: int = 2
+    poll_seconds: float = 2
+    lease_seconds: int = 30
+    heartbeat_seconds: float = 5
+
+    @classmethod
+    def from_environment(cls):
+        projects = tuple(dict.fromkeys(str(UUID(p.strip())) for p in
+            os.environ['CALCULATION_WORKER_PROJECT_IDS'].split(',')))
+        result = cls(projects, int(os.getenv('CALCULATION_WORKER_MAX_CONCURRENCY', '2')),
+                     float(os.getenv('CALCULATION_WORKER_POLL_SECONDS', '2')),
+                     int(os.getenv('CALCULATION_WORKER_LEASE_SECONDS', '30')),
+                     float(os.getenv('CALCULATION_WORKER_HEARTBEAT_SECONDS', '5')))
+        if not projects or not 1 <= result.max_concurrency <= 32:
+            raise ValueError('Configure explicit project IDs and concurrency between 1 and 32')
+        if not 0 < result.heartbeat_seconds < result.lease_seconds / 3 or result.poll_seconds <= 0:
+            raise ValueError('Heartbeat must be positive and below one third of the lease')
+        return result
+
+
+@asynccontextmanager
+async def queue_repository(project_id):
+    """Resolve current routing from metadata, never from a job payload."""
+    from app.infra.db.calculation_worker import queue_repository as open_repository
+
+    async with open_repository(project_id) as repository:
+        yield repository
+
+
+@asynccontextmanager
+async def execution_scope(row):
+    from fastapi import HTTPException
+    from app.mcp.scope import QueryMetadataRepository, QueryScope
+    from app.auth.project_dependencies import (resolve_project_context,
+        resolve_project_business_routing, get_project_pg_connection)
+    from app.infra.db.metadb.database import SessionLocal
+    from app.infra.db.project_routing import activate_project_routing
+    async with SessionLocal() as session:
+        metadata = QueryMetadataRepository(session)
+        user = await metadata.get_user_by_id(UUID(str(row['owner_user_id'])))
+        if not user or not user.is_active:
+            raise HTTPException(403, 'Calculation owner is inactive')
+        project = await resolve_project_context(str(row['project_id']), user, metadata)
+        routing = await resolve_project_business_routing(project, metadata)
+        await session.commit()
+        async with asynccontextmanager(get_project_pg_connection)(project, metadata) as conn:
+            await session.commit()
+            with activate_project_routing(routing):
+                yield QueryScope(project, user, conn, routing, metadata)
+
+
+async def execute_child(input_path: str, output_path: str):
+    from app.mcp.calculation_jobs import execute_calculation_job
+    row = json.loads(Path(input_path).read_text())
+    try:
+        async with execution_scope(row) as scope:
+            result = await execute_calculation_job(scope, row)
+        payload = {'status': 'succeeded', 'result': result.model_dump(mode='json')}
+    except Exception as exc:
+        # Do not persist exception text: DSNs, SQL, credentials or model data can occur there.
+        category = 'CALCULATION_FAILED'
+        if getattr(exc, 'status_code', None) in (401, 403):
+            category = 'AUTHORIZATION_DENIED'
+        elif getattr(exc, 'status_code', None) == 409:
+            category = 'EXECUTION_CONFLICT'
+        elif getattr(exc, 'status_code', None) in (404, 410):
+            category = 'EXECUTION_CONTEXT_UNAVAILABLE'
+        payload = {'status': 'failed', 'error': {'code': category}}
+    encoded = json.dumps(payload, allow_nan=False)
+    if len(encoded.encode()) > 8 * 1024 * 1024:
+        encoded = json.dumps({'status': 'failed', 'error': {'code': 'RESULT_TOO_LARGE'}})
+    Path(output_path).write_text(encoded)
+
+
+async def stop_process_group(process):
+    """Always kill the group, even if its original leader already exited."""
+    try:
+        os.killpg(process.pid, signal.SIGTERM)
+    except ProcessLookupError:
+        pass
+    try:
+        await asyncio.wait_for(process.wait(), 2)
+    except TimeoutError:
+        pass
+    # Native pools can ignore SIGTERM or survive the Python leader.
+    try:
+        os.killpg(process.pid, signal.SIGKILL)
+    except ProcessLookupError:
+        pass
+    await process.wait()
+    if process.stdin is not None:
+        process.stdin.close()
+
+
+@contextmanager
+def job_directory(row):
+    root = Path(os.getenv('CALCULATION_WORKER_STATE_DIR', 'temp/calculation-jobs'))
+    root.mkdir(parents=True, exist_ok=True)
+    directory = Path(tempfile.mkdtemp(prefix=f"{row.get('job_id', 'job')}-", dir=root))
+    directory.chmod(0o700)
+    (directory / 'scope.json').write_text(json.dumps({'project_id': str(row['project_id']), 'job_id': str(row.get('job_id', 'job')),
+        'lease_generation': row.get('lease_generation')}))
+    try:
+        yield str(directory.resolve())
+    finally:
+        # Failed external resource cleanup must remain operator-recoverable.
+        if not (directory / 'cleanup-failed').exists():
+            shutil.rmtree(directory)
+
+
+async def cleanup_job_resources(row, directory):
+    from app.services.calculation_cleanup import cleanup_calculation_databases
+    try:
+        await cleanup_calculation_databases(row['project_id'], directory)
+    except BaseException:
+        (Path(directory) / 'cleanup-failed').touch()
+        raise
+
+
+async def recover_job_resources(project_ids):
+    """Reconcile only this worker's explicit scopes after obtaining the singleton."""
+    root = Path(os.getenv('CALCULATION_WORKER_STATE_DIR', 'temp/calculation-jobs'))
+    if not root.exists():
+        return
+    for directory in root.iterdir():
+        if directory.is_symlink() or not directory.is_dir():
+            continue
+        scope_file = directory / 'scope.json'
+        if not scope_file.exists():
+            continue
+        row = json.loads(scope_file.read_text())
+        if str(row.get('project_id')) not in project_ids:
+            continue
+        job_id = str(UUID(row['job_id']))
+        if not directory.name.startswith(job_id + '-'):
+            raise RuntimeError('Calculation state directory identity is invalid')
+        async with queue_repository(row['project_id']) as repository:
+            stored = await repository.get(job_id)
+        if not stored or stored['lease_generation'] != row['lease_generation']:
+            raise RuntimeError('Calculation state ownership cannot be verified')
+        expiry = stored.get('lease_expires_at')
+        if expiry and expiry > datetime.now(timezone.utc) and stored['status'] in ('running', 'cancel_requested'):
+            raise RuntimeError('Previous calculation lease has not expired')
+        await cleanup_job_resources(row, str(directory.resolve()))
+        async with queue_repository(row['project_id']) as repository:
+            await repository.recover()
+        shutil.rmtree(directory)
+
+
+class CalculationWorker:
+    def __init__(self, settings, *, repository=queue_repository):
+        self.settings = settings
+        self.repository = repository
+        self.owner = str(uuid4())
+
+    async def run_job(self, row):
+        from fastapi.encoders import jsonable_encoder
+        with job_directory(row) as directory:
+            input_path = Path(directory) / 'input.json'
+            output_path = Path(directory) / 'result.json'
+            input_path.write_text(json.dumps(jsonable_encoder(row)))
+            process = await asyncio.create_subprocess_exec(
+                sys.executable, '-m', 'app.services.calculation_worker',
+                '--guard', str(input_path), str(output_path), start_new_session=True,
+                stdin=asyncio.subprocess.PIPE, stdout=asyncio.subprocess.DEVNULL,
+                stderr=asyncio.subprocess.DEVNULL,
+                env={**os.environ, 'TMPDIR': directory, 'TJWATER_CALCULATION_JOB_DIR': directory},
+            )
+            started = time.monotonic()
+            cleaned = False
+            try:
+                while True:
+                    async with asyncio.timeout(self.settings.heartbeat_seconds):
+                        async with self.repository(row['project_id']) as repository:
+                            status = await repository.heartbeat(row, self.owner, self.settings.lease_seconds)
+                    if status is None:
+                        return  # Finally stops all owned computation on a lost lease.
+                    terminal = None
+                    if status == 'cancel_requested':
+                        terminal = {'status': 'cancelled'}
+                    elif time.monotonic() - started >= row['timeout_seconds']:
+                        terminal = {'status': 'timed_out', 'error': {'code': 'EXECUTION_TIMEOUT'}}
+                    elif process.returncode is not None:
+                        try:
+                            if output_path.stat().st_size > 8 * 1024 * 1024:
+                                raise ValueError('Result exceeds persistence limit')
+                            terminal = json.loads(output_path.read_text())
+                        except (OSError, ValueError):
+                            terminal = {'status': 'failed', 'error': {'code': 'CHILD_PROCESS_FAILED'}}
+                    if terminal:
+                        await stop_process_group(process)
+                        try:
+                            await cleanup_job_resources(row, directory)
+                            cleaned = True
+                        except Exception:
+                            terminal = {'status': 'failed', 'error': {'code': 'CLEANUP_FAILED'}}
+                            cleaned = True
+                        async with self.repository(row['project_id']) as repository:
+                            finished = await repository.finish(row, self.owner, **terminal)
+                            if not finished:
+                                status = await repository.heartbeat(row, self.owner, self.settings.lease_seconds)
+                                if status == 'cancel_requested':
+                                    await repository.finish(row, self.owner, 'cancelled')
+                        return
+                    await asyncio.sleep(self.settings.heartbeat_seconds)
+            finally:
+                await stop_process_group(process)
+                if not cleaned:
+                    await cleanup_job_resources(row, directory)
+
+    async def run(self):
+        from app.infra.db.calculation_worker import calculation_supervisor_lease
+
+        # Dedicated session-level metadata lock: independent API workers never
+        # participate; duplicate supervisors exit before claiming any jobs.
+        async with calculation_supervisor_lease() as lease:
+            tasks = set()
+            project_cursor = 0
+            try:
+                # Allow any old supervisor to detect a broken lock connection and
+                # stop its children before this replacement claims capacity.
+                await asyncio.sleep(self.settings.lease_seconds)
+                await recover_job_resources(self.settings.project_ids)
+                while True:
+                    async with asyncio.timeout(self.settings.heartbeat_seconds):
+                        await lease.ping()
+                    for task in tuple(tasks):
+                        if task.done():
+                            tasks.remove(task)
+                            task.result()  # Fail closed if heartbeat/persistence failed.
+                    projects = self.settings.project_ids
+                    ordered = projects[project_cursor:] + projects[:project_cursor]
+                    project_cursor = (project_cursor + 1) % len(projects)
+                    for project_id in ordered:
+                        if len(tasks) >= self.settings.max_concurrency:
+                            break
+                        async with self.repository(project_id) as repository:
+                            await repository.recover()
+                            row = await repository.claim(self.owner, self.settings.lease_seconds)
+                        if row:
+                            tasks.add(asyncio.create_task(self.run_job(row)))
+                    await asyncio.sleep(min(self.settings.poll_seconds, self.settings.heartbeat_seconds))
+            finally:
+                for task in tasks:
+                    task.cancel()
+                await asyncio.gather(*tasks, return_exceptions=True)
+
+
+async def guard_child(input_path, output_path):
+    """A parent-death pipe kills the whole group even when the supervisor is SIGKILLed."""
+    reader = asyncio.StreamReader()
+    protocol = asyncio.StreamReaderProtocol(reader)
+    transport, _ = await asyncio.get_running_loop().connect_read_pipe(lambda: protocol, sys.stdin)
+    process = await asyncio.create_subprocess_exec(
+        sys.executable, '-m', 'app.services.calculation_worker', input_path, output_path,
+        stdin=asyncio.subprocess.DEVNULL,
+    )
+    completed = asyncio.create_task(process.wait())
+    parent_gone = asyncio.create_task(reader.read(1))
+    try:
+        done, _ = await asyncio.wait((completed, parent_gone), return_when=asyncio.FIRST_COMPLETED)
+        if parent_gone in done:
+            os.killpg(os.getpgrp(), signal.SIGKILL)
+        await completed
+    finally:
+        parent_gone.cancel()
+        transport.close()
+
+
+if __name__ == '__main__':
+    if sys.argv[1] == '--guard':
+        asyncio.run(guard_child(sys.argv[2], sys.argv[3]))
+    else:
+        asyncio.run(execute_child(sys.argv[1], sys.argv[2]))

app/services/node_coverage.py

@@ -0,0 +1,128 @@
+"""Reusable geometric coverage of verified project nodes, independent of business analysis."""
+
+import json
+from math import isfinite
+from typing import Any
+
+from pyproj import CRS, Transformer
+from shapely import MultiPoint, concave_hull, get_num_coordinates
+from shapely.geometry import mapping
+from shapely.ops import transform, unary_union
+
+from app.native.wndb.gis.network_views import get_node_coverage_links, get_node_geographic_positions
+
+CONCAVE_RATIO = 0.25
+BUFFER_METERS = 25.0
+SIMPLIFY_METERS = 2.0
+MAX_COVERAGE_NODES = 10000
+
+
+def _components(node_ids: list[str], links: list[dict[str, Any]], closed: set[str]) -> list[list[str]]:
+    parent = {node_id: node_id for node_id in node_ids}
+
+    def root(node: str) -> str:
+        while parent[node] != node:
+            parent[node] = parent[parent[node]]
+            node = parent[node]
+        return node
+
+    for link in links:
+        start, end = link["start_node_id"], link["end_node_id"]
+        if link["id"] not in closed and link.get("status") != "CLOSED" and start in parent and end in parent:
+            parent[root(start)] = root(end)
+    groups: dict[str, list[str]] = {}
+    for node_id in node_ids:
+        groups.setdefault(root(node_id), []).append(node_id)
+    return list(groups.values())
+
+
+def build_node_coverage(
+    groups: list[dict[str, Any]], positions: list[dict[str, Any]],
+    links: list[dict[str, Any]], *, split_by_connectivity: bool = False,
+    excluded_link_ids: list[str] | None = None,
+    concave_ratio: float = CONCAVE_RATIO, buffer_m: float = BUFFER_METERS,
+) -> dict[str, Any]:
+    """Keep caller-defined comparison groups and derive geometry without assigning business meaning."""
+    coordinates = {}
+    for row in positions:
+        longitude, latitude = row.get("longitude"), row.get("latitude")
+        if (isinstance(longitude, (int, float)) and not isinstance(longitude, bool)
+                and isinstance(latitude, (int, float)) and not isinstance(latitude, bool)
+                and isfinite(longitude) and isfinite(latitude)
+                and -180 <= longitude <= 180 and -90 < latitude < 90):
+            coordinates[str(row["id"])] = (longitude, latitude)
+
+    features, groups_metadata = [], {}
+    closed = set(excluded_link_ids or [])
+    simplify_m = min(SIMPLIFY_METERS, buffer_m / 4)
+    for group in groups:
+        group_id, label = group["group_id"], group["title"]
+        ids = sorted(set(group["node_ids"]))
+        mapped = [node for node in ids if node in coordinates]
+        missing = [node for node in ids if node not in coordinates]
+        metadata = {"node_ids": ids, "mapped_node_count": len(mapped), "missing_node_ids": missing}
+        groups_metadata[group_id] = {"title": label, **metadata}
+        metadata = groups_metadata[group_id]
+        if not mapped:
+            metadata["component_count"] = 0
+            continue
+        points = [coordinates[node] for node in mapped]
+        longitude = sum(point[0] for point in points) / len(points)
+        latitude = sum(point[1] for point in points) / len(points)
+        local = CRS.from_proj4(
+            f"+proj=aeqd +lat_0={latitude} +lon_0={longitude} +datum=WGS84 +units=m"
+        )
+        project = Transformer.from_crs(4326, local, always_xy=True).transform
+        unproject = Transformer.from_crs(local, 4326, always_xy=True).transform
+        envelopes = []
+        components = _components(ids, links, closed) if split_by_connectivity else [ids]
+        for nodes in components:
+            component_points = [coordinates[node] for node in nodes if node in coordinates]
+            if not component_points:
+                continue
+            cloud = transform(project, MultiPoint(component_points))
+            # A point or collinear cloud remains drawable through the same metric buffer.
+            envelope = concave_hull(cloud, ratio=concave_ratio).buffer(buffer_m)
+            envelopes.append(envelope)
+        metadata["component_count"] = len(envelopes)
+        geometry = unary_union(envelopes).simplify(simplify_m, preserve_topology=True)
+        if geometry.is_empty or not geometry.is_valid or get_num_coordinates(geometry) > 20000:
+            raise ValueError("Invalid derived coverage")
+        feature = {
+            "type": "Feature", "id": group_id,
+            "properties": {
+                "coverage_group": group_id, "label": label, "derived": True, "display_only": True,
+                "node_count": len(ids), "mapped_node_count": len(mapped),
+                "missing_node_count": len(missing), "component_count": len(envelopes),
+            },
+            "geometry": mapping(transform(unproject, geometry)),
+        }
+        features.append(feature)
+
+    return {
+        "status": "partial" if any(group["missing_node_ids"] for group in groups_metadata.values())
+        else "ready" if features else "empty",
+        "method": "concave_hull_buffer",
+        "parameters": {"concave_ratio": concave_ratio, "buffer_m": buffer_m,
+                       "split_by_connectivity": split_by_connectivity,
+                       "excluded_link_ids": sorted(closed),
+                       "simplify_m": simplify_m, "projection": "local_azimuthal_equidistant"},
+        "output_crs": "EPSG:4326", "display_only": True,
+        "position_source": "gis.node_geometries",
+        "boundary": "节点空间覆盖示意；不推断故障、污染、停水、水力影响或用户服务边界。",
+        "groups": groups_metadata,
+        "map_payload": {"type": "FeatureCollection", "features": features},
+    }
+
+
+def generate_node_coverage(network: str, groups: list[dict[str, Any]], **parameters: Any) -> dict[str, Any]:
+    node_ids = sorted({node for group in groups for node in group["node_ids"]})
+    if len(node_ids) > MAX_COVERAGE_NODES:
+        raise ValueError("Node coverage supports at most 10000 distinct nodes")
+    positions = get_node_geographic_positions(network, node_ids)
+    links = get_node_coverage_links(network, node_ids) if parameters.get("split_by_connectivity") else []
+    coverage = build_node_coverage(groups, positions, links, **parameters)
+    payload = coverage.pop("map_payload")
+    if len(json.dumps(payload, ensure_ascii=False).encode("utf-8")) > 400 * 1024:
+        raise ValueError("Coverage exceeds presentation budget; narrow the node selection")
+    return {"coverage": coverage, "map_payload": payload}

app/services/operational_presentation.py

@@ -51,12 +51,19 @@ def _geometry(value):
 def resolve_overlay(snapshot: TaskDataSnapshot, mode: str) -> dict:
     """Use recorded GeoJSON or recorded coordinates; never geocode or infer a boundary."""
     data = snapshot.data
+    # MCP persists its complete QueryResult envelope, unlike direct data queries.
+    if (isinstance(data, dict) and data.get("source") == snapshot.source
+            and isinstance(data.get("data"), (dict, list))):
+        data = data["data"]
     candidates = [data]
     if isinstance(data, dict):
         candidates.extend([data.get("map_payload"), data.get("drawing_payload")])
         latest = data.get("latest_analysis")
         if isinstance(latest, dict):
             candidates.append(latest.get("map_payload"))
+        result = data.get("result")
+        if isinstance(result, dict):
+            candidates.append(result.get("map_payload"))
     collection = next((item for item in candidates
                        if isinstance(item, dict) and item.get("type") == "FeatureCollection"), None)
     if collection is not None:
@@ -88,7 +95,7 @@ def resolve_overlay(snapshot: TaskDataSnapshot, mode: str) -> dict:
         properties = feature.get("properties") or {}
         require(isinstance(properties, dict), "OVERLAY_GEOMETRY_INVALID", "Feature properties are invalid", status=422)
         normalized = {"type": "Feature", "geometry": geometry, "properties": dict(properties)}
-        if isinstance(feature.get("id"), (str, int)) and not isinstance(feature.get("id"), bool):
+        if isinstance(feature.get("id"), str) or _number(feature.get("id")):
             normalized["id"] = feature["id"]
         if mode == "heatmap":
             weight = properties.get("weight", 1)

app/services/operational_tasks.py

@@ -19,6 +19,7 @@ from app.auth.permissions import (
     SIMULATION_VIEW,
 )
 from app.domain.domain_errors import DomainError, require
+from app.services import task_questions
 from app.domain.schemas.operational_task import (
     ActionDecisionCommand,
     ActionExecuteCommand,
@@ -65,6 +66,8 @@ from app.domain.schemas.operational_task import (
     TaskPresentationEventCommand,
     TaskPresentationEventResult,
     TaskPresentationCursor,
+    TaskPresentationHistoryItem,
+    TaskPresentationHistoryPage,
 )
 from app.domain.schemas.operational_data import TaskDataSnapshot, TaskDataSource
 from app.services.content_digest import digest
@@ -97,6 +100,15 @@ class OperationalTaskService:
     def _actor(self) -> str:
         return f"principal:{self.context.user_id}"
 
+    @staticmethod
+    def _require_run_owner(run: dict, command: AgentRunCommand) -> None:
+        require(run.get("agent_session_id") == command.source_session_id,
+                "SESSION_MISMATCH", "Agent session does not own this run", status=409)
+        owner = run.get("agent_instance_id")
+        if owner is not None:
+            require(owner == command.agent_instance_id,
+                    "AGENT_INSTANCE_MISMATCH", "Agent instance does not own this run", status=409)
+
     @property
     def _supports_runs(self) -> bool:
         return hasattr(self.repo, "get_active_run") and hasattr(self.repo, "update_run")
@@ -138,6 +150,7 @@ class OperationalTaskService:
         permissions = self.permissions
         return OperationalTaskPermissions(
             can_publish_artifacts=OPERATIONAL_TASK_EXECUTE in permissions,
+            can_answer_questions=OPERATIONAL_TASK_WRITE in permissions,
             can_propose_actions=OPERATIONAL_TASK_WRITE in permissions,
             can_sync_external_feedback=OPERATIONAL_TASK_EXTERNAL_SYNC in permissions,
             can_decide_work_orders=OPERATIONAL_ACTION_WORK_ORDER_APPROVE in permissions,
@@ -258,6 +271,7 @@ class OperationalTaskService:
             evidence=evidence,
             actions=actions,
             access_requests=access_requests,
+            questions=await task_questions.list_questions(self, task_id),
             permissions=self._permission_view(),
             work_order_snapshot=work_order_snapshot,
         )
@@ -483,6 +497,7 @@ class OperationalTaskService:
                 status="running",
                 revision=expected + 1,
                 agent_session_id=command.source_session_id,
+                agent_instance_id=command.agent_instance_id,
                 lease_expires_at=now + LEASE_DURATION,
                 heartbeat_at=now,
                 started_at=run.get("started_at") or now,
@@ -527,8 +542,7 @@ class OperationalTaskService:
                     "Run revision changed", status=409)
             require(run["status"] in {"claimed", "running"}, "INVALID_TRANSITION",
                     "Only an active run has a lease", status=409)
-            require(run.get("agent_session_id") == command.source_session_id,
-                    "SESSION_MISMATCH", "Agent session does not own this run", status=409)
+            self._require_run_owner(run, command)
             now = self._now()
             self._require_active_run_lease(run, now)
             expected = run["revision"]
@@ -580,6 +594,7 @@ class OperationalTaskService:
             replay = await self._request_replay(scope, command.request_id, body)
             if replay:
                 return await self.detail(task_id)
+            await self.repo.lock(f"task:{task_id}")
             await self.repo.lock(f"run:{run_id}")
             run = await self._run(task_id, run_id)
             require(run["revision"] == command.expected_revision, "VERSION_CONFLICT",
@@ -587,6 +602,7 @@ class OperationalTaskService:
             require(run["status"] in {"queued", "claimed", "running"},
                     "INVALID_TRANSITION", "Run is already terminal", status=409)
             now = self._now()
+            await task_questions.cancel_pending(self, task_id, now)
             expected = run["revision"]
             run.update(
                 status="cancelled", revision=expected + 1, finished_at=now,
@@ -624,6 +640,7 @@ class OperationalTaskService:
             require(await self.repo.get_active_run(str(task_id)) is None,
                     "ACTIVE_RUN_EXISTS", "Task already has an active run", status=409)
             now = self._now()
+            await task_questions.cancel_pending(self, task_id, now)
             generation = task["execution_generation"] + 1
             run = self._new_run(task_id, generation, now)
             await self.repo.insert_run(_json(run))
@@ -778,6 +795,10 @@ class OperationalTaskService:
                     "Only a running run accepts Agent artifacts", status=409)
             now = self._now()
             self._require_active_run_lease(active_run, now)
+            if command.report and (
+                command.report.stage == "final" or command.report.outcome == "insufficient_data"
+            ):
+                await self._require_no_pending_calculations(task_id, active_run)
             previous_report = await self.repo.get_latest_report(str(task_id))
             require(not previous_report or previous_report.get("execution_generation") != task["execution_generation"]
                     or previous_report["report"].get("outcome") != "insufficient_data",
@@ -982,6 +1003,27 @@ class OperationalTaskService:
                 event_cursor=cursor,
             )
 
+    async def presentation_history(
+        self, task_id: UUID, *, after: int = 0, through: int | None = None, limit: int = 100,
+    ) -> TaskPresentationHistoryPage:
+        require(0 <= after and (through is None or through >= 0) and 1 <= limit <= 200,
+                "INVALID_ARGUMENT", "Invalid presentation history page", status=422)
+        async with self.repo.transaction():
+            await self._task(task_id)
+            cursor = await self.repo.current_event_cursor()
+            boundary = min(through, cursor) if through is not None else cursor
+            rows = await self.repo.list_presentation_events(str(task_id), after, boundary, limit + 1)
+            items = [TaskPresentationHistoryItem(
+                event_seq=row["event_seq"], occurred_at=row["occurred_at"],
+                run_id=row["metadata"]["run_id"],
+                execution_generation=row["metadata"]["execution_generation"],
+                command=row["metadata"]["command"],
+            ) for row in rows[:limit]]
+            return TaskPresentationHistoryPage(
+                task_id=task_id, event_cursor=boundary, items=items,
+                next_after=items[-1].event_seq if len(rows) > limit else None,
+            )
+
     async def publish_presentation_event(
         self, task_id: UUID, command: TaskPresentationEventCommand
     ) -> TaskPresentationEventResult:
@@ -1015,7 +1057,16 @@ class OperationalTaskService:
                 evidence = StoredEvidenceRecord.model_validate(stored)
                 require(evidence.data_snapshot is not None, "SNAPSHOT_NOT_FOUND",
                         "Overlay requires a Server-ingested immutable data snapshot", status=422)
-                payload["data"] = resolve_overlay(evidence.data_snapshot, command.command.mode)
+                resolved = resolve_overlay(evidence.data_snapshot, command.command.mode)
+                if command.command.data is not None:
+                    supplied = resolve_overlay(
+                        evidence.data_snapshot.model_copy(update={
+                            "data": command.command.data.model_dump(mode="json", exclude_unset=True),
+                        }), command.command.mode,
+                    )
+                    require(supplied == resolved, "OVERLAY_DATA_MISMATCH",
+                            "Overlay data must match the immutable evidence snapshot", status=422)
+                payload["data"] = resolved
             metadata = {"task_id": str(task_id), "run_id": str(run["run_id"]),
                         "execution_generation": run["generation"], "command": payload}
             try:
@@ -1866,6 +1917,15 @@ class OperationalTaskService:
                 task_ids.append(task_id)
         return DispatchDueSchedulesResult(task_ids=task_ids, dispatched_at=now)
 
+    async def _require_no_pending_calculations(self, task_id: UUID, run: dict) -> None:
+        if hasattr(self.repo, "pending_calculation_jobs"):
+            pending = await self.repo.pending_calculation_jobs(
+                str(task_id), str(run["run_id"]), str(run["execution_request_id"]),
+            )
+            require(not pending, "PENDING_CALCULATIONS",
+                    "Wait for submitted calculations before publishing a final report or completing the run",
+                    refs=pending, status=409, recovery="retry_same_request")
+
     async def _finish_run(
         self, task_id: UUID, run_id: UUID, command: AgentRunCommand,
         *, outcome: str, failure_code: str | None = None,
@@ -1877,19 +1937,23 @@ class OperationalTaskService:
             replay = await self._request_replay(scope, command.request_id, body)
             if replay:
                 return await self.detail(task_id)
+            await self.repo.lock(f"task:{task_id}")
             await self.repo.lock(f"run:{run_id}")
             run = await self._run(task_id, run_id)
             require(run["revision"] == command.expected_revision, "VERSION_CONFLICT",
                     "Run revision changed", status=409)
             require(run["status"] in {"claimed", "running"}, "INVALID_TRANSITION",
                     "Only an active run can finish", status=409)
-            require(run.get("agent_session_id") == command.source_session_id,
-                    "SESSION_MISMATCH", "Agent session does not own this run", status=409)
+            self._require_run_owner(run, command)
             now = self._now()
             self._require_active_run_lease(run, now)
             task = await self._task(task_id)
             final_outcome = outcome
+            pending_question = any(q["status"] == "pending" and str(q["run_id"]) == str(run_id)
+                for q in await task_questions.list_questions(self, task_id))
+            actions = []
             if outcome == "succeeded":
+                await self._require_no_pending_calculations(task_id, run)
                 if hasattr(self.repo, "pending_run_messages"):
                     pending_messages = await self.repo.pending_run_messages(str(run_id))
                     require(not pending_messages, "PENDING_AGENT_MESSAGES",
@@ -1915,7 +1979,9 @@ class OperationalTaskService:
                         and plan.source_session_id == command.source_session_id
                     ))
                 )
-                if not same_generation or report.report.stage != "final":
+                if pending_question:
+                    lifecycle = "waiting_human"
+                elif not same_generation or report.report.stage != "final":
                     final_outcome = "failed"
                     lifecycle = "failed"
                     failure_code = "ARTIFACT_INCOMPLETE"
@@ -1960,6 +2026,7 @@ class OperationalTaskService:
             run_model = OperationalTaskRun.model_validate(run)
             await self.repo.update_run(_json(run_model), expected)
             if final_outcome != "succeeded":
+                await task_questions.cancel_pending(self, task_id, now)
                 await self._fail_run_messages(run_id, failure_code or "Run failed", now)
             task.update(
                 lifecycle_status=lifecycle,
@@ -1980,7 +2047,7 @@ class OperationalTaskService:
                     failure_detail or "工况任务执行失败",
                     now,
                 )
-            elif lifecycle == "waiting_human":
+            elif lifecycle == "waiting_human" and not pending_question:
                 pending_actions = any(
                     action.approval_status == "pending" for action in actions
                 )
@@ -2193,6 +2260,7 @@ class OperationalTaskService:
         task["agent_lease_expires_at"] = None
 
     async def _requeue_task(self, task: dict, now: datetime) -> None:
+        await task_questions.cancel_pending(self, UUID(str(task["task_id"])), now)
         if self._supports_runs:
             active_run = await self.repo.get_active_run(str(task["task_id"]))
             while active_run is not None:

app/services/project_inp.py

@@ -2,6 +2,7 @@ from collections.abc import Iterator
 from contextlib import contextmanager
 from pathlib import Path
 from tempfile import NamedTemporaryFile
+import os
 
 from app.native.wndb.inp.exporter import dump_inp
 
@@ -17,9 +18,11 @@ def temporary_project_inp(
     version: str = "2",
 ) -> Iterator[Path]:
     """Export one project model to an isolated INP and remove it afterwards."""
-    PROJECT_INP_DIRECTORY.mkdir(parents=True, exist_ok=True)
+    job_directory = os.environ.get("TJWATER_CALCULATION_JOB_DIR")
+    directory = Path(job_directory) / "inp" if job_directory else PROJECT_INP_DIRECTORY
+    directory.mkdir(parents=True, exist_ok=True)
     with NamedTemporaryFile(
-        dir=PROJECT_INP_DIRECTORY,
+        dir=directory,
         prefix=f"{purpose}_",
         suffix=".inp",
         delete=False,

