Add Frontend 增加一层渐变背景色

- Removed multiple failed document entries from `documents.json`.
- Added a new document entry with updated metadata and changed the index name to `regulations_dense_1024_v2`.
- Updated architecture documentation to reflect changes in the Milvus collection name.
- Adjusted requirements by removing the sqlalchemy dependency.
- Modified test cases to align with new document structure and naming conventions.
- Introduced a new test file for Milvus vector index runtime recovery and error handling.
- Updated assertions in various test files to ensure compatibility with the new schema.

.env

@@ -9,7 +9,7 @@ DEBUG=false
 # ===== Milvus向量数据库配置（已有）=====
 MILVUS_HOST=6.86.80.8
 MILVUS_PORT=19530
-MILVUS_COLLECTION=regulations_dense_1024_v1
+MILVUS_COLLECTION=regulations_dense_1024_v2
 MILVUS_DB_NAME=default
 MILVUS_INDEX_TYPE=IVF_FLAT
 MILVUS_NLIST=128

.env.development

@@ -4,7 +4,7 @@
 # ===== Milvus向量数据库配置（已有）=====
 MILVUS_HOST=6.86.80.8
 MILVUS_PORT=19530
-MILVUS_COLLECTION=regulations_dense_1024_v1
+MILVUS_COLLECTION=regulations_dense_1024_v2
 MILVUS_DB_NAME=default
 MILVUS_INDEX_TYPE=IVF_FLAT
 MILVUS_NLIST=128

.env.example

@@ -9,7 +9,7 @@ DEBUG=false
 # ===== Milvus向量数据库配置 =====
 MILVUS_HOST=6.86.80.8
 MILVUS_PORT=19530
-MILVUS_COLLECTION=regulations_dense_1024_v1
+MILVUS_COLLECTION=regulations_dense_1024_v2
 MILVUS_DB_NAME=default
 MILVUS_INDEX_TYPE=IVF_FLAT
 MILVUS_NLIST=128

.gitignore

@@ -59,3 +59,6 @@ Thumbs.db
 
 # logs files
 logs/
+
+# codex
+.agents

AGENTS.md

@@ -4,6 +4,12 @@
 
 - Backend code lives under `backend/app/`; frontend is the Vite app in `frontend/`.
 
+## Frontend UX Constraints
+
+- Frontend work in `frontend/` must target desktop Web first.
+- Do not proactively add mobile-specific adaptations, responsive reflow for small screens, or mobile-first layout compromises unless the user explicitly asks for them.
+- When desktop and mobile requirements conflict, preserve the desktop Web layout and interaction model by default.
+
 ## Entrypoints
 
 - Backend entrypoint is `backend/app/main.py`, which re-exports `app` from `app.api.main`.
@@ -39,6 +45,15 @@
 - `tests/verify_mvp.py` also expects the `BGEM3Embedder` stack to be available and explicitly mentions `FlagEmbedding`.
 - For backend-only changes, prefer focused import/startup checks unless you know the external services and model dependencies are available.
 
+## Backend Architecture Authority
+
+- `docs/architecture/backend-project-architecture.md` is the authoritative backend architecture document for ongoing backend development.
+- New backend business logic must follow `api -> application -> domain ports -> infrastructure`.
+- Treat `backend/app/shared/bootstrap.py` as the current composition root for backend dependency wiring.
+- Do not add new business orchestration to `backend/app/services/*` or `backend/app/workflows/*` unless the task is explicitly a migration step.
+- API routes must not directly access `ConversationStore`; session access should go through application services.
+- Legacy files may be patched for compatibility or bug fixes, but should not gain new long-term responsibilities.
+
 ## Backend Commenting Standard
 
 - All comments and docstrings in `backend/**/*.py` must be written in English.

QUICK_DEPLOY.md

@@ -105,7 +105,7 @@ ALIBABA_ACCESS_KEY_ID=your_aliyun_access_key_id
 ALIBABA_ACCESS_KEY_SECRET=your_aliyun_access_key_secret
 EMBEDDING_API_KEY=your_embedding_api_key_here
 EMBEDDING_MODEL=text-embedding-v3
-EMBEDDING_DIM=1536
+EMBEDDING_DIM=1024
 PARSER_BACKEND=aliyun
 CHUNK_BACKEND=aliyun
 PARSER_FAILURE_MODE=fail

README.md

@@ -1,139 +0,0 @@
-# AI+合规智能中枢 - 法律法规文档解析入库
-
-面向车企与工厂的合规智能平台，实现法规文档的解析、分块、嵌入和向量存储。
-
-## MVP功能
-
-本次实现的核心功能（最小可用版本）：
-
-- ✅ PDF/DOC/DOCX 文档解析（阿里云文档智能）
-- ✅ 基于阿里云 `vector_chunks` 的统一切片
-- ✅ OpenAI 兼容 embedding（`text-embedding-v3`，1536维）
-- ✅ Milvus 向量数据库存储与 dense-only 检索
-- ✅ FastAPI接口封装
-
-## 项目结构
-
-```text
-AIRegulation-DocAnalysis-Demo/
-├── backend/
-│   ├── app/
-│   │   ├── api/                # FastAPI 接口层
-│   │   ├── application/        # 用例编排层
-│   │   ├── domain/             # 领域模型与稳定端口
-│   │   ├── infrastructure/     # MinIO / Milvus / 阿里云 / embedding / session 适配
-│   │   ├── config/             # 配置与日志
-│   │   └── workers/
-│   ├── requirements.txt
-│   └── main.py
-├── frontend/                   # Vite React 前端
-├── tests/                      # 根级测试，导入 backend/app
-├── docker/
-│   └── docker-compose.yml
-├── pyproject.toml
-└── .env.example
-```
-
-## 快速开始
-
-### 1. 安装依赖
-
-```bash
-./dev.sh setup
-```
-
-### 2. 启动Milvus向量数据库
-
-```bash
-cd docker
-docker-compose up -d
-```
-
-等待Milvus启动完成（约30秒）：
-```bash
-docker-compose logs -f milvus
-```
-
-### 3. 启动API服务
-
-```bash
-./dev.sh start api --foreground
-```
-
-访问API文档：http://localhost:8000/docs
-
-## API接口
-
-### 上传文档
-
-```bash
-curl -X POST http://localhost:8000/api/v1/documents/upload \
-  -F "file=@your_regulation.pdf" \
-  -F "doc_name=GB 7258-2017" \
-  -F "regulation_type=车辆安全"
-```
-
-### 检索法规
-
-```bash
-curl -X POST http://localhost:8000/api/v1/knowledge/search \
-  -H "Content-Type: application/json" \
-  -d '{"query": "机动车安全技术要求", "top_k": 10}'
-```
-
-## 技术栈
-
-| 类别 | 技术 |
-|------|------|
-| 文档解析 | 阿里云文档智能 + python-docx |
-| 分块策略 | 阿里云 `vector_chunks` |
-| 嵌入模型 | `text-embedding-v3`（1536维 Dense） |
-| 向量数据库 | Milvus 2.4（本地Docker部署） |
-| 检索方式 | Dense-only 检索 |
-| API框架 | FastAPI |
-
-## 配置
-
-创建 `.env` 文件（参考 `.env.example`）：
-
-```env
-# Milvus配置
-MILVUS_HOST=localhost
-MILVUS_PORT=19530
-
-# 阿里云文档解析
-ALIBABA_ACCESS_KEY_ID=your_aliyun_access_key_id
-ALIBABA_ACCESS_KEY_SECRET=your_aliyun_access_key_secret
-PARSER_BACKEND=aliyun
-CHUNK_BACKEND=aliyun
-
-# embedding 配置
-EMBEDDING_MODEL=text-embedding-v3
-EMBEDDING_DIM=1536
-EMBEDDING_API_KEY=your_embedding_api_key_here
-
-# 分块配置
-CHUNK_SIZE=512
-```
-
-## 后续迭代（不在本次MVP范围）
-
-- LLM摘要生成（当前上传主链路默认不生成）
-- 文档上传UI界面
-- 混合检索问答功能
-- 法规变更监控与自动更新
-
-## 解析产物
-
-上传成功后，系统会把阿里云解析的中间结果持久化到 MinIO：
-
-- `artifacts/{doc_id}/layouts.json`
-- `artifacts/{doc_id}/structure_nodes.json`
-- `artifacts/{doc_id}/semantic_blocks.json`
-- `artifacts/{doc_id}/vector_chunks.json`
-
-当前默认 Milvus collection 为 `regulations_dense_1536_v2`。
-
-## 许可证
-
-MIT License

backend/README.md

@@ -2,6 +2,13 @@
 
 `backend` 是当前正式使用的 FastAPI 后端目录，入口为 `app.main:app`。
 
+## 架构约束入口
+
+- Backend authoritative architecture 文档：`docs/architecture/backend-project-architecture.md`
+- Backend migration RFC：`docs/rfc/backend-api-parsing-embedding-migration-requirements.md`
+- 后续 backend 新增功能和重构默认遵守：`api -> application -> domain ports -> infrastructure`
+- `backend/app/services/*` 与 `backend/app/workflows/*` 为迁移期 legacy 目录，除迁移或兼容修复外，不应新增业务编排逻辑。
+
 ## 启动
 
 ```bash
@@ -34,10 +41,15 @@ PYTHONPATH=backend uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
 ```text
 backend/
 ├── app/
-│   ├── api/          # FastAPI 路由与模型
-│   ├── config/       # 配置与日志
-│   ├── services/     # 文档处理、LLM、RAG、存储
-│   └── workers/      # 任务相关代码
+│   ├── api/              # FastAPI 路由与 transport models
+│   ├── application/      # 用例编排层
+│   ├── domain/           # 核心业务模型与稳定端口
+│   ├── infrastructure/   # 外部系统适配器
+│   ├── shared/           # composition root 与横切支撑
+│   ├── config/           # 配置与日志
+│   ├── services/         # legacy façade / 兼容入口
+│   ├── workflows/        # legacy workflow 入口
+│   └── workers/          # 任务相关代码
 ├── .env.example
 ├── requirements.txt
 └── main.py
@@ -46,4 +58,13 @@ backend/
 ## 说明
 
 - 路由前缀保持为 `/api/v1`，以兼容当前前端。
-- 原 `backend/app/api/routes/docs.py`、`rag.py`、`compliance.py`、`status.py` 仍保留在仓库中，但不再作为主路由入口。
+- 当前主业务链路入口是 `documents`、`knowledge`、`agent`。
+- `compliance.py` 当前仍被挂载，但尚未满足目标架构约束；在迁移前不应继续扩展业务编排。
+- `docs.py` 与 `rag.py` 为遗留/非主入口，不应继续扩展。
+
+## 开发约束
+
+- backend 开发前先阅读 `docs/architecture/backend-project-architecture.md`。
+- 新增业务能力默认落在 `application` 层，由 `api` 调用，不要直接写进 route。
+- route 不应直接访问 MinIO、Milvus、Parser SDK、LLM SDK 或 `ConversationStore`。
+- `backend/app/shared/bootstrap.py` 是当前 composition root；依赖装配优先收口到这里。

backend/aliyun_parser/.claude/settings.local.json

@@ -0,0 +1,8 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(python3 *)",
+      "Bash(PGPASSWORD=postgresql123456 psql *)"
+    ]
+  }
+}

