fix: restore heaven interpretation persistence dependency

app/backend/features/heaven/service.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import copy
 import json
 import re
+import secrets
 from datetime import date, datetime, timedelta
 from typing import Any
 

app/backend/llm/service.py

@@ -4,6 +4,7 @@ from datetime import datetime, timezone
 from typing import Any
 from urllib.parse import urlparse
 
+from backend.bootstrap.config import validate_text
 from backend.features.screener.compiler import LLMCompilerError, test_llm_connection
 
 

app/config/architecture-inventory.json

@@ -354,8 +354,8 @@
     },
     {
       "path": "backend/features/heaven/service.py",
-      "bytes": 63123,
-      "lines": 1303
+      "bytes": 63138,
+      "lines": 1304
     },
     {
       "path": "backend/features/market/insights.py",

app/tests/test_heaven_readings.py

@@ -124,6 +124,50 @@ class HeavenReadingTests(unittest.TestCase):
             len(self.database.list_heaven_readings(self.owner["id"], "fortune")), 1
         )
 
+    def test_trend_interpretation_saves_successful_agent_result(self):
+        service = DashboardService.__new__(DashboardService)
+        service.database = self.database
+        service._request_context = threading.local()
+        service._request_context.user_id = self.owner["id"]
+        setup = {
+            "trade_date": "20260723",
+            "chart": {
+                "available": True,
+                "sector": "银行",
+                "stock": {"code": "000001", "name": "平安银行"},
+                "hexagram": {
+                    "name": "中孚",
+                    "lines": [],
+                    "transformed": {"name": "小畜"},
+                },
+                "movement": {},
+            },
+        }
+
+        with patch.object(service, "heaven_setup", return_value=setup), patch.object(
+            service,
+            "_call_heaven_agent",
+            return_value=(
+                {"answer": "完整的解势结果", "model": "test", "latency_ms": 1},
+                "primary",
+            ),
+        ):
+            result = service.heaven_interpret(
+                {
+                    "mode": "trend",
+                    "trade_date": "2026-07-23",
+                    "stock_code": "000001",
+                }
+            )
+
+        self.assertFalse(result["reused"])
+        self.assertEqual(result["answer"], "完整的解势结果")
+        self.assertEqual(result["reading"]["subject"], "000001 平安银行")
+        self.assertEqual(
+            self.database.list_heaven_readings(self.owner["id"], "trend")[0]["answer"],
+            "完整的解势结果",
+        )
+
 
 if __name__ == "__main__":
     unittest.main()

app/tests/test_llm_gateway.py

@@ -1,8 +1,10 @@
 from __future__ import annotations
 
 import unittest
+from unittest.mock import patch
 
 from backend.llm import LLMGateway, LLMGatewayError
+from backend.llm.service import LLMServiceMixin
 
 
 class ProviderFailure(RuntimeError):
@@ -113,6 +115,31 @@ class LLMGatewayTests(unittest.TestCase):
         with self.assertRaisesRegex(LLMGatewayError, "额度已用完"):
             gateway.call("mentor", "mentor-v1", lambda model: "unused", (ProviderFailure,))
 
+    def test_saved_system_model_can_reach_the_connection_probe(self) -> None:
+        service = LLMServiceMixin()
+        service._system_credentials = {
+            "llm_models": [
+                {
+                    "id": "primary-model",
+                    "name": "主模型",
+                    "api_key": "secret",
+                    "base_url": "https://model.example/v1",
+                    "model": "model-name",
+                }
+            ]
+        }
+        service.llm_gateway = self.gateway()
+        expected = {"ok": True, "reply": "OK"}
+        with patch(
+            "backend.llm.service.test_llm_connection", return_value=expected
+        ) as connection_probe:
+            result = service.test_system_llm_profile("primary-model", {})
+
+        self.assertEqual(result, expected)
+        connection_probe.assert_called_once_with(
+            "secret", "https://model.example/v1", "model-name"
+        )
+
 
 if __name__ == "__main__":
     unittest.main()

docs/governance/code-reduction.md

@@ -274,6 +274,26 @@
 本批基线为`xiaobai-reduction-10-repository-ownership-20260801`；检查点为
 `xiaobai-reduction-11-background-jobs-20260801`。
 
+## CR-11人工验收修正
+
+- 2026-08-02人工验收发现本地页面可访问，但行情与LLM同时无法连接。第一原因是验收服务由受限
+  自动化会话启动，子进程继承了禁止外部网络访问的权限；重新在主机正常网络权限下启动后恢复。
+- 随后的主模型连接测试暴露`LLMServiceMixin`机械迁移时遗漏`validate_text`导入，导致请求在真正
+  访问模型前抛出`NameError`并关闭HTTP连接；恢复原依赖并增加保存模型连接探测的运行契约测试。
+- 网站自身实测`000001`返回Tushare日K 60根、分时242点；主模型
+  `MiniMax-M2.7-highspeed`在2236毫秒内回复`OK`，证明服务进程的数据与LLM出网链路均已恢复。
+- 修正后候选329项、纯`app/`导出266项通过；本次只恢复缺失导入和测试，不修改模型配置、额度、
+  提示词、回退策略、行情来源或计算逻辑。
+- 继续验收观势时发现模型已经成功返回，但问天服务在保存解势记录前关闭了HTTP连接。原因是原版
+  `server.py`已有的`secrets`导入在机械拆分到问天服务时遗漏，生成非观气记录去重键时触发
+  `NameError`。迁移版恢复该标准库依赖，并增加“模型成功返回后保存观势结果”的完整服务回归测试。
+- 使用`000001 平安银行`完成真实页面复测：六爻安全门6/6通过，解势结果正常返回并写入历史，
+  `8797`错误日志为空。该修正不改变提示词、模型选择、额度、卦象计算、记录结构或前端行为。
+
+本修正基线为`xiaobai-reduction-11-background-jobs-20260801`；检查点为
+`xiaobai-reduction-11-runtime-connectivity-fix-20260802`；后续解势修正检查点为
+`xiaobai-reduction-11-heaven-interpret-fix-20260802`。
+
 ## 人工验收记录
 
 - 2026-08-01：用户检查CR-02与CR-03运行结果，确认未发现明显异常。本记录仅表示本轮可见功能与