app/services/task_questions.py

@@ -0,0 +1,120 @@
+"""Question lifecycle sharing the operational task transaction and notification bus."""
+from datetime import datetime
+from typing import TYPE_CHECKING
+from uuid import UUID, uuid4
+
+from app.auth.permissions import OPERATIONAL_TASK_EXECUTE, OPERATIONAL_TASK_WRITE
+from app.domain.domain_errors import require
+from app.domain.schemas.operational_task import OperationalTaskDetail, OperationalTaskSummary
+from app.domain.schemas.task_question import TaskQuestionCreate, TaskQuestionReply, TaskQuestionRequest
+
+if TYPE_CHECKING:
+    from app.services.operational_tasks import OperationalTaskService
+
+
+async def list_questions(service: "OperationalTaskService", task_id: UUID) -> list[dict]:
+    # Legacy test repositories do not implement question persistence.
+    reader = getattr(service.repo, "list_questions", None)
+    return await reader(str(task_id)) if reader else []
+
+
+async def cancel_pending(service: "OperationalTaskService", task_id: UUID, now: datetime) -> None:
+    for raw in await list_questions(service, task_id):
+        question = TaskQuestionRequest.model_validate(raw)
+        if question.status != "pending":
+            continue
+        question = question.model_copy(update={"status": "cancelled", "version": question.version + 1})
+        await service.repo.save_question(str(task_id), question.model_dump(mode="json"))
+        await service._resolve_notifications(task_id, kinds=["human_input_required"],
+            subject_type="question", subject_id=question.question_request_id,
+            now=now, reason="question_superseded")
+
+
+async def create_question(service: "OperationalTaskService", task_id: UUID,
+                          command: TaskQuestionCreate) -> TaskQuestionRequest:
+    require(OPERATIONAL_TASK_EXECUTE in service.permissions, "PERMISSION_DENIED",
+            "Current user cannot publish task questions", status=403)
+    body = command.model_dump(mode="json")
+    scope = f"task:{task_id}:question:create"
+    async with service.repo.transaction():
+        await service.repo.lock(f"task:{task_id}")
+        replay = await service._request_replay(scope, command.request_id, body)
+        if replay:
+            return TaskQuestionRequest.model_validate(await service.repo.get_question(
+                str(task_id), replay["question_request_id"]))
+        task = await service._task(task_id)
+        service._require_actionable_task(task)
+        require(task["task_type"] == "event_investigation", "INVALID_TASK_TYPE",
+                "Questions require an investigation task", status=409)
+        await service.repo.lock(f"run:{command.run_id}")
+        run = await service._run(task_id, command.run_id)
+        require(run["status"] in {"claimed", "running"}
+                and run["generation"] == task["execution_generation"]
+                and run.get("agent_session_id") == command.source_session_id,
+                "SESSION_MISMATCH", "Question must belong to the active investigation run", status=409)
+        now = service._now()
+        service._require_active_run_lease(run, now)
+        await service._require_no_pending_calculations(task_id, run)
+        require(not any(q["status"] == "pending" for q in await list_questions(service, task_id)),
+                "QUESTION_PENDING", "Answer the existing questions first", status=409)
+        question = TaskQuestionRequest(question_request_id=uuid4(), task_id=task_id,
+            run_id=command.run_id, execution_generation=run["generation"],
+            source_session_id=command.source_session_id, questions=command.questions,
+            status="pending", version=1, created_at=now)
+        await service.repo.save_question(str(task_id), question.model_dump(mode="json"))
+        task.update(revision=task["revision"] + 1, updated_at=now)
+        await service.repo.update_task(OperationalTaskSummary.model_validate(task).model_dump(mode="json", exclude={"active_run", "latest_run"}))
+        await service._task_event(task, now)
+        await service._notify(task_id, "human_input_required", task["title"],
+            "；".join(q.title for q in question.questions), now,
+            required_permission=OPERATIONAL_TASK_WRITE, subject_type="question",
+            subject_id=question.question_request_id)
+        await service._save_request(scope, command.request_id, body,
+            {"question_request_id": str(question.question_request_id)})
+        return question
+
+
+async def answer_question(service: "OperationalTaskService", task_id: UUID, question_id: UUID,
+                          command: TaskQuestionReply) -> OperationalTaskDetail:
+    require(OPERATIONAL_TASK_WRITE in service.permissions, "PERMISSION_DENIED",
+            "Current user cannot answer task questions", status=403)
+    body = command.model_dump(mode="json")
+    scope = f"task:{task_id}:question:{question_id}:answer"
+    async with service.repo.transaction():
+        await service.repo.lock(f"task:{task_id}")
+        if await service._request_replay(scope, command.request_id, body):
+            return await service.detail(task_id)
+        task = await service._task(task_id)
+        service._require_actionable_task(task)
+        raw = await service.repo.get_question(str(task_id), str(question_id))
+        require(raw is not None, "QUESTION_NOT_FOUND", "Question not found", status=404)
+        question = TaskQuestionRequest.model_validate(raw)
+        require(question.status == "pending" and question.version == command.expected_version
+                and question.execution_generation == task["execution_generation"],
+                "VERSION_CONFLICT", "Question is no longer pending in the current generation", status=409)
+        require(task["lifecycle_status"] == "waiting_human"
+                and await service.repo.get_active_run(str(task_id)) is None,
+                "RUN_NOT_SETTLED", "Wait until the investigation has entered waiting_human", status=409)
+        answers = {a.question_id: a for a in command.answers}
+        known = {q.question_id for q in question.questions}
+        require(len(answers) == len(command.answers) and set(answers) <= known,
+                "ANSWER_INVALID", "Unknown or duplicate questions", status=422)
+        for q in question.questions:
+            answer = answers.get(q.question_id)
+            selected = answer.selected_option_ids if answer else []
+            text = answer.custom_text.strip() if answer else ""
+            require(len(set(selected)) == len(selected) and set(selected) <= {o.id for o in q.options}
+                    and (q.mode == "multiple" or len(selected) <= 1)
+                    and (q.allow_custom or not text) and (not q.required or bool(selected or text)),
+                    "ANSWER_INVALID", "Answer does not satisfy the question", status=422)
+        now = service._now()
+        question = question.model_copy(update={"status": "answered", "version": question.version + 1,
+            "answers": command.answers, "answered_at": now, "answered_by": service._actor()})
+        await service.repo.save_question(str(task_id), question.model_dump(mode="json"))
+        await service._resolve_notifications(task_id, kinds=["human_input_required"],
+            subject_type="question", subject_id=question_id, now=now, reason="question_answered")
+        await service._requeue_task(task, now)
+        await service.repo.update_task(OperationalTaskSummary.model_validate(task).model_dump(mode="json", exclude={"active_run", "latest_run"}))
+        await service._task_event(task, now)
+        await service._save_request(scope, command.request_id, body, {"task_id": str(task_id)})
+        return await service.detail(task_id)

app/services/valve_isolation.py

@@ -1,13 +1,22 @@
 from typing import Any
 
-from app.algorithms.valve_isolation.topology_search import valve_isolation_analysis
-from app.native.wndb.gis.network_views import get_network_link_nodes
+from app.algorithms.valve_isolation.topology_search import (
+    AccidentElementType,
+    valve_isolation_analysis,
+)
+from app.native.wndb.gis.network_views import get_valve_isolation_topology
 
 
 def analyze_valve_isolation(
     network: str,
     accident_element: str | list[str],
-    disabled_valves: list[str] = None,
+    disabled_valves: list[str] | None = None,
+    accident_element_type: AccidentElementType = "auto",
 ) -> dict[str, Any]:
-    link_entries = get_network_link_nodes(network)
-    return valve_isolation_analysis(link_entries, accident_element, disabled_valves)
+    link_entries, source_nodes, closed_links = get_valve_isolation_topology(network)
+    return valve_isolation_analysis(
+        link_entries, accident_element, disabled_valves,
+        accident_element_type=accident_element_type,
+        water_source_nodes=source_nodes,
+        closed_links=closed_links,
+    )

contracts/answer-ui-catalog.json

@@ -17,6 +17,7 @@
                 "Stack",
                 "Section",
                 "Text",
+                "Result",
                 "Metric",
                 "Table",
                 "Chart",
@@ -66,6 +67,7 @@
     "Stack",
     "Section",
     "Text",
+    "Result",
     "Metric",
     "Table",
     "Chart",
@@ -90,6 +92,21 @@
             "compact",
             "normal"
           ]
+        },
+        "intent": {
+          "type": "string",
+          "enum": [
+            "response",
+            "comparison",
+            "report"
+          ]
+        },
+        "density": {
+          "type": "string",
+          "enum": [
+            "compact",
+            "comfortable"
+          ]
         }
       },
       "required": [
@@ -116,6 +133,15 @@
               "type": "null"
             }
           ]
+        },
+        "role": {
+          "type": "string",
+          "enum": [
+            "result",
+            "analysis",
+            "evidence",
+            "follow_up"
+          ]
         }
       },
       "required": [
@@ -139,6 +165,10 @@
             "info",
             "warning"
           ]
+        },
+        "title": {
+          "type": "string",
+          "maxLength": 200
         }
       },
       "required": [
@@ -148,6 +178,47 @@
       "additionalProperties": false,
       "$schema": "http://json-schema.org/draft-07/schema#"
     },
+    "Result": {
+      "type": "object",
+      "properties": {
+        "title": {
+          "type": "string",
+          "maxLength": 200
+        },
+        "summary": {
+          "type": "string",
+          "maxLength": 4000
+        },
+        "status": {
+          "type": "string",
+          "enum": [
+            "success",
+            "partial",
+            "unconfirmed",
+            "error"
+          ]
+        },
+        "source": {
+          "anyOf": [
+            {
+              "type": "string",
+              "maxLength": 200
+            },
+            {
+              "type": "null"
+            }
+          ]
+        }
+      },
+      "required": [
+        "title",
+        "summary",
+        "status",
+        "source"
+      ],
+      "additionalProperties": false,
+      "$schema": "http://json-schema.org/draft-07/schema#"
+    },
     "Metric": {
       "type": "object",
       "properties": {

contracts/manifest.json

@@ -3,7 +3,7 @@
   "contracts": {
     "server": {
       "file": "server-v1.openapi.json",
-      "sha256": "66fbf1105aa966a590f93fac176cbc77d22a4a6cf0a94dfcd1d3c27273f73692"
+      "sha256": "2c1d757ccf06ae72b10c592e1e9b3d32b4f72cda8489e224cf2c5f2690fc6972"
     }
   }
 }

contracts/peripheral-sample.openapi.json

@@ -103,7 +103,7 @@
             "$ref": "#/components/schemas/ManagedResourceCapabilities"
           },
           "origin": {
-            "const": "synthetic",
+            "const": "system_data",
             "title": "Origin",
             "type": "string"
           },
@@ -273,6 +273,12 @@
             ],
             "title": "Location"
           },
+          "origin": {
+            "const": "system_data",
+            "default": "system_data",
+            "title": "Origin",
+            "type": "string"
+          },
           "priority": {
             "default": "normal",
             "enum": [
@@ -1843,7 +1849,7 @@
             "title": "Location"
           },
           "origin": {
-            "const": "synthetic",
+            "const": "system_data",
             "title": "Origin",
             "type": "string"
           },
@@ -2036,8 +2042,8 @@
     }
   },
   "info": {
-    "description": "测试监测与核查反馈样板，不是正式工单平台或设备控制服务。",
-    "title": "TJWater 外围系统模拟接入",
+    "description": "按项目管理业务工单与反馈、读取监测历史；测试数据保留独立来源标记。",
+    "title": "TJWater 外围系统接入",
     "version": "1.0.0"
   },
   "openapi": "3.1.0",

contracts/view-command.schema.json

@@ -0,0 +1,663 @@
+{
+  "anyOf": [
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "locate_features"
+        },
+        "layer": {
+          "type": "string",
+          "enum": [
+            "junctions",
+            "pipes",
+            "valves",
+            "pumps",
+            "tanks",
+            "reservoirs",
+            "scada"
+          ]
+        },
+        "feature_ids": {
+          "type": "array",
+          "items": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 128
+          },
+          "minItems": 1,
+          "maxItems": 100
+        }
+      },
+      "required": [
+        "action",
+        "layer",
+        "feature_ids"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "zoom_to_map"
+        },
+        "center": {
+          "type": "array",
+          "minItems": 2,
+          "maxItems": 2,
+          "items": [
+            {
+              "type": "number",
+              "minimum": -180,
+              "maximum": 180
+            },
+            {
+              "type": "number",
+              "minimum": -90,
+              "maximum": 90
+            }
+          ]
+        },
+        "zoom": {
+          "type": "number",
+          "minimum": 0,
+          "maximum": 24
+        }
+      },
+      "required": [
+        "action",
+        "center",
+        "zoom"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "render_scada_analysis"
+        },
+        "items": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "properties": {
+              "device_id": {
+                "type": "string",
+                "minLength": 1,
+                "maxLength": 128
+              },
+              "level": {
+                "type": "string",
+                "enum": [
+                  "high",
+                  "medium",
+                  "low",
+                  "unrated"
+                ]
+              }
+            },
+            "required": [
+              "device_id",
+              "level"
+            ],
+            "additionalProperties": false
+          },
+          "minItems": 1,
+          "maxItems": 100
+        }
+      },
+      "required": [
+        "action",
+        "items"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "clear_scada_analysis"
+        }
+      },
+      "required": [
+        "action"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "highlight_features"
+        },
+        "layer": {
+          "type": "string",
+          "enum": [
+            "junctions",
+            "pipes",
+            "valves",
+            "pumps",
+            "tanks",
+            "reservoirs",
+            "scada"
+          ]
+        },
+        "feature_ids": {
+          "type": "array",
+          "items": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 128
+          },
+          "minItems": 1,
+          "maxItems": 100
+        },
+        "effect": {
+          "type": "string",
+          "enum": [
+            "steady",
+            "pulse"
+          ]
+        }
+      },
+      "required": [
+        "action",
+        "layer",
+        "feature_ids",
+        "effect"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "clear_highlights"
+        }
+      },
+      "required": [
+        "action"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "open_chart"
+        },
+        "chart_id": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 128
+        },
+        "chart": {
+          "type": "object",
+          "properties": {
+            "title": {
+              "type": "string",
+              "maxLength": 200
+            },
+            "chart_type": {
+              "type": "string",
+              "enum": [
+                "line",
+                "bar"
+              ]
+            },
+            "x_data": {
+              "type": "array",
+              "items": {
+                "type": "string",
+                "maxLength": 128
+              },
+              "minItems": 1,
+              "maxItems": 100
+            },
+            "series": {
+              "type": "array",
+              "items": {
+                "type": "object",
+                "properties": {
+                  "name": {
+                    "type": "string",
+                    "maxLength": 128
+                  },
+                  "data": {
+                    "type": "array",
+                    "items": {
+                      "type": "number"
+                    },
+                    "minItems": 1,
+                    "maxItems": 100
+                  }
+                },
+                "required": [
+                  "name",
+                  "data"
+                ],
+                "additionalProperties": false
+              },
+              "minItems": 1,
+              "maxItems": 8
+            }
+          },
+          "required": [
+            "title",
+            "chart_type",
+            "x_data",
+            "series"
+          ],
+          "additionalProperties": false
+        },
+        "evidence_refs": {
+          "type": "array",
+          "items": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 128
+          },
+          "minItems": 0,
+          "maxItems": 100
+        }
+      },
+      "required": [
+        "action",
+        "chart_id",
+        "chart",
+        "evidence_refs"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "close_chart"
+        },
+        "chart_id": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 128
+        }
+      },
+      "required": [
+        "action",
+        "chart_id"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "set_overlay"
+        },
+        "overlay_id": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 128
+        },
+        "title": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 200
+        },
+        "mode": {
+          "type": "string",
+          "enum": [
+            "geometry",
+            "heatmap"
+          ]
+        },
+        "evidence_id": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 128
+        },
+        "data": {
+          "type": "object",
+          "properties": {
+            "type": {
+              "type": "string",
+              "const": "FeatureCollection"
+            },
+            "features": {
+              "type": "array",
+              "items": {
+                "type": "object",
+                "properties": {
+                  "type": {
+                    "type": "string",
+                    "const": "Feature"
+                  },
+                  "id": {
+                    "anyOf": [
+                      {
+                        "type": "string"
+                      },
+                      {
+                        "type": "number"
+                      }
+                    ]
+                  },
+                  "properties": {
+                    "anyOf": [
+                      {
+                        "type": "object",
+                        "additionalProperties": {}
+                      },
+                      {
+                        "type": "null"
+                      }
+                    ]
+                  },
+                  "geometry": {
+                    "anyOf": [
+                      {
+                        "type": "object",
+                        "properties": {
+                          "type": {
+                            "type": "string",
+                            "const": "Point"
+                          },
+                          "coordinates": {
+                            "type": "array",
+                            "minItems": 2,
+                            "items": [
+                              {
+                                "type": "number",
+                                "minimum": -180,
+                                "maximum": 180
+                              },
+                              {
+                                "type": "number",
+                                "minimum": -90,
+                                "maximum": 90
+                              }
+                            ],
+                            "additionalItems": {
+                              "type": "number"
+                            }
+                          }
+                        },
+                        "required": [
+                          "type",
+                          "coordinates"
+                        ],
+                        "additionalProperties": false
+                      },
+                      {
+                        "type": "object",
+                        "properties": {
+                          "type": {
+                            "type": "string",
+                            "const": "MultiPoint"
+                          },
+                          "coordinates": {
+                            "type": "array",
+                            "items": {
+                              "type": "array",
+                              "minItems": 2,
+                              "items": [
+                                {
+                                  "type": "number",
+                                  "minimum": -180,
+                                  "maximum": 180
+                                },
+                                {
+                                  "type": "number",
+                                  "minimum": -90,
+                                  "maximum": 90
+                                }
+                              ],
+                              "additionalItems": {
+                                "type": "number"
+                              }
+                            },
+                            "minItems": 1,
+                            "maxItems": 20000
+                          }
+                        },
+                        "required": [
+                          "type",
+                          "coordinates"
+                        ],
+                        "additionalProperties": false
+                      },
+                      {
+                        "type": "object",
+                        "properties": {
+                          "type": {
+                            "type": "string",
+                            "const": "LineString"
+                          },
+                          "coordinates": {
+                            "type": "array",
+                            "items": {
+                              "type": "array",
+                              "minItems": 2,
+                              "items": [
+                                {
+                                  "type": "number",
+                                  "minimum": -180,
+                                  "maximum": 180
+                                },
+                                {
+                                  "type": "number",
+                                  "minimum": -90,
+                                  "maximum": 90
+                                }
+                              ],
+                              "additionalItems": {
+                                "type": "number"
+                              }
+                            },
+                            "minItems": 2,
+                            "maxItems": 20000
+                          }
+                        },
+                        "required": [
+                          "type",
+                          "coordinates"
+                        ],
+                        "additionalProperties": false
+                      },
+                      {
+                        "type": "object",
+                        "properties": {
+                          "type": {
+                            "type": "string",
+                            "const": "MultiLineString"
+                          },
+                          "coordinates": {
+                            "type": "array",
+                            "items": {
+                              "type": "array",
+                              "items": {
+                                "type": "array",
+                                "minItems": 2,
+                                "items": [
+                                  {
+                                    "type": "number",
+                                    "minimum": -180,
+                                    "maximum": 180
+                                  },
+                                  {
+                                    "type": "number",
+                                    "minimum": -90,
+                                    "maximum": 90
+                                  }
+                                ],
+                                "additionalItems": {
+                                  "type": "number"
+                                }
+                              },
+                              "minItems": 2,
+                              "maxItems": 20000
+                            },
+                            "minItems": 1,
+                            "maxItems": 20000
+                          }
+                        },
+                        "required": [
+                          "type",
+                          "coordinates"
+                        ],
+                        "additionalProperties": false
+                      },
+                      {
+                        "type": "object",
+                        "properties": {
+                          "type": {
+                            "type": "string",
+                            "const": "Polygon"
+                          },
+                          "coordinates": {
+                            "type": "array",
+                            "items": {
+                              "type": "array",
+                              "items": {
+                                "type": "array",
+                                "minItems": 2,
+                                "items": [
+                                  {
+                                    "type": "number",
+                                    "minimum": -180,
+                                    "maximum": 180
+                                  },
+                                  {
+                                    "type": "number",
+                                    "minimum": -90,
+                                    "maximum": 90
+                                  }
+                                ],
+                                "additionalItems": {
+                                  "type": "number"
+                                }
+                              },
+                              "minItems": 4,
+                              "maxItems": 20000
+                            },
+                            "minItems": 1,
+                            "maxItems": 20000
+                          }
+                        },
+                        "required": [
+                          "type",
+                          "coordinates"
+                        ],
+                        "additionalProperties": false
+                      },
+                      {
+                        "type": "object",
+                        "properties": {
+                          "type": {
+                            "type": "string",
+                            "const": "MultiPolygon"
+                          },
+                          "coordinates": {
+                            "type": "array",
+                            "items": {
+                              "type": "array",
+                              "items": {
+                                "type": "array",
+                                "items": {
+                                  "type": "array",
+                                  "minItems": 2,
+                                  "items": [
+                                    {
+                                      "type": "number",
+                                      "minimum": -180,
+                                      "maximum": 180
+                                    },
+                                    {
+                                      "type": "number",
+                                      "minimum": -90,
+                                      "maximum": 90
+                                    }
+                                  ],
+                                  "additionalItems": {
+                                    "type": "number"
+                                  }
+                                },
+                                "minItems": 4,
+                                "maxItems": 20000
+                              },
+                              "minItems": 1,
+                              "maxItems": 20000
+                            },
+                            "minItems": 1,
+                            "maxItems": 20000
+                          }
+                        },
+                        "required": [
+                          "type",
+                          "coordinates"
+                        ],
+                        "additionalProperties": false
+                      }
+                    ]
+                  }
+                },
+                "required": [
+                  "type",
+                  "properties",
+                  "geometry"
+                ],
+                "additionalProperties": false
+              },
+              "minItems": 1,
+              "maxItems": 1000
+            }
+          },
+          "required": [
+            "type",
+            "features"
+          ],
+          "additionalProperties": false
+        }
+      },
+      "required": [
+        "action",
+        "overlay_id",
+        "title",
+        "mode",
+        "evidence_id",
+        "data"
+      ],
+      "additionalProperties": false
+    },
+    {
+      "type": "object",
+      "properties": {
+        "action": {
+          "type": "string",
+          "const": "clear_overlay"
+        },
+        "overlay_id": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 128
+        }
+      },
+      "required": [
+        "action",
+        "overlay_id"
+      ],
+      "additionalProperties": false
+    }
+  ],
+  "$schema": "http://json-schema.org/draft-07/schema#"
+}

docs/burst-response-workflow.md

@@ -0,0 +1,248 @@
+# 从爆管上报到处置闭环
+
+本文定义爆管调查的完整业务路线，供维护者审阅和验收。运行时指令位于 [burst-report/SKILL.md](../resources/work-order-skills/burst-report/SKILL.md)，由 Server 根据绑定任务的 `workflow=burst_report` 经 MCP 分发。
+
+重点成果是有依据的事故候选位置、隔离阀门清单和影响范围，以及经审批执行、反馈核验后的闭环记录。Agent 根据证据选择、跳过或重做分支，不按图逐节点机械调用工具。工程图覆盖多轮调查和人工业务；每轮发布的进度图只标注实际进行的工作，不复制整图冒充执行记录。
+
+## 完整工程图
+
+```mermaid
+flowchart TB
+  subgraph intake[工单进入与后台任务]
+    order[用户或外围系统提交爆管上报工单]
+    source[保存原工单及来源版本、发生时间、位置和描述]
+    sync[同步器按来源关联调查及工单投影]
+    queue[Server 保存调查任务和 queued run]
+    claim[Worker 投递，Agent 领取并维持租约]
+    context[读取任务摘要、业务 Skill、历史报告和既有动作]
+    order --> source --> sync --> queue --> claim --> context
+  end
+
+  subgraph investigation[Agent 调查与工程分析]
+    scope[确定事件时刻、地点、调查目标和真实或演练来源]
+    location[核对已有资产编号；必要时解析地点和查询附近管段]
+    monitor[核对相关测点映射、异常窗口、正常对照和数据质量]
+    cover{查询是否实际覆盖事件与对照窗口}
+    more[缩小对象和窗口，按需补页，核对缺测及零值]
+    hypothesis{是否具备定位所需观测和漏水量依据}
+    locate[计算爆管候选排序，保留假设、覆盖范围与证据]
+    candidate[整合上报、资产关联、算法与现场反馈得到候选集合]
+    usable{是否存在可用于隔离评估的候选资产}
+    isolation[逐候选分析拓扑隔离边界、必关阀及受影响节点]
+    isolateok{可隔离且阀门可用性有可核验依据}
+    alternate[纳入已知不可关闭阀门，重新评估替代隔离方案]
+    hydraulic[按需模拟候选关阀组合的压力和流量影响]
+    impact[区分拓扑停水区、水力影响区和有资料支持的用户范围]
+    findings[形成候选位置、阀门清单、影响对比和建议方案]
+    fieldbasis{是否已有足以支持具体现场核验的结论}
+    fielddraft[形成有位置、核验目标及回执要求的现场核验建议]
+    gap[真实关键数据缺失：发布 final insufficient_data 并结束本轮]
+    error[接口、权限或配置异常：记录具体失败并由宿主结束失败运行]
+    scope --> location --> candidate
+    scope --> monitor --> cover
+    cover -->|未覆盖且可补齐| more --> monitor
+    cover -->|已覆盖或确认实际缺失| hypothesis
+    hypothesis -->|具备依据| locate --> candidate
+    hypothesis -->|条件不具备，明确原因| candidate
+    candidate --> usable
+    usable -->|候选有效，标明待现场确认| isolation --> isolateok
+    isolateok -->|有可评估方案| hydraulic --> impact --> findings
+    isolateok -->|已知存在不可关阀| alternate --> isolation
+    isolateok -->|无法形成方案| fieldbasis
+    usable -->|尚无候选| fieldbasis
+    fieldbasis -->|可形成具体核验建议| fielddraft
+    fieldbasis -->|当前判断所需证据确实缺失| gap
+    monitor -->|不可恢复的查询异常| error
+    location -->|不可恢复的必要查询异常| error
+  end
+
+  subgraph approval[报告、审批与现场处置]
+    interim[保存本轮 interim concluded 阶段报告]
+    proposals[创建具体待审批草稿，关联证据和执行前置条件]
+    finalround[保存本轮 final 报告，由宿主结束本轮 run]
+    review[有权限人员审核当前草稿版本]
+    decision{审批结果}
+    revise[驳回理由保存并触发原调查新一轮]
+    submit[Server 通过配置的适配器幂等提交]
+    receipt{外部提交结果}
+    reconcile[使用同一执行请求号核对回执，不重复下发]
+    unsupported[未配置或明确失败：记录原因并重新分析]
+    workkind{本次获批的是核验还是处置}
+    fieldcheck[现场核验准确位置、漏水迹象及资产，回传实际记录]
+    notice[人工实施获批的通知、供水保障和应急安排]
+    isolatefield[现场核验阀门及实际隔离结果]
+    repair[现场定位、止水和抢修，回传实际故障管段及修复记录]
+    clean[按现场风险及获批方案清洗、消毒或冲洗]
+    restore[核验修复与适用复测，按批准方案恢复供水]
+    follow[记录恢复后压力流量、漏水复查及用户回访]
+    findings --> interim
+    fielddraft --> interim
+    interim --> proposals --> finalround --> review --> decision
+    decision -->|驳回| revise
+    decision -->|批准| submit --> receipt
+    receipt -->|结果不明或仍在执行| reconcile --> receipt
+    receipt -->|明确失败或不支持| unsupported
+    receipt -->|工单已创建或设备动作有回执| workkind
+    workkind -->|现场核验事项| fieldcheck
+    workkind -->|有依据的处置事项| notice
+    notice --> isolatefield --> repair --> clean --> restore --> follow
+  end
+
+  subgraph closure[现场反馈、复核与原事件闭环]
+    feedback[外围系统保存关联工单的新反馈和单调版本]
+    requeue[Server 校验来源并为原调查排队新 generation]
+    reread[重新读取现场事实、执行回执和最新监测证据]
+    verify[对照预测与现场结果，复核修复、恢复和风险]
+    faultstate{复核结果支持故障、误报还是尚待核验}
+    falsealarm[核对排除故障的证据及人工确认]
+    closed{处置及闭环证据是否齐全}
+    remaining[列出未完成事项及对应已有或待审批动作]
+    sourceclosed{原上报工单已结案且版本已核对}
+    readyreport[发布具备结案条件报告并结束本轮]
+    closedreport[发布闭环复核 final 报告并结束本轮]
+    closeowner[有权限人员在原工单系统核对并完成原事件结案]
+    closedcase[报告与来源事实一致，确认业务闭环]
+    follow --> feedback --> requeue
+    fieldcheck --> feedback
+    reread --> verify --> faultstate
+    faultstate -->|已确认故障| closed
+    faultstate -->|有依据排除故障| falsealarm --> sourceclosed
+    faultstate -->|尚需核验| remaining
+    closed -->|尚未齐全或复测异常| remaining --> interim
+    closed -->|证据齐全| sourceclosed
+    sourceclosed -->|是| closedreport --> closedcase
+    sourceclosed -->|否| readyreport --> closeowner --> feedback
+    reread -->|现场位置或阀门条件改变| candidate
+  end
+
+  subgraph presentation[贯穿调查的地图、图表与数据整理]
+    progress[发布真实调查进展和 Mermaid 状态]
+    map[高亮候选管段、闪烁建议关阀、展示已有证据覆盖层]
+    chart[打开异常与对照曲线，展示候选及影响对比表]
+    filepermit{会话是否已有文件写入权限}
+    jsonl[按权限整理有来源引用的 JSONL 调查数据]
+    reportui[通过报告 ui_blocks 发布 json-render 结构化界面]
+    sourceview[展示保持证据来源，失败不阻断业务调查]
+    progress --> map --> chart --> filepermit
+    filepermit -->|已获准且需要导出| jsonl --> reportui
+    filepermit -->|未获准或无需导出| reportui
+    reportui --> sourceview
+  end
+
+  context -->|首次或需要重新分析| scope
+  context -->|已有新反馈待复核| reread
+  revise --> requeue
+  unsupported --> requeue
+  requeue --> queue
+  gap -->|补充事实被 Server 接受后才重开| requeue
+  findings -->|无须新增动作| finalonly[无需新增动作时保存 final 报告并结束本轮]
+  scope -.-> progress
+  candidate -.-> map
+  impact -.-> chart
+  verify -.-> reportui
+```
+
+图中的现场处置由实际审批、工单与回执推进；顺序按现场紧迫性和获批方案调整。必要的紧急通知、现场核验和供水保障不应等待可选的完整模型计算。`clean` 为按风险适用的步骤，不能把每次爆管都判断为污染。`hydraulic` 在缺少模型、时序输入或权限时记录具体限制，仍可保留已完成的拓扑隔离结果。
+
+`requeue` 表示 Server 接受了新业务事实后分配下一轮，不是 Agent 复活旧会话或自行循环等待。原工单结案由原系统有权限人员处理，后台 Agent 目前没有直接关闭原工单的工具。`completed`、`concluded`、动作 `succeeded` 和“业务闭环”分别核对，不能互相推导。
+
+## 将业务步骤映射为现有能力
+
+以下产品查询与计算均使用 `mcp_tjwater_` 前缀。参数以本轮实际发现的工具 schema 为准。
+
+| 业务动作 | 已有入口 | 必须形成的结果 |
+| --- | --- | --- |
+| 接续原任务 | `get_task_context`，依次读取 summary、skill 及所需分节；`get_task_run_history` | 原工单来源、事件时刻、调查代次、流程版本、既有动作与现场反馈 |
+| 查找地点和资产 | `geocode` / `search_project_places`、`query_nearby_elements`、`get_element_properties`、`get_node_connections` | 实际候选管段或节点、位置来源、匹配理由；邻近不代表故障确认 |
+| 核对监测 | `list_scada_devices`、`get_scada_device`、`query_scada_readings`、`query_element_history` | 设备和模型节点的映射、异常及对照窗口、单位、缺测、分页覆盖 |
+| 压力异常侦测 | `run_burst_detection` | 至少 5 个完整压力测点及此前 15 天数据满足时才调用；不把侦测当作定位 |
+| 爆管候选定位 | `run_burst_location` | 候选排序、算法实际返回的评价值、观测范围、漏水量假设及来源 |
+| 逐候选隔离 | `analyze_valve_isolation` | `accident_elements`、`isolatable`、`must_close_valves`、`optional_valves`、`affected_nodes`、`affected_node_count` 及已知 `disabled_valves` |
+| 水力影响 | `analyze_valve_closure` | 指定关阀组合、时间窗、输出对象和模拟压力流量；参数不足时保留明确原因 |
+| 用户影响 | 已授权资产属性、区域及外围资料查询 | 受影响对象、关联依据和覆盖限制；没有用户资料不估造户数 |
+| 发布和提议 | `publish_task_progress`、`publish_task_report`、`propose_task_action` | 持久化报告、证据引用和待审批动作；动作不是执行事实 |
+| 反馈接续 | Server `external-feedback` / `human-evidence` 及 worker | 同一任务的新轮次、真实来源版本、修复和复测证据 |
+
+定位的 `burst_leakage` 必须有依据，单位为 m³/s；`pressure_scada_ids` 是模型节点 ID，不能填设备查询编号，也可省略并使用服务的项目配置。monitoring 模式正常窗可成对省略，使用前一天同窗，但这只是一项对照假设，仍要核对数据有效性。simulation 模式必须引用已有分析运行 UUID，不把在线重算当成已有分析运行。已知事故管段可直接做候选隔离；不必先完成一次没有必要的定位计算，也不能把已知答案作为算法独立定位成功的证明。
+
+隔离工具支持明确标注的候选管段，无须等到现场确认才评估方案。多个候选的比较优先分开调用；把多个 ID 一次传入表示联合事故场景，不等同于逐个比较。`disabled_valves` 仅填已有事实支持的不可关阀；空数组不证明现场阀门均可用。`isolatable=false` 不能解释为零影响，即使 `affected_nodes` 为空也须解释返回的计数和限制。
+
+## 像调查员一样查询、查看、比较和复核
+
+每一步先明确待回答的问题，再查小范围资料，观察实际结果后决定下一步。例如：核实地址，查附近管段并查看属性，关联压力点，比较事件前后及正常对照，检查候选，评估关阀影响。只把必要的查询、事实和判断依据写入进度及报告，不保存模型内部推理。
+
+- 不满足于“查过 SCADA”：必须确认返回的时间实际覆盖事件。查询 24 小时而只返回当天前 45 分钟时，应缩到事件前后的小窗，按需补页，不直接进入最终判断。
+- 正常基线、实测、模型模拟、外围演练和人工反馈分别标注；缺测不是零值，异常不是爆管确认，候选评分不是校准后的概率。
+- 查不到事故管段时先检查地址解析、附近资产与上下游关联；能查询的字段先查询，不能直接全部交给现场。
+- 既有证据通过 `get_task_evidence` 的有界分页读取，不重复拉取同一批事实。失败、权限拒绝和存储未配置不当作“无异常”或“无数据”。
+- 可以用最多两个前台 `investigation-reader` 并行核对无依赖的问题。读取、计算、发布和文件权限仍按角色执行；reader 只返回查询事实及真实证据编号。
+
+## 地图与图表操作
+
+后台调查统一调用 `publish_task_presentation`，每个不同命令使用不同 UUID `request_id`，相同命令重试复用原编号。以下为工具已有能力，不需要浏览器自动化。
+
+| 调查进展 | command.action | 取值及依据 |
+| --- | --- | --- |
+| 找到实际候选资产 | `highlight_features` | `layer=pipes` / `junctions`，`feature_ids` 为已核实的地图要素编号，`effect=steady` |
+| 正在核查候选或建议关阀 | `highlight_features` | 对应 pipes 或 valves，`effect=pulse`；闪烁只表示调查关注，不代表设备正在操作 |
+| 查看异常与对照 | `open_chart` | 有限长度的时间和值，引用真实 `evidence_refs`；不同日期的对照对齐方式在标题或来源中说明 |
+| 查看已有空间结果 | `set_overlay` | 提供 `overlay_id/title/mode/evidence_id`，Server 从证据提取几何，不接收手绘影响区 |
+| 撤下旧的临时展示 | `close_chart` / `clear_overlay` / `clear_highlights` | 清理自己记录的临时展示；清理失败不修改业务结果 |
+
+支持的图层为 junctions、pipes、valves、pumps、tanks、reservoirs、scada。先核实业务 ID 与地图 feature ID 的映射，不能把设备编号、节点编号与 GIS 编号混用。每次高亮最多 100 个要素，需要更多时说明所显示的子集；不要假定多次命令会累积成完整范围。没有几何时可以按已知 ID 高亮、用表格解释结果，不画包络多边形冒充影响范围。
+
+图表最多 100 个横轴值和 8 组数值序列，单位须在系列名称或标题中明确。不能把 null 缺测填零以通过 schema；可以按有数据的区间分图，并保留缺测说明。`recorded` 仅表示请求保存。浏览器未开 Dev 跟随时仍继续调查，不能等待地图显示回执。后台接口目前没有摄像机缩放、平移、鼠标点击或任意图层开关命令；不要调用仅供聊天的 `locate_features`、`zoom_to_map` 或原生浏览器工具替代。
+
+## JSONL 调查数据与 json-render 展示
+
+JSONL 用于整理已经取得的数据与分析结果，不补造监测、阀门、现场反馈或审批。推荐每行一个对象，记录 `record_type/origin/source_evidence_refs`，并按对象保留时间窗、单位、模型或假设、覆盖限制和实际结果。`origin` 沿用来源分类；本地派生记录引用输入证据，并说明转换方法。不要把 JSONL 文件自身当作 Server 已保存的业务证据。
+
+按内容选择 `event_context`、`monitoring_coverage`、`candidate`、`isolation`、`hydraulic_impact`、`field_feedback`、`closure_check`，无需为没有结果的类别制造空壳文件。候选记录写实际排序及评价值；隔离记录保留建议必关阀、可选阀、已知不可关阀及受影响节点；闭环记录逐项保留已核验、待核验或不适用及原因。JSONL 是导出约定，不新增 MCP 输入字段或服务端数据协议。
+
+以下两行只演示文件格式，`synthetic`、`example_only` 与空证据引用表示它们不可用于实际报告或审批，不包含真实分析结果：
+
+```jsonl
+{"record_type":"candidate","origin":"synthetic","example_only":true,"source_evidence_refs":[],"element_id":"DEMO-PIPE-01","rank":null,"evaluation_value":null,"status":"not_computed","limitations":["仅说明字段，未执行定位"]}
+{"record_type":"closure_check","origin":"synthetic","example_only":true,"source_evidence_refs":[],"item":"repair_verified","status":"not_verified","observed_at":null,"limitations":["无现场反馈，不能声称闭环"]}
+```
+
+如果当前会话已获准使用原生文件写入，则在当前 workspace 下写 `files/burst-investigation-<generation>.jsonl`，保留原始证据不变；通过允许的读取工具逐行复核 JSON、行数、对象 ID、时间和单位。只引用宿主绑定的 generation，不能从用户文本自选。不得读取兄弟会话、业务数据库、宿主源码或凭据；需要本地转换时只使用已获准的原生工具及沙箱 Shell，不引入 HTTP、SQL 或自由命令查询桥。
+
+当前没有专用的免审批 JSONL 导出工具。文件权限未授予时，不通过 Shell 绕过、不让后台无限等待批准、不声称已生成文件；继续完成报告内的结构化数据，并如实说明文件未保存。恢复写入授权须由有权限的人或宿主处理，Skill 不扩大权限。
+
+图表、表格和指标使用 `publish_task_report.report.ui_blocks`，每块为 `answer-ui@1`，包含独立 `block_id`、`fallback_text` 和 json-render 的 `spec.root/elements`，外层附真实 `evidence_refs`。最多 8 块、总计 400 KB，单块还受现有 schema 限制。优先组织以下内容：
+
+| 展示 | 建议组件 | 数据来源 |
+| --- | --- | --- |
+| 候选管段及理由 | Table、Evidence、MapLocation | 地点、资产与定位证据；明确候选和已确认的区别 |
+| 异常及正常对照 | Chart | 同单位的实际序列，说明时间对齐及缺测 |
+| 关阀与隔离范围 | Table、Metric、MapLocation | 隔离证据、阀门属性及候选场景 |
+| 水力与用户影响 | Table、Chart | 指定输出节点的模拟结果及真实用户关联资料 |
+| 处置与闭环核对 | Table、Evidence | 动作回执、现场反馈、恢复监测和原工单版本 |
+
+只用工具公布的组件属性。JSONL 可作为整理中间格式，但目前没有“把任意 JSONL 直接传给 json-render”的通道；需转换为已注册组件的 JSON 属性。后台不用 `present_answer`，不把原始 JSONL 输出到聊天正文，也不把表单提交当成已保存现场证据。小数据和简单结论可直接文字表达。
+
+## 本轮成果与爆管闭环的验收
+
+报告至少回答下列四类问题，逐项给出实际结果或经过查询证实的具体限制；无需固定排版或创建固定数量的动作。
+
+1. **位置**：事故已确认还是仅有候选？每个候选的模型/地图编号、匹配或定位依据、排名及未确认事项是什么？
+2. **隔离**：每个候选场景是否可隔离？应关、可选和已知不可关的阀门分别是什么？有无替代方案及现场前置核验？
+3. **影响**：拓扑受影响节点、水力压降或失供范围、用户关联分别有哪些实际结果？未覆盖什么？没有水力或用户资料时分别说明，不能推定为零。
+4. **处置与闭环**：当前可审批的工作是什么？已有哪些动作、审批、提交回执和现场事实？还缺什么反馈才具备结案条件？
+
+有具体结论和动作建议时，先保存 `interim/concluded`，再 `propose_task_action`，最后保存 `final`。人工执行事项使用 `work_order_type=general`，保留原爆管事件。工单描述写位置、任务内容、依赖、执行前核验、需回传的事实和责任要求；`propose_task_action` 当前 payload 没有独立的 dependencies 或 feedback_requirements 字段，不发明字段。没有客户通知适配器时，将通知作为人工工单事项，不宣称已发送。
+
+闭环复核至少核对实际故障资产及原因、隔离/阀门实施记录、抢修结果、适用的清洗及水质复测、恢复供水记录和恢复后监测、用户通知及回访、未决动作与遗留风险、原上报工单结案状态。不适用事项必须有事件依据；未收到结果写待核验，不把记录创建时间当现场完成时间。只齐备处置证据而原工单未结案时，结论为“具备结案条件，待原系统确认”，而非“已闭环”。误报或未发现故障的结案也要有支持排除的实际核查与人工确认。
+
+必要查询失败、权限拒绝和存储未配置属于执行异常，按 schema 修正可纠正问题；仍无法取得当前判断必需的资料时交宿主记录失败。已通过有效查询确认关键数据确实缺失且无法补齐，才发布 `final/insufficient_data`，提供 blocking 缺口和查询依据，建议与待办为空，不在该退出分支再生成补数据工单。能够明确形成一项现场核验结论时可提议具体核验动作，但不能把“尚未开展定位”包装为“已完成定位调查”。
+
+## 能力边界与验证
+
+本次交付为工程图和运行时 Skill 文本，使用现有 MCP、发布、审批与反馈能力；不新增自动关单、鼠标自动化、摄像机操作、JSONL 导出接口或数据库字段。Server 目前没有按 Skill 检查定位/隔离成果的强制验收器，本文的业务验收要求仍由 Agent 执行并由人复核。
+
+当前 Skill 仅通过 `section=skill` 整份分发，没有相对路径 reference 读取协议，因此运行时指令保持自包含，不能要求 Agent 读取本工程文档。内容需控制在既有内联预算内。Server 更新资源后，应按原发布方式让实际 Server 实例取得新文件；Agent 读取时核对 `name/version/content_sha256`，不把本地文件更新当成运行中实例已生效。
+
+验证包含 Mermaid 语法解析、本地路径和工具 ID 核对、Skill 分发/权限/体积测试及 `git diff --check`。后续真实行为验收应分别覆盖：有效观测和漏水量依据下的定位；已知候选但缺漏水量时的候选隔离；不可关阀替代方案；SCADA 首屏未覆盖事件时继续读取；接口失败明确终止；文件权限缺失不绕过；无浏览器时照常调查；实际反馈触发新轮及最终闭环复核。没有真实模型计算或现场回执，不报告定位准确性或业务闭环已通过。
+
+关联契约：[分析 MCP](mcp-analysis-tools.md)、[调查展示](investigation-progress-and-presentation.md)、[报告契约](investigation-report-contract.md)、[任务与多轮反馈](operational-tasks.md)、[后台工作进程](operational-task-workers.md)。