backend/aliyun_parser/parse_pdf.py

@@ -0,0 +1,475 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+阿里云文档智能 API 解析 PDF，输出三层结构 chunks
+- structure_nodes: 目录树结构
+- semantic_blocks: 语义块（章节文本、表格、图片）
+- vector_chunks: 检索块（带 overlap 切分）
+"""
+
+import argparse
+import json
+import re
+import time
+from pathlib import Path
+from typing import Dict, List
+
+from alibabacloud_docmind_api20220711.client import Client as DocmindClient
+from alibabacloud_tea_openapi import models as open_api_models
+from alibabacloud_docmind_api20220711 import models as docmind_models
+from alibabacloud_tea_util import models as util_models
+
+# ===================== 阿里云配置 =====================
+ALIBABA_ACCESS_KEY_ID = "LTAI5t6fWvAsvZkoF9WTbtys"
+ALIBABA_ACCESS_KEY_SECRET = "WX4oaE4FLYRa5L85TMQkqRPHeTJAF0"
+ALIBABA_ENDPOINT = "docmind-api.cn-hangzhou.aliyuncs.com"
+
+# ===================== 切分参数 =====================
+MAX_CHARS = 600
+OVERLAP_CHARS = 80
+
+# ===================== 布局类型常量 =====================
+TOC_TITLES = {"目次", "目录"}
+TITLE_SUBTYPES = {"doc_title", "para_title"}
+TEXT_SUBTYPES = {"para", "none"}
+FIGURE_TYPES = {"figure", "figure_name", "figure_note"}
+FIGURE_SUBTYPES = {"picture", "pic_title", "pic_caption"}
+
+
+# ===================== 阿里云 API 客户端 =====================
+def init_client() -> DocmindClient:
+    config = open_api_models.Config(
+        access_key_id=ALIBABA_ACCESS_KEY_ID,
+        access_key_secret=ALIBABA_ACCESS_KEY_SECRET,
+    )
+    config.endpoint = ALIBABA_ENDPOINT
+    return DocmindClient(config)
+
+
+def submit_job(client: DocmindClient, file_path: str) -> str:
+    """提交文档解析任务"""
+    file_name = Path(file_path).name
+    request = docmind_models.SubmitDocParserJobAdvanceRequest(
+        file_url_object=open(file_path, "rb"),
+        file_name=file_name,
+        file_name_extension=Path(file_path).suffix.lstrip("."),
+        llm_enhancement=True,
+        enhancement_mode="VLM",
+    )
+    runtime = util_models.RuntimeOptions()
+    response = client.submit_doc_parser_job_advance(request, runtime)
+    return response.body.data.id
+
+
+def query_status(client: DocmindClient, task_id: str) -> Dict:
+    """查询任务状态"""
+    request = docmind_models.QueryDocParserStatusRequest(id=task_id)
+    response = client.query_doc_parser_status(request)
+    return response.body.data.to_map() if response.body.data else None
+
+
+def wait_for_completion(client: DocmindClient, task_id: str, poll_interval: int = 5) -> bool:
+    """等待任务完成"""
+    while True:
+        status_data = query_status(client, task_id)
+        if not status_data:
+            return False
+        status = status_data.get("Status", "").lower()
+        if status == "success":
+            return True
+        elif status == "failed":
+            print(f"任务失败: {status_data}")
+            return False
+        print(f"任务状态: {status}, 等待中...")
+        time.sleep(poll_interval)
+
+
+def get_result(client: DocmindClient, task_id: str, layout_num: int = 0, layout_step_size: int = 50) -> Dict:
+    """获取解析结果"""
+    request = docmind_models.GetDocParserResultRequest(
+        id=task_id,
+        layout_step_size=layout_step_size,
+        layout_num=layout_num,
+    )
+    response = client.get_doc_parser_result(request)
+    return response.body.data if response.body.data else None
+
+
+def collect_all_results(client: DocmindClient, task_id: str, layout_step_size: int = 50) -> List[Dict]:
+    """收集所有解析结果"""
+    all_layouts = []
+    layout_num = 0
+    while True:
+        result_data = get_result(client, task_id, layout_num, layout_step_size)
+        if not result_data:
+            break
+        layouts = result_data.get("layouts", [])
+        if not layouts:
+            break
+        all_layouts.extend(layouts)
+        layout_num += len(layouts)
+        if len(layouts) < layout_step_size:
+            break
+    return all_layouts
+
+
+# ===================== 文本处理 =====================
+def normalize_text(text: str) -> str:
+    text = text.replace("\r", "\n")
+    text = text.replace(" ", " ")
+    text = re.sub(r"\n+", "\n", text)
+    text = re.sub(r"[ \t]+", " ", text)
+    return text.strip()
+
+
+def get_page(layout: Dict) -> int:
+    return layout.get("pageNum", layout.get("pageNumber", 0))
+
+
+def get_text(layout: Dict) -> str:
+    text = normalize_text(layout.get("text", ""))
+    if text:
+        return text
+    return normalize_text(layout.get("markdownContent", ""))
+
+
+# ===================== 布局类型判断 =====================
+def is_title(layout: Dict) -> bool:
+    return layout.get("type") == "title" or layout.get("subType") in TITLE_SUBTYPES
+
+
+def is_text(layout: Dict) -> bool:
+    return layout.get("type") == "text" and layout.get("subType", "none") in TEXT_SUBTYPES
+
+
+def is_figure(layout: Dict) -> bool:
+    return layout.get("type") in FIGURE_TYPES or layout.get("subType") in FIGURE_SUBTYPES
+
+
+def is_table(layout: Dict) -> bool:
+    return layout.get("type") == "table"
+
+
+def is_toc_layout(layout: Dict) -> bool:
+    text = get_text(layout)
+    if text in TOC_TITLES:
+        return True
+    if get_page(layout) == 1 and re.match(r"^\d+(\.\d+)*\s+.+[.。…]{2,}\s*\d+$", text):
+        return True
+    return False
+
+
+def extract_table_text(layout: Dict) -> str:
+    rows = []
+    for cell in layout.get("cells", []):
+        texts = []
+        for cell_layout in cell.get("layouts", []):
+            cell_text = normalize_text(cell_layout.get("text", ""))
+            if cell_text:
+                texts.append(cell_text)
+        if texts:
+            rows.append(" ".join(texts))
+    return "\n".join(rows).strip()
+
+
+# ===================== 结构层：目录树 =====================
+def build_structure_nodes(layouts: List[Dict]) -> List[Dict]:
+    nodes = []
+    for layout in layouts:
+        if not is_title(layout):
+            continue
+        text = get_text(layout)
+        if not text or text in TOC_TITLES:
+            continue
+        nodes.append(
+            {
+                "unique_id": layout.get("uniqueId"),
+                "page": get_page(layout),
+                "index": layout.get("index", 0),
+                "level": layout.get("level", 0),
+                "title": text,
+                "type": layout.get("type"),
+                "sub_type": layout.get("subType"),
+            }
+        )
+    return nodes
+
+
+# ===================== 语义层：章节内容 =====================
+def update_section_path(section_stack: List[Dict], layout: Dict) -> List[Dict]:
+    level = layout.get("level", 0)
+    title = get_text(layout)
+    while section_stack and section_stack[-1]["level"] >= level:
+        section_stack.pop()
+    section_stack.append(
+        {
+            "level": level,
+            "title": title,
+            "page": get_page(layout),
+            "unique_id": layout.get("uniqueId"),
+        }
+    )
+    return section_stack
+
+
+def section_path_titles(section_stack: List[Dict]) -> List[str]:
+    return [item["title"] for item in section_stack]
+
+
+def flush_text_block(blocks: List[Dict], semantic_blocks: List[Dict], block_id: int) -> int:
+    if not blocks:
+        return block_id
+
+    texts = [item["text"] for item in blocks if item["text"]]
+    merged_text = "\n".join(texts).strip()
+    if not merged_text:
+        return block_id
+
+    semantic_blocks.append(
+        {
+            "semantic_id": f"semantic-{block_id}",
+            "block_type": "section_text",
+            "page_start": min(item["page"] for item in blocks),
+            "page_end": max(item["page"] for item in blocks),
+            "section_path": blocks[0]["section_path"],
+            "section_level": blocks[0]["section_level"],
+            "section_title": blocks[0]["section_title"],
+            "source_ids": [item["unique_id"] for item in blocks if item.get("unique_id")],
+            "text": merged_text,
+        }
+    )
+    return block_id + 1
+
+
+def build_semantic_blocks(layouts: List[Dict]) -> List[Dict]:
+    semantic_blocks = []
+    section_stack = []
+    pending_text_blocks = []
+    block_id = 1
+    skip_toc_page = False
+
+    for layout in layouts:
+        text = get_text(layout)
+        page = get_page(layout)
+
+        if is_toc_layout(layout):
+            skip_toc_page = True
+            continue
+        if skip_toc_page and page == 1:
+            continue
+        if skip_toc_page and page != 1:
+            skip_toc_page = False
+
+        if is_title(layout):
+            block_id = flush_text_block(pending_text_blocks, semantic_blocks, block_id)
+            pending_text_blocks = []
+            section_stack = update_section_path(section_stack, layout)
+            continue
+
+        section_path = section_path_titles(section_stack)
+        section_title = section_path[-1] if section_path else "未分类"
+        section_level = len(section_path)
+
+        if is_table(layout):
+            block_id = flush_text_block(pending_text_blocks, semantic_blocks, block_id)
+            pending_text_blocks = []
+            table_text = extract_table_text(layout)
+            if table_text:
+                semantic_blocks.append(
+                    {
+                        "semantic_id": f"semantic-{block_id}",
+                        "block_type": "table",
+                        "page_start": page,
+                        "page_end": page,
+                        "section_path": section_path,
+                        "section_level": section_level,
+                        "section_title": section_title,
+                        "source_ids": [layout.get("uniqueId")],
+                        "text": table_text,
+                    }
+                )
+                block_id += 1
+            continue
+
+        if is_figure(layout):
+            block_id = flush_text_block(pending_text_blocks, semantic_blocks, block_id)
+            pending_text_blocks = []
+            if text:
+                semantic_blocks.append(
+                    {
+                        "semantic_id": f"semantic-{block_id}",
+                        "block_type": "figure",
+                        "page_start": page,
+                        "page_end": page,
+                        "section_path": section_path,
+                        "section_level": section_level,
+                        "section_title": section_title,
+                        "source_ids": [layout.get("uniqueId")],
+                        "text": text,
+                    }
+                )
+                block_id += 1
+            continue
+
+        if is_text(layout) and text:
+            pending_text_blocks.append(
+                {
+                    "page": page,
+                    "text": text,
+                    "unique_id": layout.get("uniqueId"),
+                    "section_path": section_path,
+                    "section_level": section_level,
+                    "section_title": section_title,
+                }
+            )
+
+    flush_text_block(pending_text_blocks, semantic_blocks, block_id)
+    return semantic_blocks
+
+
+# ===================== 检索层：向量 chunks =====================
+def split_text_with_overlap(text: str, max_chars: int, overlap_chars: int) -> List[str]:
+    text = text.strip()
+    if len(text) <= max_chars:
+        return [text] if text else []
+
+    parts = []
+    start = 0
+    while start < len(text):
+        end = min(len(text), start + max_chars)
+        parts.append(text[start:end].strip())
+        if end >= len(text):
+            break
+        start = max(0, end - overlap_chars)
+    return [part for part in parts if part]
+
+
+def build_vector_chunks(
+    semantic_blocks: List[Dict],
+    doc_id: str,
+    doc_title: str,
+    max_chars: int,
+    overlap_chars: int,
+) -> List[Dict]:
+    vector_chunks = []
+    chunk_index = 1
+
+    for block in semantic_blocks:
+        pieces = split_text_with_overlap(block["text"], max_chars, overlap_chars)
+        for piece_index, piece in enumerate(pieces, start=1):
+            if block["section_path"]:
+                header = f"标准：{doc_title}\n章节：{' > '.join(block['section_path'])}\n\n"
+            else:
+                header = f"标准：{doc_title}\n\n"
+            vector_chunks.append(
+                {
+                    "doc_id": doc_id,
+                    "doc_title": doc_title,
+                    "chunk_id": f"chunk-{chunk_index}",
+                    "chunk_index": chunk_index,
+                    "semantic_id": block["semantic_id"],
+                    "chunk_type": block["block_type"],
+                    "piece_index": piece_index,
+                    "page_start": block["page_start"],
+                    "page_end": block["page_end"],
+                    "section_path": block["section_path"],
+                    "section_level": block["section_level"],
+                    "section_title": block["section_title"],
+                    "source_ids": block["source_ids"],
+                    "text": piece,
+                    "embedding_text": header + piece,
+                }
+            )
+            chunk_index += 1
+
+    return vector_chunks
+
+
+# ===================== 主转换函数 =====================
+def convert_layouts(
+    layouts: List[Dict],
+    doc_id: str,
+    doc_title: str,
+    max_chars: int,
+    overlap_chars: int,
+) -> Dict:
+    structure_nodes = build_structure_nodes(layouts)
+    semantic_blocks = build_semantic_blocks(layouts)
+    vector_chunks = build_vector_chunks(
+        semantic_blocks,
+        doc_id=doc_id,
+        doc_title=doc_title,
+        max_chars=max_chars,
+        overlap_chars=overlap_chars,
+    )
+    return {
+        "doc_id": doc_id,
+        "doc_title": doc_title,
+        "structure_nodes": structure_nodes,
+        "semantic_blocks": semantic_blocks,
+        "vector_chunks": vector_chunks,
+    }
+
+
+# ===================== CLI 入口 =====================
+def main() -> None:
+    parser = argparse.ArgumentParser(description="阿里云文档智能解析 PDF，输出三层结构 chunks")
+    parser.add_argument("pdf_path", help="PDF 文件路径")
+    parser.add_argument("--out", default="vector_chunks.json", help="输出 JSON 文件路径")
+    parser.add_argument("--layouts-out", dest="layouts_output", help="输出原始 layouts JSON")
+    parser.add_argument("--doc-id", default="GB14747-2006", help="文档 ID")
+    parser.add_argument("--doc-title", default="GB 14747—2006 儿童三轮车安全要求", help="文档标题")
+    parser.add_argument("--max-chars", type=int, default=MAX_CHARS, help="单个检索 chunk 最大字符数")
+    parser.add_argument("--overlap-chars", type=int, default=OVERLAP_CHARS, help="相邻检索 chunk 重叠字符数")
+    parser.add_argument("--poll-interval", type=int, default=5, help="轮询间隔（秒）")
+    args = parser.parse_args()
+
+    pdf_path = Path(args.pdf_path).expanduser().resolve()
+    if not pdf_path.exists():
+        raise FileNotFoundError(f"PDF 文件不存在: {pdf_path}")
+
+    # 1. 提交阿里云任务
+    client = init_client()
+    print(f"提交任务: {pdf_path}")
+    task_id = submit_job(client, str(pdf_path))
+    print(f"任务 ID: {task_id}")
+
+    # 2. 等待完成
+    print("等待任务完成...")
+    if not wait_for_completion(client, task_id, args.poll_interval):
+        print("任务失败，退出")
+        return
+
+    # 3. 获取 layouts
+    print("获取解析结果...")
+    layouts = collect_all_results(client, task_id)
+    print(f"获取到 {len(layouts)} 个布局块")
+
+    # 4. 输出原始 layouts（可选）
+    if args.layouts_output:
+        layouts_path = Path(args.layouts_output).expanduser().resolve()
+        layouts_path.write_text(json.dumps(layouts, ensure_ascii=False, indent=2), encoding="utf-8")
+        print(f"原始 layouts 已写入: {layouts_path}")
+
+    # 5. 转换为三层结构
+    print("转换为三层结构...")
+    data = convert_layouts(
+        layouts,
+        doc_id=args.doc_id,
+        doc_title=args.doc_title,
+        max_chars=args.max_chars,
+        overlap_chars=args.overlap_chars,
+    )
+
+    # 6. 输出结果
+    output_path = Path(args.out).expanduser().resolve()
+    output_path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    print(f"结构层节点数: {len(data['structure_nodes'])}")
+    print(f"语义层块数: {len(data['semantic_blocks'])}")
+    print(f"检索层块数: {len(data['vector_chunks'])}")
+    print(f"输出文件: {output_path}")
+
+
+if __name__ == "__main__":
+    main()