docs/calculation-queue.md

@@ -0,0 +1,94 @@
+# MCP 后台计算队列
+
+API 保留 `uvicorn --workers 4`。长计算在项目 BizDB 持久化入队，由独立计算进程领取；所有 API 进程读取相同任务状态，不依赖粘性会话。HTTP 返回预算由宿主管理，Server 默认 120 秒、Agent 默认 180 秒。Server 通过工具目录 `_meta["io.tjwater/request-timeout"]` 统一发布 `defaultSeconds` 和 `maxSeconds`，所有工具（包括状态和结果读取）使用同一配置来源。Server 兼容保留 `timeout_seconds` 参数；Agent 对模型隐藏并拒绝该参数，按目录预算代填。Server 请求上限为 240 秒，Agent 加 60 秒返回余量并封顶 270 秒，原生 MCP 与本地连接的总时限为 300 秒；上游 HTTP 的 5 分钟空闲超时保留。提交、查询状态和读取结果都应及时返回，不等待整个计算。模拟时间仍由 `duration` 表示。
+
+后台执行时限由管理员配置 `TJWATER_MCP_MAX_CALCULATION_TIMEOUT_SECONDS`（默认 3600 秒），入队时单独写入任务记录。HTTP 参数不参与作业幂等参数比较，也不会缩短或延长后台执行。任务表的内部 `timeout_seconds` 列仍保存执行时限。HTTP 超时或断开只结束本次等待，后台作业需通过独立取消工具取消。
+
+```text
+前端 / Agent → API × 4 → 项目 BizDB calculation_jobs
+                              ↑
+                      独立计算 supervisor
+                              ↓
+                    受管理的计算子进程组
+```
+
+## 工具契约
+
+原工具名继续使用。以下工具提交后返回 `QueryResult`，其中 `semantics=task`，`data` 包含 `job_id`、`status`、`tool` 和 `poll_after_ms`；这只证明提交状态，不是分析结果或调查证据：
+
+- `run_burst_location`、`run_leakage_identification`、`run_burst_detection`
+- `run_simulation`、`run_burst_simulation`、`run_contaminant_simulation`
+- `analyze_valve_closure`、`analyze_flushing`、`analyze_pressure_regulation`
+- `optimize_sensor_placement`
+- `save_simulation_scenario`、`create_sensor_placement` 的 `dry_run=false` 执行阶段
+
+后两项保留预览、摘要校验和原有审批；提交与执行时均校验权限及预览摘要，已有写入回执继续约束重复请求。`analyze_valve_isolation`、`calculate_demand_allocation` 保持同步执行。普通查询和其他写入不进入此队列。
+
+| MCP 工具 | 输入 | 返回 |
+| --- | --- | --- |
+| `get_calculation_job` | `job_id` | 状态、工具、时间和安全错误码 |
+| `get_calculation_result` | `job_id`，可选 `data_path/limit/offset` | 成功后读取原分析结果；未成功返回任务状态 |
+| `list_calculation_jobs` | 可选 `status/limit/offset` | 当前授权会话及调查轮次的任务列表 |
+| `cancel_calculation_job` | `job_id` | 排队任务可直接取消；运行任务先进入 `cancel_requested` |
+
+状态为 `queued/running/cancel_requested/succeeded/failed/cancelled/timed_out/interrupted`。返回创建、开始、结束和排队/结果到期时间；`result_available` 表示结果是否仍可读取。任务取消不等于在线模型写入回滚；发生失败、超时或中断时，需要核查已有写入，不能自动重算整个窗口。
+
+当前使用普通 MCP 业务工具承载队列契约，没有宣称已实现协议级 Tasks 扩展。Server 和 Agent 目录必须配套更新，旧 Agent 不应接收新的队列返回契约。
+
+## 身份、幂等和结果
+
+任务记录保存提交人的元数据用户 ID、项目、原工具参数和受限授权快照，不保存 Bearer、数据库 DSN 或短期 `context_handle`。执行和读取结果时重新查询当前用户、成员关系、项目路由、权限及调查运行租约。任务不会因为短期 MCP 句柄失效而丢失；也不会借持久化任务绕过撤销的权限。
+
+普通计算使用宿主从原生工具调用记录解析出的稳定调用 ID；同一次传输重试复用任务，不将新的业务计算自动按参数合并。重写工具同时保留原审批 `request_id` 幂等语义。不同参数复用同一操作编号会被拒绝。
+
+调查的排队回执不写成分析证据。成功结果读取时，按原计算来源、任务及运行验证，并幂等地保存证据。未完成的计算阻止提交最终报告、`insufficient_data` 或将运行标记成功；中间进度仍可保存。
+
+Agent 调查执行在原生回合空闲后由宿主短轮询等待，继续维持调查心跳。新作业进入终态即触发模型读取结果，其余作业继续后台运行。状态读取遇到网络、限流或临时服务错误，最多重试 3 次，采用指数退避和随机抖动；权限拒绝、上下文失效立即停止，重试耗尽明确报错，不重新提交或取消计算。普通聊天展示任务编号，可在当前会话后续查询。异常流结束时若无法安全继续处理新结果，会明确失败而不是虚报完成；Server 中的计算结果仍保留。
+
+## 排队期限与结果保留
+
+每项目最多 100 个未结束作业（queued/running/cancel_requested），跨 API 进程用数据库事务锁串行执行幂等检查与容量准入；同一请求的重试优先返回已有作业。排队最长 3600 秒，超过后标为 `timed_out`，错误码为 `QUEUE_TIMEOUT`；开始执行后使用独立的执行期限。提交、查询和 Worker 领取都会核查到期任务，所以 Worker 停止时，查询也能结束过期排队。部署监控仍须覆盖 Worker 进程状态和最老排队任务年龄，队列到期机制不代替服务可用性监控。
+
+结果正文在终态后保留 7 天，到期由队列维护清除，读取返回 `RESULT_EXPIRED`。任务状态、输入、幂等标识和已归档调查证据不随之删除，也不会因结果过期重新计算。状态列表只读取状态字段，不加载完整计算结果。
+
+## 并发、取消和恢复
+
+一个 supervisor 默认全项目合计并发 2 项，独立于 API 的 4 个进程。它持有元数据库的会话级 advisory lock，第二个 supervisor 会拒绝启动；当前不是多台计算节点各自扩容的模式。各项目轮转领取，使用 PostgreSQL 短事务 `SKIP LOCKED`，租约代次约束心跳和结果写回。
+
+每项任务有独立子进程组；取消、执行超时或失去租约时先停止进程组，再写终态。守护管道在 supervisor 被强杀后关闭，负责终止仍在运行的算法进程。API 重启不影响计算 supervisor；计算进程崩溃后，旧任务标为中断，不自动重放。
+
+原生临时文件写入独立任务目录。临时模型数据库创建前登记到该目录的账本，正常结束或取消后只清理明确登记的临时数据库。清理失败返回 `CLEANUP_FAILED` 并保留恢复目录；启动时检查项目、任务和执行代次，确认旧租约过期后恢复清理。无法验证归属或清理失败时停止领取新工作。
+
+## 部署
+
+1. 先对每个目标项目的 BizDB 依次应用 `resources/sql/026_calculation_jobs.sql` 和 `resources/sql/027_calculation_job_lifecycle.sql`，包括未来项目使用的相应业务模板。已有 026 的库只需追加 027。不要应用到 TimescaleDB。027 增加时间字段与索引，不清除结果；已有结果从首次迁移起获得完整 7 天保留期。必须先迁移，再加载新 API/Worker 代码。
+2. 配置 `CALCULATION_WORKER_PROJECT_IDS` 为明确的项目 UUID 列表，配置现有元数据库连接及数据库路由解密密钥。Worker 不需要新 OAuth 密钥；仅部署侧工作进程读取这些设置。
+3. 同步发布 Server 和 Agent，在接入长计算前启动独立 Worker。不要从 API `lifespan` 启动，不要把调查 Worker 当作计算 Worker。
+
+从 Server 仓库启动：
+
+```bash
+conda run -n server python scripts/run_calculation_worker.py
+```
+
+systemd 模板：`infra/systemd/tjwater-calculation-worker.service`。Docker 使用同一 Server 镜像的独立服务：
+
+```bash
+docker compose -f infra/docker/docker-compose.yml --profile calculations up -d calculation-worker
+```
+
+Compose 使用 `../../.env` 注入配置，需确保其中数据库地址在容器内可达。`calculation-state` volume 保存任务恢复账本，不能当作临时缓存删除。非容器启动可用 `CALCULATION_WORKER_STATE_DIR` 指定持久目录。
+
+并发默认 2、轮询默认 2 秒、租约 30 秒、心跳 5 秒。算法内部并行度仍消耗额外 CPU/内存；调整并发必须依据实际资源测量。
+
+回滚前暂停接收新的队列计算并等待或明确取消在途任务，核对有副作用任务的写入。保留任务表、结果和恢复目录；旧同步版本不支持继续处理这些作业，不能直接删除记录后重新提交。
+
+## 验证
+
+```bash
+conda run -n server python -m pytest tests/unit/test_mcp_calculation_jobs.py tests/unit/test_calculation_worker.py tests/unit/test_calculation_cleanup.py tests/unit/test_mcp_query.py tests/unit/test_mcp_analysis.py tests/unit/test_operational_tasks.py -q
+```
+
+真实 PostgreSQL 并发测试为 `tests/integration/test_calculation_queue_storage.py`。它要求 `CALCULATION_QUEUE_TEST_DSN` 指向名称以 `tjwater_calc_test_` 开头的专用可丢弃数据库，不能指向业务库。覆盖四连接重复提交、参数冲突、并发领取、项目和用户隔离、取消竞争、失效代次拒绝和过期租约恢复。
+
+这些验证不代替真实项目算法运行、浏览器断线重连或部署验收。生产接入还应观察超过 5 分钟的真实计算、恢复后结果读取和原生临时资源清理。

docs/investigation-progress-and-presentation.md

@@ -55,21 +55,60 @@ Server 校验当前 run、绑定会话、有效租约和幂等身份，记录 `o
 
 | command.action | 参数 |
 | --- | --- |
-| highlight_features | layer、feature_ids、effect：steady 或 pulse |
+| locate_features | layer、feature_ids |
+| zoom_to_map | center：WGS84 经度纬度、zoom：0–24 |
+| highlight_features | layer、feature_ids、必填 effect：steady 或 pulse |
 | clear_highlights | 无 |
 | open_chart | chart_id、chart、非空 evidence_refs |
 | close_chart | chart_id |
-| set_overlay | overlay_id、title、mode：geometry 或 heatmap、evidence_id |
+| set_overlay | overlay_id、title、mode：geometry 或 heatmap、evidence_id，可选 data |
 | clear_overlay | overlay_id |
+| render_scada_analysis | items：device_id 与 high/medium/low/unrated 等级，设备编号不重复 |
+| clear_scada_analysis | 无 |
+
+以上十种动作与 Chat 的 `present_view`、Dev JSONL 共用扁平 `command: {action, ...字段}`。Chat 使用 `frontend-action@2` 和执行回执；任务仍只记录展示请求，Dev 只做本地预览。三者使用独立的地图源、SCADA 图层和图表状态。Dev 的动作数据行必须为 `{"record_type":"presentation","command":{...}}`；界面块行使用 `record_type: ui_block`，新示例明确带 `phase: final`。普通调查数据保持来源字段，不执行其中的 action。仓库示例见[前端统一展示夹具](../../next-tjwater-frontend/tests/fixtures/unified-presentation.jsonl)，其中编号来自测试夹具。
 
 图表限定折线和柱形，包含 `title/chart_type/x_data/series`。最多 100 个横轴值和 8 组有限数值序列，
 序列长度与横轴对齐，引用证据必须属于当前任务。地图高亮来源沿用上述七类。
-覆盖层输入不接受模型提供的 `data`。Server 从证据的不可变数据快照提取现有 GeoJSON，
+覆盖层的权威数据来自不可变证据。省略 `data` 时，Server 从证据快照提取现有 GeoJSON，
 或使用已记录的 `longitude/latitude` 点位，将 `FeatureCollection` 加入事件中的 `command.data`。
+若传入 `data`，必须与同一证据快照解析出的内容一致；Server 始终发布快照解析结果，不能用调用参数覆盖证据。
 热力图只接受记录中的点，权重为有限非负数，未记录权重时取 1。
 没有坐标的地址、空数据、非法几何、模型自绘区域均不能生成覆盖层。该接口不做地址解析、
 范围推算或污染传播计算。展示事件总大小上限为 512 KiB，超限返回 422。
 
+`generate_node_coverage` 是独立的节点空间覆盖工具，适用于普通项目聊天和后台调查，
+不绑定爆管、阀门隔离或固定执行顺序。调用者提供需要比较的节点分组，Server 从当前项目
+`gis.node_geometries` 读取真实位置。监测设备、投诉、地址等须先关联模型节点，
+不接受模型自由生成的坐标。具体适用范围和参数见[节点覆盖工具](node-coverage-tool.md)。
+
+计算先投影到局部米制坐标，取凹包（默认 ratio=0.25）、加展示缓冲（默认 25 m），
+再以 2 m 容差简化并转回 WGS84；凹包比例与缓冲距离可在有界参数内选择。
+默认展示各组的地理包络，不判断连接关系；需要时可按模型连接拆分，额外指定视为断开的连接。
+少量点、重合点和共线点同样通过缓冲形成面。多组结果一起保存在 `result.map_payload`，
+避免连续 `set_overlay` 覆盖前一组；`result.coverage` 保存来源节点、映射完整性及算法参数。
+该面不推断故障、污染、实际停水、水力影响或用户服务边界，缓冲距离也不是供水服务半径。
+缺失几何时标记 `partial`，没有可用位置时不生成面。输入最多 8 组、合计 10000 个不同节点；
+几何输出遵守覆盖层顶点和大小限制，超限要求缩小范围，不静默抽样或改变原分析结果。
+
+MCP 保存完整查询响应作为不可变证据，展示解析器同时支持响应信封中的
+`data.result.map_payload`。Agent 使用 `set_overlay` 引用本次 `evidence_id` 并省略 `data`。
+超过内联返回预算时，响应的 `data.presentation.command` 仍保留可发布的展示命令，
+`data.presentation.groups` 保留覆盖完整性计数，
+完整几何留在证据中，不能把缩略返回误认为没有空间成果。调查按实际步骤同步发布展示：
+地址得到明确坐标后缩放定位，取得候选资产后定位候选，检查影响节点后生成并展示覆盖示意，
+形成或调整关阀方案后定位阀门，读取监测对照后展示点位及序列。
+图表同样随结果返回发布：监测趋势与事件对照用折线图，可比候选评价或方案用柱状图，
+水力及污染传播等模拟按实际返回的时刻、对象和指标画图。压力、流量、浓度等不同单位分图，
+注明来源、范围与实测或模型属性；没有同口径基准时不计算或绘制改善量，缺测不补零。
+过程图表通过 `open_chart` 提供 `chart_id/chart/evidence_refs`，每图最多100个横轴值、8组序列，
+超限时分图或明确说明抽样、聚合范围；单值或无可比数据用文字或表格。
+关键趋势与比较结论保留到报告中，最终报告的图表不替代过程发布。
+业务 Skill 说明本步骤应展示什么，调查指令统一说明发布方法、非空几何检查、去重和失败处理。
+同一阶段批量展示，数据或方案没有变化时不重复请求；不为地图动作增加无关业务分析。
+报告的 `map_targets` 仅为手动定位入口。关键步骤的数据缺失或展示失败需记录具体限制，
+没有浏览器连接也不阻止分析或报告收尾。
+
 `GET /api/v1/operational-tasks/{task_id}/presentation-cursor` 在同一事务中读取
 `task_id/execution_generation/active_run/event_cursor`，其中活跃 run 仅返回 `run_id/generation/status`。
 浏览器打开 Dev 联动、切换任务或断线重连时，以此建立新基线，只消费之后的展示事件。
@@ -98,7 +137,7 @@ Server 校验当前 run、绑定会话、有效租约和幂等身份，记录 `o
 实际 worker 曾成功投递指定调查，模型发布进展和报告，但因 service 账号的任务权限过滤
 导致 `approved_data_scopes` 为空，误作数据不足结束。这不是业务数据确实不足。
 用户确认开放规则后，任务 API 已移除该额外过滤，沿用账号已有的项目读取权限。
-在线上下文返回七类来源，同一身份实际读取 SCADA 设备和外围合成工单成功；没有新增
+在线上下文返回七类来源，同一身份实际读取 SCADA 设备和外围系统工单成功；没有新增
 角色权限或放宽跨项目、审批和设备控制边界。旧委托路径保留兼容，已有权限无需委托凭据。
 用户级 `tjwater-operational-worker.service` 已重新启用。旧报告保留，未自动重新调查，
 不能把旧报告视为修复后的业务验收。任务创建时间下界和并发配置见

docs/investigation-report-contract.md

@@ -43,7 +43,8 @@
 完整图语法的解析失败只影响图示，正文、依据和审核仍可阅读。
 
 每条 node_ref 包含 `node_id`、`label`、`evidence_refs`、`action_refs`；后两项默认空数组。
-node_id 为最多 80 字符的 ASCII 标识符，必须以字母或下划线开头，在本图引用内唯一；
+node_id 为最多 80 字符的 ASCII 标识符，必须以字母或下划线开头，在本图引用内唯一，
+并与 Mermaid 源码中实际节点使用完全相同的 ID；禁止为同一节点另造引用 ID；
 最多 100 条引用。证据必须来自报告 evidence_refs，动作必须属于当前任务。
 节点可以说明候选方向与条件关系；图中的分支、状态文字不能证明实际并发、动作执行或现场完成。
 图的颜色、字体和交互由前端统一配置。正文不包含 SVG 或 HTML。

docs/mcp-analysis-tools.md

@@ -52,17 +52,37 @@ Agent 原生工具统一使用 `mcp_tjwater_` 前缀。Server目录共67项：42
 - 数据源和计算假设必须有依据。定位漏水量和DMA总漏损约束使用m³/s；冲洗使用服务实际接收的m³/h，不沿用旧REST误标的L/s。工频泵序列0/1，变频泵序列为Hz；每项对应模型水力步长。冲洗当前限LPS/CMH模型。选址当前推荐压力点，不冒充污染溯源优化。
 - 调查结果经现有快照/证据/回执链保存，只有返回evidence_id后才引用为系统证据；普通聊天不自动保存调查证据。模拟结果始终是analysis而非实测。已知模拟事件源管段是输入背景，不是算法定位准确性的证明。
 
+## 关阀隔离输入与影响范围
+
+REST `/api/v1/valve-isolation-analyses` 与 MCP `analyze_valve_isolation` 共用同一服务。
+`accident_element_type` 默认为 `auto`：先识别管段 ID，再识别节点 ID；指定 `link` 时只匹配管段，指定 `node` 时只匹配节点。
+节点表和管段表分别保证 ID 唯一，但不同类型可以重号；分析同名节点必须显式指定 `node`。
+多个 ID 表示联合事故，不表示分别比较；类型参数作用于本次调用中的所有 ID。
+
+服务从模型主表的一致快照读取连接关系、初始开关状态和水源，不依赖 GIS 几何或物化视图刷新。
+不可关阀按保持现状处理；已关闭的管段、泵和阀门不参与连通，已关闭阀门不再列为建议关阀。
+未知的事故 ID、未知的不可关阀 ID 和空事故列表会被拒绝。
+
+- `affected_nodes` / `affected_node_count` 仍表示最近可关阀门围成的直接隔离区。不可隔离时列表为空，计数保留。
+- `supply_affected_nodes` / `supply_affected_node_count` 表示按建议关阀后新增失去水源连接的节点，包含间接受影响区域，排除原本就不连通水源的节点。不可隔离或缺少水源资料时为 `null`。
+- `source_check_performed` 表示是否已执行水源连通检查。`isolatable=true` 要求存在关阀边界，且关闭这些阀门后直接隔离区不再连通任何水源。
+- `isolation_failure_reason` 可为 `connected_to_water_source`、`water_sources_unavailable`、`no_boundary_valves`；可隔离时为 `null`。
+
+以上是模型初始开关状态下的无向拓扑分析，水库和水池均按潜在水源处理；不模拟泵或止回阀的单向性、时变控制、水池蓄水量、压力与流量。
+因此“失去水源连接”不能直接当作经过水力计算或现场核实的停水范围。已保存的旧调查证据不会自动重算，需重新调用隔离分析生成新结果。
+本次无需数据库迁移。
+
 ## 资源边界与发布
 
-12项计算和13项预览式写入工具均提供 `timeout_seconds`，由Agent根据模型规模、模拟步数及已有耗时按次选择；它表示实际等待预算，与 `duration` 表示的模拟时间不同。未填写默认120秒。服务端上限由 `TJWATER_MCP_MAX_CALCULATION_TIMEOUT_SECONDS` 控制，默认3600秒，管理员可设120至86400秒；`tools/list` schema 发布实际有效上限，超限在计算前拒绝。
+长计算采用[后台计算队列](calculation-queue.md)。10项分析计算与2项重计算写入的正式执行阶段快速返回任务编号，再使用 `get_calculation_job`、`get_calculation_result`、`list_calculation_jobs`、`cancel_calculation_job` 管理。原有预览与审批仍有效；排队回执不作为分析证据。
 
-例如 `duration=3600, timeout_seconds=900` 表示模拟1小时、允许等待15分钟。Agent的上游HTTP/SDK预算为所选预算加5秒，整个宿主调用另留60秒处理认证、上下文绑定和清理；OpenCode连接上限从工具目录计算并写入宿主连接文件。本地MCP请求关闭Bun空闲断连，仍受上述调用截止时间控制。普通查询继续使用原30秒Server预算与35秒宿主预算。
+HTTP 预算由宿主从目录元数据读取，Server 默认 120 秒、Agent 默认 180 秒；爆管定位不再设置单独的长超时。`timeout_seconds` 保留在 Server 兼容入口，Agent 对模型隐藏并拒绝覆盖。Server 上限 240 秒，Agent 加 60 秒余量并封顶 270 秒，外层原生 MCP 保留 300 秒总时限与上游 Bun 的 5 分钟空闲超时。后台执行时限由管理员配置 `TJWATER_MCP_MAX_CALCULATION_TIMEOUT_SECONDS`（默认 3600 秒），不受该 HTTP 参数控制；模拟时间仍由 `duration` 表示。
 
-计算在线程执行，不阻塞事件循环，单Server进程最多两项计算。超时不能杀死底层原生计算，槽位一直保留到计算真正退出；超时结果不会补写为调查成功结果，不自动反复重试。本次未增加后台计算作业队列。普通聊天大结果仍受现有12KB内联限制，调查结果可经证据分页读取。
+`analyze_valve_isolation` 和 `calculate_demand_allocation` 保持同步。队列消费者独立于Uvicorn的4个API进程，默认全部配置项目合计并发2项，并通过元数据库单实例锁防止重复启动。取消、超时和租约失效需要停止计算子进程组；成功结果按原运行授权保存证据，未完成计算阻止调查提前结案。
 
 在线重算逐个时刻写入基线，超时或中途失败可能已有部分时刻更新，不能把工具失败解释为整个窗口已回滚；不要自动重试整个窗口。
 
-无需新增数据库迁移或模型配置。须同步发布Server、Agent和前端证据枚举；先更新Server目录，再重启Agent，client模式还需重载原生连接。保持既有OAuth、MCP开关及数据库路由配置，不自动更改权限或重启调查。
+队列接入需要在目标项目BizDB依次应用 `resources/sql/026_calculation_jobs.sql`、`resources/sql/027_calculation_job_lifecycle.sql` 并启动独立计算Worker。Server与Agent需配套发布，client模式还需重载原生连接。既有OAuth与项目权限保持不变；完整部署与回滚步骤见后台计算队列文档。
 
 本次验证限schema、权限、协议目录和mock业务函数，不运行真实模拟、优化或付费模型。真实项目的模板同步状态、运算耗时、SCADA覆盖和结果准确性仍需另行验收。
 

docs/mcp-query-implementation.md

@@ -1,10 +1,10 @@
 # 产品 MCP 查询：本地实施与启用
 
-实现使用 MCP 2026-07-28、Python mcp 2.3.0、TypeScript @modelcontextprotocol/client 与 @modelcontextprotocol/server 2.3.1。Server 提供 `/mcp` 和67项工具（42项查询、12项计算、13项预览式写入）；Agent 将其投影合并到本地 MCP，由 OpenCode 原生发现为 `mcp_tjwater_*`，普通项目会话、调查主 Agent 和已验证的直接只读子 Agent 共享42项查询；12项计算仅向非reader会话开放，13项写入仅向项目成员的普通聊天开放。个人会话不能获取项目凭据。旧 CLI 只作为能力清单，没有运行时 CLI 桥接。
+实现使用 MCP 2026-07-28、Python mcp 2.3.0、TypeScript @modelcontextprotocol/client 与 @modelcontextprotocol/server 2.3.1。Server 提供 `/mcp` 和71项工具（42项原查询、12项计算、13项预览式写入、4项计算任务管理）；Agent 将其投影合并到本地 MCP，由 OpenCode 原生发现为 `mcp_tjwater_*`，普通项目会话、调查主 Agent 和已验证的直接只读子 Agent 共享42项查询；12项计算仅向非reader会话开放，13项写入仅向项目成员的普通聊天开放。个人会话不能获取项目凭据。旧 CLI 只作为能力清单，没有运行时 CLI 桥接。
 
 `run_simulation` 是在线重算，复用 REST 在线计算服务而不复制数据库；按水力步长更新模型条件和实时基线，不创建正式方案。MCP 时长为秒，原 REST 时长仍为分钟。其 annotation 为 `readOnlyHint=false`、`destructiveHint=true`；只读 reader 不能调用，其他方案模拟继续使用隔离临时模型。详细边界见 [MCP 分析工具说明](mcp-analysis-tools.md)。
 
-计算工具支持Agent按次指定 `timeout_seconds`，默认120秒；计算上限默认1小时，由 `TJWATER_MCP_MAX_CALCULATION_TIMEOUT_SECONDS` 配置并在工具schema公布。Agent、OpenCode和本地HTTP调用预算随目录同步，普通查询仍保持原有短超时。等待超时不意味着底层线程已停止或在线基线已回滚，禁止自动重试整个写入窗口。
+长计算使用持久化队列，原工具快速返回任务编号，新增状态、结果、列表和取消工具。HTTP 返回预算由目录元数据和宿主管理，Server 默认120秒、Agent默认180秒；模型不再提供 `timeout_seconds`，Server保留兼容入口。后台执行时限由管理员独立配置。提交和读取保持短调用，不延长上游5分钟空闲超时。独立计算Worker、迁移与权限边界见[后台计算队列](calculation-queue.md)。
 
 ## 认证与数据边界
 
@@ -14,13 +14,13 @@ Agent 根据401 challenge 或路径/根 well-known 发现受保护资源元数
 
 调查数据返回前，在短事务内追加现有任务证据并保存调用回执。同一主体、会话、任务、run 和原生 call ID 的重试复用证据；不同参数或失效范围明确报错。七类原有访问申请 scope 保持原契约，MCP 的新增来源只扩充快照来源枚举。没有新增报告发布、审批、控制、Shell 或递归委派权限。
 
-普通查询默认20条；目录类最多500条，时序最多10000条、20个标识、24小时；offset 最大10000。Agent 首次优先单对象、必要的短时间窗和小页，仅在证据不足时按需扩大，不自动遍历全网。SQL 有 limit/offset 和 statement_timeout。历史对比在同一条SQL中按序列、时间对数据点分页，返回 page.has_more；非零offset也标明覆盖不完整。历史曲线展示调用初始每页500点，`RESULT_TOO_LARGE` 时原位减半重试，最多读取1000条、发出16次请求，共用120秒超时（单次普通查询仍为30秒），图表最多100个有效点；这不改变模型首次查询默认值或时序接口上限。
+普通查询默认20条；目录类最多500条，时序最多10000条、20个标识、24小时；offset 最大10000。Agent 首次优先单对象、必要的短时间窗和小页，仅在证据不足时按需扩大，不自动遍历全网。SQL 有 limit/offset 和 statement_timeout。历史对比在同一条SQL中按序列、时间对数据点分页，返回 page.has_more；非零offset也标明覆盖不完整。历史曲线展示调用初始每页500点，`RESULT_TOO_LARGE` 时原位减半重试，最多读取1000条、发出16次请求，共用120秒超时（单次请求同时受该剩余总预算约束），图表最多100个有效点；这不改变模型首次查询默认值或时序接口上限。
 
 单份结果超过8MiB报错；超过12KB的调查工具消息只返回证据资源引用，完整快照仍保存，Server 的 `resources/read` 可以读取它。Agent 本地 MCP 只提供工具，将上游资源引用转为 `evidence_read` 中的 `mcp_tjwater_get_task_evidence` 参数，不向 OpenCode 泄露不可在本地读取的链接。`get_task_evidence` 在数据库中按 data_path、limit、offset 选页，嵌套集合和超过256字符的字符串返回路径描述；字符串按 Unicode 字符分页。页和完整资源均在数据库返回前检查8MiB上限，保留原证据编号和合成来源；普通查询要求缩小范围。其他结果仍保留执行后的整体8MiB检查，不能把它理解为所有查询均有数据库端字节限制。原始/清洗实测、模型基线、分析结果、比较和 synthetic 均明确标记。非有限数值转为 null 并记录数据缺口。
 