backend/aliyun_parser/rebuild_milvus_collection.py

@@ -0,0 +1,115 @@
+"""Rebuild the migrated Milvus collection from saved vector chunks."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+from pymilvus import Collection, CollectionSchema, DataType, FieldSchema, connections, utility
+
+
+DEFAULT_COLLECTION = "regulations_dense_1024_v2"
+DEFAULT_DIM = 1024
+
+
+def build_collection(name: str, dim: int) -> Collection:
+    """Create the migrated Milvus collection from scratch."""
+    if utility.has_collection(name):
+        utility.drop_collection(name)
+
+    schema = CollectionSchema(
+        fields=[
+            FieldSchema(name="id", dtype=DataType.VARCHAR, max_length=128, is_primary=True, auto_id=False),
+            FieldSchema(name="doc_id", dtype=DataType.VARCHAR, max_length=64),
+            FieldSchema(name="doc_title", dtype=DataType.VARCHAR, max_length=256),
+            FieldSchema(name="chunk_id", dtype=DataType.VARCHAR, max_length=128),
+            FieldSchema(name="chunk_index", dtype=DataType.INT64),
+            FieldSchema(name="piece_index", dtype=DataType.INT64),
+            FieldSchema(name="text", dtype=DataType.VARCHAR, max_length=65535),
+            FieldSchema(name="embedding_text", dtype=DataType.VARCHAR, max_length=65535),
+            FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=dim),
+            FieldSchema(name="semantic_id", dtype=DataType.VARCHAR, max_length=128),
+            FieldSchema(name="chunk_type", dtype=DataType.VARCHAR, max_length=64),
+            FieldSchema(name="page_start", dtype=DataType.INT64),
+            FieldSchema(name="page_end", dtype=DataType.INT64),
+            FieldSchema(name="section_level", dtype=DataType.INT64),
+            FieldSchema(name="source_ids", dtype=DataType.VARCHAR, max_length=4096),
+            FieldSchema(name="section_path", dtype=DataType.VARCHAR, max_length=4096),
+            FieldSchema(name="section_title", dtype=DataType.VARCHAR, max_length=512),
+            FieldSchema(name="metadata_json", dtype=DataType.VARCHAR, max_length=65535),
+            FieldSchema(name="created_at", dtype=DataType.INT64),
+        ],
+        description="Dense-only regulations index",
+        enable_dynamic_field=False,
+    )
+    collection = Collection(name=name, schema=schema)
+    collection.create_index(
+        field_name="embedding",
+        index_params={
+            "metric_type": "COSINE",
+            "index_type": "IVF_FLAT",
+            "params": {"nlist": 128},
+        },
+    )
+    return collection
+
+
+def load_chunks(payload_path: Path) -> list[dict]:
+    """Load vector chunks emitted by the Aliyun parser pipeline."""
+    payload = json.loads(payload_path.read_text(encoding="utf-8"))
+    if isinstance(payload, dict):
+        chunks = payload.get("vector_chunks", [])
+    else:
+        chunks = payload
+    if not isinstance(chunks, list):
+        raise ValueError("vector chunk payload must be a list or a dict containing vector_chunks")
+    return chunks
+
+
+def main() -> None:
+    """Rebuild the target collection from a vector chunk payload."""
+    parser = argparse.ArgumentParser(description="Rebuild the migrated Milvus collection.")
+    parser.add_argument("--host", default="127.0.0.1", help="Milvus host")
+    parser.add_argument("--port", default="19530", help="Milvus port")
+    parser.add_argument("--collection", default=DEFAULT_COLLECTION, help="Milvus collection name")
+    parser.add_argument("--dim", type=int, default=DEFAULT_DIM, help="Embedding dimension")
+    parser.add_argument("--payload", required=True, help="Path to vector_chunks.json or a compatible JSON file")
+    args = parser.parse_args()
+
+    connections.connect("default", host=args.host, port=args.port)
+    collection = build_collection(args.collection, args.dim)
+    chunks = load_chunks(Path(args.payload))
+    if not chunks:
+        print("No vector chunks found; collection was created but remains empty.")
+        return
+
+    data = [
+        [chunk["chunk_id"] for chunk in chunks],
+        [chunk["doc_id"] for chunk in chunks],
+        [chunk["doc_title"] for chunk in chunks],
+        [chunk["chunk_id"] for chunk in chunks],
+        [int(chunk.get("chunk_index", 0) or 0) for chunk in chunks],
+        [int(chunk.get("piece_index", 0) or 0) for chunk in chunks],
+        [str(chunk.get("text", ""))[:65535] for chunk in chunks],
+        [str(chunk.get("embedding_text", chunk.get("text", "")))[:65535] for chunk in chunks],
+        [chunk["embedding"] for chunk in chunks],
+        [str(chunk.get("semantic_id", "")) for chunk in chunks],
+        [str(chunk.get("chunk_type", "")) for chunk in chunks],
+        [int(chunk.get("page_start", 0) or 0) for chunk in chunks],
+        [int(chunk.get("page_end", 0) or 0) for chunk in chunks],
+        [int(chunk.get("section_level", 0) or 0) for chunk in chunks],
+        [json.dumps(chunk.get("source_ids", []), ensure_ascii=False) for chunk in chunks],
+        [json.dumps(chunk.get("section_path", []), ensure_ascii=False) for chunk in chunks],
+        [str(chunk.get("section_title", "")) for chunk in chunks],
+        [json.dumps(chunk, ensure_ascii=False) for chunk in chunks],
+        [int(chunk.get("created_at", 0) or 0) for chunk in chunks],
+    ]
+    collection.insert(data)
+    collection.flush()
+    collection.load()
+    print(f"Rebuilt collection {args.collection} with {len(chunks)} chunks.")
+
+
+if __name__ == "__main__":
+    main()

backend/aliyun_parser/schema.sql

@@ -0,0 +1,122 @@
+-- 法规文档向量检索系统数据库表结构
+-- PostgreSQL
+
+-- ==================== 文档表 ====================
+CREATE TABLE documents (
+    id SERIAL PRIMARY KEY,
+    doc_id VARCHAR(128) UNIQUE NOT NULL,       -- 文档唯一标识，如 "GB14747-2006"
+    title VARCHAR(512) NOT NULL,               -- 文档标题
+    doc_type VARCHAR(32),                      -- 文档类型：标准/法规/规范
+    standard_number VARCHAR(64),               -- 标准编号：如 "GB 14747-2006"
+    publish_date DATE,                         -- 发布日期
+    implement_date DATE,                       -- 实施日期
+    status VARCHAR(32),                        -- 状态：现行/废止/修订
+    source_url VARCHAR(512),                   -- 来源 URL
+    file_path VARCHAR(512),                    -- 本地 PDF 文件路径
+    file_size INT,                             -- 文件大小（字节）
+    upload_time TIMESTAMP DEFAULT NOW(),       -- 上传时间
+    created_at TIMESTAMP DEFAULT NOW(),
+    updated_at TIMESTAMP DEFAULT NOW()
+);
+
+COMMENT ON TABLE documents IS '文档元数据表';
+COMMENT ON COLUMN documents.doc_id IS '文档唯一标识，用于关联 Milvus 和其他表';
+COMMENT ON COLUMN documents.standard_number IS '标准编号，如 GB 14747-2006';
+
+-- ==================== 章节结构表 ====================
+CREATE TABLE sections (
+    id SERIAL PRIMARY KEY,
+    doc_id VARCHAR(128) NOT NULL,
+    unique_id VARCHAR(64) NOT NULL,            -- 阿里云返回的唯一标识
+    level INT NOT NULL,                        -- 层级：1, 2, 3...
+    title VARCHAR(512) NOT NULL,               -- 章节标题
+    page INT,                                  -- 所在页码
+    index INT,                                 -- 页内顺序
+    parent_id INT,                             -- 父章节 ID（树形结构）
+    created_at TIMESTAMP DEFAULT NOW(),
+
+    CONSTRAINT fk_sections_doc_id FOREIGN KEY (doc_id) REFERENCES documents(doc_id),
+    CONSTRAINT fk_sections_parent_id FOREIGN KEY (parent_id) REFERENCES sections(id),
+    CONSTRAINT uq_sections_doc_unique UNIQUE (doc_id, unique_id)
+);
+
+COMMENT ON TABLE sections IS '章节结构表，用于目录导航';
+COMMENT ON COLUMN sections.parent_id IS '父章节 ID，构建树形结构';
+COMMENT ON COLUMN sections.level IS '层级深度，1 为最顶层';
+
+-- ==================== 语义块表 ====================
+CREATE TABLE semantic_blocks (
+    id SERIAL PRIMARY KEY,
+    doc_id VARCHAR(128) NOT NULL,
+    semantic_id VARCHAR(64) NOT NULL,          -- 语义块唯一标识
+    block_type VARCHAR(32) NOT NULL,           -- 类型：section_text/table/figure
+    page_start INT NOT NULL,                   -- 起始页码
+    page_end INT NOT NULL,                     -- 结束页码
+    section_id INT,                            -- 所属章节
+    section_title VARCHAR(512),                -- 章节标题（冗余，方便查询）
+    section_level INT,                         -- 章节层级
+    source_ids JSONB,                          -- 原始 layout IDs（JSON 数组）
+    text TEXT NOT NULL,                        -- 完整内容（未被切分）
+    created_at TIMESTAMP DEFAULT NOW(),
+
+    CONSTRAINT fk_semantic_blocks_doc_id FOREIGN KEY (doc_id) REFERENCES documents(doc_id),
+    CONSTRAINT fk_semantic_blocks_section_id FOREIGN KEY (section_id) REFERENCES sections(id),
+    CONSTRAINT uq_semantic_blocks_doc_semantic UNIQUE (doc_id, semantic_id)
+);
+
+COMMENT ON TABLE semantic_blocks IS '语义块表，用于邻域扩展，恢复完整内容';
+COMMENT ON COLUMN semantic_blocks.block_type IS '类型：section_text（正文）、table（表格）、figure（图示）';
+COMMENT ON COLUMN semantic_blocks.source_ids IS '原始阿里云 layout 的 uniqueId 数组';
+COMMENT ON COLUMN semantic_blocks.text IS '完整语义内容，未被切分';
+
+-- ==================== 向量块元数据表 ====================
+CREATE TABLE vector_chunks (
+    id SERIAL PRIMARY KEY,
+    doc_id VARCHAR(128) NOT NULL,
+    chunk_id VARCHAR(64) NOT NULL,             -- Milvus 主键
+    semantic_id VARCHAR(64) NOT NULL,          -- 关联语义块
+    chunk_index INT NOT NULL,                  -- 切片序号（全局）
+    piece_index INT,                           -- 同语义块内的切片序号
+    page_start INT,
+    page_end INT,
+    section_title VARCHAR(512),
+    text VARCHAR(2048),                        -- 切片文本（可选，缩短版用于展示）
+    source_ids JSONB,                          -- 原始 layout IDs（JSON 数组）
+    created_at TIMESTAMP DEFAULT NOW(),
+
+    CONSTRAINT fk_vector_chunks_doc_id FOREIGN KEY (doc_id) REFERENCES documents(doc_id),
+    CONSTRAINT fk_vector_chunks_semantic_id FOREIGN KEY (doc_id, semantic_id)
+        REFERENCES semantic_blocks(doc_id, semantic_id),
+    CONSTRAINT uq_vector_chunks_doc_chunk UNIQUE (doc_id, chunk_id)
+);
+
+COMMENT ON TABLE vector_chunks IS '向量块元数据表，用于快速关联查询';
+COMMENT ON COLUMN vector_chunks.chunk_id IS 'Milvus 向量库主键';
+COMMENT ON COLUMN vector_chunks.piece_index IS '同语义块内的切片序号，用于按序拼接';
+
+-- ==================== 索引 ====================
+CREATE INDEX idx_sections_doc_id ON sections(doc_id);
+CREATE INDEX idx_sections_parent_id ON sections(parent_id);
+CREATE INDEX idx_sections_level ON sections(level);
+
+CREATE INDEX idx_semantic_blocks_doc_id ON semantic_blocks(doc_id);
+CREATE INDEX idx_semantic_blocks_section_id ON semantic_blocks(section_id);
+CREATE INDEX idx_semantic_blocks_block_type ON semantic_blocks(block_type);
+CREATE INDEX idx_semantic_blocks_semantic_id ON semantic_blocks(semantic_id);
+
+CREATE INDEX idx_vector_chunks_doc_id ON vector_chunks(doc_id);
+CREATE INDEX idx_vector_chunks_semantic_id ON vector_chunks(semantic_id);
+CREATE INDEX idx_vector_chunks_chunk_id ON vector_chunks(chunk_id);
+
+-- ==================== 触发器：自动更新 updated_at ====================
+CREATE OR REPLACE FUNCTION update_updated_at()
+RETURNS TRIGGER AS $$
+BEGIN
+    NEW.updated_at = NOW();
+    RETURN NEW;
+END;
+$$ LANGUAGE plpgsql;
+
+CREATE TRIGGER tr_documents_updated_at
+    BEFORE UPDATE ON documents
+    FOR EACH ROW EXECUTE FUNCTION update_updated_at();

backend/aliyun_parser/upload_to_milvus.py

@@ -0,0 +1,327 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+将 vector_chunks.json 向量化并上传到 Milvus 和 PostgreSQL
+使用中转站的 OpenAI 兼容 API
+"""
+
+import argparse
+import json
+import time
+from pathlib import Path
+from typing import List, Dict
+
+import psycopg2
+from psycopg2.extras import execute_values
+from pymilvus import (
+    connections,
+    Collection,
+    FieldSchema,
+    CollectionSchema,
+    DataType,
+    utility,
+)
+from openai import OpenAI
+
+# ===================== 配置 =====================
+# 中转站配置
+RELAY_BASE_URL = "http://6.86.80.4:30080/v1"
+RELAY_API_KEY = "sk-5HeY7gfSIlyZMacfuXOf5cphpymsNqufEu1ou4U3avbULcyY"
+EMBEDDING_MODEL = "text-embedding-v3"  # 中转站支持的 embedding 模型
+
+# Milvus 配置
+MILVUS_HOST = "localhost"
+MILVUS_PORT = "19530"
+COLLECTION_NAME = "regulation_chunks"
+
+# PostgreSQL 配置
+PG_HOST = "6.86.80.10"
+PG_PORT = 5432
+PG_USER = "postgresql"
+PG_PASSWORD = "postgresql123456"
+PG_DATABASE = "postgres"
+
+
+# ===================== Embedding =====================
+def get_openai_client(api_key: str, base_url: str) -> OpenAI:
+    """创建 OpenAI 客户端连接到中转站"""
+    return OpenAI(api_key=api_key, base_url=base_url)
+
+
+def get_embeddings_batch(client: OpenAI, texts: List[str], batch_size: int = 10) -> List[List[float]]:
+    """批量获取文本向量"""
+    all_embeddings = []
+
+    for i in range(0, len(texts), batch_size):
+        batch = texts[i:i + batch_size]
+        print(f"Embedding batch {i // batch_size + 1}/{(len(texts) - 1) // batch_size + 1}...")
+
+        response = client.embeddings.create(
+            model=EMBEDDING_MODEL,
+            input=batch,
+        )
+
+        embeddings = [item.embedding for item in response.data]
+        all_embeddings.extend(embeddings)
+
+    return all_embeddings
+
+
+# ===================== Milvus =====================
+def init_milvus(host: str, port: str):
+    connections.connect("default", host=host, port=port)
+    print(f"已连接 Milvus: {host}:{port}")
+
+
+def create_collection(name: str, dim: int) -> Collection:
+    """创建或获取 collection"""
+    if utility.has_collection(name):
+        print(f"Collection '{name}' 已存在，删除重建")
+        utility.drop_collection(name)
+
+    fields = [
+        FieldSchema(name="chunk_id", dtype=DataType.VARCHAR, max_length=64, is_primary=True),
+        FieldSchema(name="doc_id", dtype=DataType.VARCHAR, max_length=128),
+        FieldSchema(name="doc_title", dtype=DataType.VARCHAR, max_length=512),
+        FieldSchema(name="chunk_index", dtype=DataType.INT64),
+        FieldSchema(name="semantic_id", dtype=DataType.VARCHAR, max_length=64),
+        FieldSchema(name="chunk_type", dtype=DataType.VARCHAR, max_length=32),
+        FieldSchema(name="page_start", dtype=DataType.INT64),
+        FieldSchema(name="page_end", dtype=DataType.INT64),
+        FieldSchema(name="section_title", dtype=DataType.VARCHAR, max_length=512),
+        FieldSchema(name="text", dtype=DataType.VARCHAR, max_length=2048),
+        FieldSchema(name="source_ids", dtype=DataType.VARCHAR, max_length=4096),  # JSON 字符串
+        FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=dim),
+    ]
+
+    schema = CollectionSchema(fields, description="法规文档检索 chunks")
+    collection = Collection(name, schema)
+
+    # 创建向量索引（IVF_FLAT，适合中小规模）
+    index_params = {
+        "metric_type": "COSINE",
+        "index_type": "IVF_FLAT",
+        "params": {"nlist": 128},
+    }
+    collection.create_index("embedding", index_params)
+    print(f"Collection '{name}' 创建完成，索引已建立")
+
+    return collection
+
+
+def insert_chunks(collection: Collection, chunks: List[Dict], embeddings: List[List[float]]):
+    """插入 chunks 到 Milvus"""
+    data = [
+        [c["chunk_id"] for c in chunks],
+        [c["doc_id"] for c in chunks],
+        [c["doc_title"] for c in chunks],
+        [c["chunk_index"] for c in chunks],
+        [c["semantic_id"] for c in chunks],
+        [c["chunk_type"] for c in chunks],
+        [c["page_start"] for c in chunks],
+        [c["page_end"] for c in chunks],
+        [c["section_title"] for c in chunks],
+        [c["text"] for c in chunks],
+        [json.dumps(c.get("source_ids", [])) for c in chunks],  # JSON 字符串
+        embeddings,
+    ]
+
+    collection.insert(data)
+    collection.flush()
+    print(f"已插入 {len(chunks)} 个 chunks")
+
+
+def load_collection(collection: Collection):
+    """加载 collection 到内存（搜索前必须）"""
+    collection.load()
+    print(f"Collection 已加载到内存")
+
+
+# ===================== PostgreSQL =====================
+def get_pg_connection(host: str, port: int, user: str, password: str, database: str):
+    """获取 PostgreSQL 连接"""
+    conn = psycopg2.connect(
+        host=host,
+        port=port,
+        user=user,
+        password=password,
+        database=database,
+    )
+    print(f"已连接 PostgreSQL: {host}:{port}/{database}")
+    return conn
+
+
+def insert_chunks_to_pg(conn, chunks: List[Dict], doc_data: Dict):
+    """插入 chunks 和相关数据到 PostgreSQL"""
+    cursor = conn.cursor()
+
+    try:
+        # 1. 插入文档
+        cursor.execute("""
+            INSERT INTO documents (doc_id, title, standard_number, upload_time)
+            VALUES (%s, %s, %s, NOW())
+            ON CONFLICT (doc_id) DO UPDATE SET title = EXCLUDED.title, updated_at = NOW()
+        """, (doc_data["doc_id"], doc_data["doc_title"], doc_data.get("standard_number")))
+
+        # 2. 插入语义块
+        semantic_blocks = doc_data.get("semantic_blocks", [])
+        if semantic_blocks:
+            block_rows = [
+                (
+                    doc_data["doc_id"],
+                    block["semantic_id"],
+                    block["block_type"],
+                    block["page_start"],
+                    block["page_end"],
+                    block.get("section_title"),
+                    block.get("section_level"),
+                    json.dumps(block.get("source_ids", [])),
+                    block["text"],
+                )
+                for block in semantic_blocks
+            ]
+            execute_values(
+                cursor,
+                """
+                INSERT INTO semantic_blocks
+                (doc_id, semantic_id, block_type, page_start, page_end, section_title, section_level, source_ids, text)
+                VALUES %s
+                ON CONFLICT (doc_id, semantic_id) DO UPDATE SET text = EXCLUDED.text
+                """,
+                block_rows,
+            )
+            print(f"已插入 {len(semantic_blocks)} 个语义块")
+
+        # 3. 插入向量块元数据
+        chunk_rows = [
+            (
+                doc_data["doc_id"],
+                chunk["chunk_id"],
+                chunk["semantic_id"],
+                chunk["chunk_index"],
+                chunk.get("piece_index"),
+                chunk["page_start"],
+                chunk["page_end"],
+                chunk.get("section_title"),
+                chunk["text"],
+                json.dumps(chunk.get("source_ids", [])),
+            )
+            for chunk in chunks
+        ]
+        execute_values(
+            cursor,
+            """
+            INSERT INTO vector_chunks
+            (doc_id, chunk_id, semantic_id, chunk_index, piece_index, page_start, page_end, section_title, text, source_ids)
+            VALUES %s
+            ON CONFLICT (doc_id, chunk_id) DO UPDATE SET text = EXCLUDED.text
+            """,
+            chunk_rows,
+        )
+        print(f"已插入 {len(chunks)} 个向量块元数据")
+
+        conn.commit()
+        print("PostgreSQL 数据插入完成")
+
+    except Exception as e:
+        conn.rollback()
+        raise e
+    finally:
+        cursor.close()
+
+
+# ===================== 主流程 =====================
+def load_data(file_path: Path) -> Dict:
+    """加载 vector_chunks.json，返回完整数据"""
+    data = json.loads(file_path.read_text(encoding="utf-8"))
+    return data
+
+
+def upload_to_milvus_and_pg(
+    chunks_file: str,
+    api_key: str,
+    base_url: str,
+    milvus_host: str,
+    milvus_port: str,
+    collection_name: str,
+    batch_size: int,
+    pg_host: str,
+    pg_port: int,
+    pg_user: str,
+    pg_password: str,
+    pg_database: str,
+):
+    # 1. 加载完整数据
+    chunks_path = Path(chunks_file).expanduser().resolve()
+    if not chunks_path.exists():
+        raise FileNotFoundError(f"文件不存在: {chunks_path}")
+
+    data = load_data(chunks_path)
+    chunks = data.get("vector_chunks", [])
+    if not chunks:
+        raise ValueError("vector_chunks 为空")
+    print(f"加载 {len(chunks)} 个 chunks")
+
+    # 2. 初始化连接
+    client = get_openai_client(api_key, base_url)
+    init_milvus(milvus_host, milvus_port)
+    pg_conn = get_pg_connection(pg_host, pg_port, pg_user, pg_password, pg_database)
+
+    # 3. 获取 embeddings
+    texts = [c["embedding_text"] for c in chunks]
+    embeddings = get_embeddings_batch(client, texts, batch_size)
+    print(f"生成 {len(embeddings)} 个向量")
+
+    # 4. 获取 embedding 维度
+    embedding_dim = len(embeddings[0])
+    print(f"Embedding 维度: {embedding_dim}")
+
+    # 5. 创建 collection 并插入 Milvus
+    collection = create_collection(collection_name, embedding_dim)
+    insert_chunks(collection, chunks, embeddings)
+    load_collection(collection)
+
+    # 6. 插入 PostgreSQL
+    insert_chunks_to_pg(pg_conn, chunks, data)
+
+    # 7. 关闭连接
+    pg_conn.close()
+
+    print("上传完成！")
+
+
+# ===================== CLI =====================
+def main():
+    parser = argparse.ArgumentParser(description="将 vector_chunks 向量化并上传到 Milvus 和 PostgreSQL")
+    parser.add_argument("chunks_file", help="vector_chunks.json 文件路径")
+    parser.add_argument("--api-key", default=RELAY_API_KEY, help="中转站 API Key")
+    parser.add_argument("--base-url", default=RELAY_BASE_URL, help="中转站 Base URL")
+    parser.add_argument("--milvus-host", default=MILVUS_HOST, help="Milvus host")
+    parser.add_argument("--milvus-port", default=MILVUS_PORT, help="Milvus port")
+    parser.add_argument("--collection", default=COLLECTION_NAME, help="Milvus collection 名称")
+    parser.add_argument("--batch-size", type=int, default=10, help="Embedding 批量大小（中转站限制最大10）")
+    parser.add_argument("--pg-host", default=PG_HOST, help="PostgreSQL host")
+    parser.add_argument("--pg-port", type=int, default=PG_PORT, help="PostgreSQL port")
+    parser.add_argument("--pg-user", default=PG_USER, help="PostgreSQL user")
+    parser.add_argument("--pg-password", default=PG_PASSWORD, help="PostgreSQL password")
+    parser.add_argument("--pg-database", default=PG_DATABASE, help="PostgreSQL database")
+    args = parser.parse_args()
+
+    upload_to_milvus_and_pg(
+        chunks_file=args.chunks_file,
+        api_key=args.api_key,
+        base_url=args.base_url,
+        milvus_host=args.milvus_host,
+        milvus_port=args.milvus_port,
+        collection_name=args.collection,
+        batch_size=args.batch_size,
+        pg_host=args.pg_host,
+        pg_port=args.pg_port,
+        pg_user=args.pg_user,
+        pg_password=args.pg_password,
+        pg_database=args.pg_database,
+    )
+
+
+if __name__ == "__main__":
+    main()