 `get_task_context` 默认 section=summary，只查固定元数据、160字符任务标题及200字符结论预览，并返回 available_sections；preview_only 和 truncated 明确说明不是完整报告。按 section=task/report/evidence/actions/progress/access_requests/plan/run 读取详情，report_revision 保留选择旧报告的语义。data_path 选择嵌套值，limit 默认20、最多500，offset 最大10000；证据、动作和授权申请集合以稳定 ID 为键，按 ID 字典序分页。字符串按字符分页。详情在数据库返回前检查8KB预算，超限缩页或选更深路径。缺少可选进度或授权申请表时返回 metadata.available=false，不冒充空事实。绑定权限范围只读取七类既有 scope，不再加载整个任务上下文。不增加工具数量、数据库表或迁移，也不改变原 REST 完整上下文契约。
 
-外围 MCP 直接读取所选项目 BizDB 的样板事实和既有时序。它不执行外围样板 HTTP 的故障场景注入；混有合成记录的结果按 synthetic 保守标记。DEMO-* 观测在数据库端拒绝超过8MiB的完整样本，预算内仍需读全样本、核验摘要后筛选并保留 synthetic；此限制不能通过缩小查询窗口解决。监测点列表沿用设备目录，不把未登记的合成点冒充实测设备。
+外围 MCP 直接读取所选项目 BizDB 的工单事实和既有时序。外围工单统一返回 `system_data`；测试数据和 DEMO-* 观测仍按各自数据来源处理。DEMO-* 观测在数据库端拒绝超过8MiB的完整样本，预算内仍需读全样本、核验摘要后筛选；此限制不能通过缩小查询窗口解决。监测点列表沿用设备目录，不把未登记的测试点冒充实测设备。
 
 ## 启用步骤
 

docs/node-coverage-tool.md

@@ -0,0 +1,43 @@
+# 节点空间覆盖工具
+
+`generate_node_coverage` 增加 Agent 可选择的空间表达能力，将已取得的模型节点集合转为
+可解释、可复核的覆盖示意。它可以在项目聊天、任务协作和后台调查中使用，
+不要求爆管任务、隔离成功或特定报告。生成、发布展示和业务分析是三个独立操作。
+
+## 适用范围
+
+| 问题 | 输入依据 | 输出含义 |
+| --- | --- | --- |
+| 候选对象分布在哪里 | 已核实的候选模型节点 | 候选节点空间包络 |
+| 两组监测点覆盖有何不同 | 已核对的设备到模型节点映射 | 已接入节点分布比较，不证明监测可观测性 |
+| 区域成员分布在哪里 | 已查询的区域成员节点 | 成员节点包络，不替代已有行政区或 DMA 边界 |
+| 哪些节点属于模型影响集合 | 已有拓扑或模拟结果中的节点集合 | 结果节点的空间示意，不重新计算影响 |
+| 投诉关联的节点如何分布 | 已解析并核对的投诉到模型节点关联 | 已关联节点分布，不证明投诉或污染真实边界 |
+
+只有地址或未关联设备时，先查询位置和资产关联；已有权威区域几何时优先展示原始边界。
+本工具不接收任意坐标、HTML、SQL、脚本或调用者指定的项目身份，不执行关阀、模拟或方案保存。
+
+## 参数和结果
+
+- `groups`：最多 8 组，每组提供唯一 `group_id`、`title` 和 `node_ids`；组之间可重叠，
+  合计最多 10000 个不同节点。不按组名自动赋予隔离、污染或服务覆盖语义。
+- `concave_ratio`：0–1，默认 0.25；越接近 1 越接近凸包。分散的点取单一包络可能包含较大空白，
+  应按实际比较目标分组，并明确示意含义。
+- `buffer_m`：大于 0、最多 1000 m，默认 25 m，仅是展示缓冲。
+- `split_by_connectivity`：默认 false。true 时按所选节点的模型连接拆分，并排除初始 CLOSED 连接。
+  `excluded_link_ids` 可额外指定最多 200 个断开连接，适用于已有关阀建议等有依据的绘图情景。
+  这不会修改模型，也不能替代完整拓扑分析。
+
+`result.map_payload` 是多组共存的 GeoJSON；`result.coverage` 记录算法参数、节点 ID、
+已映射数量、缺失几何节点、分组数量和 `display_only` 边界。缺少部分位置时返回 `partial`，
+不补造坐标。单点、重合点或共线点通过米制缓冲生成可绘制面。
+
+## 与展示工具配合
+
+生成工具不会移动地图。项目聊天可用 `present_view` 展示本次返回的几何；后台调查用
+`publish_task_presentation` 的 `set_overlay`，引用生成结果自己的 `evidence_id` 并省略 `data`，
+Server 从不可变证据解析完整几何。大结果内联缩略时，`data.presentation.command` 保留可发布命令。
+`recorded` 仍只表示请求已记录，浏览器显示和后台调查完成各自独立。
+
+本次爆管报告只是一个验证样例：可将 24 个直接隔离节点、295 个供水影响节点作为两组，
+按已有三阀建议设置连通分组；不将这种调用固定进 `analyze_valve_isolation`。

docs/operational-task-workers.md

@@ -72,8 +72,8 @@ OPERATIONAL_WORKER_AGENT_URL=http://127.0.0.1:8787
 OPERATIONAL_WORKER_TOKEN_URL=https://<identity-host>/realms/<realm>/protocol/openid-connect/token
 OPERATIONAL_WORKER_CLIENT_ID=<dedicated-service-client>
 OPERATIONAL_WORKER_CLIENT_SECRET=<inject-securely>
-OPERATIONAL_ACTION_PERIPHERAL_URL=http://127.0.0.1:8011
-OPERATIONAL_DATA_PERIPHERAL_URL=http://127.0.0.1:8011
+OPERATIONAL_ACTION_PERIPHERAL_URL=http://127.0.0.1:8000/peripheral
+OPERATIONAL_DATA_PERIPHERAL_URL=http://127.0.0.1:8000/peripheral
 ```
 
 选择 OAuth 时不要同时配置 `OPERATIONAL_WORKER_BEARER`，否则固定 Bearer 优先。不要配置设备命令网关进行本次合成联调。
@@ -109,8 +109,9 @@ Server 校验该会话属于当前活动运行。跨用户投递查询与任务
 原对话用户可以确认收到的回复，但不能改写目标会话或原生消息 ID。
 
 既有 worker 每轮最多对账四页、每页 50 条私有待投递消息，即使调查并发已满仍继续对账，
-不增加进程或配置项。临时网络错误保留消息等待重试；权限拒绝、不存在等确定性
-拒绝保存失败结果。消息正文、发送人和会话归属不写入项目事件日志。
+不增加进程或配置项。临时网络错误保留消息等待重试；400、403、410、422 等确定性
+拒绝保存失败结果。404 可能表示目标会话属于另一个 Agent 实例，因此保持待投递，
+不把路由未命中误记为消息终态失败。消息正文、发送人和会话归属不写入项目事件日志。
 待投递接口按创建时间与消息 ID 返回 keyset 游标，worker 跨轮保留扫描位置；
 扫到末页后下一轮从头重试。长期空闲对话的待收消息不会一直占住首屏、阻止后续调查投递。
 
@@ -122,6 +123,8 @@ Server 校验该会话属于当前活动运行。跨用户投递查询与任务
 
 `resources/sql/015_workbench_events_and_task_runs.sql` 在项目 BizDB 中增加 `domain_event_log` 和 `operational_task_runs`。后续迁移默认只应用到 `tjwater_v2`，扩展规则见 [`workbench-database-rollout.md`](workbench-database-rollout.md)。事件与业务投影在同一事务提交，`LISTEN/NOTIFY` 只负责唤醒；浏览器与 worker 始终按持久游标补读。每个 API 进程对活跃项目最多保留一条 LISTEN 连接，无订阅者 60 秒后关闭。
 
+`resources/sql/030_operational_run_agent_ownership.sql` 为运行增加 `agent_instance_id`。Worker 从目标 Agent 的 `/health` 读取实例标识，只恢复、续期该实例拥有的本地会话，并在投递时要求 Agent 回显同一标识。Server 在领取、心跳和完成时保存或校验归属，避免多个独立 Agent 共用项目队列时，一个实例反复刷新另一个实例并不存在的 session。部署顺序是：先迁移各目标 BizDB，再发布 Server、Agent 与 Worker；每个 Agent 的 `AGENT_INSTANCE_ID` 必须唯一且跨重启稳定。旧运行的归属保持为空，不会由任意 Worker 猜测接管。
+
 事件默认保留 7 天。清理脚本会从 Metadata DB 解析项目路由，并在发现多个项目共享同一 BizDB 时拒绝运行。先确认 `OPERATIONAL_WORKER_PROJECT_ID` 对应 `tjwater_v2`，然后运行定向 dry-run：
 
 ```bash
@@ -138,7 +141,7 @@ conda run -n server python scripts/cleanup_workbench_events.py \
 
 禁止省略 `--project-id` 执行清理。数据库维护脚本同样遵循先在 `tjwater_v2` 验证、再逐库扩展的规则。
 
-浏览器使用 `GET /api/v1/workbench/sync-snapshot` 建立原子快照，并以该游标连接 `GET /api/v1/workbench/events`。游标早于保留窗口、超前或对应事件已全部清理时返回 `reset_required`，客户端必须重建快照。SSE 需要 Bearer 与 `X-Project-Id`，且 `CORS_ALLOWED_ORIGINS` 必须配置为明确来源。
+浏览器使用 `GET /api/v1/workbench/sync-snapshot` 建立原子快照，并以该游标连接 `GET /api/v1/workbench/events`。游标早于保留窗口、超前或对应事件已全部清理时返回 `reset_required`，客户端必须重建快照。SSE 需要 Bearer 与 `X-Project-Id`，且 `CORS_ALLOWED_ORIGINS` 必须配置为明确来源，不允许 `*`。配置说明见 [README](../README.md#环境配置)。
 
 ## 外围样板工单同步
 
@@ -147,11 +150,11 @@ conda run -n server python scripts/sync_peripheral_work_orders.py --once
 conda run -n server python scripts/sync_peripheral_work_orders.py
 ```
 
-使用相同工作身份，并配置 `OPERATIONAL_ACTION_PERIPHERAL_URL`。同步器订阅源工单变更事件，收到消息即读取分页工单，将外围版本投影到 `external_work_order` 任务；默认每 300 秒兜底对账。合成来源在摘要和证据中明确标注；工单完成只表示该外部任务完成，不代表其他调查自动完成。事件协议、`tjwater_v2` 范围与启用步骤见 [外围工单主动同步](peripheral-work-order-events.md)。
+使用相同工作身份，并配置 `OPERATIONAL_ACTION_PERIPHERAL_URL`。同步器订阅源工单变更事件，收到消息即读取分页工单，将外围版本按系统数据投影到 `external_work_order` 任务；默认每 300 秒兜底对账。工单完成只表示该外部任务完成，不代表其他调查自动完成。事件协议、`tjwater_v2` 范围与启用步骤见 [外围工单主动同步](peripheral-work-order-events.md)。
 
 工况任务面板的刷新按钮只重新读取当前项目的工作台快照，不直接触发外围系统全量对账。外围工单由独立同步器使用服务身份响应后端事件，避免多标签页或普通用户刷新放大全量扫描，也避免浏览器项目与同步器固定项目不一致。
 
-通用合成检查工单不属于预约任务。早期种子脚本创建的通信、压力和零流量检查工单已经退役，种子脚本不再恢复它们；同步器按不可编辑的专用 seed actor 跳过旧种子工单。投诉、爆管事件以及调查动作实际创建的现场工单继续同步。`018_remove_synthetic_check_task_projections.sql` 只清理专用 seed actor 创建的旧检查投影，不按用户可编辑的标题或状态删除任务。三张独立演示工单已迁入外围样板，由同一同步器维护。
+早期种子脚本创建的通信、压力和零流量检查工单已经退役，种子脚本不再恢复它们；`018_remove_synthetic_check_task_projections.sql` 清理了当时错误生成的预约投影。当前同步器投影所有有效外围工单，`029_peripheral_work_orders_system_data.sql` 将既有工单及任务快照统一为系统数据。投诉、爆管事件以及调查动作实际创建的现场工单继续同步。
 
 水质投诉和爆管上报类型的外围工单同时创建关联调查，即使当时没有浏览器打开也会入队。两类事件从外围样板测试面板模拟上报；工况页负责展示调查。调查中创建的外围工单通过回执编号关联，后续反馈送回原调查，由 Agent 重新分析。
 
@@ -165,7 +168,7 @@ conda run -n server python scripts/sync_peripheral_work_orders.py
 
 ## 人工确认后的动作
 
-- `OPERATIONAL_ACTION_PERIPHERAL_URL`：使用既有 `/api/sample/v1/work-orders` 创建合成外围工单。成功回执说明工单已创建，不表示现场工作已经完成。
+- `OPERATIONAL_ACTION_PERIPHERAL_URL`：使用既有 `/api/sample/v1/work-orders` 创建外围工单。成功回执说明工单已创建，不表示现场工作已经完成。
 - `OPERATIONAL_ACTION_COMMAND_GATEWAY_URL`：可选设备命令网关。未配置时明确返回不支持执行，不能生成成功回执。
 - `OPERATIONAL_ACTION_TIMEOUT_SECONDS`：外部请求超时，默认 10 秒。
 

docs/peripheral-sample.md

@@ -1,8 +1,12 @@
-# 外围样板服务
+# 外围系统接入服务
 
-该独立 FastAPI 服务提供历史监测回放、固定合成数据，以及 synthetic 测试工单和测试数据的版本化 CRUD。管理入口位于供水工作台“外围样板”。它不控制设备，也不修改正式业务资产。
+该 FastAPI 子应用接入真实业务工单，提供版本化 CRUD、处理反馈和事件调查入口，同时保留独立的历史监测回放与演练数据接口。生产环境挂载在主 Server 的 `/peripheral`，随 `next-backend` 单容器发布；源码仍可独立启动用于本地隔离调试。管理入口位于供水工作台“外围系统”。
 
-浏览器通过 `TJWATER_PERIPHERAL_SAMPLE_API_BASE_URL` 连接样板服务。服务复用主系统的 Bearer 身份和 `X-Project-Id`，读取需要 `simulation.view`，写入需要 `simulation.run`。
+外围工单的 `origin` 固定为 `system_data`，前端录入、同步器、任务快照、执行回执和工单查询均使用该来源，不再提供演练来源分支。业务录入代表系统中的上报，不代表投诉内容或故障原因已核实，Agent 仍需查询实际证据。
+
+此次变更不修改表结构；部署时需在每个项目业务库执行 `resources/sql/029_peripheral_work_orders_system_data.sql`，统一修正源工单、修订记录、任务快照和外围反馈证据。API 路径、配置键中的 `sample` 为兼容名称，不代表工单是模拟数据。更新需发布 Server、同步器、Agent 与前端对应版本。
+
+浏览器通过 `TJWATER_PERIPHERAL_SAMPLE_API_BASE_URL` 连接外围服务。服务复用主系统的 Bearer 身份和 `X-Project-Id`，读取需要 `simulation.view`，写入需要 `simulation.run`。权限键沿用现有契约，不用于判断数据来源。
 
 生产环境不需要为外围服务新增域名。前端设置
 `TJWATER_PERIPHERAL_SAMPLE_API_BASE_URL=/peripheral`，在现有前端 NPMplus Proxy Host 中增加以下
@@ -10,17 +14,16 @@ Custom Location：
 
 | 配置项 | 值 |
 | --- | --- |
-| Location | `/peripheral/`，必须保留末尾斜杠 |
+| Location | `/peripheral`，不添加末尾斜杠 |
 | Scheme | `http` |
-| Forward Hostname / IP | NPMplus 能访问的应用机地址，通常与现有前端上游相同 |
-| Forward Port | `8011` |
+| Forward Hostname / IP | `192.168.1.114`，不追加路径或末尾斜杠 |
+| Forward Port | `8001` |
 | Forward Path | 留空 |
 
-NPMplus 会去掉 `/peripheral/` 前缀，因此浏览器请求
-`/peripheral/api/sample/v1/context` 时，上游收到 `/api/sample/v1/context`。不要在 Forward
-Hostname 后再追加 `/api/sample/v1`。外围服务应只向 NPMplus 所在网络开放 `8011`，不要直接开放公网。
-本机开发和预览模式也提供等价的 `/peripheral` 代理，可通过
-`TJWATER_PERIPHERAL_SAMPLE_UPSTREAM` 覆盖默认上游 `http://127.0.0.1:8011`。
+容器内部仍监听 `8000`，生产发布在宿主机映射为 `8001`，避免与已有服务冲突。该配置保留 `/peripheral` 前缀，因此浏览器请求
+`/peripheral/api/sample/v1/context` 时，主 Server 收到相同路径并交给挂载的子应用。不要在
+Forward Hostname 后追加 `/api/sample/v1`。本机开发和预览模式也将 `/peripheral` 代理到
+`TJWATER_SERVER_API_BASE_URL`，默认是 `http://127.0.0.1:8000`。
 
 外围 API 支持已授权项目切换。每次请求先由主系统验证项目成员身份和权限，再从 Metadata DB 解析该项目的 `biz_data` 路由；历史观测另解析 `iot_data` 路由。逻辑项目代码、业务物理库名和时序物理库名可以不同，连接沿用各自完整 DSN。路由缺失返回 `503 PROJECT_STORAGE_NOT_CONFIGURED`，不会回退到其他项目数据库。未配置存储表的项目仍需单独完成结构部署；API 不自动创建表或复制演示数据。
 
@@ -38,8 +41,8 @@ Hostname 后再追加 `/api/sample/v1`。外围服务应只向 NPMplus 所在网
 
 演示数据通过 `scripts/seed_peripheral_sample.py` 写入 `tjwater_v2` 的 `peripheral_sample` schema。当前固定窗口为 2026-09-17，所有记录标记为 `synthetic`，并使用 `DEMO-` 前缀。
 
-## 水质投诉 Agent 演示
+## 工单事件调查
 
-`POST /api/sample/v1/work-orders` 可创建 `work_order_type=water_quality_complaint` 的合成投诉事件，或 `work_order_type=burst_report` 的合成爆管上报。二者由外围同步器转换为统一事件调查任务，再由后端 worker 可靠交给 Agent；浏览器不再负责启动调查。事件和执行请求 ID 均由稳定业务身份确定，回执恢复或重试不会重复创建任务。
+`POST /api/sample/v1/work-orders` 可创建 `work_order_type=water_quality_complaint` 的水质投诉事件，或 `work_order_type=burst_report` 的爆管上报，均默认使用业务来源。二者由外围同步器转换为统一事件调查任务，再由后端 worker 可靠交给 Agent；浏览器不再负责启动调查。事件和执行请求 ID 均由稳定业务身份确定，回执恢复或重试不会重复创建任务。普通业务录入无需绑定演练场景码，也不会向 Agent 注入预设监测结果或邻区投诉情况。
 
-样板服务只保存工单，不直接调用 Agent。`scripts/sync_peripheral_work_orders.py` 使用受认证的后台身份同步外部版本并创建调查任务，`scripts/run_operational_worker.py` 再将持久化的待执行任务交给 Agent，因此浏览器关闭后流程仍可继续。
+外围服务只保存工单，不直接调用 Agent。`scripts/sync_peripheral_work_orders.py` 使用受认证的后台身份同步外部版本并创建调查任务，`scripts/run_operational_worker.py` 再将持久化的待执行任务交给 Agent，因此浏览器关闭后流程仍可继续。

infra/docker/docker-compose.yml

@@ -21,6 +21,26 @@ services:
       - timescaledb
       - postgis
 
+  # Enable after provisioning calculation storage and explicit project scopes.
+  calculation-worker:
+    profiles: [calculations]
+    image: ${TJWATER_SERVER_IMAGE:-tjwater-server:local}
+    init: true
+    restart: on-failure
+    command: ["python", "scripts/run_calculation_worker.py"]
+    env_file:
+      - ../../.env
+    environment:
+      - PYTHONPATH=/app
+    volumes:
+      - ../../app:/app/app
+      - ../../resources:/app/resources
+      - calculation-state:/app/temp/calculation-jobs
+    stop_grace_period: 30s
+    depends_on:
+      - timescaledb
+      - postgis
+
   # ==========================================
   # Infrastructure Services
   # ==========================================
@@ -127,6 +147,9 @@ services:
     volumes:
       - ./mapservice/geoserver_data:/opt/geoserver_data
 
+volumes:
+  calculation-state:
+
 networks:
   default:
     driver: bridge

infra/systemd/tjwater-calculation-worker.service

@@ -0,0 +1,17 @@
+[Unit]
+Description=TJWater queued calculation worker
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+WorkingDirectory=%h/NextTJWater/next-tjwater-server
+ExecStart=%h/miniconda3/envs/server/bin/python scripts/run_calculation_worker.py
+Restart=on-failure
+RestartSec=5
+KillMode=control-group
+TimeoutStopSec=30
+UMask=0077
+
+[Install]
+WantedBy=default.target

resources/sql/015_workbench_events_and_task_runs.sql

@@ -48,6 +48,7 @@ CREATE TABLE IF NOT EXISTS workbench.operational_task_runs (
     )),
     revision bigint NOT NULL CHECK (revision >= 1),
     agent_session_id text,
+    agent_instance_id text,
     lease_expires_at timestamptz,
     heartbeat_at timestamptz,
     started_at timestamptz,

resources/sql/026_calculation_jobs.sql

@@ -0,0 +1,23 @@
+-- Apply independently to each project's business database.
+CREATE SCHEMA IF NOT EXISTS workbench;
+CREATE TABLE IF NOT EXISTS workbench.calculation_jobs (
+    job_id uuid PRIMARY KEY,
+    project_id uuid NOT NULL,
+    owner_user_id uuid NOT NULL,
+    request_key text NOT NULL,
+    tool text NOT NULL,
+    parameters jsonb NOT NULL,
+    binding jsonb NOT NULL,
+    timeout_seconds integer NOT NULL CHECK (timeout_seconds BETWEEN 1 AND 86400),
+    status text NOT NULL DEFAULT 'queued' CHECK (status IN
+      ('queued','running','cancel_requested','succeeded','failed','cancelled','timed_out','interrupted')),
+    result jsonb,
+    error jsonb,
+    lease_generation bigint NOT NULL DEFAULT 0,
+    lease_owner text,
+    lease_expires_at timestamptz,
+    created_at timestamptz NOT NULL DEFAULT clock_timestamp(),
+    updated_at timestamptz NOT NULL DEFAULT clock_timestamp(),
+    UNIQUE(project_id,owner_user_id,request_key)
+);
+CREATE INDEX IF NOT EXISTS calculation_jobs_queue ON workbench.calculation_jobs(project_id,status,created_at);

resources/sql/027_calculation_job_lifecycle.sql

@@ -0,0 +1,21 @@
+-- Apply after 026, independently to every project business database.
+-- Preserve job identity, deduplication and audit metadata, including legacy results.
+ALTER TABLE workbench.calculation_jobs
+    ADD COLUMN IF NOT EXISTS started_at timestamptz,
+    ADD COLUMN IF NOT EXISTS finished_at timestamptz,
+    ADD COLUMN IF NOT EXISTS queue_expires_at timestamptz,
+    ADD COLUMN IF NOT EXISTS result_expires_at timestamptz;
+UPDATE workbench.calculation_jobs SET queue_expires_at=created_at+interval '1 hour'
+WHERE queue_expires_at IS NULL;
+UPDATE workbench.calculation_jobs SET finished_at=updated_at
+WHERE finished_at IS NULL AND status IN ('succeeded','failed','cancelled','timed_out','interrupted');
+-- Give pre-deployment results a complete retention window, even on reruns.
+UPDATE workbench.calculation_jobs SET result_expires_at=clock_timestamp()+interval '7 days'
+WHERE result IS NOT NULL AND result_expires_at IS NULL;
+ALTER TABLE workbench.calculation_jobs
+    ALTER COLUMN queue_expires_at SET DEFAULT (clock_timestamp()+interval '1 hour'),
+    ALTER COLUMN queue_expires_at SET NOT NULL;
+CREATE INDEX IF NOT EXISTS calculation_jobs_queue_expiry
+ON workbench.calculation_jobs(project_id,queue_expires_at) WHERE status='queued';
+CREATE INDEX IF NOT EXISTS calculation_jobs_result_expiry
+ON workbench.calculation_jobs(project_id,result_expires_at) WHERE result IS NOT NULL;

resources/sql/028_task_questions.sql

@@ -0,0 +1,15 @@
+-- Apply to every project business database before enabling task questions.
+CREATE TABLE IF NOT EXISTS workbench.operational_task_questions (
+    project_id uuid NOT NULL,
+    task_id uuid NOT NULL,
+    question_request_id uuid NOT NULL,
+    document jsonb NOT NULL,
+    PRIMARY KEY (project_id, question_request_id),
+    FOREIGN KEY (project_id, task_id)
+        REFERENCES workbench.operational_tasks(project_id, task_id) ON DELETE CASCADE
+);
+CREATE INDEX IF NOT EXISTS operational_task_questions_task
+    ON workbench.operational_task_questions(project_id, task_id);
+CREATE UNIQUE INDEX IF NOT EXISTS operational_task_questions_pending
+    ON workbench.operational_task_questions(project_id, task_id)
+    WHERE document->>'status' = 'pending';

resources/sql/029_peripheral_work_orders_system_data.sql

@@ -0,0 +1,141 @@
+-- Peripheral work orders are business-system records. Remove the former
+-- synthetic provenance branch from source rows and their task projections.
+
+BEGIN;
+SET LOCAL lock_timeout = '5s';
+
+CREATE FUNCTION pg_temp.normalize_peripheral_work_order_text(value jsonb)
+RETURNS jsonb
+LANGUAGE plpgsql
+IMMUTABLE
+AS $$
+DECLARE
+    normalized text := value::text;
+BEGIN
+    normalized := replace(normalized, '对应外围记录明确标为 synthetic（系统测试）', '对应外围记录为 system_data 系统数据');
+    normalized := replace(normalized, '有一条演练上报', '有一条系统上报');
+    normalized := replace(normalized, '证据覆盖为 synthetic 上报工单快照', '证据覆盖为 system_data 上报工单快照');
+    normalized := replace(normalized, '记录快照带 synthetic 标识', '记录快照为 system_data');
+    normalized := replace(normalized, '外围工单结果注明为synthetic演练数据', '外围工单为system_data系统数据');
+    normalized := replace(normalized, '工单为synthetic演练', '工单为system_data系统数据');
+    normalized := replace(normalized, 'synthetic 爆管上报记录', '系统爆管上报记录');
+    normalized := replace(normalized, '合成上报记录', '系统上报记录');
+    normalized := replace(normalized, '外围模拟工单', '外围系统工单');
+    normalized := replace(normalized, '外围记录为synthetic', '外围记录为system_data');
+    normalized := replace(normalized, '证据快照整体为synthetic演练来源', '工单快照为system_data系统来源');
+    normalized := replace(normalized, '来源标记为synthetic，仅能作为演练资料', '来源为system_data系统数据');
+    normalized := replace(normalized, 'synthetic演练记录', '系统记录');
+    normalized := replace(normalized, 'synthetic演练来源', 'system_data系统来源');
+    normalized := replace(normalized, 'synthetic 演练来源', 'system_data 系统来源');
+    normalized := replace(normalized, '外围样板合成摘要', '外围工单摘要');
+    normalized := replace(normalized, '合成外围演练', '外围系统记录');
+    normalized := replace(normalized, '合成外围', '外围');
+    normalized := replace(normalized, '合成联调', '系统联调');
+    normalized := replace(normalized, '合成演练', '系统测试');
+    normalized := replace(normalized, '合成事件', '系统测试事件');
+    RETURN normalized::jsonb;
+END
+$$;
+
+UPDATE peripheral_sample.work_orders
+SET document = pg_temp.normalize_peripheral_work_order_text(
+    jsonb_set(document, '{origin}', '"system_data"'::jsonb, true)
+)
+WHERE document->>'origin' IS DISTINCT FROM 'system_data'
+   OR document::text ~* '(synthetic.*演练|synthetic 爆管上报|外围记录为synthetic|外围样板合成摘要|外围模拟工单|合成上报记录|合成外围|合成联调|合成事件)';
+
+UPDATE peripheral_sample.work_order_revisions
+SET document = pg_temp.normalize_peripheral_work_order_text(
+    jsonb_set(document, '{origin}', '"system_data"'::jsonb, true)
+)
+WHERE document->>'origin' IS DISTINCT FROM 'system_data'
+   OR document::text ~* '(synthetic.*演练|synthetic 爆管上报|外围记录为synthetic|外围样板合成摘要|外围模拟工单|合成上报记录|合成外围|合成联调|合成事件)';
+
+UPDATE peripheral_sample.requests
+SET result = pg_temp.normalize_peripheral_work_order_text(
+    jsonb_set(result, '{origin}', '"system_data"'::jsonb, true)
+)
+WHERE command_key IN ('create_work_order', 'update_work_order')
+  AND jsonb_typeof(result) = 'object'
+  AND result ? 'work_order_id'
+  AND (
+      result->>'origin' IS DISTINCT FROM 'system_data'
+      OR result::text ~* '(synthetic.*演练|synthetic 爆管上报|外围记录为synthetic|外围样板合成摘要|外围模拟工单|合成上报记录|合成外围|合成联调|合成事件)'
+  );
+
+UPDATE workbench.operational_task_external_feedback
+SET document = pg_temp.normalize_peripheral_work_order_text(jsonb_set(
+    jsonb_set(
+        jsonb_set(
+            jsonb_set(
+                document,
+                '{work_order_snapshot,origin}',
+                '"system_data"'::jsonb,
+                false
+            ),
+            '{evidence,kind}',
+            '"external_feedback"'::jsonb,
+            false
+        ),
+        '{evidence,source_label}',
+        '"外围工单"'::jsonb,
+        false
+    ),
+    '{detail}',
+    to_jsonb(replace(COALESCE(document->>'detail', ''), '合成外围工单', '外围工单')),
+    false
+))
+WHERE document ? 'work_order_snapshot';
+
+UPDATE workbench.operational_task_evidence
+SET document = pg_temp.normalize_peripheral_work_order_text(jsonb_set(
+    jsonb_set(
+        document,
+        '{kind}',
+        '"external_feedback"'::jsonb,
+        false
+    ),
+    '{source_label}',
+    '"外围工单"'::jsonb,
+    false
+))
+WHERE document->>'source_ref' LIKE 'peripheral_sample:%'
+   OR evidence_id LIKE 'provenance-correction:peripheral_sample:%';
+
+UPDATE workbench.operational_tasks AS task
+SET document = pg_temp.normalize_peripheral_work_order_text(jsonb_set(
+    task.document,
+    '{summary}',
+    to_jsonb(
+        replace(
+            replace(COALESCE(task.document->>'summary', ''), '合成外围工单', '外围工单'),
+            '合成外围事件',
+            '外围事件'
+        )
+    ),
+    false
+))
+FROM peripheral_sample.work_orders AS source_order
+WHERE task.project_id = source_order.project_id
+  AND (
+      task.document#>>'{source,external_id}' = source_order.work_order_id::text
+      OR task.document#>>'{source,external_id}' =
+          'peripheral_sample:' || source_order.work_order_id::text
+  );
+
+UPDATE workbench.operational_task_reports AS report
+SET document = pg_temp.normalize_peripheral_work_order_text(report.document)
+FROM workbench.operational_tasks AS task
+JOIN peripheral_sample.work_orders AS source_order
+  ON task.project_id = source_order.project_id
+ AND (
+     task.document#>>'{source,external_id}' = source_order.work_order_id::text
+     OR task.document#>>'{source,external_id}' =
+         'peripheral_sample:' || source_order.work_order_id::text
+ )
+WHERE report.project_id = task.project_id
+  AND report.task_id = task.task_id
+  AND report.document::text ~*
+      '(synthetic.*演练|synthetic 爆管上报|外围记录为synthetic|外围样板合成摘要|外围模拟工单|合成上报记录|合成外围|合成联调|合成事件)';
+
+COMMIT;

resources/sql/030_operational_run_agent_ownership.sql

@@ -0,0 +1,13 @@
+-- Bind durable task runs to the Agent instance that owns their local session.
+
+BEGIN;
+
+ALTER TABLE workbench.operational_task_runs
+    ADD COLUMN IF NOT EXISTS agent_instance_id text;
+
+CREATE INDEX IF NOT EXISTS operational_task_runs_agent_owner_idx
+    ON workbench.operational_task_runs(project_id,agent_instance_id,status,created_at,run_id)
+    WHERE agent_instance_id IS NOT NULL
+      AND status IN ('claimed','running');
+
+COMMIT;

resources/work-order-skills/appointment-follow-up/SKILL.md

@@ -1,16 +1,36 @@
 ---
 name: appointment-follow-up
-description: 调查关联预约任务的执行状态与反馈，核对审批、派发和现场结果，避免把预约任务当成新的爆管或投诉事件。
+description: 核查关联预约工单的审批、提交、现场执行和反馈，分别判断预约履约与原事件闭环。
 metadata:
-  version: "1.0.0"
+  version: "2.0.4"
 ---
 
 # 预约任务跟进
 
-预约工单承载已安排的人工工作。先读取原任务、已审批动作、预约时间、派发状态和现场反馈；需要时追溯关联的投诉或爆管事件。预约时间、工单派发时间、现场完成时间和事件发生时间分别说明。
+目标是说明预约安排了什么、流程到哪一步、现场实际做了什么、证据是否满足要求，以及还有哪些责任事项。预约工单承载已安排的人工工作，不自动代表新的爆管或水质事件，也不因存在待办而反复创建跟进工单。
 
-核对预约工作的内容、执行条件和实际回执。只有现场反馈与证据支持时，才能认定采样、通知、抢修、冲洗或复测完成；工单创建或派发成功仅代表流程状态变化。对尚未完成的工作，报告责任事项、已有进展、缺少的反馈和下一步依赖。
+## 接续与核查
 
-预约工单不自动触发隔离停水或污染水排放，也不生成同类型的投诉、爆管事件。确有新增处置需要且已有证据支持时，保存阶段报告，再提议具体的关联预约工单（work_order_type=general），供人工审批；先核查已有有效草稿，避免在跟进过程中递归建单。
+读取当前任务、关联工单、已有报告、证据、动作和最新反馈，核对预约时间、地点、负责人、工作内容及回执要求；需要时追溯关联的投诉或爆管。新一轮优先检查新增事实和旧未决项，保留本流程名称和版本。
 
-如发现新的爆管或水质风险，明确记录有依据的发现及待核验项，不擅自更改当前任务类型。审批、现场操作、业务重新排队和恢复供水均遵守现有 Server 流程。正式报告保留所用流程名称和版本，便于追溯。
+围绕原工作目标完成必要核查：
+
+1. 将要求与来源工单最新版本逐项对照，区分预约、派发、到场、完成和原事件发生时间，不用创建或更新时间替代现场时间，也不因预约已过推定已经执行或失约。
+2. 核对动作的审批、提交、执行状态和回执。`approved` 只表示允许提交，`submitted` 只表示已接收，`succeeded` 的建单动作只证明工单创建成功；这些状态均不证明采样、通知、抢修、冲洗或复测完成。`unknown` 由 Server 复核原执行请求，不重复下发。
+3. 用 `query_peripheral_data` 查询外围工单、点位和现场观察，按来源 ID、关联动作及版本核对；目录首屏不能证明没有反馈。区分缺失、重复、旧版本和相互冲突的记录。
+4. 按原目标查询位置、资产和工作前后的监测。地点和资产明确后展示正在核查的工作位置；取得监测对照后展示核查点位，并画出实际取得的工作前后趋势或对比图，核对现场发生时间与数据覆盖，不能把计划时间当实际执行分界；已有影响结果时展示其真实几何或节点覆盖示意。核实设备与模型要素、时间覆盖、单位和数据质量；缺测不是零值，局部正常不证明整项工作完成。
+5. 将现场描述、来源系统附件或照片说明、监测变化和原工作要求交叉核对。没有图像查看能力时不声称已检查照片内容。每项要求标为已核验、待核验或不适用并给出依据；状态与证据冲突时明确列出。
+
+## 结果与后续
+
+报告分别说明原预约要求、审批及派发事实、已核实的现场工作和效果、冲突或缺口、当前责任事项，以及它与原事件的关系。等待回执不等于调查完成，更不等于现场工作完成。
+
+位置、资产、监测和已有影响结果随核查步骤同步展示，遵循后台地图与图表规则，不用最终报告代替过程展示；单个数值或无可比数据时用文字、指标或表格，不为凑图新增业务分析。无数据或展示失败时说明具体限制，调查不等待浏览器。
+
+只有证据支持新的具体工作时才提议动作。先核对已有有效动作和关联工单，优先沿原工单的反馈与人工修改流程推进；不要因缺少一项反馈就新建催办或复查工单。新增人工工作使用关联的 `work_order_type=general` 草稿，写明地点、内容、依赖、执行前核验和回执要求。
+
+本流程不自动发起隔离停水、污染水排放或设备命令。发现新的爆管或水质风险时，列出证据、紧迫事项和建议，由原事件的合法调查继续专业分析；不要改写当前任务类型或递归创建同类事件。
+
+预约履约完成须有与原要求对应的现场记录、结果与时间、必要回执、适用的监测或复测证据，以及未决事项核对。由有权限人员在来源系统确认工单完成，核对其新版本后才能认定履约完成。
+
+预约完成不等于关联爆管或水质投诉闭环。报告须分别说明预约履约和原事件处置；只具备履约证据时，不宣称原事件风险已消除或已经结案。

resources/work-order-skills/burst-report/SKILL.md

@@ -1,26 +1,41 @@
 ---
 name: burst-report
-description: 调查供水爆管上报，形成隔离停水、用户通知、抢修和恢复供水的处置方案及待审批动作。
+description: 调查供水爆管上报，形成有依据的候选位置、隔离阀门、影响范围和后续处置建议。
 metadata:
-  version: "1.0.0"
+  version: "2.0.6"
 ---
 
 # 爆管上报处置
 
-依据上报位置、时间、现场描述、相关管段台账和同窗压力流量证据，区分上报事实、爆管候选、已确认故障与演练假设。完成初步风险判断不要求抢修、复测和回访已经发生；这些属于后续执行反馈。演练数据可支持明确标注的演练判断，不能证明生产故障。
+目标是形成有依据的候选位置、隔离方案、影响判断和可审批处置，并在后续轮次根据真实反馈复核事件闭环。区分候选方案、现场确诊和事件闭环；后两者不能提前阻断当前可完成的分析。演练数据可以用于比较，但必须标注 `synthetic`，不能证明生产故障。
 
-## 调查与处置路线
+## 接续与证据
 
-1. 核对上报对象与时间窗，查询相关管段、周边监测和既有处置反馈。先确定实际覆盖范围；目录首屏或查询错误不能证明全网没有异常。具备输入时使用爆管侦测或定位分析，候选结果仍需现场核验。
-2. 形成隔离停水方案。已确认目标管段时，按需分析隔离阀门、关阀影响、失供范围和供水保障。资产映射、阀门状态或影响范围不清时，明确待现场核验条件；不得编造阀门编号或直接生成无目标的设备命令。
-3. 安排用户通知和供水保障。草稿写明已知影响区域、拟停水时段或“待确认”、通知内容、通知渠道及责任岗位；关阀前通知与紧急处置后的补通知取决于现场紧迫性和获批方案。未知用户数量不填虚构值。通知草稿或内部调查消息均不能说成用户已收到通知。
-4. 安排现场定位、止水与抢修。说明施工位置、隔离前置条件、抢修目标和需要回传的现场事实。存在管内污染侵入风险时，再安排适用的清洗、消毒或冲洗评估，不能把每次爆管都判断为污染事件。
-5. 恢复供水前核验修复、压力、漏水复查及适用的水质复测结果，按获批方案恢复并通知用户；缺少现场反馈时保留“待执行／待核验”，不能宣称恢复完成。
+读取当前任务、事件时刻、地点、来源、已有报告、证据、动作和反馈。新一轮优先检查新增事实和上轮未决项，按时间、来源、对象及输入比较新旧结果，不按 ID 推断先后，也不因原工单未更新就断言没有新证据。保留本流程名称和版本。
 
-## 报告与动作草稿
+## 定位与影响分析
 
-报告说明当前判断、证据覆盖、隔离影响、通知安排、抢修及恢复条件。能够形成有边界的初步判断时，不因后续维修或复测尚未完成而把整个调查判为关键数据不足。确实缺少支撑当前判断的关键证据时，遵守 insufficient_data 契约，不强造结论或动作。
+根据现有线索自主选择必要步骤：
 
-在已有证据足以支持具体处置建议时，先保存阶段报告，再通过现有动作提议入口保存待人工审批的具体草稿。现场核验、停水通知、抢修等人工执行事项使用关联预约工单（work_order_type=general）；原爆管上报事件保持独立。草稿含位置、工作内容、先后依赖、执行前核验条件和回执要求，避免为每个图节点机械建单。没有实际客户通知工具时，将通知列为待执行的人工工单步骤。
+1. 核对已有管段或节点。只有地点描述时，先解析位置，再查询附近资产、属性和连接。地点明确后展示调查位置，候选资产取得后展示正在核查的管段或节点。空间邻近只支持候选，不能单独证明故障。
+2. 核对 SCADA 设备与模型节点映射，用 `query_scada_readings` 或 `query_element_history` 查询事件前后及正常对照。先用少量对象和合适时间窗，确认返回时间实际覆盖事件；必要时缩窗或翻页。同步展示核查点位，并将取得的事件、对照压力或流量序列画成过程曲线，分清指标和单位；保留缺测、真实零值和覆盖限制。
+3. 有异常观测时评估爆管定位输入。`run_burst_location` 必须有 `burst_leakage`（m³/s）；先查询已有估计及方法依据，或结合边界流量、用水和储量变化判断能否估计。不能把单表增量、压力下降或任意数值直接当漏量。
+4. 有依据的漏量估计或明确的情景假设可用于条件定位，须说明来源、方法和不确定性。`pressure_scada_ids` 使用模型节点 ID；监测模式读取 SCADA，模拟模式须引用已有分析运行。取得多个可比候选的评价值后，用柱状图展示实际排序及评价含义。压力点和历史数据满足工具要求时可选 `run_burst_detection`，但侦测不代替定位，评价值也不表述为概率。缺少定位参数时说明已查来源和影响，继续其他可独立完成的分析。
+5. 对已确认或有依据的候选资产使用 `analyze_valve_isolation` 评估阀门隔离，无需等待现场确诊。逐候选比较；多个事故 ID 表示联合场景。保留可隔离性、必关阀、可选阀及受影响节点。结果取得后展示直接隔离与供水影响节点的覆盖示意，形成或调整关阀方案时展示对应阀门；区分两种影响及必关、可选阀。未知阀门状态不能写成可用，不可隔离或空结果也不能解释为零影响；已知阀门失效时评估替代边界。
+6. 输入具备时用 `analyze_valve_closure` 分析关阀水力影响，核对阀门组合、起算时间、时长和输出对象。结果返回后，用曲线展示请求节点的压力、请求管段的流量；有同口径方案或基准时再画对比图，不自造前后变化。不同单位分图，标明局部模型结果。区分拓扑范围、水力影响和有依据的用户影响，不虚构户数，不用局部结果代表全网。
 
-有真实且可控制的设备目标、明确命令及充分依据时，才提议设备命令。审批和执行由 Server 控制，调查不直接关闭阀门或发送用户通知。复用已存在的有效草稿，结合实际回执继续调查，不重复提议同一动作。正式报告保留所用流程名称和版本，便于追溯。
+查询错误、权限拒绝或接口故障不是缺数证据。必要分析无法执行时，说明失败步骤和已经完成的部分，不借有限摘要宣布调查完成。
+
+## 处置与报告
+
+报告分别给出位置候选及依据、隔离阀门及可用性、替代方案、拓扑与水力影响、用户影响边界，以及处置和闭环缺口。只有初步结果时明确范围，不宣称已经定位、确诊或闭环。位置、候选、监测、影响和阀门方案按实际调查步骤同步展示，遵循后台展示规则，不用报告定位链接代替过程展示。覆盖面保留分组、缺失几何和示意边界，不能当作实际停水、水力影响或用户服务边界。无数据或展示失败时说明具体限制；展示不替代证据，也不要求等待浏览器。
+
+形成具体建议后，先保存阶段结论，再提议必要动作。人工核验、通知、定位隔离、抢修、适用的清洗复测和恢复回访使用关联的 `work_order_type=general` 草稿；原爆管事件保持独立。草稿写明位置、任务、依赖、执行前核验和回执要求。没有通知适配器时明确由人工实施。设备命令仅用于真实可控目标，审批和执行由 Server 控制。
+
+抢修或复测尚未发生不阻止有边界的初步方案。现场阀门状态应写为执行前核验条件；工单创建、审批或派发均不证明现场完成。
+
+## 闭环复核
+
+后续轮次核对实际故障资产和原因、隔离实施、抢修记录、适用的清洗与水质复测、恢复供水及恢复后监测、用户通知回访、未决动作和遗留风险。每项标明已核验、待核验或不适用及依据。
+
+证据齐全时报告具备结案条件，由来源系统的有权限人员确认关闭。原工单未结案时写明待确认，读取其新版本后才能认定闭环；误报结案也须有足够的排除证据和人工确认。

resources/work-order-skills/water-quality-complaint/SKILL.md

@@ -1,26 +1,41 @@
 ---
 name: water-quality-complaint
-description: 调查供水水质投诉，按证据和风险提出采样、隔离、用户通知、污染水处置及复测恢复方案。
+description: 调查供水水质投诉，核查投诉与监测覆盖，评估候选原因、影响和采样处置建议。
 metadata:
-  version: "1.0.0"
+  version: "2.0.4"
 ---
 
 # 水质投诉处置
 
-区分用户感官投诉、监测异常、已确认的水质问题与污染源假设。单个投诉不能直接证明供水管网污染；上游监测正常也不能排除末梢或二次供水问题。不预设异常必然来自二次供水，不把模拟扩散结果当成污染源反演。
+目标是说明投诉和异常的实际覆盖、可能原因及依据、候选影响范围、采样或处置建议和后续复核条件。区分感官投诉、实测异常、污染源假设和已确认问题；不预设异常来自上游或二次供水。演练结果须标注 `synthetic`，不能证明生产风险。
 
-## 调查与处置路线
+## 接续与证据
 
-1. 核对投诉时间、位置、异常描述与同类投诉，查询供水路径、相关水质和压力监测、适用的历史记录及现场采样反馈。分别说明实际覆盖的上游、末梢和二次供水范围，不混用模型基线与实测。
-2. 评估是否需要隔离停水及饮用水风险告知。已确认或有充分证据支持局部供水风险时，提出与风险相称的隔离范围和保障措施，核对阀门与影响区域。尚不能确定风险时，先提出有依据的现场核验或采样安排，不因一条投诉直接建议全网停水，也不承诺水可安全饮用。
-3. 将用户通知纳入处置方案：明确已知影响范围、需人工确认的用水提示、通知渠道、责任岗位和更新条件。没有实际发送工具时形成待审批的人工通知步骤，不把调查消息当成客户通知。
-4. 对已支持的污染水风险，安排排查来源、控制污染进入并评估受影响管段的冲洗或排放。仅在来源参数、目标管段或冲洗节点等必需输入已知时使用扩散或冲洗分析；参数不足不虚构源浓度、排放量或节点。具体排水位置、承接条件及现场操作要求需核验后审批执行，不默认污染水可任意排放。
-5. 清洗或冲洗后取得适用采样和复测反馈，再评估恢复供水与用户回访。现场处置、复测和恢复是下游执行步骤，其尚未发生不能单独作为拒绝初步风险分析的理由。
+读取当前任务、投诉时刻、位置、现象、来源版本、已有报告、证据、采样、动作和处置反馈。新一轮先核对新增事实和上轮未决项，保留本流程名称和版本。
 
-## 报告与动作草稿
+## 风险与工程分析
 
-报告列明当前风险判断、证据覆盖、隔离与通知条件、污染水处置路线和复测恢复条件。演练输入只支持标注为演练的判断。真正缺少支撑当前判断的关键证据时，遵守 insufficient_data 契约，不用通用处置模板强造已确认污染结论。
+根据当前证据选择必要步骤：
 
-已能形成有边界的处置建议时，保存阶段报告后，通过现有动作提议入口生成待人工审批的关联预约工单（work_order_type=general）。保留原投诉事件，草稿写明实际位置、采样或处置内容、责任要求、执行依赖、现场待核验条件和回执要求；可将通知、采样、冲洗和复测组织成合理的人工执行事项，不机械逐节点建单。
+1. 核实投诉位置及相近时段的同类投诉，用 `query_peripheral_data` 按实际时间、位置和来源标识核对外围工单或观察。目录首屏不能证明没有其他投诉；未知点位先查目录，不猜编号或把演示点当真实资产。
+2. 解析未知地址，查询附近资产、属性和连接，区分上游、管网、末梢、二次供水和用户端的实际覆盖。地点明确后展示投诉位置，取得关联资产后展示正在核查的管段或节点。拓扑连接不等于实际流向，方向判断需要同窗模型或监测依据。
+3. 核对 SCADA 指标、单位和模型映射，用 `query_scada_readings` 或 `query_element_history` 查询投诉前后及对照，检查时间覆盖、在线状态、漂移、长期缺测和质量标识。同步展示核查点位，将实际取得的水质、压力等指标画成趋势或对比图，不同单位分图。缺测不是零值，上游正常不能排除末梢问题，压力正常不能证明水质合格。
+4. 综合投诉分布、监测、供水关联和实际采样，列出候选原因、支持证据、反证和区分原因所需的核验。未证实污染源时保留假设，不把感官描述换算成浓度。水质限值只能引用已取得且适用的依据，不能自行承诺安全饮用。
+5. 有合理目标管段或节点时，可用 `analyze_valve_isolation` 和 `analyze_valve_closure` 评估候选隔离及关阀影响，无需等待污染确诊，但不能据此直接执行停水。逐候选区分可隔离性、必关阀、可选阀、受影响节点和未知阀门状态。取得影响节点后展示覆盖示意，形成或调整关阀方案时展示对应阀门。拓扑隔离区、水力影响、污染范围和用户影响分别说明；节点覆盖面不能当作污染传播边界，无用户资料时不虚构户数。
+6. 只有假设源节点、浓度及必要模型输入有依据时，才用 `run_contaminant_simulation` 正向比较传播；它不是源头反演，未模拟范围不能宣布安全。具备实际排放节点、流量和前置条件时才用 `analyze_flushing` 评估冲洗。取得模拟结果后，按实际返回的对象、时刻和指标展示浓度、压力或流量曲线；同口径的隔离、冲洗方案可画对比图，标明情景假设和局部覆盖。冲洗模拟不证明消毒合格，排放位置和承接条件仍需现场核验与审批。
 
-设备命令只针对真实且可控制的目标，审批与执行由 Server 控制。草稿保存或工单派发不等于采样、通知、冲洗、复测完成。核对已有草稿和现场反馈后再继续，避免重复提议。正式报告保留所用流程名称和版本，便于追溯。
+缺少源参数不阻止已有证据支持的采样、候选隔离或人工告知建议。查询错误、权限拒绝或接口故障属于执行异常，不能写成无数据或未见异常。
+
+## 处置与报告
+
+报告分别说明投诉与监测覆盖、候选原因和未排除路径、隔离及影响、通知保障、采样或污染水处置，以及复测与恢复条件。每项给出实际结果或经查询确认的限制；区分当前风险判断与尚未发生的未来处置。投诉位置、监测、候选隔离和影响按实际步骤同步展示，遵循后台展示规则，不用报告定位链接代替过程展示。无数据或展示失败时说明具体限制；展示不替代证据，也不要求等待浏览器。
+
+形成具体建议后，先保存阶段结论，再提议关联的 `work_order_type=general` 草稿。写明实际位置、采样或处置目标、时间要求、依赖、现场待核验条件和回执要求。通知、采样、清洗冲洗与复测按依赖合理组织，不机械逐节点建单，也不重复已有有效动作。没有客户通知适配器时明确由人工实施；调查消息不能当作用户通知。设备命令仅针对真实可控目标，审批与执行由 Server 控制。
+
+采样、处置和回访尚未发生不阻止有边界的初步建议。工单创建、审批或派发不证明采样、通知、冲洗或复测完成。
+
+## 闭环复核
+
+后续轮次核对实际采样位置、时间、结果和判定依据，以及污染进入控制、隔离与通知、清洗冲洗、复测、恢复供水、用户回访、未决动作和遗留风险。不适用项说明依据，缺少反馈保留待核验。
+
+证据齐全时报告具备结案条件，由来源系统的有权限人员确认关闭。核对来源新版本后才能认定闭环；误报或未发现异常也须有足够核查和人工确认，不能用有限正常数据、模拟结果或流程状态证明水质安全。

samples/peripheral/api.py

@@ -1,4 +1,4 @@
-"""Independent sample API for synthetic work orders and monitoring data."""
+"""Project-scoped peripheral work orders, monitoring history and explicit test data."""
 from typing import get_args
 from uuid import UUID
 
@@ -34,8 +34,8 @@ async def store(request: Request, who: Identity = Depends(identity),
 
 
 def create_app(settings: SampleSettings | None = None) -> FastAPI:
-    app = FastAPI(title="TJWater 外围系统模拟接入", version="1.0.0",
-                  description="测试监测与核查反馈样板，不是正式工单平台或设备控制服务。")
+    app = FastAPI(title="TJWater 外围系统接入", version="1.0.0",
+                  description="按项目管理业务工单与反馈、读取监测历史；测试数据保留独立来源标记。")
     app.state.settings = settings or SampleSettings()
     app.add_middleware(CORSMiddleware, allow_origins=app.state.settings.cors_origins,
         allow_credentials=False, allow_methods=["GET", "POST", "PATCH", "DELETE"],
@@ -51,7 +51,7 @@ def create_app(settings: SampleSettings | None = None) -> FastAPI:
 
     @app.get("/health")
     def health():
-        return {"service": "peripheral-sample", "origin": "synthetic", "storage": "postgresql",
+        return {"service": "peripheral-sample", "origin": "system_data", "storage": "postgresql",
                 "status": "up", "storage_verified": False}
 
     prefix = "/api/sample/v1"
@@ -59,7 +59,7 @@ def create_app(settings: SampleSettings | None = None) -> FastAPI:
 
     @app.get(prefix + "/context", response_model=Context, responses=errors)
     def context(request: Request, who: Identity = Depends(identity), db=Depends(store)):
-        return {"protocol_version": "tjwater-peripheral-sample@1", "origin": "synthetic",
+        return {"protocol_version": "tjwater-peripheral-sample@1", "origin": "system_data",
                 "project_id": who.project_id, "user_id": who.user_id, "can_write": who.can_write,
                 "scenarios": list(get_args(Scenario)) if request.app.state.settings.enable_faults else ["normal"],
                 "state_transitions": TRANSITIONS, "managed_resources": db.managed_resource_capabilities()}

samples/peripheral/schemas.py

@@ -122,6 +122,7 @@ WorkOrderStatus = Literal[
 ]
 Priority = Literal["low", "normal", "high", "urgent"]
 WorkOrderType = Literal["general", "water_quality_complaint", "burst_report"]
+WorkOrderOrigin = Literal["system_data"]
 class CreateWorkOrder(Model):
     request_id: UUID
     title: Annotated[str, Field(min_length=1, max_length=200)]
@@ -131,6 +132,7 @@ class CreateWorkOrder(Model):
     assignee: Annotated[str, Field(min_length=1, max_length=200)] | None = None
     scheduled_at: WireDateTime | None = None
     work_order_type: WorkOrderType = "general"
+    origin: WorkOrderOrigin = "system_data"
     location: Annotated[str, Field(min_length=1, max_length=200)] | None = None
     scenario_code: Annotated[str, Field(min_length=1, max_length=128)] | None = None
     feedback_requirements: list[Text] = Field(default_factory=list)
@@ -193,7 +195,7 @@ class WorkOrder(Model):
     priority: Priority
     assignee: str | None
     scheduled_at: WireDateTime | None
-    origin: Literal["synthetic"]
+    origin: WorkOrderOrigin
     created_at: WireDateTime
     updated_at: WireDateTime
     created_by: str
@@ -208,7 +210,11 @@ class WorkOrder(Model):
     @classmethod
     def accept_legacy_document(cls, value):
         if isinstance(value, dict):
-            return {key: item for key, item in value.items() if key != "subtasks"}
+            return {
+                key: "system_data" if key == "origin" else item
+                for key, item in value.items()
+                if key != "subtasks"
+            }
         return value
 
 
@@ -314,7 +320,7 @@ class ManagedResourceCapabilities(Model):
 
 class Context(Model):
     protocol_version: Literal["tjwater-peripheral-sample@1"]
-    origin: Literal["synthetic"]
+    origin: Literal["system_data"]
     project_id: UUID
     user_id: UUID
     can_write: StrictBool

samples/peripheral/store.py

@@ -369,7 +369,7 @@ class PostgresStore:
             now = datetime.now(timezone.utc).isoformat()
             document = command.model_dump(mode="json", exclude={"request_id"}) | {
                 "work_order_id": str(uuid4()), "project_id": who.project_id, "revision": 1,
-                "origin": "synthetic", "created_at": now, "updated_at": now,
+                "origin": "system_data", "created_at": now, "updated_at": now,
                 "created_by": who.user_id, "updated_by": who.user_id,
                 "activity": [{"sequence": 1, "occurred_at": now, "actor": who.user_id,
                               "kind": "created", "title": "工单创建",

scripts/provision_operational_worker_identity.py

@@ -309,8 +309,8 @@ def _update_worker_environment(
         "OPERATIONAL_WORKER_SERVER_URL": "http://127.0.0.1:8000",
         "OPERATIONAL_WORKER_AGENT_URL": "http://127.0.0.1:18787",
         "OPERATIONAL_WORKER_TOKEN_REFRESH_SECONDS": "300",
-        "OPERATIONAL_ACTION_PERIPHERAL_URL": "http://127.0.0.1:8011",
-        "OPERATIONAL_DATA_PERIPHERAL_URL": "http://127.0.0.1:8011",
+        "OPERATIONAL_ACTION_PERIPHERAL_URL": "http://127.0.0.1:8000/peripheral",
+        "OPERATIONAL_DATA_PERIPHERAL_URL": "http://127.0.0.1:8000/peripheral",
         "OPERATIONAL_DATA_SERVER_URL": "http://127.0.0.1:8000",
     }
     for key, value in values.items():

scripts/run_calculation_worker.py

@@ -0,0 +1,30 @@
+#!/usr/bin/env python3
+"""Run the independent calculation supervisor (never an API lifespan task)."""
+import argparse
+import asyncio
+from pathlib import Path
+import signal
+import sys
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from dotenv import load_dotenv
+
+from app.services.calculation_worker import CalculationWorker, WorkerSettings
+
+
+async def main():
+    load_dotenv(Path(__file__).resolve().parents[1] / '.env', override=False)
+    task = asyncio.create_task(CalculationWorker(WorkerSettings.from_environment()).run())
+    loop = asyncio.get_running_loop()
+    for signum in (signal.SIGTERM, signal.SIGINT):
+        loop.add_signal_handler(signum, task.cancel)
+    try:
+        await task
+    except asyncio.CancelledError:
+        pass
+
+
+if __name__ == '__main__':
+    argparse.ArgumentParser(description=__doc__).parse_args()
+    asyncio.run(main())

scripts/run_operational_worker.py

@@ -124,6 +124,20 @@ class OperationalWorker:
     async def tick(self) -> int:
         headers = await self.credentials.headers(self.client)
         server = self.settings.server_url.rstrip("/")
+        agent = self.settings.agent_url.rstrip("/")
+        target_agent_instance_id: str | None = None
+
+        async def target_instance_id() -> str:
+            nonlocal target_agent_instance_id
+            if target_agent_instance_id is None:
+                response = await self.client.get(f"{agent}/health")
+                response.raise_for_status()
+                value = response.json().get("agent_instance_id")
+                if not isinstance(value, str) or not value.strip():
+                    raise ValueError("Agent health response has no instance identity")
+                target_agent_instance_id = value.strip()
+            return target_agent_instance_id
+
         scope = {"task_type": "event_investigation"}
         if self.settings.task_created_from:
             scope["task_created_from"] = self.settings.task_created_from
@@ -136,10 +150,13 @@ class OperationalWorker:
         for status_value in ("claimed", "running"):
             response = await self.client.get(
                 f"{server}/api/v1/operational-task-runs", headers=headers,
-                params={"status": status_value, "limit": self.settings.max_concurrency, **scope},
+                params={"status": status_value, "limit": 100, **scope},
             )
             response.raise_for_status()
             for item in response.json()["items"]:
+                owner = item.get("agent_instance_id")
+                if not isinstance(owner, str) or owner != await target_instance_id():
+                    continue
                 run_id = str(item["run_id"])
                 active_run_ids.add(run_id)
                 session_id = item.get("agent_session_id")
@@ -152,7 +169,7 @@ class OperationalWorker:
             # Refresh the Agent vault with the worker's existing OAuth provider.
             # Background runs never depend on an interactive browser refresh.
             renewed = await self.client.post(
-                f"{self.settings.agent_url.rstrip('/')}/api/v1/session/{session_id}/credentials",
+                f"{agent}/api/v1/session/{session_id}/credentials",
                 headers=headers, json={},
             )
             if renewed.status_code not in (202, 404, 409):
@@ -167,6 +184,7 @@ class OperationalWorker:
         )
         response.raise_for_status()
         queued = response.json()["items"]
+        dispatch_instance_id = await target_instance_id() if queued else None
 
         async def dispatch(run: dict) -> bool:
             if run.get("status") != "queued":
@@ -176,18 +194,21 @@ class OperationalWorker:
             request_id = str(UUID(run["execution_request_id"]))
             try:
                 result = await self.client.post(
-                    f"{self.settings.agent_url.rstrip('/')}/api/v1/task-executions",
+                    f"{agent}/api/v1/task-executions",
                     headers=headers,
                     json={
                         "task_id": task_id,
                         "run_id": run_id,
                         "execution_request_id": request_id,
+                        "agent_instance_id": dispatch_instance_id,
                     },
                 )
                 if result.is_success:
                     payload = result.json()
                     if payload.get("task_id") != task_id or payload.get("run_id") != run_id:
                         raise ValueError("Agent returned a different task run")
+                    if payload.get("agent_instance_id") != dispatch_instance_id:
+                        raise ValueError("Agent returned a different instance identity")
                     session_id = payload.get("session_id")
                     if isinstance(session_id, str) and session_id:
                         self._agent_sessions[run_id] = session_id
@@ -221,7 +242,12 @@ class OperationalWorker:
                         f"{self.settings.agent_url.rstrip('/')}/api/v1/task-messages/deliver",
                         headers=headers, json={"message_id": message_id},
                     )
-                    if delivered.status_code in {400, 403, 404, 410, 422}:
+                    if delivered.status_code == 404:
+                        # A different Agent instance may own the target session.
+                        # Keep the durable message pending instead of turning a
+                        # routing miss into a terminal delivery failure.
+                        logger.info("Agent message target is not local; delivery remains pending: %s", message_id)
+                    elif delivered.status_code in {400, 403, 410, 422}:
                         failed = await self.client.post(
                             f"{self.settings.server_url.rstrip('/')}/api/v1/agent-messages/{message_id}/delivery",
                             headers=headers, json={

scripts/sync_peripheral_work_orders.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""Synchronize the explicitly synthetic peripheral work orders into task projections."""
+"""Synchronize peripheral work orders as project-owned system data."""
 from __future__ import annotations
 
 import argparse
@@ -15,7 +15,6 @@ import httpx
 
 sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
 
-from samples.peripheral.seed_identity import SEED_ACTOR_ID
 from app.domain.schemas.workbench_events import WorkbenchEvent
 
 try:
@@ -42,23 +41,15 @@ def peripheral_order_id(external_id: str | None) -> str | None:
 
 
 def should_project_work_order(order: dict, linked_order_ids: set[str]) -> bool:
-    """Exclude retired seed checks while preserving user and workflow orders."""
-    work_order_id = order["work_order_id"]
-    work_order_type = order.get("work_order_type", "general")
-    if work_order_type in EVENT_WORK_ORDER_TYPES or work_order_id in linked_order_ids:
-        return True
-    return not (
-        work_order_type == "general"
-        and order.get("origin") == "synthetic"
-        and order.get("created_by") == str(SEED_ACTOR_ID)
-    )