backend/aliyun_parser/嵌入和召回.md

@@ -0,0 +1,263 @@
+# 文档解析与向量检索说明
+
+## 相关文件
+
+- `aliyun_doc_parser.py`：调用阿里云文档智能解析 PDF，生成原始 `layouts.json`
+- `layouts_to_vector_chunks.py`：把 `layouts.json` 转成适合向量数据库入库的三层结构
+- `layouts.json`：阿里云返回的原始布局结果
+- `vector_chunks.json`：转换后的结构化输出
+
+## 一、`layouts.json` 的结构
+
+`layouts.json` 顶层是一个数组，每个元素代表一个布局块（layout）。常见字段如下：
+
+- `type`：主类型，例如 `title`、`text`、`table`、`figure`
+- `subType`：更细的语义类型，例如 `doc_title`、`para_title`、`para`、`picture`、`pic_title`、`pic_caption`
+- `text`：当前布局块的纯文本
+- `markdownContent`：带 markdown 标记的文本
+- `pageNum`：页码
+- `index`：页内顺序
+- `level`：标题层级
+- `uniqueId`：布局块唯一标识
+- `blocks`：更细粒度的文本与样式信息
+- `cells`：表格单元格，仅 `table` 类型存在
+
+这个结构不是简单 OCR 文本流，而是已经带有版面理解和语义分类的结构化数据。
+
+## 二、推荐的三层转换结构
+
+### 1. 结构层 `structure_nodes`
+
+结构层用于恢复文档标题树，不直接作为最终向量检索单元。
+
+示例：
+
+- `1 范围`
+- `2 规范性引用文件`
+- `3 术语和定义`
+  - `3.1 儿童三轮车`
+  - `3.2 轮距`
+
+结构层主要用于给下游 chunk 绑定 `section_path`。
+
+### 2. 语义层 `semantic_blocks`
+
+语义层是按文档意义聚合后的内容块，主要分为三类：
+
+- `section_text`：同一章节下连续正文聚合而成
+- `table`：表格内容单独成块
+- `figure`：图、图名、图注等单独成块
+
+这一层比单 layout 更适合做语义理解，也适合后续做上下文扩展。
+
+### 3. 检索层 `vector_chunks`
+
+检索层是最终写进向量数据库的 chunk。
+
+处理方式：
+
+- 对 `semantic_blocks` 中较短的块直接入库
+- 对较长的块按 `max_chars` 再切分
+- 相邻切片保留 `overlap_chars` 重叠
+- 每个 chunk 都带完整 metadata，便于后续过滤、重排和邻域扩展
+
+## 三、当前转换脚本做了什么
+
+`layouts_to_vector_chunks.py` 当前已经实现：
+
+1. 过滤目录页噪声（如 `目次`）
+2. 根据标题层级维护章节路径
+3. 将正文聚合成 `section_text`
+4. 将表格单独转成 `table`
+5. 将图相关内容单独转成 `figure`
+6. 对长文本继续切分为最终 `vector_chunks`
+7. 为每个检索 chunk 生成 `embedding_text`
+
+## 四、为什么不要直接按 layout 入库
+
+如果把 `layouts.json` 的每条 layout 直接做向量：
+
+- 颗粒度太碎
+- 标题和正文容易分离
+- 表格会丢失结构上下文
+- 图示信息无法完整表达
+- 检索命中结果噪声较大
+
+对于标准文档，最合适的单位通常不是“句子”，而是“条款语义块”。
+
+## 五、建议的入库字段
+
+建议向量数据库每条记录至少保存：
+
+- `embedding_text`：用于生成向量
+- `text`：原始 chunk 文本
+- `chunk_id`
+- `semantic_id`
+- `chunk_type`：`section_text` / `table` / `figure`
+- `section_path`
+- `section_title`
+- `section_level`
+- `page_start`
+- `page_end`
+- `doc_id`
+- `doc_title`
+- `source_ids`
+
+其中：
+
+- 向量化字段：`embedding_text`
+- 展示字段：`text`
+- 检索增强字段：其余 metadata
+
+## 六、推荐的检索方式
+
+不要只做最简单的 top-k 向量搜索，建议采用：
+
+**向量召回 + metadata 重排 + 邻域扩展**
+
+### 1. 向量召回
+
+使用 `vector_chunks[*].embedding_text` 做 embedding，并在向量数据库中检索 top 10 ~ 15 条。
+
+查询时可以对用户问题做轻微改写，例如：
+
+原问题：
+
+`儿童三轮车的定义是什么？`
+
+可改写为：
+
+`请检索 GB 14747—2006 儿童三轮车安全要求 中关于“儿童三轮车定义”的条款、术语、表格或图示说明。`
+
+这样更适合标准文档检索。
+
+### 2. metadata 重排
+
+向量召回后，根据 metadata 做轻量规则重排。
+
+常见规则：
+
+- `chunk_type == section_text`：对定义类、要求类问题优先级更高
+- `section_path` 命中查询关键词：例如查询“定义”时，`术语和定义` 章节优先
+- `chunk_type == table`：对“尺寸 / 参数 / 数值 / 对照 / 要求”类问题加权
+- `chunk_type == figure`：对“图 / 结构 / 状态 / 示意”类问题加权
+
+### 3. 邻域扩展
+
+检索命中的是最终切片，但回答往往需要更完整上下文。
+
+建议命中某个 `vector_chunk` 后：
+
+1. 优先回捞同一个 `semantic_id` 下的所有 chunk
+2. 如果还不够，再补充同 `section_path`、相邻页码或相邻 `chunk_index` 的内容
+
+这样可以恢复完整条款，而不是只给模型一小段碎片。
+
+## 七、不同问题的检索重点
+
+### 1. 定义类问题
+
+例如：
+
+- `儿童三轮车的定义是什么？`
+- `轮距是什么意思？`
+
+优先检索：
+
+- `section_text`
+- `section_path` 中包含 `术语和定义` 的内容
+
+### 2. 要求类问题
+
+例如：
+
+- `外露突出物有什么要求？`
+- `辅助推杆有哪些安全要求？`
+
+优先检索：
+
+- `section_text`
+- `table`
+
+### 3. 数值 / 尺寸 / 对照类问题
+
+例如：
+
+- `鞍座到脚蹬距离要求是什么？`
+- `哪些项目需要满足规定尺寸？`
+
+优先检索：
+
+- `table`
+- `section_text`
+
+### 4. 图示说明类问题
+
+例如：
+
+- `正常乘骑状态是什么意思？`
+- `图1表示什么？`
+
+优先检索：
+
+- `figure`
+- 同章节相邻 `section_text`
+
+## 八、推荐的最终检索流程
+
+建议采用以下固定流程：
+
+1. 用 `vector_chunks.embedding_text` 做 embedding 检索
+2. 取 top 10 ~ 15 条候选
+3. 按 `chunk_type + section_path` 做规则重排
+4. 以 `semantic_id` 为中心回捞完整语义块
+5. 选 3 ~ 5 组上下文提供给大模型回答
+
+## 九、给大模型的上下文组织方式
+
+最终不要直接把原始 JSON 扔给模型，建议整理成如下格式：
+
+```text
+[命中片段 1]
+章节：3 术语和定义 > 3.1 儿童三轮车
+页码：1-2
+类型：section_text
+内容：
+......
+
+[命中片段 2]
+章节：4 要求 > 4.3 外露突出物
+页码：5
+类型：section_text
+内容：
+......
+
+[命中片段 3]
+章节：5 试验方法
+页码：8
+类型：table
+内容：
+......
+```
+
+这种格式更利于模型稳定回答并引用出处。
+
+## 十、转换命令
+
+生成三层结构：
+
+```bash
+python3 /home/huaci/dev/ai/SuperMew/tests/layouts_to_vector_chunks.py \
+  --layouts /home/huaci/dev/ai/SuperMew/tests/layouts.json \
+  --out /home/huaci/dev/ai/SuperMew/tests/vector_chunks.json
+```
+
+自定义切片大小：
+
+```bash
+python3 /home/huaci/dev/ai/SuperMew/tests/layouts_to_vector_chunks.py \
+  --layouts /home/huaci/dev/ai/SuperMew/tests/layouts.json \
+  --out /home/huaci/dev/ai/SuperMew/tests/vector_chunks.json \
+  --max-chars 500 \
+  --overlap-chars 80
+```

backend/app/api/main.py

@@ -3,6 +3,7 @@
 from contextlib import asynccontextmanager
 
 from fastapi import FastAPI, Request
+from fastapi.encoders import jsonable_encoder
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import JSONResponse
 from loguru import logger
@@ -11,7 +12,8 @@ from app.api.models import ErrorResponse
 from app.api.routes import api_router
 from app.config.logging import setup_logging
 from app.config.settings import settings
-from app.services.llm.llm_factory import LLMFactory
+from app.shared.bootstrap import cleanup_runtime_dependencies, preload_runtime_dependencies
+from app.shared.errors import VectorStoreSchemaError
 # Keep module behavior explicit so the backend flow stays easy to audit.
 
 
@@ -24,12 +26,12 @@ async def lifespan(app: FastAPI):
     logger.info(f"启动 {settings.app_name} v{settings.app_version}")
     logger.info(f"调试模式: {settings.debug}")
     logger.info("预加载LLM客户端...")
-    LLMFactory.preload_clients(["qwen", "deepseek"])
+    preload_runtime_dependencies()
 
     yield
 
     logger.info("应用关闭，执行清理...")
-    LLMFactory.cleanup()
+    cleanup_runtime_dependencies()
 
 
 app = FastAPI(
@@ -55,16 +57,33 @@ app.add_middleware(
 app.include_router(api_router, prefix="/api/v1")
 
 
+@app.exception_handler(VectorStoreSchemaError)
+async def vector_store_schema_exception_handler(request: Request, exc: VectorStoreSchemaError):
+    """Return a stable JSON response for vector store schema/runtime errors."""
+    logger.error(f"向量库 schema 异常: {exc}")
+    return JSONResponse(
+        status_code=500,
+        content=jsonable_encoder(
+            ErrorResponse(
+                error="VectorStoreSchemaError",
+                message=str(exc),
+            )
+        ),
+    )
+
+
 @app.exception_handler(Exception)
 async def global_exception_handler(request: Request, exc: Exception):
     """Global exception handler."""
     logger.error(f"未处理的异常: {exc}")
     return JSONResponse(
         status_code=500,
-        content=ErrorResponse(
-            error="InternalServerError",
-            message=str(exc),
-        ).model_dump(),
+        content=jsonable_encoder(
+            ErrorResponse(
+                error="InternalServerError",
+                message=str(exc),
+            )
+        ),
     )
 
 

backend/app/api/routes/__init__.py

@@ -6,6 +6,8 @@ from .documents import router as documents_router
 from .knowledge import router as knowledge_router
 from .agent import router as agent_router
 from .status import router as status_router
+from .perception import router as perception_router
+from .rag import router as rag_router
 # Keep package boundaries explicit so backend imports stay predictable.
 
 
@@ -18,6 +20,8 @@ api_router.include_router(knowledge_router)
 api_router.include_router(agent_router)
 api_router.include_router(compliance_router)
 api_router.include_router(status_router)
+api_router.include_router(perception_router)
+api_router.include_router(rag_router)
 
 __all__ = [
     "api_router",
@@ -26,4 +30,6 @@ __all__ = [
     "agent_router",
     "compliance_router",
     "status_router",
+    "perception_router",
+    "rag_router",
 ]

backend/app/api/routes/agent.py

@@ -20,7 +20,7 @@ from app.api.models import (
 )
 from app.config.settings import settings
 from app.shared.async_utils import iter_in_thread
-from app.shared.bootstrap import get_agent_conversation_service, get_conversation_store
+from app.shared.bootstrap import get_agent_conversation_service, get_agent_session_service
 # Keep route handlers close to their transport-layer wiring for easier auditing.
 
 
@@ -65,7 +65,7 @@ async def chat_with_session(request: ChatRequest):
             model=request.model or settings.llm_model,
             top_k=request.top_k or settings.rag_top_k,
         )
-        session = get_conversation_store().get_session(session_id)
+        session = get_agent_session_service().get_session(session_id)
         return ChatResponse(
             session_id=session_id,
             answer=result.answer,
@@ -133,45 +133,52 @@ async def chat_stream(request: ChatRequest):
 @router.get("/session/{session_id}", response_model=SessionInfo)
 async def get_session_info(session_id: str):
     """Return session info."""
-    session = get_conversation_store().get_session(session_id)
-    if not session:
-        raise HTTPException(status_code=404, detail="会话不存在或已过期")
-    return SessionInfo(
-        session_id=session.session_id,
-        message_count=len(session.messages),
-        created_at=session.created_at,
-        updated_at=session.updated_at,
-    )
+    try:
+        session = get_agent_session_service().get_session(session_id)
+        return SessionInfo(
+            session_id=session.session_id,
+            message_count=len(session.messages),
+            created_at=session.created_at,
+            updated_at=session.updated_at,
+        )
+    except ValueError as exc:
+        raise HTTPException(status_code=404, detail=str(exc))
 
 
 @router.get("/session/{session_id}/history")
 async def get_session_history(session_id: str, max_turns: int = 5):
     """Return session history."""
-    session = get_conversation_store().get_session(session_id)
-    if not session:
-        raise HTTPException(status_code=404, detail="会话不存在或已过期")
-    history = [{"role": msg.role, "content": msg.content} for msg in session.messages[-(max_turns * 2):]]
-    return {"session_id": session_id, "history": history}
+    try:
+        history = get_agent_session_service().get_history(session_id=session_id, max_turns=max_turns)
+        return {"session_id": session_id, "history": history}
+    except ValueError as exc:
+        raise HTTPException(status_code=404, detail=str(exc))
 
 
 @router.delete("/session/{session_id}")
 async def delete_session(session_id: str):
     """Delete session."""
-    if not get_conversation_store().delete_session(session_id):
-        raise HTTPException(status_code=404, detail="会话不存在")
-    return {"message": "会话已删除", "session_id": session_id}
+    try:
+        get_agent_session_service().delete_session(session_id)
+        return {"message": "会话已删除", "session_id": session_id}
+    except ValueError as exc:
+        raise HTTPException(status_code=404, detail=str(exc))
 
 
 @router.get("/sessions", response_model=List[SessionInfo])
 async def list_sessions():
     """List sessions."""
-    return [SessionInfo(**item) for item in get_conversation_store().list_sessions()]
+    return [SessionInfo(**item) for item in get_agent_session_service().list_sessions()]
 
 
 @router.post("/feedback")
 async def submit_feedback(request: FeedbackRequest):
     """Submit feedback."""
-    session = get_conversation_store().get_session(request.session_id)
-    if not session:
-        raise HTTPException(status_code=404, detail="会话不存在")
-    return {"message": "反馈已提交", "session_id": request.session_id, "message_index": request.message_index}
+    try:
+        result = get_agent_session_service().submit_feedback(
+            session_id=request.session_id,
+            message_index=request.message_index,
+        )
+        return {"message": "反馈已提交", "session_id": result.session_id, "message_index": result.message_index}
+    except ValueError as exc:
+        raise HTTPException(status_code=404, detail=str(exc))

backend/app/api/routes/knowledge.py

@@ -29,14 +29,19 @@ async def search_knowledge(request: SearchRequest):
         results=[
             SearchResultItem(
                 id=index + 1,
-                content=item.content,
+                content=item.text,
                 score=item.score,
                 metadata={
                     "doc_id": item.doc_id,
-                    "doc_name": item.doc_name,
+                    "doc_title": item.doc_title,
                     "chunk_id": item.chunk_id,
+                    "chunk_type": item.chunk_type,
                     "section_title": item.section_title,
-                    "page_number": item.page_number,
+                    "page_start": item.page_start,
+                    "page_end": item.page_end,
+                    "section_level": item.section_level,
+                    "chunk_index": item.chunk_index,
+                    "piece_index": item.piece_index,
                     **item.metadata,
                 },
             )