+    """Project every active work order returned by the peripheral system."""
+    return True
 
 
 def work_order_snapshot(order: dict) -> dict:
     """Preserve the flat order and its recorded activity without inventing stages."""
     required = (
         "work_order_id", "revision", "title", "summary", "status", "priority",
-        "origin", "updated_at",
+        "updated_at",
     )
     snapshot = {key: order[key] for key in required}
     snapshot.update({key: order.get(key) for key in (
@@ -67,6 +58,7 @@ def work_order_snapshot(order: dict) -> dict:
     snapshot["feedback_requirements"] = order.get("feedback_requirements") or []
     snapshot["activity"] = order.get("activity") or []
     snapshot["work_order_type"] = order.get("work_order_type", "general")
+    snapshot["origin"] = "system_data"
     return snapshot
 
 
@@ -148,8 +140,9 @@ async def sync_once(client: httpx.AsyncClient, settings: WorkerSettings, periphe
         work_order_ids, deleted_ids,
     ):
         for order in page["items"]:
-            if order["project_id"] != settings.project_id or order["origin"] != "synthetic":
+            if order["project_id"] != settings.project_id or order["origin"] != "system_data":
                 raise ValueError("peripheral work order scope or origin mismatch")
+            source_label = "外围工单"
             seen_order_ids.add(order["work_order_id"])
             if not should_project_work_order(order, linked_order_ids):
                 continue
@@ -165,12 +158,13 @@ async def sync_once(client: httpx.AsyncClient, settings: WorkerSettings, periphe
                 event_request_id = str(uuid5(NAMESPACE_URL, f"{settings.project_id}:{external_id}:investigation"))
                 event = await client.post(tasks_url, headers=headers, json={
                     "request_id": event_request_id, "task_type": "event_investigation", "workflow": event_kind,
-                    "title": "水质投诉调查" if event_kind == "water_quality_complaint" else "爆管事件调查", "summary": f"合成外围事件：{order['summary']}",
+                    "title": "水质投诉调查" if event_kind == "water_quality_complaint" else "爆管事件调查",
+                    "summary": f"外围事件：{order['summary']}",
                     "priority": "urgent" if order["priority"] in {"high", "urgent"} else "normal",
                     "source": {"kind": "external_system", "external_id": order["work_order_id"],
                                "external_revision": order["revision"], "event_kind": event_kind},
                     "trigger": {"kind": "event", "occurred_at": order.get("scheduled_at") or order["created_at"],
-                                "detail": f"外围样板事件上报 · {order.get('location') or '位置待确认'}"},
+                                "detail": f"外围事件上报 · {order.get('location') or '位置待确认'}"},
                 })
                 event.raise_for_status()
                 event_tasks[order["work_order_id"]] = event.json()
@@ -179,12 +173,12 @@ async def sync_once(client: httpx.AsyncClient, settings: WorkerSettings, periphe
                 request_id = str(uuid5(NAMESPACE_URL, f"{settings.project_id}:{external_id}:create"))
                 result = await client.post(tasks_url, headers=headers, json={
                     "request_id": request_id, "task_type": "external_work_order", "workflow": "external-work-order",
-                    "title": order["title"], "summary": f"合成外围工单：{order['summary']}",
+                    "title": order["title"], "summary": f"{source_label}：{order['summary']}",
                     "priority": "urgent" if order["priority"] in {"high", "urgent"} else "normal",
                     "source": {"kind": "external_system", "external_id": external_id,
                                "event_kind": order.get("work_order_type", "general")},
                     "trigger": {"kind": "external_sync", "occurred_at": order["created_at"],
-                                "detail": "外围样板工单同步，进度以外部系统反馈为准"},
+                                "detail": "外围工单同步，进度以外部系统反馈为准"},
                     "scheduled_for": order["scheduled_at"],
                 })
                 result.raise_for_status()
@@ -225,9 +219,9 @@ async def sync_once(client: httpx.AsyncClient, settings: WorkerSettings, periphe
                 "request_id": str(uuid5(NAMESPACE_URL, event_id)), "source_event_id": external_id,
                 "source_revision": order["revision"], "status": STATUS[order["status"]],
                 "occurred_at": order["updated_at"],
-                "detail": f"合成外围工单 {order['work_order_id']}：{order['status']}",
-                "evidence": {"evidence_id": event_id, "kind": "synthetic", "summary": evidence_summary,
-                             "source_label": "外围样板合成工单", "source_ref": external_id,
+                "detail": f"{source_label} {order['work_order_id']}：{order['status']}",
+                "evidence": {"evidence_id": event_id, "kind": "external_feedback", "summary": evidence_summary,
+                             "source_label": source_label, "source_ref": external_id,
                              "observed_at": order["updated_at"]},
             }
             # A child order's creation is already represented by its dispatch

tests/api/test_openapi_contract.py

@@ -254,14 +254,16 @@ def test_committed_write_refresh_failure_has_explicit_http_contract() -> None:
     assert response.json()["code"] == "materialized_view_refresh_failed_after_commit"
 
 
-def test_valve_isolation_runtime_accepts_frontend_query(monkeypatch) -> None:
+@pytest.mark.parametrize("element_type", ["auto", "link", "node"])
+def test_valve_isolation_runtime_accepts_frontend_query(monkeypatch, element_type) -> None:
     captured: dict[str, object] = {}
 
-    def fake_analyze_valve_isolation(network, accident_element, disabled_valves):
+    def fake_analyze_valve_isolation(network, accident_element, disabled_valves, accident_element_type):
         captured.update(
             network=network,
             accident_element=accident_element,
             disabled_valves=disabled_valves,
+            accident_element_type=accident_element_type,
             business_dsn=get_project_pgconn_string(network),
             timescale_dsn=get_project_timescale_pgconn_string(network),
         )
@@ -288,6 +290,7 @@ def test_valve_isolation_runtime_accepts_frontend_query(monkeypatch) -> None:
             ("accident_element", "P-1"),
             ("accident_element", "P-2"),
             ("disabled_valves", "V-9"),
+            *([] if element_type == "auto" else [("accident_element_type", element_type)]),
         ],
     )
 
@@ -297,6 +300,7 @@ def test_valve_isolation_runtime_accepts_frontend_query(monkeypatch) -> None:
         "network": "fengyang",
         "accident_element": ["P-1", "P-2"],
         "disabled_valves": ["V-9"],
+        "accident_element_type": element_type,
         "business_dsn": "postgresql://user:password@biz/fengyang",
         "timescale_dsn": "postgresql://user:password@timescale/fengyang",
     }

tests/api/test_operational_task_dynamic_flow.py

@@ -33,6 +33,7 @@ def test_dynamic_investigation_routes_are_in_the_public_contract():
     assert "get" in paths["/api/v1/operational-tasks/{task_id}/reports"]
     assert "get" in paths["/api/v1/operational-tasks/{task_id}/reports/{revision}"]
     assert "post" in paths["/api/v1/operational-tasks/{task_id}/presentation-events"]
+    assert "get" in paths["/api/v1/operational-tasks/{task_id}/presentation-events"]
     assert "get" in paths["/api/v1/operational-tasks/{task_id}/presentation-cursor"]
     context_params = paths["/api/v1/operational-tasks/{task_id}/agent-context"]["get"]["parameters"]
     assert any(item["name"] == "report_revision" and item["in"] == "query" for item in context_params)
@@ -123,3 +124,21 @@ async def test_report_endpoints_preserve_interactive_blocks_and_version_context(
     schemas = app.openapi()["components"]["schemas"]
     assert "ui_blocks" in schemas["InvestigationReportInput"]["properties"]
     assert "ui_blocks" in schemas["InvestigationReport"]["properties"]
+
+
+@pytest.mark.anyio
+async def test_presentation_history_route_forwards_read_only_paging_and_domain_errors(monkeypatch):
+    service = SimpleNamespace(presentation_history=AsyncMock(return_value={"items": []}))
+    monkeypatch.setattr(endpoints, "_service", lambda _ctx, _conn: service)
+    task_id = uuid4()
+    await endpoints.get_operational_task_presentation_history(
+        task_id, after=7, through=30, limit=20, ctx=_service_context(), conn=object(),
+    )
+    service.presentation_history.assert_awaited_once_with(task_id, after=7, through=30, limit=20)
+    from app.domain.domain_errors import DomainError
+    service.presentation_history.side_effect = DomainError("TASK_NOT_FOUND", "missing", status=404)
+    with pytest.raises(HTTPException) as error:
+        await endpoints.get_operational_task_presentation_history(
+            task_id, after=0, through=None, limit=100, ctx=_service_context(), conn=object(),
+        )
+    assert error.value.status_code == 404

tests/api/test_peripheral_mount.py

@@ -0,0 +1,23 @@
+from fastapi.testclient import TestClient
+
+from app.main import app
+
+
+def test_main_server_mounts_peripheral_health_endpoint():
+    response = TestClient(app).get("/peripheral/health")
+
+    assert response.status_code == 200
+    assert response.json() == {
+        "service": "peripheral-sample",
+        "origin": "system_data",
+        "storage": "postgresql",
+        "status": "up",
+        "storage_verified": False,
+    }
+
+
+def test_mounted_peripheral_api_preserves_its_authentication_boundary():
+    response = TestClient(app).get("/peripheral/api/sample/v1/context")
+
+    assert response.status_code == 401
+    assert response.json()["code"] == "AUTHENTICATION_REQUIRED"

tests/api/test_task_question_endpoints.py

@@ -0,0 +1,39 @@
+import asyncio
+from uuid import uuid4
+
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from app.api.v1.endpoints import operational_tasks as endpoints
+from tests.unit.test_task_questions import setup, finish
+
+
+def test_question_http_contract_waits_and_resumes_original_task(monkeypatch):
+    service, repo, task_id, command = setup()
+    app = FastAPI()
+    app.include_router(endpoints.router)
+    monkeypatch.setattr(endpoints, "_service", lambda *_args: service)
+    # Exercise HTTP parsing and response contracts; service tests cover authorization.
+    for route in app.routes:
+        if getattr(route, "name", "") in {"create_operational_task_question", "answer_operational_task_question"}:
+            for dependency in route.dependant.dependencies:
+                app.dependency_overrides[dependency.call] = lambda: None
+    with TestClient(app) as client:
+        path = f"/operational-tasks/{task_id}/questions"
+        invalid = client.post(path, json={**command.model_dump(mode="json"), "project_id": str(uuid4())})
+        assert invalid.status_code == 422
+        created = client.post(path, json=command.model_dump(mode="json"))
+        assert created.status_code == 200, created.text
+        question_id = created.json()["question_request_id"]
+        asyncio.run(finish(service, repo, task_id))
+        answered = client.post(f"{path}/{question_id}/answer", json={
+            "request_id": str(uuid4()), "expected_version": 1,
+            "answers": [{"question_id": "flow", "selected_option_ids": ["unknown"], "custom_text": ""}],
+        })
+        assert answered.status_code == 200, answered.text
+        result = answered.json()
+        assert result["task_id"] == str(task_id) and result["execution_generation"] == 2
+        assert result["active_run"]["status"] == "queued"
+        assert result["questions"][0]["status"] == "answered"
+        assert result["questions"][0]["answered_by"] == service._actor()
+        assert result["evidence"] == []

tests/integration/test_calculation_queue_storage.py

@@ -0,0 +1,127 @@
+"""Opt-in real PostgreSQL queue races, using an explicitly disposable database."""
+import asyncio
+import os
+from pathlib import Path
+from uuid import uuid4
+
+import pytest
+from fastapi import HTTPException
+from psycopg import AsyncConnection
+from psycopg.conninfo import conninfo_to_dict
+from psycopg.rows import dict_row
+
+from app.infra.db.postgresql.calculation_jobs import CalculationJobRepository
+
+
+def test_concurrent_submission_claims_fencing_and_project_isolation():
+    dsn = os.getenv("CALCULATION_QUEUE_TEST_DSN")
+    if not dsn:
+        pytest.skip("Set CALCULATION_QUEUE_TEST_DSN to a disposable PostgreSQL database")
+    assert conninfo_to_dict(dsn)["dbname"].startswith("tjwater_calc_test_")
+
+    async def run():
+        project, other, owner = uuid4(), uuid4(), uuid4()
+        async with await AsyncConnection.connect(dsn, row_factory=dict_row) as conn:
+            migrations = Path(__file__).parents[2] / "resources/sql"
+            await conn.execute((migrations / "026_calculation_jobs.sql").read_text())
+            legacy_job = uuid4()
+            await conn.execute("""INSERT INTO workbench.calculation_jobs
+                (job_id,project_id,owner_user_id,request_key,tool,parameters,binding,timeout_seconds,status,result,updated_at)
+                VALUES (%s,%s,%s,'legacy','run_simulation','{}','{}',1800,'succeeded','{"preserved":true}',now()-interval '30 days')""",
+                (legacy_job, uuid4(), owner))
+            lifecycle = (migrations / "027_calculation_job_lifecycle.sql").read_text()
+            await conn.execute(lifecycle)
+            cursor = await conn.execute("SELECT result,result_expires_at,result_expires_at>now()+interval '6 days' AS grace FROM workbench.calculation_jobs WHERE job_id=%s", (legacy_job,))
+            first = await cursor.fetchone()
+            assert first["result"] == {"preserved": True} and first["grace"]
+            await conn.execute(lifecycle)
+            cursor = await conn.execute("SELECT result_expires_at FROM workbench.calculation_jobs WHERE job_id=%s", (legacy_job,))
+            assert (await cursor.fetchone())["result_expires_at"] == first["result_expires_at"]
+
+        async def submit(key, parameters=None, selected_project=project):
+            async with await AsyncConnection.connect(dsn, row_factory=dict_row) as conn:
+                return await CalculationJobRepository(conn, selected_project).enqueue(
+                    owner_user_id=owner, request_key=key, tool="run_burst_location",
+                    parameters=parameters or {"x": 1}, binding={"native_session_id": "session"}, timeout_seconds=600)
+
+        # Four independent API connections, same logical submission.
+        submissions = await asyncio.gather(*(submit("same") for _ in range(4)))
+        assert len({str(row["job_id"]) for row in submissions}) == 1
+        with pytest.raises(HTTPException) as conflict:
+            await submit("same", {"x": 2})
+        assert conflict.value.status_code == 409
+        second = await submit("second")
+        foreign = await submit("foreign", selected_project=other)
+
+        async def claim(worker):
+            async with await AsyncConnection.connect(dsn, row_factory=dict_row) as conn:
+                return await CalculationJobRepository(conn, project).claim(worker, 60)
+
+        claimed = await asyncio.gather(*(claim(f"worker-{index}") for index in range(4)))
+        rows = [row for row in claimed if row]
+        assert len(rows) == 2
+        assert len({str(row["job_id"]) for row in rows}) == 2
+        assert str(foreign["job_id"]) not in {str(row["job_id"]) for row in rows}
+
+        async with await AsyncConnection.connect(dsn, row_factory=dict_row) as conn:
+            repo = CalculationJobRepository(conn, project)
+            assert await repo.get(foreign["job_id"], owner) is None
+            assert await repo.get(second["job_id"], uuid4()) is None
+            row = rows[0]
+            stale = {**row, "lease_generation": row["lease_generation"] - 1}
+            assert await repo.heartbeat(stale, row["lease_owner"], 60) is None
+            assert not await repo.finish(stale, row["lease_owner"], "succeeded", result={})
+            cancelled = await repo.cancel(row["job_id"], owner)
+            assert cancelled["status"] == "cancel_requested"
+            assert not await repo.finish(row, row["lease_owner"], "succeeded", result={})
+            assert await repo.finish(row, row["lease_owner"], "cancelled")
+            expired = rows[1]
+            await conn.execute("UPDATE workbench.calculation_jobs SET lease_expires_at=now()-interval '1 second' WHERE job_id=%s", (expired["job_id"],))
+            await repo.recover()
+            assert (await repo.get(expired["job_id"], owner))["status"] == "interrupted"
+            assert not await repo.finish(expired, expired["lease_owner"], "succeeded", result={})
+            assert await repo.claim("replacement", 60) is None
+
+            # Queued cancellation is terminal immediately, with no later claim.
+            queued = await repo.enqueue(owner_user_id=owner, request_key="cancel-before-start",
+                tool="run_simulation", parameters={}, binding={}, timeout_seconds=30)
+            cancelled = await repo.cancel(queued["job_id"], owner)
+            assert cancelled["status"] == "cancelled" and cancelled["finished_at"] is not None
+            assert cancelled["started_at"] is None
+            assert await repo.claim("replacement", 60) is None
+
+            # A polling client expires queued jobs without any worker process.
+            waiting = await repo.enqueue(owner_user_id=owner, request_key="expired-queue",
+                tool="run_simulation", parameters={}, binding={}, timeout_seconds=30)
+            await conn.execute("UPDATE workbench.calculation_jobs SET queue_expires_at=now()-interval '1 second' WHERE job_id=%s", (waiting["job_id"],))
+            status = await repo.get(waiting["job_id"], owner)
+            assert status["status"] == "timed_out" and status["error"]["code"] == "QUEUE_TIMEOUT"
+            assert status["finished_at"] is not None
+            assert "result" not in status
+
+            # Retention scrubs only payload, keeping deduplication and status.
+            completed = await repo.enqueue(owner_user_id=owner, request_key="retained-identity",
+                tool="run_simulation", parameters={}, binding={}, timeout_seconds=30)
+            await conn.execute("""UPDATE workbench.calculation_jobs SET status='succeeded',
+                result='{"large":"payload"}'::jsonb,finished_at=now(),
+                result_expires_at=now()-interval '1 second' WHERE job_id=%s""", (completed["job_id"],))
+            expired_result = await repo.get(completed["job_id"], owner, include_result=True)
+            assert expired_result["result"] is None and expired_result["status"] == "succeeded"
+            duplicate = await repo.enqueue(owner_user_id=owner, request_key="retained-identity",
+                tool="run_simulation", parameters={}, binding={}, timeout_seconds=30, max_active_jobs=0)
+            assert duplicate["job_id"] == completed["job_id"]
+
+        # Concurrent unique requests cannot exceed an otherwise empty project's cap.
+        capacity_project = uuid4()
+        async def bounded_submit(index):
+            try:
+                async with await AsyncConnection.connect(dsn, row_factory=dict_row) as conn:
+                    return await CalculationJobRepository(conn, capacity_project).enqueue(
+                        owner_user_id=owner, request_key=str(index), tool="run_simulation",
+                        parameters={}, binding={}, timeout_seconds=30, max_active_jobs=2)
+            except HTTPException as exc:
+                return exc
+        bounded = await asyncio.gather(*(bounded_submit(i) for i in range(8)))
+        assert sum(isinstance(value, dict) for value in bounded) == 2
+        assert all(isinstance(value, dict) or value.status_code == 503 for value in bounded)
+    asyncio.run(run())

tests/integration/test_operational_tasks_storage.py

@@ -293,7 +293,7 @@ async def test_postgres_task_artifacts_approval_and_project_isolation_roll_back(
                 assert request.url.path.endswith(execution_ids[0])
                 return httpx.Response(200, json={"request_id": execution_ids[0], "result": {
                     "work_order_id": order_id, "project_id": str(context.project_id),
-                    "created_at": now.isoformat(), "origin": "synthetic", "status": "open",
+                    "created_at": now.isoformat(), "origin": "system_data", "status": "open",
                 }})
 
             dispatcher = HttpOperationalActionDispatcher(

tests/unit/test_calculation_cleanup.py

@@ -0,0 +1,25 @@
+import json
+
+import pytest
+
+from app.services.calculation_cleanup import _names, record_calculation_database
+from app.algorithms._utils import calculation_temp_directory
+
+
+def test_job_owned_database_ledger_survives_until_successful_cleanup(tmp_path, monkeypatch):
+    monkeypatch.setenv("TJWATER_CALCULATION_JOB_DIR", str(tmp_path))
+    name = "tjw_tmp_run_project_" + "a" * 32
+    record_calculation_database(name)
+    record_calculation_database(name)
+    assert _names(tmp_path) == [name]
+    assert json.loads((tmp_path / "databases.json").read_text()) == [name]
+    record_calculation_database(name, removed=True)
+    assert _names(tmp_path) == []
+    assert calculation_temp_directory("leakage") == str(tmp_path / "leakage")
+
+
+@pytest.mark.parametrize("names", [["production"], ["tjw_tmp_not_random"], {"database": "production"}])
+def test_cleanup_rejects_non_job_database_names(tmp_path, names):
+    (tmp_path / "databases.json").write_text(json.dumps(names))
+    with pytest.raises(ValueError):
+        _names(tmp_path)

tests/unit/test_calculation_queue_lifecycle.py

@@ -0,0 +1,63 @@
+"""Queue lifecycle SQL contracts; real races are covered by the opt-in PG suite."""
+import asyncio
+from contextlib import asynccontextmanager
+
+import pytest
+
+from app.infra.db.postgresql.calculation_jobs import CalculationJobRepository
+from app.mcp.errors import CalculationCapacityExceeded
+
+
+class Connection:
+    def __init__(self, rows=()):
+        self.rows = iter(rows)
+        self.statements = []
+        self.in_transaction = False
+
+    @asynccontextmanager
+    async def transaction(self):
+        self.in_transaction = True
+        try:
+            yield
+        finally:
+            self.in_transaction = False
+
+    async def execute(self, sql, params):
+        self.statements.append((sql, params, self.in_transaction))
+        row = next(self.rows, None) if sql.lstrip().startswith(('SELECT job_id', 'SELECT count', 'INSERT')) else None
+        class Cursor:
+            async def fetchone(self):
+                return row
+            async def fetchall(self):
+                return []
+        return Cursor()
+
+
+def test_status_projection_excludes_result_and_expires_without_worker():
+    connection = Connection()
+    repo = CalculationJobRepository(connection, 'project')
+    asyncio.run(repo.get('job'))
+    assert 'QUEUE_TIMEOUT' in connection.statements[0][0]
+    assert 'SET result=NULL' in connection.statements[1][0]
+    assert ',result FROM' not in connection.statements[2][0]
+    asyncio.run(repo.get('job', include_result=True))
+    assert ',result FROM' in connection.statements[-1][0]
+
+
+def test_duplicate_submission_bypasses_full_queue_inside_transaction():
+    existing = {'tool': 'tool', 'parameters': {'timeout_seconds': 3600}, 'binding': {}, 'job_id': 'existing'}
+    connection = Connection([existing])
+    row = asyncio.run(CalculationJobRepository(connection, 'project').enqueue(
+        owner_user_id='owner', request_key='key', tool='tool', parameters={}, binding={}, timeout_seconds=30))
+    assert row['job_id'] == 'existing'
+    assert 'pg_advisory_xact_lock' in connection.statements[0][0]
+    assert all(statement[2] for statement in connection.statements)
+    assert not any('count(*)' in sql for sql, _, _ in connection.statements)
+
+
+def test_capacity_rejects_new_job_before_insert():
+    connection = Connection([None, {'active': 100}])
+    with pytest.raises(CalculationCapacityExceeded):
+        asyncio.run(CalculationJobRepository(connection, 'project').enqueue(
+            owner_user_id='owner', request_key='key', tool='tool', parameters={}, binding={}, timeout_seconds=30))
+    assert not any('INSERT' in sql for sql, _, _ in connection.statements)

tests/unit/test_calculation_worker.py

@@ -0,0 +1,241 @@
+"""Worker boundaries use real managed subprocesses, without external databases."""
+import asyncio
+from functools import wraps
+from types import SimpleNamespace
+from contextlib import asynccontextmanager
+import os
+from pathlib import Path
+import signal
+import sys
+
+import pytest
+
+from app.services.calculation_worker import CalculationWorker, WorkerSettings, stop_process_group
+
+
+@pytest.fixture(autouse=True)
+def no_external_cleanup(monkeypatch):
+    async def cleanup(*args):
+        pass
+    monkeypatch.setattr('app.services.calculation_worker.cleanup_job_resources', cleanup)
+
+
+def run_async(function):
+    @wraps(function)
+    def wrapper(*args, **kwargs):
+        return asyncio.run(function(*args, **kwargs))
+    return wrapper
+
+
+def test_worker_settings_require_explicit_valid_scope(monkeypatch):
+    monkeypatch.setenv('CALCULATION_WORKER_PROJECT_IDS', '6b39b128-2ba8-454f-9bbe-e9553902ab53')
+    settings = WorkerSettings.from_environment()
+    assert settings.max_concurrency == 2
+    monkeypatch.setenv('CALCULATION_WORKER_HEARTBEAT_SECONDS', '30')
+    with pytest.raises(ValueError):
+        WorkerSettings.from_environment()
+
+
+@run_async
+async def test_process_group_cleanup_kills_native_style_descendant(tmp_path):
+    pid_file = tmp_path / 'pid'
+    code = ('import subprocess,time,pathlib,sys; '
+            'p=subprocess.Popen([sys.executable,"-c","import time; time.sleep(60)"]); '
+            'pathlib.Path(sys.argv[1]).write_text(str(p.pid)); time.sleep(60)')
+    process = await asyncio.create_subprocess_exec(sys.executable, '-c', code,
+        str(pid_file), start_new_session=True)
+    try:
+        for _ in range(100):
+            if pid_file.exists():
+                break
+            await asyncio.sleep(.02)
+        child_pid = int(pid_file.read_text())
+        await stop_process_group(process)
+        assert process.returncode is not None
+        # Linux may retain a dead grandchild as a zombie until PID 1 reaps it.
+        stat_path = Path(f'/proc/{child_pid}/stat')
+        for _ in range(100):
+            if not stat_path.exists() or stat_path.read_text().split()[2] == 'Z':
+                break
+            await asyncio.sleep(.01)
+        assert not stat_path.exists() or stat_path.read_text().split()[2] == 'Z'
+    finally:
+        await stop_process_group(process)
+
+
+@run_async
+@pytest.mark.parametrize('lease_status, expected', [('cancel_requested', 'cancelled'), (None, None)])
+async def test_cancel_and_lost_lease_stop_process_before_persisting(monkeypatch, lease_status, expected):
+    processes = []
+    original = asyncio.create_subprocess_exec
+    async def spawn(*args, **kwargs):
+        process = await original(sys.executable, '-c', 'import time; time.sleep(60)',
+                                 start_new_session=True)
+        processes.append(process)
+        return process
+    monkeypatch.setattr(asyncio, 'create_subprocess_exec', spawn)
+    finishes = []
+    class Repository:
+        async def heartbeat(self, *args):
+            return lease_status
+        async def finish(self, row, owner, status, **kwargs):
+            assert processes[0].returncode is not None
+            finishes.append(status)
+            return True
+    @asynccontextmanager
+    async def repository(project):
+        yield Repository()
+    worker = CalculationWorker(WorkerSettings(('project',), heartbeat_seconds=.05), repository=repository)
+    await worker.run_job({'project_id': 'project', 'job_id': 'job', 'timeout_seconds': 60})
+    assert finishes == ([expected] if expected else [])
+    assert processes[0].returncode is not None
+
+
+@run_async
+async def test_timeout_stops_process_before_recording(monkeypatch):
+    original = asyncio.create_subprocess_exec
+    processes = []
+    async def spawn(*args, **kwargs):
+        p = await original(sys.executable, '-c', 'import time; time.sleep(60)', start_new_session=True)
+        processes.append(p)
+        return p
+    monkeypatch.setattr(asyncio, 'create_subprocess_exec', spawn)
+    statuses = []
+    class Repository:
+        async def heartbeat(self, *args):
+            return 'running'
+        async def finish(self, row, owner, status, **kwargs):
+            assert processes[0].returncode is not None
+            statuses.append(status)
+            return True
+    @asynccontextmanager
+    async def repository(project):
+        yield Repository()
+    worker = CalculationWorker(WorkerSettings(('project',), heartbeat_seconds=.02), repository=repository)
+    await worker.run_job({'project_id': 'project', 'timeout_seconds': 0})
+    assert statuses == ['timed_out']
+
+
+@run_async
+async def test_duplicate_supervisor_never_claims(monkeypatch):
+    from app.infra.db.metadb import database
+    class Lock:
+        async def scalar(self, statement):
+            assert 'pg_try_advisory_lock' in str(statement)
+            return False
+        async def commit(self):
+            pass
+    @asynccontextmanager
+    async def connection():
+        yield Lock()
+    monkeypatch.setattr(database, 'engine', SimpleNamespace(connect=connection))
+    @asynccontextmanager
+    async def repository(project):
+        pytest.fail('A duplicate supervisor must never open the queue')
+        yield
+    with pytest.raises(RuntimeError, match='already active'):
+        await CalculationWorker(WorkerSettings(('project',)), repository=repository).run()
+
+
+@run_async
+async def test_repository_terminal_updates_are_fenced():
+    from app.infra.db.postgresql.calculation_jobs import CalculationJobRepository
+    statements = []
+    class Cursor:
+        async def fetchone(self):
+            return None
+    class Connection:
+        async def execute(self, sql, parameters):
+            statements.append((sql, parameters))
+            return Cursor()
+    repository = CalculationJobRepository(Connection(), 'project')
+    row = {'job_id': 'job', 'lease_generation': 7}
+    assert await repository.finish(row, 'owner', 'succeeded', result={'original': True}) is False
+    sql, parameters = statements[-1]
+    assert 'lease_generation=%s' in sql and 'lease_owner=%s' in sql
+    assert 'lease_expires_at>clock_timestamp()' in sql
+    assert "status!='cancel_requested'" in sql
+    assert 7 in parameters
+
+
+@run_async
+async def test_guard_parent_death_pipe_kills_entire_group(tmp_path):
+    pid_file = tmp_path / 'guard-child'
+    child = 'import os,time,pathlib,sys; pathlib.Path(sys.argv[1]).write_text(str(os.getpid())); time.sleep(60)'
+    launcher = '''import asyncio,sys
+from app.services.calculation_worker import guard_child
+original = asyncio.create_subprocess_exec
+async def spawn(*args, **kwargs):
+    return await original(sys.executable, '-c', sys.argv[2], sys.argv[1])
+asyncio.create_subprocess_exec = spawn
+asyncio.run(guard_child('unused','unused'))
+'''
+    guard = await asyncio.create_subprocess_exec(sys.executable, '-c', launcher,
+        str(pid_file), child, stdin=asyncio.subprocess.PIPE, start_new_session=True)
+    try:
+        for _ in range(100):
+            if pid_file.exists():
+                break
+            await asyncio.sleep(.02)
+        child_pid = int(pid_file.read_text())
+        guard.stdin.close()
+        await asyncio.wait_for(guard.wait(), 3)
+        assert guard.returncode == -signal.SIGKILL
+        stat = Path(f'/proc/{child_pid}/stat')
+        for _ in range(100):
+            if not stat.exists() or stat.read_text().split()[2] == 'Z':
+                break
+            await asyncio.sleep(.01)
+        assert not stat.exists() or stat.read_text().split()[2] == 'Z'
+    finally:
+        await stop_process_group(guard)
+
+
+@run_async
+async def test_cleanup_failure_is_not_reported_cancelled(monkeypatch, tmp_path):
+    monkeypatch.setenv('CALCULATION_WORKER_STATE_DIR', str(tmp_path))
+    async def cleanup(row, directory):
+        (Path(directory) / 'cleanup-failed').touch()
+        raise RuntimeError('external resource unavailable')
+    monkeypatch.setattr('app.services.calculation_worker.cleanup_job_resources', cleanup)
+    original = asyncio.create_subprocess_exec
+    async def spawn(*args, **kwargs):
+        return await original(sys.executable, '-c', 'import time; time.sleep(60)', start_new_session=True)
+    monkeypatch.setattr(asyncio, 'create_subprocess_exec', spawn)
+    results = []
+    class Repository:
+        async def heartbeat(self, *args):
+            return 'cancel_requested'
+        async def finish(self, row, owner, status, **kwargs):
+            results.append((status, kwargs['error']['code']))
+            return True
+    @asynccontextmanager
+    async def repository(project):
+        yield Repository()
+    worker = CalculationWorker(WorkerSettings(('project',)), repository=repository)
+    await worker.run_job({'project_id': 'project', 'timeout_seconds': 60})
+    assert results == [('failed', 'CLEANUP_FAILED')]
+    assert len(list(tmp_path.glob('*/cleanup-failed'))) == 1
+
+
+@run_async
+async def test_startup_recovery_checks_persisted_generation(monkeypatch, tmp_path):
+    import json
+    from app.services.calculation_worker import recover_job_resources
+    project = '6b39b128-2ba8-454f-9bbe-e9553902ab53'
+    job = '2d3a6231-a8ee-4acb-bb13-58328d29b88f'
+    directory = tmp_path / (job + '-owned')
+    directory.mkdir()
+    (directory / 'scope.json').write_text(json.dumps({
+        'project_id': project, 'job_id': job, 'lease_generation': 1}))
+    monkeypatch.setenv('CALCULATION_WORKER_STATE_DIR', str(tmp_path))
+    class Repository:
+        async def get(self, job_id):
+            return {'lease_generation': 2}
+    @asynccontextmanager
+    async def repository(project):
+        yield Repository()
+    monkeypatch.setattr('app.services.calculation_worker.queue_repository', repository)
+    with pytest.raises(RuntimeError, match='ownership'):
+        await recover_job_resources((project,))
+    assert directory.exists()

tests/unit/test_cors.py

@@ -0,0 +1,73 @@
+import pytest
+from fastapi import FastAPI
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.testclient import TestClient
+
+from app.core.config import Settings
+
+
+ALLOWED_ORIGINS = [
+    "https://next-demo.waternetwork.cn",
+    "http://192.168.1.133",
+    "https://192.168.1.133",
+    "http://192.168.1.110",
+    "https://192.168.1.110",
+]
+
+
+@pytest.mark.parametrize("origins", ["*", " *, http://localhost:5173", "http://localhost:5173,*"])
+def test_wildcard_origins_are_rejected(origins: str):
+    with pytest.raises(ValueError, match="must list explicit origins"):
+        Settings(_env_file=None, CORS_ALLOWED_ORIGINS=origins).cors_allowed_origins
+
+
+@pytest.mark.parametrize(
+    "origin",
+    ALLOWED_ORIGINS + [
+        "http://192.168.1.114",
+        "http://192.168.1.133:5173",
+        "https://demo.waternetwork.cn",
+        "http://192.168.1.133.evil.example",
+        "null",
+    ],
+)
+def test_explicit_origin_preflight(origin: str):
+    settings = Settings(_env_file=None, CORS_ALLOWED_ORIGINS=", ".join(ALLOWED_ORIGINS))
+    app = FastAPI()
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=settings.cors_allowed_origins,
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+    with TestClient(app) as client:
+        response = client.options(
+            "/api/v1/projects",
+            headers={
+                "Origin": origin,
+                "Access-Control-Request-Method": "GET",
+                "Access-Control-Request-Headers": "authorization,x-project-id",
+            },
+        )
+    allowed = origin in ALLOWED_ORIGINS
+    assert response.status_code == (200 if allowed else 400)
+    assert response.headers.get("access-control-allow-origin") == (origin if allowed else None)
+    if allowed:
+        assert response.headers["access-control-allow-credentials"] == "true"
+        assert "authorization" in response.headers["access-control-allow-headers"]
+
+
+def test_default_origins_are_explicit_local_sources(monkeypatch):
+    monkeypatch.delenv("CORS_ALLOWED_ORIGINS", raising=False)
+    settings = Settings(_env_file=None)
+    assert settings.cors_allowed_origins == [
+        "http://127.0.0.1:5173", "http://localhost:5173",
+        "http://127.0.0.1:15173", "http://localhost:15173",
+    ]
+
+
+def test_environment_controls_origins(monkeypatch):
+    monkeypatch.setenv("CORS_ALLOWED_ORIGINS", "http://192.168.1.133,https://192.168.1.110")
+    settings = Settings(_env_file=None)
+    assert settings.cors_allowed_origins == ["http://192.168.1.133", "https://192.168.1.110"]

tests/unit/test_investigation_diagram_status.py

@@ -0,0 +1,94 @@
+from uuid import uuid4
+
+import pytest
+from pydantic import ValidationError
+
+from app.domain.schemas.operational_task import (
+    AgentArtifactsCommand,
+    InvestigationDiagramNodeRef,
+    InvestigationProgress,
+    InvestigationReportInput,
+)
+
+
+def test_waiting_and_unknown_node_states_survive_progress_and_final_report():