backend/app/api/routes/perception.py

@@ -0,0 +1,67 @@
+"""Define API routes for perception (regulatory intelligence)."""
+
+from __future__ import annotations
+
+import json
+
+from fastapi import APIRouter, Query
+from fastapi.responses import StreamingResponse
+
+from app.shared.bootstrap import get_perception_service
+from app.shared.async_utils import iter_in_thread
+
+router = APIRouter(prefix="/perception", tags=["智能感知"])
+
+
+@router.get("/stats")
+async def get_perception_stats():
+    """Return KPI statistics for the perception dashboard."""
+    return get_perception_service().get_stats()
+
+
+@router.get("/events")
+async def list_events(
+    source: str | None = Query(default=None, description="来源筛选 (MIIT/UN-ECE/ISO/国标委/EUR-Lex/IATF)"),
+    impact_level: str | None = Query(default=None, description="影响等级 (high/medium/low)"),
+    limit: int = Query(default=50, ge=1, le=100),
+):
+    """Return regulatory events with optional filters."""
+    events = get_perception_service().list_events(
+        source=source,
+        impact_level=impact_level,
+        limit=limit,
+    )
+    return {"events": events, "total": len(events)}
+
+
+@router.get("/events/{event_id}")
+async def get_event(event_id: str):
+    """Return a single regulatory event by ID."""
+    event = get_perception_service().get_event(event_id)
+    if event is None:
+        from fastapi import HTTPException
+        raise HTTPException(status_code=404, detail=f"Event {event_id} not found")
+    return event
+
+
+@router.post("/events/{event_id}/analyze")
+async def analyze_event(event_id: str):
+    """Stream SSE impact analysis for a regulatory event."""
+    service = get_perception_service()
+
+    async def event_stream():
+        async for item in iter_in_thread(service.analyze_event(event_id)):
+            event_name = item.get("event", "message")
+            data = item.get("data", "")
+            if isinstance(data, (dict, list)):
+                data = json.dumps(data, ensure_ascii=False)
+            yield f"event: {event_name}\ndata: {data}\n\n"
+
+    return StreamingResponse(
+        event_stream(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "X-Accel-Buffering": "no",
+        },
+    )

backend/app/api/routes/rag.py

@@ -50,8 +50,8 @@ async def rag_chat(request: RagChatRequest):
                     {
                         "id": str(s.get("chunk_id") or s.get("doc_id") or idx + 1),
                         "score": s.get("score", 0),
-                        "preview": s.get("content", "")[:200],
-                        "doc_name": s.get("doc_name", ""),
+                        "preview": s.get("text", s.get("content", ""))[:200],
+                        "doc_name": s.get("doc_title", s.get("doc_name", "")),
                         "clause": s.get("section_title", "法规片段"),
                         "doc_id": s.get("doc_id"),
                         "download_url": (

backend/app/application/agent/__init__.py

@@ -1,7 +1,7 @@
 """Initialize the app.application.agent package."""
 
-from .services import AgentConversationService
+from .services import AgentConversationService, AgentSessionFeedbackResult, AgentSessionService
 # Keep package boundaries explicit so backend imports stay predictable.
 
 
-__all__ = ["AgentConversationService"]
+__all__ = ["AgentConversationService", "AgentSessionFeedbackResult", "AgentSessionService"]

backend/app/application/agent/services.py

@@ -1,7 +1,8 @@
-"""Implement application-layer logic for services."""
+"""Implement application-layer logic for agent services."""
 
 from __future__ import annotations
 
+from dataclasses import dataclass
 from typing import Generator
 
 from app.domain.conversation import AnswerGenerator, AnswerResult, ConversationStore
@@ -143,3 +144,48 @@ class AgentConversationService:
             )
         
         return session.session_id, event_stream()
+
+
+@dataclass
+class AgentSessionFeedbackResult:
+    """Represent the result of storing session feedback."""
+
+    session_id: str
+    message_index: int
+
+
+class AgentSessionService:
+    """Provide application-layer access to session management workflows."""
+
+    def __init__(self, *, conversation_store: ConversationStore) -> None:
+        """Initialize the Agent Session Service instance."""
+        self.conversation_store = conversation_store
+
+    def get_session(self, session_id: str):
+        """Return a session by id or raise when it does not exist."""
+        session = self.conversation_store.get_session(session_id)
+        if not session:
+            raise ValueError("会话不存在或已过期")
+        return session
+
+    def get_history(self, *, session_id: str, max_turns: int = 5) -> list[dict[str, str]]:
+        """Return the recent conversation history for a session."""
+        session = self.get_session(session_id)
+        return [{"role": msg.role, "content": msg.content} for msg in session.messages[-(max_turns * 2):]]
+
+    def delete_session(self, session_id: str) -> None:
+        """Delete a session or raise when it does not exist."""
+        if not self.conversation_store.delete_session(session_id):
+            raise ValueError("会话不存在")
+
+    def list_sessions(self) -> list[dict]:
+        """Return the list of visible sessions."""
+        return self.conversation_store.list_sessions()
+
+    def submit_feedback(self, *, session_id: str, message_index: int) -> AgentSessionFeedbackResult:
+        """Validate feedback targets and return a normalized feedback result."""
+        session = self.get_session(session_id)
+        if message_index < 0 or message_index >= len(session.messages):
+            raise ValueError("消息索引不存在")
+        # Preserve the existing API behavior until a persistent feedback store is introduced.
+        return AgentSessionFeedbackResult(session_id=session_id, message_index=message_index)

backend/app/application/documents/services.py

@@ -7,16 +7,22 @@ import tempfile
 import uuid
 import json
 from dataclasses import dataclass
+from datetime import UTC, datetime
 
 from loguru import logger
+from app.config.settings import settings
 
 from app.domain.documents import (
     ChunkBuilder,
     Document,
+    DocumentArtifact,
     DocumentBinaryStore,
     DocumentParser,
+    DocumentProcessingRun,
+    DocumentProcessingStore,
     DocumentRepository,
     DocumentStatus,
+    DocumentStatusEvent,
     ParseArtifactStore,
     ParsedDocument,
 )
@@ -39,6 +45,7 @@ class DocumentProcessResult:
 
 class DocumentCommandService:
     """Provide the Document Command Service service."""
+
     def __init__(
         self,
         *,
@@ -49,6 +56,7 @@ class DocumentCommandService:
         embedding_provider: EmbeddingProvider,
         vector_index: VectorIndex,
         parse_artifact_store: ParseArtifactStore | None = None,
+        document_processing_store: DocumentProcessingStore | None = None,
     ) -> None:
         """Initialize the Document Command Service instance."""
         self.document_repository = document_repository
@@ -58,6 +66,11 @@ class DocumentCommandService:
         self.embedding_provider = embedding_provider
         self.vector_index = vector_index
         self.parse_artifact_store = parse_artifact_store
+        self.document_processing_store = document_processing_store
+
+    def _utcnow(self) -> datetime:
+        """Return the current UTC timestamp for persisted processing metadata."""
+        return datetime.now(UTC)
 
     def _save_parse_artifacts(self, *, doc_id: str, parsed_document: ParsedDocument) -> dict[str, str]:
         """Persist parse artifacts so troubleshooting does not depend on provider retention windows."""
@@ -80,6 +93,143 @@ class DocumentCommandService:
             artifact_keys[name] = object_name
         return artifact_keys
 
+    def _safe_create_processing_run(self, *, doc_id: str, trigger_type: str, generate_summary: bool) -> str | None:
+        """Create a processing run record when the optional store is available."""
+        if not self.document_processing_store:
+            return None
+        run = DocumentProcessingRun(
+            run_id=str(uuid.uuid4()),
+            doc_id=doc_id,
+            trigger_type=trigger_type,
+            run_status="running",
+            parser_backend=settings.parser_backend,
+            chunk_backend=settings.chunk_backend,
+            embedding_model=settings.embedding_model,
+            metadata={"generate_summary": generate_summary},
+        )
+        try:
+            created = self.document_processing_store.create_run(run)
+            return created.run_id
+        except Exception:
+            logger.warning("DocumentProcessingStore.create_run failed for doc_id={}", doc_id)
+            return None
+
+    def _safe_append_status_event(
+        self,
+        *,
+        doc_id: str,
+        run_id: str | None,
+        from_status: str,
+        to_status: str,
+        stage: str,
+        message: str = "",
+        metadata: dict | None = None,
+    ) -> None:
+        """Append a status event without allowing auxiliary persistence failures to abort processing."""
+        if not self.document_processing_store or not run_id:
+            return
+        event = DocumentStatusEvent(
+            event_id=str(uuid.uuid4()),
+            doc_id=doc_id,
+            run_id=run_id,
+            from_status=from_status,
+            to_status=to_status,
+            stage=stage,
+            message=message,
+            metadata=metadata or {},
+        )
+        try:
+            self.document_processing_store.append_status_event(event)
+        except Exception:
+            logger.warning(
+                "DocumentProcessingStore.append_status_event failed for doc_id={}, run_id={}",
+                doc_id,
+                run_id,
+            )
+
+    def _safe_mark_run_stored(self, *, doc_id: str, run_id: str | None) -> None:
+        """Mark the processing run as stored without affecting the main workflow."""
+        if not self.document_processing_store or not run_id:
+            return
+        try:
+            self.document_processing_store.mark_run_stored(run_id, stored_at=self._utcnow())
+        except Exception:
+            logger.warning("DocumentProcessingStore.mark_run_stored failed for doc_id={}, run_id={}", doc_id, run_id)
+
+    def _safe_mark_run_parsed(self, *, doc_id: str, run_id: str | None, parsed_document: ParsedDocument) -> None:
+        """Persist parse completion details without failing the document pipeline."""
+        if not self.document_processing_store or not run_id:
+            return
+        try:
+            self.document_processing_store.mark_run_parsed(
+                run_id,
+                parser_backend=parsed_document.parser_name,
+                layout_count=int(parsed_document.metadata.get("layout_count", len(parsed_document.raw_layouts)) or 0),
+                structure_node_count=len(parsed_document.structure_nodes),
+                semantic_block_count=len(parsed_document.semantic_blocks),
+                vector_chunk_count=len(parsed_document.vector_chunks),
+                parsed_at=self._utcnow(),
+                metadata={"parse_task_id": parsed_document.metadata.get("task_id", "")},
+            )
+        except Exception:
+            logger.warning("DocumentProcessingStore.mark_run_parsed failed for doc_id={}, run_id={}", doc_id, run_id)
+
+    def _safe_replace_processing_artifacts(self, *, doc_id: str, run_id: str | None, artifact_keys: dict[str, str]) -> None:
+        """Store artifact references without turning persistence drift into a user-visible failure."""
+        if not self.document_processing_store or not run_id:
+            return
+        artifacts = [
+            DocumentArtifact(
+                artifact_id=str(uuid.uuid4()),
+                doc_id=doc_id,
+                run_id=run_id,
+                artifact_type=artifact_type,
+                object_name=object_name,
+                content_type="application/json",
+                byte_size=0,
+                checksum="",
+            )
+            for artifact_type, object_name in artifact_keys.items()
+        ]
+        try:
+            self.document_processing_store.replace_artifacts_for_run(run_id, artifacts)
+        except Exception:
+            logger.warning(
+                "DocumentProcessingStore.replace_artifacts_for_run failed for doc_id={}, run_id={}",
+                doc_id,
+                run_id,
+            )
+
+    def _safe_mark_run_indexed(self, *, doc_id: str, run_id: str | None, chunk_count: int, index_name: str) -> None:
+        """Mark the processing run as indexed without affecting the success path."""
+        if not self.document_processing_store or not run_id:
+            return
+        now = self._utcnow()
+        try:
+            self.document_processing_store.mark_run_indexed(
+                run_id,
+                chunk_count=chunk_count,
+                index_name=index_name,
+                indexed_at=now,
+                finished_at=now,
+            )
+        except Exception:
+            logger.warning("DocumentProcessingStore.mark_run_indexed failed for doc_id={}, run_id={}", doc_id, run_id)
+
+    def _safe_mark_run_failed(self, *, doc_id: str, run_id: str | None, failure_stage: str, error_message: str) -> None:
+        """Mark the processing run as failed without masking the original error handling path."""
+        if not self.document_processing_store or not run_id:
+            return
+        try:
+            self.document_processing_store.mark_run_failed(
+                run_id,
+                failure_stage=failure_stage,
+                error_message=error_message,
+                finished_at=self._utcnow(),
+            )
+        except Exception:
+            logger.warning("DocumentProcessingStore.mark_run_failed failed for doc_id={}, run_id={}", doc_id, run_id)
+
     def upload_and_process(
         self,
         *,
@@ -91,11 +241,15 @@ class DocumentCommandService:
         regulation_type: str,
         version: str,
         generate_summary: bool,
+        trigger_type: str = "upload",
     ) -> DocumentProcessResult:
         """Handle upload and process for the Document Command Service instance."""
         doc_id = doc_id or str(uuid.uuid4())[:8]
         final_doc_name = doc_name or file_name
         object_name = f"{doc_id}/{file_name}"
+        run_id: str | None = None
+        current_status = DocumentStatus.PENDING
+        current_stage = "store"
 
         document = Document(
             doc_id=doc_id,
@@ -109,6 +263,19 @@ class DocumentCommandService:
             metadata={"generate_summary": generate_summary},
         )
         self.document_repository.create(document)
+        run_id = self._safe_create_processing_run(
+            doc_id=doc_id,
+            trigger_type=trigger_type,
+            generate_summary=generate_summary,
+        )
+        self._safe_append_status_event(
+            doc_id=doc_id,
+            run_id=run_id,
+            from_status="",
+            to_status=DocumentStatus.PENDING.value,
+            stage="document_created",
+            message="Document record created",
+        )
 
         temp_path = ""
         try:
@@ -119,6 +286,17 @@ class DocumentCommandService:
                 metadata={"doc_id": doc_id},
             )
             self.document_repository.update_status(doc_id, DocumentStatus.STORED)
+            current_status = DocumentStatus.STORED
+            current_stage = "parse"
+            self._safe_mark_run_stored(doc_id=doc_id, run_id=run_id)
+            self._safe_append_status_event(
+                doc_id=doc_id,
+                run_id=run_id,
+                from_status=DocumentStatus.PENDING.value,
+                to_status=DocumentStatus.STORED.value,
+                stage="store",
+                message="Source file stored",
+            )
 