+    diagram = {
+        "title": "等待系统结果",
+        "mermaid": "flowchart LR\nDONE --> WAIT --> UNKNOWN --> UNSET",
+        "node_refs": [
+            {"node_id": "DONE", "label": "查询完成", "status": "completed"},
+            {"node_id": "WAIT", "label": "等待系统结果", "status": "waiting"},
+            {"node_id": "UNKNOWN", "label": "未知", "status": None},
+            {"node_id": "UNSET", "label": "未确认"},
+        ],
+    }
+    progress = InvestigationProgress(summary="等待系统结果", diagram=diagram)
+    report = InvestigationReportInput(
+        title="报告", current_conclusion="已确认当前结果", conclusion_boundary="仅限查询范围",
+        stage="final", outcome="concluded", data_gaps=[], diagram=diagram,
+    )
+    assert report.diagram == progress.diagram
+    assert [node.status for node in report.diagram.node_refs] == ["completed", "waiting", None, None]
+    assert "status" not in report.diagram.node_refs[-1].model_fields_set
+
+
+def test_unknown_node_state_is_rejected():
+    with pytest.raises(ValidationError):
+        InvestigationDiagramNodeRef(node_id="A", label="查询", status="blocked")
+
+
+def test_class_attached_nodes_are_valid_declarations_without_accepting_partial_ids():
+    payload = {"request_id": uuid4(), "source_session_id": "session-1", "progress": {
+        "summary": "核查", "diagram": {"title": "路线", "mermaid": "flowchart TD\nQUERY:::running --> REVIEW:::pending",
+            "node_refs": [{"node_id": "QUERY", "label": "核查", "status": "running"},
+                          {"node_id": "REVIEW", "label": "复核", "status": "pending"}]},
+    }}
+    assert AgentArtifactsCommand.model_validate(payload).progress.diagram.node_refs[0].status == "running"
+    payload["progress"]["diagram"]["node_refs"][0]["node_id"] = "QUER"
+    with pytest.raises(ValidationError, match="QUER"):
+        AgentArtifactsCommand.model_validate(payload)
+
+
+@pytest.mark.parametrize("artifact", ["progress", "report"])
+def test_artifact_write_rejects_node_refs_missing_from_mermaid(artifact):
+    diagram = {
+        "title": "当前路线",
+        "mermaid": "flowchart TD\nA[绑定任务] --> B[核查证据]",
+        "node_refs": [
+            {"node_id": "ctx", "label": "绑定任务", "status": "completed"},
+            {"node_id": "B", "label": "核查证据", "status": "running"},
+        ],
+    }
+    payload = (
+        {"summary": "正在核查", "diagram": diagram}
+        if artifact == "progress"
+        else {
+            "title": "调查报告",
+            "current_conclusion": "已确认当前范围",
+            "conclusion_boundary": "仅限已核查资料",
+            "stage": "final",
+            "outcome": "concluded",
+            "data_gaps": [],
+            "diagram": diagram,
+        }
+    )
+
+    with pytest.raises(ValidationError, match="ctx"):
+        AgentArtifactsCommand.model_validate({
+            "request_id": uuid4(),
+            "source_session_id": "session-1",
+            artifact: payload,
+        })
+
+
+def test_historical_progress_keeps_read_compatibility_for_undeclared_node_refs():
+    progress = InvestigationProgress.model_validate({
+        "summary": "旧进度",
+        "diagram": {
+            "title": "旧图",
+            "mermaid": "flowchart TD\nA[绑定任务]",
+            "node_refs": [{"node_id": "ctx", "label": "绑定任务", "status": "completed"}],
+        },
+    })
+
+    assert progress.diagram.node_refs[0].node_id == "ctx"

tests/unit/test_mcp_analysis.py

@@ -35,6 +35,7 @@ INPUTS = {
     "run_burst_detection": {"target_time": START, "sensor_nodes": ["1", "2", "3", "4", "5"]},
     "run_burst_location": {"data_source": "simulation", "simulation_run_id": str(uuid4()), "burst_leakage": 0.02, "pressure_scada_ids": ["1"], "scada_burst_start": START, "scada_burst_end": END},
     "analyze_valve_isolation": {"accident_element": ["540672"]},
+    "generate_node_coverage": {"groups": [{"group_id": "monitoring", "title": "监测节点分布", "node_ids": ["J1"]}]},
     "run_leakage_identification": {"scada_start": START, "scada_end": END, "sensor_nodes": ["1"], "q_sum": 0.02, "start_time": 12, "dma_count": 1},
 }
 
@@ -129,7 +130,7 @@ def test_missing_timescale_route_stops_online_simulation_before_model_changes(mo
     calculate.assert_not_awaited()
 
 
-@pytest.mark.parametrize("name", ["analyze_valve_isolation", "optimize_sensor_placement"])
+@pytest.mark.parametrize("name", ["analyze_valve_isolation", "optimize_sensor_placement", "generate_node_coverage"])
 def test_business_only_calculations_do_not_require_timescale(monkeypatch, routed_calculation_scope, name):
     scope, _, events = routed_calculation_scope
 
@@ -192,7 +193,7 @@ def test_catalog_and_snapshots_accept_calculations_but_reject_model_owned_author
 
 @pytest.mark.parametrize("name", INPUTS)
 def test_calculation_permission_and_reader_boundary(monkeypatch, name):
-    permissions = {"burst.run", "simulation.run", "simulation.view", "scada.view", "optimization.run"}
+    permissions = {"burst.run", "simulation.run", "simulation.view", "scada.view", "optimization.run", "webgis.view"}
     monkeypatch.setattr(query_service, "permissions_for_context", lambda _: permissions)
     scope = SimpleNamespace(project=SimpleNamespace(project_role="member"))
     binding = SimpleNamespace(tool_cap=list(TOOLS), purpose="task_execution", permission_cap=list(permissions), task_id=uuid4(), task_scope_cap=["scada_readings"])
@@ -218,7 +219,7 @@ def test_time_sources_and_finite_inputs_are_explicit():
         ("run_simulation", {"duration": 0}),
         ("run_simulation", {"duration": 86401}),
         ("run_simulation", {"timeout_seconds": 0}),
-        ("run_simulation", {"timeout_seconds": 86401}),
+        ("run_simulation", {"timeout_seconds": 241}),
         ("run_simulation", {"timeout_seconds": 30.5}),
         ("run_simulation", {"timeout_seconds": True}),
         ("run_simulation", {"output_node_ids": []}),
@@ -316,9 +317,9 @@ def test_burst_location_optional_inputs_still_reject_partial_or_invalid_values(o
 
 
 @pytest.mark.parametrize("name", INPUTS)
-def test_calculations_accept_agent_execution_budget_separately_from_duration(name):
-    args = arguments(name, timeout_seconds=900)
-    assert args.timeout_seconds == 900
+def test_calculations_accept_http_response_budget_separately_from_duration(name):
+    args = arguments(name, timeout_seconds=180)
+    assert args.timeout_seconds == 180
     if hasattr(args, "duration"):
         assert args.duration == INPUTS[name].get("duration", 900)
 
@@ -350,6 +351,25 @@ def test_adapter_uses_existing_service_without_scheme_or_shared_output(monkeypat
         assert call["n_workers"] == 1
 
 
+@pytest.mark.parametrize("element_type", ["auto", "link", "node"])
+def test_isolation_adapter_preserves_explicit_element_type(monkeypatch, element_type):
+    calls = []
+    monkeypatch.setattr(
+        "app.services.valve_isolation.analyze_valve_isolation",
+        lambda **kwargs: calls.append(kwargs) or {"isolatable": True},
+    )
+    args = arguments("analyze_valve_isolation", accident_element_type=element_type)
+
+    asyncio.run(analysis_service.calculate(
+        "analyze_valve_isolation", args, network="trusted-project", username="actor",
+    ))
+
+    assert calls[0]["accident_element_type"] == element_type
+    assert calls[0]["network"] == "trusted-project"
+    with pytest.raises(ValidationError):
+        arguments("analyze_valve_isolation", accident_element_type="unknown")
+
+
 @pytest.mark.parametrize("name", ["run_burst_location", "run_simulation"])
 def test_dispatch_ends_read_transaction_and_preserves_analysis_provenance(monkeypatch, name):
     order = []

tests/unit/test_mcp_calculation_jobs.py

@@ -0,0 +1,206 @@
+import asyncio
+from datetime import datetime, timezone
+from uuid import uuid4
+from types import SimpleNamespace
+import pytest
+from fastapi import HTTPException
+from app.mcp.models import BoundContext
+from app.mcp.calculation_jobs import durable_binding, restore_binding, check_binding, should_queue, QUEUED_TOOLS
+
+
+def binding():
+    return BoundContext(context_handle='secret-handle', project_id=uuid4(), subject='user', client_id='host', native_session_id='session', purpose='ordinary_chat', expires_at=datetime.now(timezone.utc))
+
+
+def test_durable_binding_never_stores_expiring_secret_handle():
+    original = binding()
+    safe = durable_binding(original)
+    assert 'context_handle' not in safe
+    assert 'expires_at' not in safe
+    assert restore_binding(safe).subject == original.subject
+
+
+def test_original_session_boundary_is_enforced():
+    original = binding()
+    row = {'binding': durable_binding(original)}
+    check_binding(original, row)
+    original.native_session_id = 'other'
+    with pytest.raises(HTTPException) as error:
+        check_binding(original, row)
+    assert error.value.status_code == 403
+
+
+def test_only_executed_expensive_mutations_are_queued():
+    tool = SimpleNamespace(name='save_simulation_scenario', mutating=True)
+    assert not should_queue(tool, SimpleNamespace(dry_run=True))
+    assert should_queue(tool, SimpleNamespace(dry_run=False))
+    assert 'analyze_valve_isolation' not in QUEUED_TOOLS
+    assert 'calculate_demand_allocation' not in QUEUED_TOOLS
+    assert len(QUEUED_TOOLS) == 12
+
+
+def test_result_pages_keep_original_semantics_and_provenance(monkeypatch):
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs
+    from app.mcp.calculation_models import CalculationResultInput
+    from app.mcp.models import QueryResult
+    original = binding()
+    result = QueryResult(source='run_burst_location', source_ref='calculation:one', queried_at=datetime.now(timezone.utc), semantics='analysis', data={'result': {'items': list(range(10))}}, evidence_id='evidence-one')
+    row = {'job_id': uuid4(), 'binding': durable_binding(original), 'status': 'succeeded', 'result': result.model_dump(mode='json')}
+    repo = SimpleNamespace(get=AsyncMock(return_value=row))
+    monkeypatch.setattr(jobs, 'CalculationJobRepository', lambda *a: repo)
+    monkeypatch.setattr(jobs, 'authorize_original', AsyncMock())
+    monkeypatch.setattr(jobs, 'attach_calculation_evidence', AsyncMock(side_effect=lambda scope, row, result: result))
+    scope = SimpleNamespace(connection=None, project=SimpleNamespace(project_id=original.project_id, user_id=uuid4()))
+    args = CalculationResultInput(context_handle='x' * 32, job_id=row['job_id'], data_path=['result', 'items'], limit=3, offset=2)
+    page = asyncio.run(jobs.read_calculation_job(scope, original, SimpleNamespace(name='get_calculation_result'), args))
+    assert page.data == [2, 3, 4]
+    assert page.semantics == 'analysis'
+    assert page.source == 'run_burst_location'
+    assert page.evidence_id == 'evidence-one'
+    assert page.truncated and page.page.has_more and page.page.total == 10
+
+
+def test_pending_result_is_task_receipt_without_evidence(monkeypatch):
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs
+    from app.mcp.calculation_models import CalculationResultInput
+    original = binding()
+    row = {'job_id': uuid4(), 'binding': durable_binding(original), 'status': 'running', 'tool': 'run_simulation'}
+    monkeypatch.setattr(jobs, 'CalculationJobRepository', lambda *a: SimpleNamespace(get=AsyncMock(return_value=row)))
+    monkeypatch.setattr(jobs, 'authorize_original', AsyncMock())
+    attach = AsyncMock()
+    monkeypatch.setattr(jobs, 'attach_calculation_evidence', attach)
+    scope = SimpleNamespace(connection=None, project=SimpleNamespace(project_id=original.project_id, user_id=uuid4()))
+    result = asyncio.run(jobs.read_calculation_job(scope, original, SimpleNamespace(name='get_calculation_result'), CalculationResultInput(context_handle='x' * 32, job_id=row['job_id'])))
+    assert result.semantics == 'task' and result.evidence_id is None
+    assert result.data['status'] == 'running'
+    attach.assert_not_awaited()
+
+
+def test_mutation_submission_cannot_bypass_existing_approval_receipt(monkeypatch):
+    from contextlib import asynccontextmanager
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs, query_service
+    from app.mcp.catalog import TOOLS
+    from app.mcp.extended_models import SensorCreateInput
+    from app.mcp.mutation_service import mutation_arguments
+    from app.infra.db.postgresql import mcp_queries
+    args = SensorCreateInput(context_handle='x' * 32, request_id=uuid4(), dry_run=False, expected_digest='a' * 64, name='approved', method='kmeans', sensor_count=2)
+    @asynccontextmanager
+    async def transaction():
+        yield
+    previous = {'parameters': {'tool': 'create_sensor_placement', 'arguments': mutation_arguments(args, preview=False)}, 'result': {'status': 'started'}}
+    receipt = SimpleNamespace(lock_receipt=AsyncMock(), receipt=AsyncMock(return_value=previous))
+    monkeypatch.setattr(mcp_queries, 'McpQueryRepository', lambda *a: receipt)
+    monkeypatch.setattr(query_service, 'authorize_query', AsyncMock())
+    queue = AsyncMock()
+    monkeypatch.setattr(jobs, 'CalculationJobRepository', queue)
+    scope = SimpleNamespace(connection=SimpleNamespace(transaction=transaction), project=SimpleNamespace(project_id=uuid4(), user_id=uuid4()))
+    with pytest.raises(HTTPException) as error:
+        asyncio.run(jobs.submit_calculation(scope, binding(), TOOLS['create_sensor_placement'], args, 'host-call'))
+    assert error.value.status_code == 409
+    assert 'already started' in str(error.value.detail)
+    queue.assert_not_called()
+    changed = args.model_copy(update={'name': 'changed'})
+    with pytest.raises(HTTPException) as conflict:
+        asyncio.run(jobs.submit_calculation(scope, binding(), TOOLS['create_sensor_placement'], changed, 'host-call'))
+    assert conflict.value.detail == 'request_id already has different parameters'
+    queue.assert_not_called()
+
+
+def test_evidence_receipt_commits_in_root_operational_transaction(monkeypatch):
+    from contextlib import asynccontextmanager
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs, query_service, contexts
+    from app.infra.db.postgresql import mcp_queries
+    from app.mcp.models import QueryResult
+    original = binding().model_copy(update={'task_id': uuid4(), 'run_id': uuid4()})
+    result = QueryResult(source='run_simulation', source_ref='original', queried_at=datetime.now(timezone.utc), semantics='analysis', data={})
+    events = []
+    async def commit():
+        events.append('commit-reads')
+    @asynccontextmanager
+    async def transaction():
+        events.append('begin-root-operational')
+        yield
+        events.append('commit-evidence')
+    task_repo = SimpleNamespace(transaction=transaction, lock=AsyncMock())
+    receipt_repo = SimpleNamespace(lock_receipt=AsyncMock(), receipt=AsyncMock(return_value=None), save_receipt=AsyncMock())
+    service = SimpleNamespace(repo=task_repo, record_data_snapshot=AsyncMock(return_value=(SimpleNamespace(evidence_id='saved'), None)))
+    monkeypatch.setattr(jobs, 'authorize_original', AsyncMock(return_value=(original, SimpleNamespace(name='run_simulation'), None)))
+    monkeypatch.setattr(contexts, 'verify_task', AsyncMock())
+    monkeypatch.setattr(query_service, 'task_service', lambda scope: service)
+    monkeypatch.setattr(mcp_queries, 'McpQueryRepository', lambda *a: receipt_repo)
+    scope = SimpleNamespace(connection=SimpleNamespace(commit=commit), project=SimpleNamespace(project_id=original.project_id))
+    attached = asyncio.run(jobs.attach_calculation_evidence(scope, {'job_id': uuid4()}, result))
+    assert attached.evidence_id == 'saved'
+    assert events == ['commit-reads', 'begin-root-operational', 'commit-evidence']
+    receipt_repo.save_receipt.assert_awaited_once()
+
+
+@pytest.mark.parametrize('job_budget', [3600, 7200])
+def test_http_timeout_does_not_limit_queued_execution(monkeypatch, job_budget):
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs, query_service
+    from app.mcp.catalog import TOOLS
+    tool = TOOLS['run_simulation']
+    args = tool.model.model_validate({
+        'context_handle': 'x' * 32, 'timeout_seconds': 1,
+        'start_time': '2026-10-08T12:00:00+08:00', 'duration': 3600,
+        'output_node_ids': ['J1'],
+    })
+    repo = SimpleNamespace(enqueue=AsyncMock(return_value={
+        'job_id': uuid4(), 'status': 'queued', 'tool': tool.name,
+    }))
+    monkeypatch.setattr(jobs, 'CalculationJobRepository', lambda *a: repo)
+    monkeypatch.setattr(query_service, 'authorize_query', AsyncMock())
+    scope = SimpleNamespace(connection=SimpleNamespace(commit=AsyncMock()),
+                            project=SimpleNamespace(project_id=uuid4(), user_id=uuid4()))
+    result = asyncio.run(jobs.submit_calculation(
+        scope, binding(), tool, args, 'host-call', job_timeout_seconds=job_budget,
+    ))
+    assert result.data['status'] == 'queued'
+    assert repo.enqueue.await_args.kwargs['timeout_seconds'] == job_budget
+    assert 'timeout_seconds' not in repo.enqueue.await_args.kwargs['parameters']
+
+
+def test_expired_result_is_safe_explicit_error(monkeypatch):
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs
+    from app.mcp.calculation_models import CalculationResultInput
+    original = binding()
+    row = {'job_id': uuid4(), 'binding': durable_binding(original), 'status': 'succeeded',
+           'result': None, 'result_expires_at': datetime(2020, 1, 1, tzinfo=timezone.utc)}
+    repo = SimpleNamespace(get=AsyncMock(return_value=row))
+    monkeypatch.setattr(jobs, 'CalculationJobRepository', lambda *a: repo)
+    monkeypatch.setattr(jobs, 'authorize_original', AsyncMock())
+    attach = AsyncMock()
+    monkeypatch.setattr(jobs, 'attach_calculation_evidence', attach)
+    scope = SimpleNamespace(connection=None, project=SimpleNamespace(project_id=original.project_id, user_id=uuid4()))
+    with pytest.raises(HTTPException) as error:
+        asyncio.run(jobs.read_calculation_job(scope, original, SimpleNamespace(name='get_calculation_result'),
+            CalculationResultInput(context_handle='x' * 32, job_id=row['job_id'])))
+    assert error.value.status_code == 410
+    assert jobs.public_job(row)['result_available'] is False
+    assert repo.get.await_args.kwargs['include_result'] is True
+    attach.assert_not_awaited()
+
+
+def test_legacy_execution_timeout_does_not_block_owner_authorization(monkeypatch):
+    from unittest.mock import AsyncMock
+    from app.mcp import calculation_jobs as jobs, query_service
+    from app.mcp.catalog import TOOLS
+    original = binding()
+    owner = uuid4()
+    row = {'binding': durable_binding(original), 'tool': 'run_simulation',
+           'owner_user_id': owner, 'parameters': {'timeout_seconds': 3600,
+           'start_time': '2026-10-29T00:00:00Z', 'output_node_ids': ['1']}}
+    authorize = AsyncMock()
+    monkeypatch.setattr(query_service, 'authorize_query', authorize)
+    scope = SimpleNamespace(project=SimpleNamespace(project_id=original.project_id, user_id=owner))
+    _, tool, args = asyncio.run(jobs.authorize_original(scope, row))
+    assert tool is TOOLS['run_simulation']
+    assert args.timeout_seconds <= 240
+    assert row['parameters']['timeout_seconds'] == 3600
+    authorize.assert_awaited_once()

tests/unit/test_mcp_config.py

@@ -23,9 +23,15 @@ def test_explicit_private_development_issuer():
     assert local_settings().issuer_uri == ISSUER
 
 
-def test_calculation_limit_covers_default_without_extending_ordinary_queries():
+def test_execution_limit_is_separate_from_http_request_budget():
     config = local_settings(max_calculation_timeout_seconds=120)
-    assert config.query_timeout_seconds == 30
+    assert config.query_timeout_seconds == 120
+    assert config.max_request_timeout_seconds == 240
+    assert local_settings().max_calculation_timeout_seconds == 3600
+    with pytest.raises(ValidationError):
+        local_settings(query_timeout_seconds=241)
+    with pytest.raises(ValidationError):
+        local_settings(max_request_timeout_seconds=241)
     with pytest.raises(ValidationError):
         local_settings(max_calculation_timeout_seconds=119)
 

tests/unit/test_mcp_extended.py

@@ -292,7 +292,7 @@ def test_artifacts_are_integrity_checked_bounded_binary_envelopes():
 
 def test_no_health_prediction_tool_and_every_write_requests_native_permission():
     assert not any("health" in name for name in TOOLS)
-    assert len(TOOLS) == 67
+    assert len(TOOLS) == 71
     assert sum(t.mutating for t in TOOLS.values()) == 13
     for tool in TOOLS.values():
         if tool.mutating:

tests/unit/test_mcp_presentation_preview.py

@@ -0,0 +1,36 @@
+from datetime import datetime, timezone
+from uuid import uuid4
+
+from app.domain.schemas.operational_task import TaskPresentationEventCommand
+from app.mcp.models import QueryResult
+from app.mcp.presentation_preview import saved_presentation_preview
+
+
+def test_abbreviated_coverage_keeps_publishable_command_without_copying_geometry():
+    result = QueryResult(
+        source="generate_node_coverage", source_ref="mcp:analysis", evidence_id="mcp:analysis",
+        queried_at=datetime.now(timezone.utc), semantics="analysis", origin="system_data", truncated=False,
+        data={"result": {"map_payload": {"type": "FeatureCollection", "features": [{"large": "geometry"}]},
+                         "coverage": {"status": "partial", "groups": {"isolation": {
+                             "node_ids": ["A", "B"], "mapped_node_count": 1, "missing_node_ids": ["B"],
+                         }}}}},
+    )
+    hint = saved_presentation_preview(result)
+    assert hint == saved_presentation_preview(result)
+    assert hint["groups"]["isolation"] == {"node_count": 2, "mapped_node_count": 1, "missing_node_count": 1}
+    command = TaskPresentationEventCommand(
+        request_id=uuid4(), run_id=uuid4(), source_session_id="session", command=hint["command"],
+    )
+    assert command.command.evidence_id == result.evidence_id and command.command.data is None
+    assert "geometry" not in str(hint["groups"]) and result.data["result"]["map_payload"]["features"]
+    result.evidence_id = None
+    assert saved_presentation_preview(result) is None
+
+
+def test_missing_geometry_does_not_offer_a_broken_overlay_command():
+    result = QueryResult(
+        source="generate_node_coverage", source_ref="mcp:analysis", evidence_id="mcp:analysis",
+        queried_at=datetime.now(timezone.utc), semantics="analysis", origin="system_data", truncated=False,
+        data={"result": {"coverage": {"status": "unavailable"}}},
+    )
+    assert saved_presentation_preview(result) is None

tests/unit/test_mcp_query.py

@@ -37,36 +37,28 @@ PERMISSIONS = {
 }
 
 
-def test_calculation_budget_outlives_query_default_and_still_times_out(runtime, monkeypatch):
+def test_calculation_submission_returns_receipt_without_running_native_code(runtime, monkeypatch):
+    from app.mcp import calculation_jobs
     app, client, _, _ = runtime
-    app.settings.query_timeout_seconds = 1
     monkeypatch.setattr(application, "authorize_query", AsyncMock())
-
-    async def execute(scope, bound, tool, args):
-        await asyncio.sleep(1.05)
-        return QueryResult(
-            source=tool.name, source_ref="test:calculation", semantics="analysis",
-            queried_at=datetime.now(timezone.utc), data={"completed": True},
-        )
-
-    app.executor = execute
-    handle = binding().context_handle
-    arguments = {
-        "context_handle": handle, "start_time": "2026-10-08T12:00:00+08:00",
+    receipt = calculation_jobs.envelope("run_simulation", {"job_id": str(uuid4()), "status": "queued"})
+    submit = AsyncMock(return_value=receipt)
+    monkeypatch.setattr(calculation_jobs, "submit_calculation", submit)
+    app.executor = AsyncMock(side_effect=AssertionError("native work entered API process"))
+    response = request(client, "tools/call", {"name": "run_simulation", "arguments": {
+        "context_handle": binding().context_handle, "start_time": "2026-10-08T12:00:00+08:00",
         "duration": 3600, "output_node_ids": ["J1"], "timeout_seconds": 2,
-    }
-    response = request(client, "tools/call", {"name": "run_simulation", "arguments": arguments}).json()
-    assert response["result"].get("isError") is not True
-    assert response["result"]["structuredContent"]["data"] == {"completed": True}
-    arguments["timeout_seconds"] = 1
-    response = request(client, "tools/call", {"name": "run_simulation", "arguments": arguments}).json()
-    assert response["result"]["isError"] is True
-    assert response["result"]["structuredContent"]["error"]["code"] == "QUERY_TIMEOUT"
+    }}).json()
+    assert response["result"]["structuredContent"]["data"]["status"] == "queued"
+    assert response["result"]["structuredContent"]["semantics"] == "task"
+    app.executor.assert_not_awaited()
+    assert submit.await_args.args[3].timeout_seconds == 2
+    assert submit.await_args.kwargs["job_timeout_seconds"] == 3600
 
 
 def test_calculation_limit_is_advertised_and_enforced_before_execution(runtime):
     app, client, calls, _ = runtime
-    app.settings.max_calculation_timeout_seconds = 120
+    app.settings.max_request_timeout_seconds = 120
     tools = request(client).json()["result"]["tools"]
     schema = next(tool["inputSchema"] for tool in tools if tool["name"] == "run_simulation")
     assert schema["properties"]["timeout_seconds"]["maximum"] == 120
@@ -80,45 +72,47 @@ def test_calculation_limit_is_advertised_and_enforced_before_execution(runtime):
     assert calls == []
 
 
-def test_calculation_capacity_has_an_explicit_mcp_error(runtime, monkeypatch):
-    from threading import BoundedSemaphore
-    from app.mcp import analysis_service
-
-    app, client, _, _ = runtime
+def test_queue_submission_errors_preserve_mcp_error_mapping(runtime, monkeypatch):
+    from app.mcp import calculation_jobs
+    _, client, _, _ = runtime
     monkeypatch.setattr(application, "authorize_query", AsyncMock())
-    slots = BoundedSemaphore(2)
-    monkeypatch.setattr(analysis_service, "_slots", slots)
-    async def execute(*unused):
-        return await analysis_service.bounded_calculation(lambda: pytest.fail("occupied calculation entered"))
-    app.executor = execute
-    assert slots.acquire(blocking=False) and slots.acquire(blocking=False)
-    try:
-        response = request(client, "tools/call", {"name": "run_simulation", "arguments": {
-            "context_handle": binding().context_handle, "start_time": "2026-10-08T12:00:00+08:00",
-            "duration": 3600, "output_node_ids": ["J1"],
-        }}).json()
-    finally:
-        slots.release()
-        slots.release()
-    error = response["result"]["structuredContent"]["error"]
-    assert response["result"]["isError"] is True
-    assert error["code"] == "CALCULATION_CAPACITY_EXCEEDED"
-    assert "not started" in error["message"]
+    monkeypatch.setattr(calculation_jobs, "submit_calculation", AsyncMock(side_effect=HTTPException(409, "Calculation request_key was used with different inputs")))
+    response = request(client, "tools/call", {"name": "run_simulation", "arguments": {
+        "context_handle": binding().context_handle, "start_time": "2026-10-08T12:00:00+08:00",
+        "duration": 3600, "output_node_ids": ["J1"],
+    }}).json()
+    assert response["result"]["structuredContent"]["error"]["code"] == "QUERY_CONTEXT_STALE"
 
 
-def test_burst_location_default_inputs_pass_protocol_validation(runtime, monkeypatch):
+def test_expired_calculation_result_has_distinct_mcp_error(runtime):
+    app, client, _, _ = runtime
+    app.executor = AsyncMock(side_effect=HTTPException(410, "Calculation result retention period has expired"))
+    response = request(client, "tools/call", {"name": "get_calculation_result", "arguments": {
+        "context_handle": binding().context_handle, "job_id": str(uuid4()),
+    }}).json()
+    assert response["result"]["structuredContent"]["error"]["code"] == "RESULT_EXPIRED"
+
+
+@pytest.mark.parametrize("maximum,expected", [(240, 120), (120, 120)])
+def test_burst_location_default_inputs_pass_protocol_validation(runtime, monkeypatch, maximum, expected):
     app, client, calls, _ = runtime
+    app.settings.max_request_timeout_seconds = maximum
+    from app.mcp import calculation_jobs
+    executor = AsyncMock(return_value=calculation_jobs.envelope("run_burst_location", {"job_id": str(uuid4()), "status": "queued"}))
+    monkeypatch.setattr(calculation_jobs, "submit_calculation", executor)
     monkeypatch.setattr(application, "authorize_query", AsyncMock())
     schema = next(tool["inputSchema"] for tool in request(client).json()["result"]["tools"]
                   if tool["name"] == "run_burst_location")
     assert "pressure_scada_ids" not in schema["required"]
+    assert schema["properties"]["timeout_seconds"]["default"] == expected
     response = request(client, "tools/call", {"name": "run_burst_location", "arguments": {
         "context_handle": binding().context_handle, "burst_leakage": 0.0274,
         "scada_burst_start": "2026-10-08T12:00:00+08:00",
         "scada_burst_end": "2026-10-08T13:00:00+08:00",
     }}).json()
     assert response["result"].get("isError") is not True
-    assert calls == [("session-1", "run_burst_location")]
+    assert calls == []
+    assert executor.await_args.args[3].timeout_seconds == expected
 
 
 def settings():
@@ -306,6 +300,37 @@ def binding(task=False):
     return value
 
 
+@pytest.mark.parametrize("seconds", [30, 120, 240])
+def test_all_tools_advertise_the_effective_host_request_budget(runtime, seconds):
+    app, client, _, _ = runtime
+    app.settings.query_timeout_seconds = seconds
+    for tool in request(client).json()["result"]["tools"]:
+        assert tool["_meta"]["io.tjwater/request-timeout"] == {
+            "defaultSeconds": seconds, "maxSeconds": 240,
+        }
+        timeout = tool["inputSchema"]["properties"].get("timeout_seconds")
+        if timeout:
+            assert timeout["default"] == seconds
+
+
+def test_omitted_calculation_http_budget_uses_server_configuration(runtime, monkeypatch):
+    from app.mcp import calculation_jobs
+    app, client, _, _ = runtime
+    app.settings.query_timeout_seconds = 240
+    monkeypatch.setattr(application, "authorize_query", AsyncMock())
+    submit = AsyncMock(return_value=calculation_jobs.envelope(
+        "run_simulation", {"job_id": str(uuid4()), "status": "queued"},
+    ))
+    monkeypatch.setattr(calculation_jobs, "submit_calculation", submit)
+    response = request(client, "tools/call", {"name": "run_simulation", "arguments": {
+        "context_handle": binding().context_handle, "start_time": "2026-10-08T12:00:00+08:00",
+        "duration": 3600, "output_node_ids": ["J1"],
+    }}).json()
+    assert response["result"].get("isError") is not True
+    assert submit.await_args.args[3].timeout_seconds == 240
+    assert submit.await_args.kwargs["job_timeout_seconds"] == 3600
+
+
 def test_modern_discovery_catalog_and_metadata(runtime):
     _, client, _, _ = runtime
     result = request(client, "server/discover").json()
@@ -318,7 +343,7 @@ def test_modern_discovery_catalog_and_metadata(runtime):
         assert (timeout is not None) == (TOOLS[tool["name"]].computation or TOOLS[tool["name"]].mutating)
         if timeout:
             assert timeout["default"] == 120
-            assert timeout["maximum"] == 3600
+            assert timeout["maximum"] == 240
     assert listed["ttlMs"] == 300000 and listed["cacheScope"] == "public"
     assert all(
         tool["inputSchema"]["additionalProperties"] is False for tool in listed["tools"]
@@ -518,6 +543,42 @@ def test_evidence_is_saved_before_return_and_call_conflicts_fail(runtime, monkey
     assert len(saved) == 1
 
 
+def test_large_node_coverage_retains_publishable_hint_and_saves_complete_geometry(runtime, monkeypatch):
+    app, client, _, _ = runtime
+    value = binding(task=True)
+    saved = []
+    monkeypatch.setattr(application, "verify_task", AsyncMock())
+    monkeypatch.setattr(contexts, "verify_task", AsyncMock())
+
+    class TaskService:
+        repo = SimpleNamespace(lock=lambda key: asyncio.sleep(0))
+
+        async def record_data_snapshot(self, task_id, snapshot):
+            saved.append(snapshot)
+            return SimpleNamespace(evidence_id=snapshot.source_ref), snapshot
+
+    monkeypatch.setattr(application, "task_service", lambda scope: TaskService())
+    geometry = {"type": "FeatureCollection", "features": [{
+        "type": "Feature", "properties": {"label": "x" * 13000},
+        "geometry": {"type": "Polygon", "coordinates": [[[117, 32], [118, 32], [118, 33], [117, 32]]]},
+    }]}
+    app.executor = AsyncMock(return_value=QueryResult(
+        source="generate_node_coverage", source_ref="mcp:before-persisting", queried_at=datetime.now(timezone.utc),
+        semantics="analysis", data={"result": {"map_payload": geometry, "coverage": {
+            "status": "ready", "groups": {"compare": {"node_ids": ["A"], "mapped_node_count": 1}},
+        }}},
+    ))
+    output = request(client, "tools/call", {"name": "generate_node_coverage", "arguments": {
+        "context_handle": value.context_handle,
+        "groups": [{"group_id": "compare", "title": "节点比较", "node_ids": ["A"]}],
+    }}).json()["result"]["structuredContent"]
+    assert output["truncated"] is True
+    command = output["data"]["presentation"]["command"]
+    assert command["action"] == "set_overlay" and command["evidence_id"] == saved[0].source_ref
+    assert "data" not in command
+    assert saved[0].data["data"]["result"]["map_payload"] == geometry
+
+
 def test_business_skill_is_delivered_over_mcp_without_recording_event_evidence(runtime, monkeypatch):
     app, client, _, _ = runtime
     value = binding(task=True)

tests/unit/test_node_coverage.py

@@ -0,0 +1,118 @@
+from datetime import datetime, timezone
+
+import pytest
+from fastapi.encoders import jsonable_encoder
+from pydantic import ValidationError
+from shapely.geometry import Point, shape
+
+from app.domain.schemas.operational_data import TaskDataSnapshot
+from app.mcp.analysis_models import NodeCoverageInput
+from app.services.node_coverage import build_node_coverage
+from app.services.operational_presentation import resolve_overlay
+
+
+def position(node, longitude, latitude=32.75):
+    return {"id": node, "longitude": longitude, "latitude": latitude}
+
+
+def group(ids, group_id="selected"):
+    return {"group_id": group_id, "title": "节点覆盖比较", "node_ids": ids}
+
+
+def link(link_id, start, end, status=None):
+    return {"id": link_id, "start_node_id": start, "end_node_id": end, "status": status}
+
+
+def test_coverage_can_compare_arbitrary_groups_without_assigning_isolation_semantics():
+    positions = [position("A", 117.24), position("B", 117.241), position("C", 117.26)]
+    coverage = build_node_coverage([group(["A", "B"], "monitoring"), group(["C"], "candidate")], positions, [])
+    assert coverage["status"] == "ready"
+    assert set(coverage["groups"]) == {"monitoring", "candidate"}
+    assert [f["properties"]["coverage_group"] for f in coverage["map_payload"]["features"]] == ["monitoring", "candidate"]
+    assert coverage["parameters"]["split_by_connectivity"] is False
+
+
+def test_connectivity_option_does_not_bridge_disconnected_groups():
+    positions = [position("A", 117.24), position("B", 117.241),
+                 position("C", 117.26), position("D", 117.261)]
+    coverage = build_node_coverage([group(["A", "B", "C", "D"])], positions,
+                                  [link("P1", "A", "B"), link("P2", "C", "D")],
+                                  split_by_connectivity=True)
+    feature = coverage["map_payload"]["features"][0]
+    geometry = shape(feature["geometry"])
+    assert geometry.geom_type == "MultiPolygon" and geometry.is_valid
+    assert feature["properties"]["component_count"] == 2
+    assert all(geometry.covers(Point(row["longitude"], row["latitude"])) for row in positions)
+    assert not geometry.covers(Point(117.25, 32.75))
+
+
+@pytest.mark.parametrize("status,excluded", [("CLOSED", []), (None, ["V"])])
+def test_connectivity_excludes_initially_closed_and_explicitly_excluded_links(status, excluded):
+    coverage = build_node_coverage([group(["A", "B"])], [position("A", 117.24), position("B", 117.26)],
+                                  [link("V", "A", "B", status)], split_by_connectivity=True,
+                                  excluded_link_ids=excluded)
+    assert coverage["groups"]["selected"]["component_count"] == 2
+
+
+def test_partial_geometries_are_reported_and_single_point_buffer_uses_meters():
+    coverage = build_node_coverage([group(["A", "A", "missing", "invalid"])],
+                                  [position("A", 117.24), position("invalid", float("nan"))], [])
+    assert coverage["status"] == "partial"
+    metadata = coverage["groups"]["selected"]
+    assert metadata["mapped_node_count"] == 1 and metadata["missing_node_ids"] == ["invalid", "missing"]
+    feature = coverage["map_payload"]["features"][0]
+    geometry = shape(feature["geometry"])
+    assert geometry.covers(Point(117.24, 32.75))
+    assert 0.0004 < geometry.bounds[2] - geometry.bounds[0] < 0.001
+    assert feature["properties"]["display_only"] is True
+
+
+def test_duplicate_and_collinear_positions_are_valid_and_parameters_are_preserved():
+    coverage = build_node_coverage([group(["A", "B", "C"])],
+                                  [position("A", 117.24), position("B", 117.24), position("C", 117.241)], [],