             suffix = os.path.splitext(file_name)[1]
             with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as temp_file:
@@ -130,7 +308,13 @@ class DocumentCommandService:
                 doc_id=doc_id,
                 doc_name=final_doc_name,
             )
-            artifact_keys = self._save_parse_artifacts(doc_id=doc_id, parsed_document=parsed_document)
+            self._safe_mark_run_parsed(doc_id=doc_id, run_id=run_id, parsed_document=parsed_document)
+
+            artifact_keys: dict[str, str] = {}
+            try:
+                artifact_keys = self._save_parse_artifacts(doc_id=doc_id, parsed_document=parsed_document)
+            except Exception:
+                logger.warning("Parse artifact binary persistence failed for doc_id={}", doc_id)
             self.document_repository.update_status(
                 doc_id,
                 DocumentStatus.PARSED,
@@ -146,6 +330,18 @@ class DocumentCommandService:
                     "processing_stage": "parsed",
                 },
             )
+            current_status = DocumentStatus.PARSED
+            current_stage = "embed"
+            self._safe_replace_processing_artifacts(doc_id=doc_id, run_id=run_id, artifact_keys=artifact_keys)
+            self._safe_append_status_event(
+                doc_id=doc_id,
+                run_id=run_id,
+                from_status=DocumentStatus.STORED.value,
+                to_status=DocumentStatus.PARSED.value,
+                stage="parse",
+                message="Document parsed",
+                metadata={"artifact_count": len(artifact_keys)},
+            )
             if self.parse_artifact_store:
                 try:
                     self.parse_artifact_store.save(
@@ -165,6 +361,7 @@ class DocumentCommandService:
                 raise ValueError("解析完成但没有生成可入库的 chunks")
 
             vectors = self.embedding_provider.embed_texts([chunk.embedding_text for chunk in chunks])
+            current_stage = "index"
             inserted = self.vector_index.upsert(chunks, vectors)
             if inserted != len(chunks):
                 logger.warning("Milvus upsert count mismatched: inserted={}, chunks={}", inserted, len(chunks))
@@ -182,6 +379,23 @@ class DocumentCommandService:
                     "processing_stage": "indexed",
                 },
             )
+            current_status = DocumentStatus.INDEXED
+            index_name = health.get("collection_name", "")
+            self._safe_mark_run_indexed(
+                doc_id=doc_id,
+                run_id=run_id,
+                chunk_count=len(chunks),
+                index_name=index_name,
+            )
+            self._safe_append_status_event(
+                doc_id=doc_id,
+                run_id=run_id,
+                from_status=DocumentStatus.PARSED.value,
+                to_status=DocumentStatus.INDEXED.value,
+                stage="index",
+                message="Document indexed",
+                metadata={"chunk_count": len(chunks), "index_name": index_name},
+            )
             stored = self.document_repository.get(doc_id)
             return DocumentProcessResult(
                 doc_id=doc_id,
@@ -194,6 +408,7 @@ class DocumentCommandService:
             )
         except Exception as exc:
             logger.exception("文档处理失败: doc_id={}", doc_id)
+            failure_stage = current_stage
             self.document_repository.update_status(
                 doc_id,
                 DocumentStatus.FAILED,
@@ -201,8 +416,23 @@ class DocumentCommandService:
                 metadata={
                     "failure_reason": str(exc),
                     "processing_stage": "failed",
+                    "failure_stage": failure_stage,
                 },
             )
+            self._safe_mark_run_failed(
+                doc_id=doc_id,
+                run_id=run_id,
+                failure_stage=failure_stage,
+                error_message=str(exc),
+            )
+            self._safe_append_status_event(
+                doc_id=doc_id,
+                run_id=run_id,
+                from_status=current_status.value,
+                to_status=DocumentStatus.FAILED.value,
+                stage=failure_stage,
+                message=str(exc),
+            )
             return DocumentProcessResult(
                 doc_id=doc_id,
                 doc_name=final_doc_name,
@@ -235,6 +465,11 @@ class DocumentCommandService:
                 self.parse_artifact_store.delete(doc_id)
             except Exception:
                 logger.warning("ParseArtifactStore delete failed for doc_id={}", doc_id)
+        if self.document_processing_store:
+            try:
+                self.document_processing_store.delete_by_document(doc_id)
+            except Exception:
+                logger.warning("DocumentProcessingStore delete failed for doc_id={}", doc_id)
         self.document_repository.delete(doc_id)
         return True
 
@@ -253,6 +488,7 @@ class DocumentCommandService:
             regulation_type=document.regulation_type,
             version=document.version,
             generate_summary=bool(document.metadata.get("generate_summary", False)),
+            trigger_type="retry",
         )
 
 
@@ -272,7 +508,7 @@ class DocumentQueryService:
         """Return documents with real-time state from Milvus as the authoritative source.
 
         Algorithm:
-        1. Query Milvus for all doc metadata (doc_id, doc_name, chunk_count, …).
+        1. Query Milvus for all doc metadata (doc_id, doc_title, chunk_count, …).
         2. Load JSON/PG metadata records and index them by doc_id.
         3. Merge: Milvus-present docs get status=INDEXED and live chunk_count;
            metadata-only docs with status=INDEXED are demoted to FAILED.
@@ -300,8 +536,8 @@ class DocumentQueryService:
                 doc.chunk_count = row["chunk_count"]
                 doc.status = DocumentStatus.INDEXED
                 # Backfill fields that may be missing from older JSON records.
-                if not doc.doc_name and row.get("doc_name"):
-                    doc.doc_name = row["doc_name"]
+                if not doc.doc_name and row.get("doc_title"):
+                    doc.doc_name = row["doc_title"]
                 if not doc.regulation_type and row.get("regulation_type"):
                     doc.regulation_type = row["regulation_type"]
                 if not doc.version and row.get("version"):
@@ -317,8 +553,8 @@ class DocumentQueryService:
             if doc_id not in meta_by_id:
                 synthetic = Document(
                     doc_id=doc_id,
-                    doc_name=row.get("doc_name", doc_id),
-                    file_name=row.get("doc_name", doc_id),
+                    doc_name=row.get("doc_title", doc_id),
+                    file_name=row.get("doc_title", doc_id),
                     object_name="",
                     content_type="",
                     size_bytes=0,

backend/app/application/knowledge/services.py

@@ -29,11 +29,16 @@ def _reciprocal_rank_fusion(
         RetrievedChunk(
             chunk_id=chunk_map[ck].chunk_id,
             doc_id=chunk_map[ck].doc_id,
-            doc_name=chunk_map[ck].doc_name,
-            content=chunk_map[ck].content,
+            doc_title=chunk_map[ck].doc_title,
+            text=chunk_map[ck].text,
             score=scores[ck],
+            chunk_type=chunk_map[ck].chunk_type,
             section_title=chunk_map[ck].section_title,
-            page_number=chunk_map[ck].page_number,
+            page_start=chunk_map[ck].page_start,
+            page_end=chunk_map[ck].page_end,
+            section_level=chunk_map[ck].section_level,
+            chunk_index=chunk_map[ck].chunk_index,
+            piece_index=chunk_map[ck].piece_index,
             metadata=chunk_map[ck].metadata,
         )
         for ck in sorted_keys

backend/app/application/perception/__init__.py

@@ -0,0 +1 @@
+"""Perception application package."""

backend/app/application/perception/services.py

@@ -0,0 +1,143 @@
+"""Perception application service — event listing and streaming impact analysis."""
+
+from __future__ import annotations
+
+import json
+from typing import Generator
+
+from app.application.knowledge.services import KnowledgeRetrievalService
+from app.infrastructure.perception.mock_event_store import MockEventStore
+from app.services.llm.llm_factory import get_llm_client
+from app.config.settings import settings
+
+
+_ANALYSIS_SYSTEM_PROMPT = (
+    "你是汽车行业法规合规专家，专注于中国国家标准（GB）、国际法规（UN-ECE、ISO）"
+    "及欧盟法规（EUR-Lex）的解读与合规建议。回答需专业、简洁、结构清晰。"
+)
+
+
+class PerceptionService:
+    """Orchestrate regulatory event queries and streaming impact analysis."""
+
+    def __init__(
+        self,
+        event_store: MockEventStore,
+        retrieval_service: KnowledgeRetrievalService,
+    ) -> None:
+        self._store = event_store
+        self._retrieval = retrieval_service
+
+    # ------------------------------------------------------------------
+    # Queries
+    # ------------------------------------------------------------------
+
+    def list_events(
+        self,
+        *,
+        source: str | None = None,
+        impact_level: str | None = None,
+        limit: int = 50,
+    ) -> list[dict]:
+        return self._store.filter(source=source, impact_level=impact_level, limit=limit)
+
+    def get_event(self, event_id: str) -> dict | None:
+        return self._store.get(event_id)
+
+    def get_stats(self) -> dict:
+        return self._store.stats()
+
+    # ------------------------------------------------------------------
+    # Streaming analysis
+    # ------------------------------------------------------------------
+
+    def analyze_event(self, event_id: str) -> Generator[dict, None, None]:
+        """Yield SSE-ready dicts: sources → content chunks → done."""
+        event = self._store.get(event_id)
+        if not event:
+            yield {"event": "error", "data": f"事件 {event_id} 不存在"}
+            return
+
+        # --- 1. RAG retrieval: find related library documents ---
+        query = event["title"] + " " + " ".join(event["tags"])
+        chunks: list = []
+        affected_docs: list[dict] = []
+        try:
+            chunks = self._retrieval.retrieve(query=query, top_k=5)
+            seen: set[str] = set()
+            for chunk in chunks:
+                if chunk.doc_id not in seen:
+                    seen.add(chunk.doc_id)
+                    affected_docs.append(
+                        {
+                            "doc_id": chunk.doc_id,
+                            "doc_title": chunk.doc_title,
+                            "score": round(float(chunk.score), 4),
+                            "snippet": (chunk.text or "")[:180],
+                            "clause": getattr(chunk, "section_title", "") or "",
+                        }
+                    )
+        except Exception:  # noqa: BLE001
+            pass
+
+        yield {"event": "sources", "data": json.dumps(affected_docs, ensure_ascii=False)}
+
+        # --- 2. Build context from retrieved chunks ---
+        context_parts = [
+            f"[文档{i}: {c.doc_title}]\n{(c.text or '')[:400]}"
+            for i, c in enumerate(chunks[:5], 1)
+        ]
+        context = "\n\n".join(context_parts) if context_parts else "（知识库中暂无相关文档）"
+
+        # --- 3. Build prompt ---
+        effective = event.get("effective_at") or "待定"
+        user_content = f"""请对以下法规动态进行专业影响分析。
+
+【法规动态】
+标准编号：{event['standard_code']}
+标题：{event['title']}
+来源：{event['source_label']}
+摘要：{event['summary']}
+生效日期：{effective}
+分类：{event['category']}
+关键词：{', '.join(event['tags'])}
+
+【知识库关联文档】
+{context}
+
+请用 Markdown 格式，从以下四个维度进行分析：
+
+## 核心变化
+列出本次法规更新最关键的 3-5 项变化（用 - 列表）
+
+## 业务影响
+分析对现有产品、认证流程、技术文档的具体影响
+
+## 整改建议
+给出优先级排序的行动清单（标注 🔴高 🟡中 🟢低 优先级）
+
+## 时间节点
+关键合规时间表与里程碑提醒"""
+
+        messages = [
+            {"role": "system", "content": _ANALYSIS_SYSTEM_PROMPT},
+            {"role": "user", "content": user_content},
+        ]
+
+        # --- 4. Stream LLM response ---
+        try:
+            client = get_llm_client(
+                provider=settings.llm_provider,
+                model=settings.llm_model,
+            )
+            if hasattr(client, "stream_chat"):
+                for chunk in client.stream_chat(messages):
+                    yield {"event": "content", "data": chunk}
+            else:
+                response = client.chat(messages)
+                yield {"event": "content", "data": response.content or ""}
+        except Exception as exc:  # noqa: BLE001
+            yield {"event": "error", "data": str(exc)}
+            return
+
+        yield {"event": "done", "data": "{}"}

backend/app/config/settings.py

@@ -33,7 +33,7 @@ class Settings(BaseSettings):
     # Keep configuration setup explicit so runtime behavior is easy to reason about.
     milvus_host: str = Field(default="6.86.80.8", description="Milvus服务地址")
     milvus_port: int = Field(default=19530, description="Milvus服务端口")
-    milvus_collection: str = Field(default="regulations_dense_1024_v1", description="法规向量集合名称")
+    milvus_collection: str = Field(default="regulations_dense_1024_v2", description="法规向量集合名称")
     milvus_db_name: str = Field(default="default", description="Milvus数据库名称")
 
     # Keep configuration setup explicit so runtime behavior is easy to reason about.
@@ -78,6 +78,7 @@ class Settings(BaseSettings):
     chunk_overlap: int = Field(default=50, description="分块重叠大小")
     max_file_size_mb: int = Field(default=100, description="最大文件大小(MB)")
     document_metadata_path: str = Field(default="backend/data/documents.json", description="文档元数据存储路径")
+    document_processing_metadata_path: str = Field(default="backend/data/document_processing.json", description="文档处理历史存储路径")
     parser_backend: str = Field(default="aliyun", description="解析后端(local/aliyun)")
     chunk_backend: str = Field(default="aliyun", description="分块后端(local/aliyun)")
     document_repository_backend: str = Field(default="json", description="文档元数据存储后端 (json/postgres)")

backend/app/core/config.py

@@ -27,7 +27,7 @@ class Settings(BaseSettings):
     # Milvus
     milvus_host: str = "6.86.80.8"
     milvus_port: int = 19530
-    milvus_collection: str = "regulations_dense_1024_v1"
+    milvus_collection: str = "regulations_dense_1024_v2"
 
     # LLM / embedding defaults aligned with the migrated backend path.
     llm_model: str = "qwen-max"
@@ -47,7 +47,7 @@ class Settings(BaseSettings):
     api_port: int = 8000
 
     # Legacy aliases retained for old utility modules.
-    regulations_collection: str = "regulations_dense_1024_v1"
+    regulations_collection: str = "regulations_dense_1024_v2"
     compliance_collection: str = "compliance_cache"
 