+                                  concave_ratio=0.5, buffer_m=40)
+    assert coverage["status"] == "ready"
+    assert coverage["parameters"]["concave_ratio"] == 0.5 and coverage["parameters"]["buffer_m"] == 40
+    assert shape(coverage["map_payload"]["features"][0]["geometry"]).is_valid
+
+
+def test_missing_positions_do_not_fabricate_an_area():
+    coverage = build_node_coverage([group(["missing"])], [], [])
+    assert coverage["status"] == "partial" and coverage["map_payload"]["features"] == []
+
+
+@pytest.mark.parametrize("mcp_envelope", [False, True])
+def test_derived_payload_round_trips_through_saved_evidence_overlay(mcp_envelope):
+    coverage = build_node_coverage([group(["A"])], [position("A", 117.24)], [])
+    data = jsonable_encoder({"result": {"map_payload": coverage["map_payload"]}})
+    if mcp_envelope:
+        data = {"source": "generate_node_coverage", "source_ref": "mcp:coverage", "data": data}
+    snapshot = TaskDataSnapshot(source="generate_node_coverage", source_ref="mcp:coverage",
+                                queried_at=datetime.now(timezone.utc), origin="system_data",
+                                truncated=False, data=data)
+    overlay = resolve_overlay(snapshot, "geometry")
+    assert overlay["features"][0]["properties"]["coverage_group"] == "selected"
+    assert shape(overlay["features"][0]["geometry"]).covers(Point(117.24, 32.75))
+
+
+def test_service_resolves_only_selected_project_nodes_and_reads_topology_only_when_requested(monkeypatch):
+    from app.services import node_coverage
+    def positions(network, ids):
+        assert network == "bound-project" and ids == ["A", "B"]
+        return [position("A", 117.24), position("B", 117.241)]
+    monkeypatch.setattr(node_coverage, "get_node_geographic_positions", positions)
+    monkeypatch.setattr(node_coverage, "get_node_coverage_links", lambda *args: pytest.fail("unnecessary topology query"))
+    result = node_coverage.generate_node_coverage("bound-project", [group(["A", "B"])])
+    assert result["coverage"]["status"] == "ready"
+    assert len(result["map_payload"]["features"]) == 1
+
+
+@pytest.mark.parametrize("extra", [
+    {"groups": []}, {"groups": [group(["A"]), group(["B"])]},
+    {"buffer_m": 0}, {"buffer_m": 1001}, {"concave_ratio": float("nan")},
+    {"excluded_link_ids": ["V"]}, {"groups": [{**group(["A"]), "coordinates": [117, 32]}]},
+    {"groups": [group([str(i) for i in range(10000)]), group(["extra"], "other")]},
+])
+def test_tool_rejects_invalid_parameters_and_unbounded_or_arbitrary_coordinate_inputs(extra):
+    with pytest.raises(ValidationError):
+        NodeCoverageInput.model_validate({"context_handle": "x" * 43, "groups": [group(["A"])], **extra})

tests/unit/test_operational_action_adapters.py

@@ -48,12 +48,13 @@ async def test_create_work_order_confirms_creation_not_field_completion():
         calls.append(request)
         return httpx.Response(201, json={"work_order_id": str(uuid4()),
             "project_id": str(data["context"].project_id), "created_at": "2026-09-23T08:00:00Z",
-            "origin": "synthetic", "status": "open"})
+            "origin": "system_data", "status": "open"})
     adapter = HttpOperationalActionDispatcher(settings(), transport=httpx.MockTransport(handler))
     result = await adapter.dispatch(**data)
     assert result.status == "succeeded"
     assert result.receipt.external_status == "open"
     assert "现场工作进度" in result.detail
+    assert result.detail == "外围工单已创建，现场工作进度由外围系统继续同步"
     assert calls[0].url.path == "/api/sample/v1/work-orders"
     assert calls[0].headers["X-Project-Id"] == str(data["context"].project_id)
 
@@ -66,7 +67,7 @@ async def test_work_order_receipt_accepts_awaiting_reply_status():
             "work_order_id": str(uuid4()),
             "project_id": str(data["context"].project_id),
             "created_at": "2026-09-23T08:00:00Z",
-            "origin": "synthetic",
+            "origin": "system_data",
             "status": "awaiting_reply",
         })
     adapter = HttpOperationalActionDispatcher(
@@ -90,7 +91,7 @@ async def test_action_work_order_is_not_reclassified_as_a_new_inbound_event():
             "work_order_id": str(uuid4()),
             "project_id": str(data["context"].project_id),
             "created_at": "2026-09-23T08:00:00Z",
-            "origin": "synthetic",
+            "origin": "system_data",
             "status": "open",
         })
 
@@ -111,7 +112,7 @@ async def test_lost_create_response_recovers_same_request_without_second_write()
             raise httpx.ReadTimeout("lost", request=request)
         return httpx.Response(200, json={"request_id": request.url.path.split("/")[-1],
             "result": {"work_order_id": str(uuid4()), "project_id": str(data["context"].project_id),
-                       "created_at": "2026-09-23T08:00:00Z", "origin": "synthetic", "status": "open"}})
+                       "created_at": "2026-09-23T08:00:00Z", "origin": "system_data", "status": "open"}})
     result = await HttpOperationalActionDispatcher(settings(), transport=httpx.MockTransport(handler)).dispatch(**data)
     assert result.status == "succeeded"
     assert methods == ["POST", "GET"]
@@ -122,7 +123,7 @@ async def test_unconfirmed_receipt_remains_unknown_not_failed_or_succeeded():
     data = inputs()
     def handler(request):
         return httpx.Response(200, json={"work_order_id": str(uuid4()), "project_id": str(uuid4()),
-            "created_at": "2026-09-23T08:00:00Z", "origin": "synthetic", "status": "open"})
+            "created_at": "2026-09-23T08:00:00Z", "origin": "system_data", "status": "open"})
     result = await HttpOperationalActionDispatcher(settings(), transport=httpx.MockTransport(handler)).dispatch(**data)
     assert result.status == "unknown"
     assert result.receipt is None

tests/unit/test_operational_presentation.py

@@ -54,3 +54,35 @@ def test_overlay_rejects_missing_or_invalid_recorded_spatial_data(data):
 def test_presentation_input_rejects_free_form_or_invalid_commands(command):
     with pytest.raises(ValidationError):
         TaskPresentationEventCommand(request_id=uuid4(), run_id=uuid4(), source_session_id="session", command=command)
+
+
+@pytest.mark.parametrize("command", [
+    {"action": "locate_features", "layer": "scada", "feature_ids": ["D-1"]},
+    {"action": "zoom_to_map", "center": [121.4, 31.1], "zoom": 24},
+    {"action": "render_scada_analysis", "items": [{"device_id": "D-1", "level": "unrated"}]},
+    {"action": "clear_scada_analysis"},
+])
+def test_unified_presentation_commands_preserve_flat_shape(command):
+    event = TaskPresentationEventCommand(request_id=uuid4(), run_id=uuid4(), source_session_id="session", command=command)
+    assert event.command.model_dump(mode="json") == command
+
+
+@pytest.mark.parametrize("command", [
+    {"action": "highlight_features", "layer": "pipes", "feature_ids": ["P-1"]},
+    {"action": "locate_features", "layer": "pipes", "feature_ids": []},
+    {"action": "zoom_to_map", "center": [181, 31], "zoom": 1},
+    {"action": "zoom_to_map", "center": [121, 91], "zoom": 1},
+    {"action": "zoom_to_map", "center": [121, 31, 0], "zoom": 1},
+    {"action": "zoom_to_map", "center": [True, 31], "zoom": 1},
+    {"action": "zoom_to_map", "center": [121, 31], "zoom": 25},
+    {"action": "zoom_to_map", "center": [121, 31], "zoom": -1},
+    {"action": "render_scada_analysis", "items": [{"device_id": "D", "level": "urgent"}]},
+    {"action": "render_scada_analysis", "items": [{"device_id": "D", "level": "high"}] * 2},
+    {"action": "render_scada_analysis", "items": [{"device_id": str(i), "level": "low"} for i in range(101)]},
+    {"action": "open_chart", "chart_id": "X", "evidence_refs": [], "chart": {
+        "title": "X", "chart_type": "line", "x_data": ["t"], "series": [{"name": "s", "data": [1]}],
+    }},
+])
+def test_unified_presentation_rejects_invalid_commands(command):
+    with pytest.raises(ValidationError):
+        TaskPresentationEventCommand(request_id=uuid4(), run_id=uuid4(), source_session_id="session", command=command)

tests/unit/test_operational_task_storage_readiness.py

@@ -17,12 +17,19 @@ def anyio_backend():
 
 
 @pytest.mark.anyio
-@pytest.mark.parametrize("available", [False, True])
-async def test_execution_requires_message_storage(available):
-    cursor = SimpleNamespace(fetchone=AsyncMock(return_value={"available": available}))
+@pytest.mark.parametrize("available,agent_ownership_available,ready", [
+    (False, False, False), (True, False, False), (True, True, True),
+])
+async def test_execution_requires_message_and_agent_ownership_storage(
+    available, agent_ownership_available, ready,
+):
+    cursor = SimpleNamespace(fetchone=AsyncMock(return_value={
+        "available": available,
+        "agent_ownership_available": agent_ownership_available,
+    }))
     connection = SimpleNamespace(execute=AsyncMock(return_value=cursor))
     repository = OperationalTaskRepository(connection, uuid4())
-    if available:
+    if ready:
         await repository.require_execution_storage()
     else:
         with pytest.raises(DomainError) as failure:
@@ -30,9 +37,31 @@ async def test_execution_requires_message_storage(available):
         assert failure.value.status == 503
         assert failure.value.code == "DEPENDENCY_UNAVAILABLE"
         assert "024_agent_messages.sql" in failure.value.message
+        assert "030_operational_run_agent_ownership.sql" in failure.value.message
     assert "to_regclass('workbench.agent_messages')" in connection.execute.call_args.args[0]
 
 
+@pytest.mark.anyio
+@pytest.mark.parametrize("available", [False, True])
+async def test_run_update_uses_agent_owner_column_only_after_migration(available):
+    availability = SimpleNamespace(fetchone=AsyncMock(return_value={"available": available}))
+    updated = SimpleNamespace(rowcount=1)
+    connection = SimpleNamespace(execute=AsyncMock(side_effect=[availability, updated]))
+    repository = OperationalTaskRepository(connection, uuid4())
+    run_id = str(uuid4())
+    await repository.update_run({
+        "run_id": run_id, "status": "running", "revision": 2,
+        "agent_session_id": "session-1", "agent_instance_id": "agent-a",
+        "lease_expires_at": None, "heartbeat_at": None, "started_at": None,
+        "finished_at": None, "failure_code": None, "failure_detail": None,
+        "dispatch_attempts": 0, "next_dispatch_at": None, "updated_at": None,
+    }, 1)
+
+    query, params = connection.execute.call_args.args
+    assert ("agent_instance_id=%s" in query) is available
+    assert ("agent-a" in params) is available
+
+
 @pytest.mark.anyio
 async def test_missing_message_storage_rejects_claim_before_state_changes():
     @asynccontextmanager
@@ -56,3 +85,16 @@ async def test_missing_message_storage_rejects_claim_before_state_changes():
     repository.lock.assert_not_awaited()
     repository.update_run.assert_not_awaited()
     repository.update_task.assert_not_awaited()
+
+
+@pytest.mark.anyio
+async def test_presentation_history_storage_filters_project_task_and_committed_cursor():
+    project_id, task_id = uuid4(), uuid4()
+    connection = SimpleNamespace(execute=AsyncMock(return_value=SimpleNamespace(fetchall=AsyncMock(return_value=[]))))
+    repository = OperationalTaskRepository(connection, project_id)
+    assert await repository.list_presentation_events(str(task_id), 7, 30, 21) == []
+    query, params = connection.execute.call_args.args
+    assert params == (str(project_id), str(task_id), 7, 30, 21)
+    assert "project_id=%s" in query and "aggregate_id=%s" in query
+    assert "operational.presentation.requested" in query and "event_seq<=%s" in query
+    assert "ORDER BY event_seq LIMIT %s" in query

tests/unit/test_operational_tasks.py

@@ -402,6 +402,9 @@ class RunTaskRepository(TaskRepository):
     async def get_run(self, run_id):
         return self.run if run_id == self.run["run_id"] else None
 
+    async def require_execution_storage(self):
+        return None
+
     async def update_run(self, run, expected_revision):
         assert run["revision"] == expected_revision + 1
         self.run = run
@@ -416,6 +419,11 @@ class RunTaskRepository(TaskRepository):
     async def current_event_cursor(self):
         return len(self.events)
 
+    async def list_presentation_events(self, task_id, after, through, limit):
+        return [{"event_seq": index, **event} for index, event in enumerate(self.events, 1)
+                if after < index <= through and event["topic"] == "operational.presentation.requested"
+                and event["aggregate_id"] == task_id][:limit]
+
 
 def _active_run(task, now):
     return OperationalTaskRun(
@@ -427,6 +435,30 @@ def _active_run(task, now):
     ).model_dump(mode="json")
 
 
+@pytest.mark.anyio
+async def test_run_claim_persists_agent_instance_ownership():
+    now = datetime(2026, 9, 23, 8, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=None)
+    task.update(status="queued", agent_session_id=None, agent_lease_expires_at=None)
+    run = _active_run(task, now)
+    run.update(status="queued", revision=1, agent_session_id=None,
+               lease_expires_at=None, heartbeat_at=None, started_at=None)
+    repository = RunTaskRepository(task, run)
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+
+    claimed = await service.claim_run(
+        task["task_id"], run["run_id"],
+        AgentRunCommand(
+            request_id=run["execution_request_id"], expected_revision=1,
+            source_session_id="session-1", agent_instance_id="agent-a",
+        ),
+    )
+
+    assert claimed.agent_session_id == "session-1"
+    assert claimed.agent_instance_id == "agent-a"
+
+
 @pytest.mark.anyio
 async def test_progress_can_publish_without_conclusion_and_replaces_the_complete_graph():
     now = datetime(2026, 10, 10, tzinfo=timezone.utc)
@@ -544,6 +576,64 @@ async def test_presentation_cursor_and_event_are_read_only_and_idempotent():
         await service.presentation_cursor(uuid4())
 
 
+@pytest.mark.anyio
+@pytest.mark.parametrize("payload", [
+    {"action": "locate_features", "layer": "scada", "feature_ids": ["D-1"]},
+    {"action": "zoom_to_map", "center": [121.4, 31.1], "zoom": 15},
+    {"action": "render_scada_analysis", "items": [{"device_id": "D-1", "level": "high"}]},
+    {"action": "clear_scada_analysis"},
+])
+async def test_unified_commands_record_canonical_events_without_mutating_task(payload):
+    now = datetime(2026, 10, 10, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=now + timedelta(minutes=1))
+    repository = RunTaskRepository(task, _active_run(task, now))
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+    command = TaskPresentationEventCommand(
+        request_id=uuid4(), run_id=repository.run["run_id"], source_session_id="session-1", command=payload,
+    )
+    result = await service.publish_presentation_event(task["task_id"], command)
+    assert result.recorded and repository.events[0]["metadata"]["command"] == payload
+    assert await service.publish_presentation_event(task["task_id"], command) == result
+    assert len(repository.events) == 1 and repository.task["revision"] == task["revision"]
+
+
+@pytest.mark.anyio
+async def test_saved_mcp_coverage_envelope_publishes_groups_without_agent_geometry():
+    from fastapi.encoders import jsonable_encoder
+    from app.services.node_coverage import build_node_coverage
+
+    now = datetime(2026, 10, 10, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=now + timedelta(minutes=1))
+    repository = RunTaskRepository(task, _active_run(task, now))
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+    coverage = build_node_coverage(
+        [{"group_id": "one", "title": "第一组", "node_ids": ["A"]},
+         {"group_id": "two", "title": "第二组", "node_ids": ["A", "B"]}],
+        [{"id": "A", "longitude": 117.24, "latitude": 32.75},
+         {"id": "B", "longitude": 117.241, "latitude": 32.75}],
+        [],
+    )
+    snapshot = TaskDataSnapshot(
+        source="generate_node_coverage", source_ref="mcp:node-coverage", queried_at=now,
+        origin="system_data", truncated=False,
+        data=jsonable_encoder({"source": "generate_node_coverage", "data": {
+            "result": {"map_payload": coverage["map_payload"]},
+        }}),
+    )
+    evidence, _ = await service.record_data_snapshot(task["task_id"], snapshot)
+    await service.publish_presentation_event(task["task_id"], TaskPresentationEventCommand(
+        request_id=uuid4(), run_id=repository.run["run_id"], source_session_id="session-1",
+        command={"action": "set_overlay", "overlay_id": "isolation-coverage", "title": "节点覆盖示意",
+                 "mode": "geometry", "evidence_id": evidence.evidence_id},
+    ))
+    features = repository.events[-1]["metadata"]["command"]["data"]["features"]
+    assert {feature["properties"]["coverage_group"] for feature in features} == {"one", "two"}
+    assert all(feature["properties"]["display_only"] for feature in features)
+    assert [feature["properties"]["node_count"] for feature in features] == [1, 2]
+
+
 @pytest.mark.anyio
 async def test_overlay_event_uses_stored_coordinates_and_rejects_untrusted_or_oversized_data():
     now = datetime(2026, 10, 10, tzinfo=timezone.utc)
@@ -566,6 +656,16 @@ async def test_overlay_event_uses_stored_coordinates_and_rejects_untrusted_or_ov
     emitted = repository.events[-1]["metadata"]["command"]["data"]
     assert emitted["features"][0]["geometry"]["coordinates"] == [121.4, 31.1]
     assert emitted["features"][0]["properties"]["weight"] == 1
+    await service.publish_presentation_event(task["task_id"], TaskPresentationEventCommand(
+        request_id=uuid4(), **identity, command={**overlay, "evidence_id": evidence.evidence_id, "data": emitted},
+    ))
+    spoofed = {**emitted, "features": [{**emitted["features"][0],
+        "geometry": {"type": "Point", "coordinates": [120, 30]}}]}
+    with pytest.raises(DomainError) as spoof:
+        await service.publish_presentation_event(task["task_id"], TaskPresentationEventCommand(
+            request_id=uuid4(), **identity, command={**overlay, "evidence_id": evidence.evidence_id, "data": spoofed},
+        ))
+    assert spoof.value.code == "OVERLAY_DATA_MISMATCH"
     with pytest.raises(DomainError) as missing:
         await service.publish_presentation_event(task["task_id"], TaskPresentationEventCommand(
             request_id=uuid4(), **identity, command={**overlay, "evidence_id": "not-this-task"},
@@ -580,7 +680,7 @@ async def test_overlay_event_uses_stored_coordinates_and_rejects_untrusted_or_ov
             request_id=uuid4(), **identity, command={**overlay, "evidence_id": evidence.evidence_id},
         ))
     assert oversized.value.code == "PRESENTATION_TOO_LARGE"
-    assert len(repository.events) == 1
+    assert len(repository.events) == 2
 
 
 @pytest.mark.anyio
@@ -699,6 +799,28 @@ async def test_run_heartbeat_does_not_change_task_business_revision():
     assert repository.events == []
 
 
+@pytest.mark.anyio
+async def test_run_heartbeat_rejects_a_different_agent_instance_with_same_session_id():
+    now = datetime(2026, 9, 23, 8, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=now + timedelta(minutes=1))
+    run = _active_run(task, now)
+    run["agent_instance_id"] = "agent-a"
+    repository = RunTaskRepository(task, run)
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+
+    with pytest.raises(DomainError) as mismatch:
+        await service.heartbeat_run(
+            task["task_id"], repository.run["run_id"],
+            AgentRunCommand(
+                request_id=uuid4(), expected_revision=2,
+                source_session_id="session-1", agent_instance_id="agent-b",
+            ),
+        )
+
+    assert mismatch.value.code == "AGENT_INSTANCE_MISMATCH"
+
+
 @pytest.mark.anyio
 async def test_complete_without_same_generation_artifacts_fails_run_explicitly():
     now = datetime(2026, 9, 23, 8, tzinfo=timezone.utc)
@@ -1537,7 +1659,7 @@ def _work_order_snapshot(work_order_id, now):
         "feedback_requirements": ["上传检测结果"],
         "activity": [{"sequence": 1, "occurred_at": now.isoformat(), "actor": "水质班组",
                       "kind": "feedback", "title": "现场反馈", "detail": "上传复测结果"}],
-        "origin": "synthetic",
+        "origin": "system_data",
         "updated_at": now.isoformat(),
     }
 
@@ -1842,3 +1964,126 @@ async def test_interactive_report_publication_is_idempotent_and_versioned_with_e
             request_id=uuid4(), source_session_id="session-1", report=report,
         ))
     assert error.value.code == "EVIDENCE_NOT_FOUND"
+
+
+@pytest.mark.anyio
+async def test_presentation_history_is_task_scoped_paginated_and_readable_after_run_finishes():
+    now = datetime(2026, 10, 10, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now)
+    repository = RunTaskRepository(task, _active_run(task, now))
+    repository.run["status"] = "succeeded"
+    for index in range(4):
+        await repository.append_event(
+            topic="operational.presentation.requested", aggregate_type="operational_task_presentation",
+            aggregate_id=task["task_id"] if index != 1 else str(uuid4()), aggregate_revision=1,
+            occurred_at=now, metadata={"run_id": repository.run["run_id"],
+            "execution_generation": 1 if index < 2 else 2,
+            "command": {"action": "clear_highlights"}},
+        )
+    service = OperationalTaskService(repository, context, {OPERATIONAL_TASK_VIEW}, clock=lambda: now)
+    first = await service.presentation_history(task["task_id"], limit=1)
+    assert [item.event_seq for item in first.items] == [1]
+    assert first.event_cursor == 4 and first.next_after == 1
+    # An append during pagination is outside the captured boundary.
+    repository.events.append(repository.events[-1])
+    second = await service.presentation_history(task["task_id"], after=first.next_after, through=first.event_cursor)
+    assert [item.event_seq for item in second.items] == [3, 4]
+    assert [item.execution_generation for item in second.items] == [2, 2]
+    assert second.next_after is None and second.event_cursor == 4
+    assert repository.run["status"] == "succeeded" and repository.task["revision"] == task["revision"]
+    with pytest.raises(DomainError):
+        await service.presentation_history(uuid4())
+    assert len(repository.events) == 5
+
+
+@pytest.mark.anyio
+@pytest.mark.parametrize("report_factory", [_current_report, _insufficient_report])
+async def test_final_report_rejected_while_original_run_has_pending_calculations(report_factory):
+    now = datetime(2026, 9, 23, 8, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=now + timedelta(minutes=1))
+    repository = RunTaskRepository(task, _active_run(task, now))
+    pending_job = str(uuid4())
+    repository.pending_calculation_jobs = AsyncMock(return_value=[pending_job])
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+    with pytest.raises(DomainError) as error:
+        await service.publish_artifacts(task['task_id'], AgentArtifactsCommand(
+            request_id=uuid4(), source_session_id='session-1', run_id=repository.run['run_id'], report=report_factory(),
+        ))
+    assert error.value.code == 'PENDING_CALCULATIONS'
+    assert error.value.status == 409
+    assert error.value.recovery == 'retry_same_request'
+    assert error.value.refs == [pending_job]
+    repository.pending_calculation_jobs.assert_awaited_once_with(
+        str(task['task_id']), str(repository.run['run_id']), str(repository.run['execution_request_id']),
+    )
+    assert not repository.reports
+    assert repository.run['status'] in {'claimed', 'running'}
+
+
+@pytest.mark.anyio
+async def test_completion_rejected_until_calculations_reach_terminal_state():
+    now = datetime(2026, 9, 23, 8, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=now + timedelta(minutes=1))
+    repository = RunTaskRepository(task, _active_run(task, now))
+    repository.pending_calculation_jobs = AsyncMock(return_value=[str(uuid4())])
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+    command = AgentRunCommand(request_id=uuid4(), expected_revision=repository.run['revision'], source_session_id='session-1')
+    with pytest.raises(DomainError) as error:
+        await service.complete_run(task['task_id'], repository.run['run_id'], command)
+    assert error.value.code == 'PENDING_CALCULATIONS'
+    assert repository.run['status'] in {'claimed', 'running'}
+    repository.pending_calculation_jobs.return_value = []
+    # Once terminal, the normal final-artifact requirement runs again.
+    completed = await service.complete_run(task['task_id'], repository.run['run_id'], command)
+    assert repository.run['failure_code'] == 'ARTIFACT_INCOMPLETE'
+    assert completed.lifecycle_status == 'failed'
+
+
+@pytest.mark.anyio
+async def test_pending_calculation_query_matches_exact_original_execution_binding():
+    job_id = uuid4()
+    connection = SimpleNamespace(execute=AsyncMock())
+    connection.execute.return_value.fetchall = AsyncMock(return_value=[{'job_id': job_id}])
+    project_id, task_id, run_id, request_id = (uuid4() for _ in range(4))
+    repo = OperationalTaskRepository(connection, project_id)
+    assert await repo.pending_calculation_jobs(str(task_id), str(run_id), str(request_id)) == [str(job_id)]
+    sql, parameters = connection.execute.await_args.args
+    assert parameters == (str(project_id), str(task_id), str(run_id), str(request_id))
+    assert "status IN ('queued','running','cancel_requested')" in sql
+    assert "binding->>'execution_request_id'=%s" in sql
+
+
+@pytest.mark.anyio
+async def test_interim_report_can_explain_progress_while_calculation_is_pending():
+    now = datetime(2026, 9, 23, 8, tzinfo=timezone.utc)
+    context = _context()
+    task = _task_document(context, now, lease=now + timedelta(minutes=1))
+    repository = RunTaskRepository(task, _active_run(task, now))
+    repository.pending_calculation_jobs = AsyncMock(return_value=[str(uuid4())])
+    service = OperationalTaskService(repository, context, set(), clock=lambda: now)
+    result = await service.publish_artifacts(task['task_id'], AgentArtifactsCommand(
+        request_id=uuid4(), source_session_id='session-1', run_id=repository.run['run_id'],
+        report={**_current_report(), 'stage': 'interim'},
+    ))
+    assert result.report_revision == 1
+    repository.pending_calculation_jobs.assert_not_awaited()
+
+
+@pytest.mark.anyio
+async def test_missing_calculation_migration_fails_finalization_closed(monkeypatch):
+    from psycopg.errors import UndefinedTable
+    from app.infra.db.postgresql import operational_tasks as storage
+    @asynccontextmanager
+    async def transaction():
+        yield
+    connection = SimpleNamespace(transaction=transaction, execute=AsyncMock(side_effect=UndefinedTable('calculation_jobs missing')))
+    monkeypatch.setattr(storage, 'lock_event_transaction', AsyncMock())
+    repo = OperationalTaskRepository(connection, uuid4())
+    with pytest.raises(DomainError) as error:
+        async with repo.transaction():
+            await repo.pending_calculation_jobs(str(uuid4()), str(uuid4()), str(uuid4()))
+    assert error.value.code == 'DEPENDENCY_UNAVAILABLE'
+    assert error.value.status == 503

tests/unit/test_operational_worker.py

@@ -90,8 +90,7 @@ async def test_worker_reconciles_messages_at_capacity_and_preserves_transport_re
     async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
         assert await OperationalWorker(settings, client).tick() == 0
     assert submitted == [row["message_id"] for row in pending]
-    assert len(failures) == 1 and pending[1]["message_id"] in failures[0][0]
-    assert failures[0][1]["status"] == "failed"
+    assert failures == []
 
 
 @pytest.mark.anyio
@@ -126,6 +125,8 @@ async def test_persisted_queue_retries_with_backend_execution_identity():
     task_id, run_id, execution_id = str(uuid4()), str(uuid4()), str(uuid4())
     submitted = []
     def handler(request):
+        if request.url.path == "/health":
+            return httpx.Response(200, json={"ok": True, "agent_instance_id": "agent-a"})
         assert request.headers["Authorization"] == "Bearer TEST_ONLY"
         assert not request.url.path.endswith("schedules/dispatch-due")
         if request.url.path.endswith("leases/recover"):
@@ -140,7 +141,10 @@ async def test_persisted_queue_retries_with_backend_execution_identity():
         body = json.loads(request.content)
         assert str(request.url) == "http://agent.test/api/v1/task-executions"
         submitted.append(body)
-        return httpx.Response(202, json={"task_id": task_id, "run_id": run_id, "session_id": "execution-session"})
+        return httpx.Response(202, json={
+            "task_id": task_id, "run_id": run_id, "session_id": "execution-session",
+            "agent_instance_id": "agent-a",
+        })
     settings = WorkerSettings("http://server.test", "http://agent.test", str(uuid4()), bearer="TEST_ONLY")
     async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
         worker = OperationalWorker(settings, client)
@@ -149,6 +153,7 @@ async def test_persisted_queue_retries_with_backend_execution_identity():
     assert submitted[0] == submitted[1]
     assert submitted[0] == {
         "task_id": task_id, "run_id": run_id, "execution_request_id": execution_id,
+        "agent_instance_id": "agent-a",
     }
     assert "credential" not in submitted[0]
 
@@ -178,6 +183,8 @@ async def test_bounded_queue_continues_after_one_agent_handoff_fails():
     limits, submitted, failures = [], [], []
 
     def handler(request):
+        if request.url.path == "/health":
+            return httpx.Response(200, json={"ok": True, "agent_instance_id": "agent-a"})
         if request.method == "GET":
             if request.url.params.get("status") != "queued":
                 return httpx.Response(200, json={"items": []})
@@ -192,7 +199,10 @@ async def test_bounded_queue_continues_after_one_agent_handoff_fails():
         submitted.append(body)
         if body["task_id"] == runs[0]["task_id"]:
             return httpx.Response(503, json={"detail": "temporary failure"})
-        return httpx.Response(202, json={"task_id": body["task_id"], "run_id": body["run_id"]})
+        return httpx.Response(202, json={
+            "task_id": body["task_id"], "run_id": body["run_id"],
+            "agent_instance_id": "agent-a",
+        })
 
     settings = WorkerSettings("http://server.test", "http://agent.test", str(uuid4()), bearer="TEST_ONLY")
     async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
@@ -257,10 +267,16 @@ async def test_worker_refreshes_running_agent_credentials_after_restart():
     renewed = []
 
     def handler(request):
+        if request.url.path == "/health":
+            return httpx.Response(200, json={"ok": True, "agent_instance_id": "agent-a"})
         if request.url.path.endswith("leases/recover"):
             return httpx.Response(200, json={"items": []})
         if request.method == "GET":
-            items = [{"run_id": run_id, "agent_session_id": "existing-session"}] if request.url.params.get("status") == "running" else []
+            items = [{
+                "run_id": run_id,
+                "agent_session_id": "existing-session",
+                "agent_instance_id": "agent-a",
+            }] if request.url.params.get("status") == "running" else []
             return httpx.Response(200, json={"items": items})
         assert request.url.host == "agent.test"
         assert str(request.url) == "http://agent.test/api/v1/session/existing-session/credentials"
@@ -274,3 +290,34 @@ async def test_worker_refreshes_running_agent_credentials_after_restart():
     async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
         assert await OperationalWorker(settings, client).tick() == 0
     assert len(renewed) == 1
+
+
+@pytest.mark.anyio
+async def test_worker_only_restores_sessions_owned_by_target_agent_instance():
+    own_run_id, foreign_run_id, project_id = str(uuid4()), str(uuid4()), str(uuid4())
+    renewed = []
+
+    def handler(request):
+        if request.url.path == "/health":
+            return httpx.Response(200, json={"ok": True, "agent_instance_id": "agent-a"})
+        if request.url.path.endswith("leases/recover"):
+            return httpx.Response(200, json={"items": []})
+        if request.method == "GET":
+            items = [{
+                "run_id": own_run_id,
+                "agent_session_id": "own-session",
+                "agent_instance_id": "agent-a",
+            }, {
+                "run_id": foreign_run_id,
+                "agent_session_id": "foreign-session",
+                "agent_instance_id": "agent-b",
+            }] if request.url.params.get("status") == "running" else []
+            return httpx.Response(200, json={"items": items})
+        assert str(request.url) == "http://agent.test/api/v1/session/own-session/credentials"
+        renewed.append(request)
+        return httpx.Response(202, json={"accepted": True})
+
+    settings = WorkerSettings("http://server.test", "http://agent.test", project_id, bearer="TEST_ONLY")
+    async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
+        assert await OperationalWorker(settings, client).tick() == 0
+    assert len(renewed) == 1

tests/unit/test_peripheral_managed_projection.py

@@ -86,7 +86,7 @@ def test_active_work_order_projection_removes_only_tombstone_fields():
     assert "deleted_at" not in result
     assert "deleted_by" not in result
     assert stored_document == document
-    WorkOrder.model_validate(result)
+    assert WorkOrder.model_validate(result).origin == "system_data"
 
 
 @pytest.mark.parametrize(

tests/unit/test_peripheral_managed_resources.py

@@ -24,7 +24,7 @@ class ManagedStore:
         now = "2026-09-18T00:00:00Z"
         return command.model_dump(mode="json", exclude={"request_id"}) | {
             "work_order_id": str(uuid4()), "project_id": str(PROJECT_ID), "revision": 1,
-            "origin": "synthetic", "created_at": now, "updated_at": now,
+            "origin": getattr(command, "origin", "system_data"), "created_at": now, "updated_at": now,
             "created_by": str(USER_ID), "updated_by": str(USER_ID), "activity": [],
         }
 
@@ -88,12 +88,17 @@ def test_managed_work_order_and_data_routes_are_typed_crud():
             "scenario_code": "WQ-SECONDARY-SUPPLY-001",
             "feedback_requirements": ["上传检测结果"],
         })
-        assert created.status_code == 201 and created.json()["origin"] == "synthetic"
+        assert created.status_code == 201 and created.json()["origin"] == "system_data"
         assert created.json()["work_order_type"] == "water_quality_complaint"
         assert created.json()["location"] == "演示小区"
         assert created.json()["scenario_code"] == "WQ-SECONDARY-SUPPLY-001"
         assert created.json()["feedback_requirements"] == ["上传检测结果"]
         assert "subtasks" not in created.json()
+        rejected = client.post("/api/sample/v1/work-orders", json={
+            "request_id": str(uuid4()), "title": "非法来源", "summary": "不接受演练来源",
+            "origin": "synthetic",
+        })
+        assert rejected.status_code == 422
         work_order_id = created.json()["work_order_id"]
         changed = client.patch(f"/api/sample/v1/work-orders/{work_order_id}", json={
             "request_id": str(uuid4()), "expected_revision": 1, "status": "awaiting_reply",

tests/unit/test_peripheral_provenance.py

@@ -0,0 +1,42 @@
+from types import SimpleNamespace
+from uuid import uuid4
+
+import httpx
+import pytest
+
+from app.auth.project_dependencies import ProjectContext
+from app.domain.schemas.operational_data import TaskDataQuery
+from app.infra.integrations.operational_data import TaskDataSettings, read_task_data
+from app.mcp import peripheral
+
+
+@pytest.fixture
+def anyio_backend():
+    return "asyncio"
+
+
+@pytest.mark.anyio
+@pytest.mark.parametrize("origins", [
+    [], ["system_data"], ["synthetic"], ["system_data", "synthetic"], [None], ["unverified"],
+])
+async def test_http_and_mcp_normalize_work_orders_as_system_data(monkeypatch, origins):
+    project_id = uuid4()
+    items = [{"work_order_id": str(uuid4()), "project_id": str(project_id), "origin": origin} for origin in origins]
+
+    async def rows(*_args, **_kwargs):
+        return [{"document": item} for item in items]
+
+    monkeypatch.setattr(peripheral, "read_rows", rows)
+    query = TaskDataQuery(source="peripheral_work_orders")
+    context = ProjectContext(project_id=project_id, user_id=uuid4(), project_code="test", project_role="member")
+    kwargs = dict(context=context, credential="Bearer TEST_ONLY", query=query,
+                  settings=TaskDataSettings(_env_file=None, peripheral_url="http://peripheral.test"),
+                  transport=httpx.MockTransport(lambda _: httpx.Response(200, json={"items": items})))
+    scope = SimpleNamespace(connection=object(), project=SimpleNamespace(project_id=project_id))
+    args = SimpleNamespace(source=query.source, limit=20, offset=0)
+    snapshots = [await read_task_data(**kwargs), await peripheral.read_peripheral_query(scope, args)]
+    for snapshot in snapshots:
+        assert snapshot.origin == "system_data"
+        assert [item["origin"] for item in snapshot.data["items"]] == ["system_data"] * len(origins)
+        assert "provenance_scope" not in snapshot.data
+        assert "source_origins" not in snapshot.data