 # Preserve the legacy module API while keeping env resolution centralized at the repo root.

backend/app/domain/conversation/models.py

@@ -8,18 +8,91 @@ from typing import Any
 
 
 
-@dataclass
+@dataclass(init=False)
 class AnswerSource:
-    """Represent answer source data."""
+    """Represent answer source data with legacy aliases."""
+
     doc_id: str
-    doc_name: str
+    doc_title: str
     chunk_id: str
+    chunk_type: str
     section_title: str
-    page_number: int
+    page_start: int
+    page_end: int
+    section_level: int
+    chunk_index: int
+    piece_index: int
     score: float
-    content: str
+    text: str
     metadata: dict[str, Any] = field(default_factory=dict)
 
+    def __init__(
+        self,
+        *,
+        doc_id: str,
+        doc_title: str | None = None,
+        chunk_id: str,
+        chunk_type: str = "",
+        section_title: str = "",
+        page_start: int = 0,
+        page_end: int = 0,
+        section_level: int = 0,
+        chunk_index: int = 0,
+        piece_index: int = 0,
+        score: float = 0.0,
+        text: str | None = None,
+        metadata: dict[str, Any] | None = None,
+        doc_name: str | None = None,
+        content: str | None = None,
+        page_number: int | None = None,
+        **_: Any,
+    ) -> None:
+        """Initialize the answer source while accepting legacy field names."""
+        self.doc_id = doc_id
+        self.doc_title = doc_title if doc_title is not None else (doc_name or "")
+        self.chunk_id = chunk_id
+        self.chunk_type = chunk_type
+        self.section_title = section_title
+        self.page_start = int(page_start or page_number or 0)
+        self.page_end = int(page_end or self.page_start)
+        self.section_level = int(section_level or 0)
+        self.chunk_index = int(chunk_index or 0)
+        self.piece_index = int(piece_index or 0)
+        self.score = float(score)
+        self.text = text if text is not None else (content or "")
+        self.metadata = dict(metadata or {})
+
+    @property
+    def doc_name(self) -> str:
+        """Return the legacy document name alias."""
+        return self.doc_title
+
+    @doc_name.setter
+    def doc_name(self, value: str) -> None:
+        """Update the legacy document name alias."""
+        self.doc_title = value
+
+    @property
+    def content(self) -> str:
+        """Return the legacy content alias."""
+        return self.text
+
+    @content.setter
+    def content(self, value: str) -> None:
+        """Update the legacy content alias."""
+        self.text = value
+
+    @property
+    def page_number(self) -> int:
+        """Return the legacy page number alias."""
+        return self.page_start
+
+    @page_number.setter
+    def page_number(self, value: int) -> None:
+        """Update the legacy page number alias."""
+        self.page_start = value
+        self.page_end = max(self.page_end, value)
+
 
 @dataclass
 class ConversationMessage:

backend/app/domain/documents/__init__.py

@@ -1,18 +1,29 @@
 """Initialize the app.domain.documents package."""
 
-from .models import Chunk, Document, DocumentStatus, ParsedDocument
-from .ports import ChunkBuilder, DocumentBinaryStore, DocumentParser, DocumentRepository, ParseArtifactStore
+from .models import Chunk, Document, DocumentArtifact, DocumentProcessingRun, DocumentStatus, DocumentStatusEvent, ParsedDocument
+from .ports import (
+    ChunkBuilder,
+    DocumentBinaryStore,
+    DocumentParser,
+    DocumentProcessingStore,
+    DocumentRepository,
+    ParseArtifactStore,
+)
 # Keep package boundaries explicit so backend imports stay predictable.
 
 
 __all__ = [
     "Chunk",
     "Document",
+    "DocumentArtifact",
+    "DocumentProcessingRun",
     "DocumentStatus",
+    "DocumentStatusEvent",
     "ParsedDocument",
     "ChunkBuilder",
     "DocumentBinaryStore",
     "DocumentParser",
+    "DocumentProcessingStore",
     "DocumentRepository",
     "ParseArtifactStore",
 ]

backend/app/domain/documents/models.py

@@ -60,19 +60,171 @@ class ParsedDocument:
     metadata: dict[str, Any] = field(default_factory=dict)
 
 
-@dataclass
+@dataclass(init=False)
 class Chunk:
-    """Represent the Chunk type."""
+    """Represent one retrieval chunk with backward-compatible aliases."""
+
     chunk_id: str
     doc_id: str
-    doc_name: str
-    content: str
+    doc_title: str
+    text: str
     embedding_text: str
+    chunk_type: str = ""
+    chunk_index: int = 0
+    piece_index: int = 0
+    page_start: int = 0
+    page_end: int = 0
     section_title: str = ""
     section_path: list[str] = field(default_factory=list)
-    page_number: int = 0
+    section_level: int = 0
+    source_ids: list[str] = field(default_factory=list)
     regulation_type: str = ""
     version: str = ""
     semantic_id: str = ""
-    block_type: str = ""
     metadata: dict[str, Any] = field(default_factory=dict)
+
+    def __init__(
+        self,
+        *,
+        chunk_id: str,
+        doc_id: str,
+        doc_title: str | None = None,
+        text: str | None = None,
+        embedding_text: str = "",
+        chunk_type: str = "",
+        chunk_index: int = 0,
+        piece_index: int = 0,
+        page_start: int = 0,
+        page_end: int = 0,
+        section_title: str = "",
+        section_path: list[str] | None = None,
+        section_level: int = 0,
+        source_ids: list[str] | None = None,
+        regulation_type: str = "",
+        version: str = "",
+        semantic_id: str = "",
+        metadata: dict[str, Any] | None = None,
+        doc_name: str | None = None,
+        content: str | None = None,
+        page_number: int | None = None,
+        block_type: str | None = None,
+        **_: Any,
+    ) -> None:
+        """Initialize the chunk while accepting legacy field names."""
+        self.chunk_id = chunk_id
+        self.doc_id = doc_id
+        self.doc_title = doc_title if doc_title is not None else (doc_name or "")
+        self.text = text if text is not None else (content or "")
+        self.embedding_text = embedding_text or self.text
+        self.chunk_type = chunk_type or (block_type or "")
+        self.chunk_index = int(chunk_index or 0)
+        self.piece_index = int(piece_index or 0)
+        self.page_start = int(page_start or page_number or 0)
+        self.page_end = int(page_end or self.page_start)
+        self.section_title = section_title
+        self.section_path = list(section_path or [])
+        self.section_level = int(section_level or 0)
+        self.source_ids = list(source_ids or [])
+        self.regulation_type = regulation_type
+        self.version = version
+        self.semantic_id = semantic_id
+        self.metadata = dict(metadata or {})
+
+    @property
+    def doc_name(self) -> str:
+        """Return the legacy document name alias."""
+        return self.doc_title
+
+    @doc_name.setter
+    def doc_name(self, value: str) -> None:
+        """Update the legacy document name alias."""
+        self.doc_title = value
+
+    @property
+    def content(self) -> str:
+        """Return the legacy content alias."""
+        return self.text
+
+    @content.setter
+    def content(self, value: str) -> None:
+        """Update the legacy content alias."""
+        self.text = value
+
+    @property
+    def page_number(self) -> int:
+        """Return the legacy page number alias."""
+        return self.page_start
+
+    @page_number.setter
+    def page_number(self, value: int) -> None:
+        """Update the legacy page number alias."""
+        self.page_start = value
+        self.page_end = max(self.page_end, value)
+
+    @property
+    def block_type(self) -> str:
+        """Return the legacy block type alias."""
+        return self.chunk_type
+
+    @block_type.setter
+    def block_type(self, value: str) -> None:
+        """Update the legacy block type alias."""
+        self.chunk_type = value
+
+
+@dataclass
+class DocumentProcessingRun:
+    """Represent one processing attempt for a document."""
+
+    run_id: str
+    doc_id: str
+    trigger_type: str
+    run_status: str
+    parser_backend: str = ""
+    chunk_backend: str = ""
+    embedding_model: str = ""
+    index_name: str = ""
+    started_at: datetime = field(default_factory=utcnow)
+    stored_at: datetime | None = None
+    parsed_at: datetime | None = None
+    indexed_at: datetime | None = None
+    finished_at: datetime | None = None
+    layout_count: int = 0
+    structure_node_count: int = 0
+    semantic_block_count: int = 0
+    vector_chunk_count: int = 0
+    chunk_count: int = 0
+    failure_stage: str = ""
+    error_message: str = ""
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class DocumentStatusEvent:
+    """Represent a document lifecycle event emitted during processing."""
+
+    event_id: str
+    doc_id: str
+    run_id: str
+    from_status: str
+    to_status: str
+    stage: str
+    message: str = ""
+    metadata: dict[str, Any] = field(default_factory=dict)
+    occurred_at: datetime = field(default_factory=utcnow)
+
+
+@dataclass
+class DocumentArtifact:
+    """Represent a persisted artifact reference for one processing run."""
+
+    artifact_id: str
+    doc_id: str
+    run_id: str
+    artifact_type: str
+    object_name: str
+    content_type: str
+    byte_size: int = 0
+    checksum: str = ""
+    metadata: dict[str, Any] = field(default_factory=dict)
+    created_at: datetime = field(default_factory=utcnow)

backend/app/domain/documents/ports.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from abc import ABC, abstractmethod
 
-from .models import Chunk, Document, DocumentStatus, ParsedDocument
+from .models import Chunk, Document, DocumentArtifact, DocumentProcessingRun, DocumentStatus, DocumentStatusEvent, ParsedDocument
 # Keep domain contracts explicit so adapters can swap implementations cleanly.
 
 
@@ -128,3 +128,111 @@ class ParseArtifactStore(ABC):
     def get_structure_nodes(self, doc_id: str) -> list[dict]:
         """Return all structure nodes for a document."""
         pass
+
+
+class DocumentProcessingStore(ABC):
+    """Persist document processing runs, events, and artifact references."""
+
+    @abstractmethod
+    def create_run(self, run: DocumentProcessingRun) -> DocumentProcessingRun:
+        """Create a new processing run record."""
+        pass
+
+    @abstractmethod
+    def mark_run_stored(
+        self,
+        run_id: str,
+        *,
+        stored_at: object | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as having persisted the source file."""
+        pass
+
+    @abstractmethod
+    def mark_run_parsed(
+        self,
+        run_id: str,
+        *,
+        parser_backend: str,
+        layout_count: int,
+        structure_node_count: int,
+        semantic_block_count: int,
+        vector_chunk_count: int,
+        parsed_at: object | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Record parse completion details for a run."""
+        pass
+
+    @abstractmethod
+    def mark_run_indexed(
+        self,
+        run_id: str,
+        *,
+        chunk_count: int,
+        index_name: str,
+        indexed_at: object | None = None,
+        finished_at: object | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as successfully indexed."""
+        pass
+
+    @abstractmethod
+    def mark_run_failed(
+        self,
+        run_id: str,
+        *,
+        failure_stage: str,
+        error_message: str,
+        finished_at: object | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as failed."""
+        pass
+
+    @abstractmethod
+    def append_status_event(self, event: DocumentStatusEvent) -> DocumentStatusEvent:
+        """Append a document status event."""
+        pass
+
+    @abstractmethod
+    def replace_artifacts_for_run(self, run_id: str, artifacts: list[DocumentArtifact]) -> list[DocumentArtifact]:
+        """Replace all artifacts for a run with the provided list."""
+        pass
+
+    @abstractmethod
+    def delete_by_document(self, doc_id: str) -> None:
+        """Delete all processing data for a document."""
+        pass
+
+    @abstractmethod
+    def list_runs_by_document(self, doc_id: str) -> list[DocumentProcessingRun]:
+        """List all processing runs for a document."""
+        pass
+
+    @abstractmethod
+    def get_run(self, run_id: str) -> DocumentProcessingRun | None:
+        """Return one processing run by identifier."""
+        pass
+
+    @abstractmethod
+    def list_status_events_by_document(self, doc_id: str) -> list[DocumentStatusEvent]:
+        """List status events for a document."""
+        pass
+
+    @abstractmethod
+    def list_status_events_by_run(self, run_id: str) -> list[DocumentStatusEvent]:
+        """List status events for a run."""
+        pass
+
+    @abstractmethod
+    def list_artifacts_by_document(self, doc_id: str) -> list[DocumentArtifact]:
+        """List artifact references for a document."""
+        pass
+
+    @abstractmethod
+    def list_artifacts_by_run(self, run_id: str) -> list[DocumentArtifact]:
+        """List artifact references for a run."""
+        pass

backend/app/domain/retrieval/models.py

@@ -16,14 +16,88 @@ class RetrievalQuery:
     filters: str | None = None
 
 
-@dataclass
+@dataclass(init=False)
 class RetrievedChunk:
-    """Represent the Retrieved Chunk type."""
+    """Represent the retrieved chunk payload with legacy aliases."""
+
     chunk_id: str
     doc_id: str
-    doc_name: str
-    content: str
+    doc_title: str
+    text: str
     score: float
+    chunk_type: str = ""
     section_title: str = ""
-    page_number: int = 0
+    page_start: int = 0
+    page_end: int = 0
+    section_level: int = 0
+    chunk_index: int = 0
+    piece_index: int = 0
     metadata: dict[str, Any] = field(default_factory=dict)
+
+    def __init__(
+        self,
+        *,
+        chunk_id: str,
+        doc_id: str,
+        doc_title: str | None = None,
+        text: str | None = None,
+        score: float = 0.0,
+        chunk_type: str = "",
+        section_title: str = "",
+        page_start: int = 0,
+        page_end: int = 0,
+        section_level: int = 0,
+        chunk_index: int = 0,
+        piece_index: int = 0,
+        metadata: dict[str, Any] | None = None,
+        doc_name: str | None = None,
+        content: str | None = None,
+        page_number: int | None = None,
+        block_type: str | None = None,
+        **_: Any,
+    ) -> None:
+        """Initialize the retrieved chunk while accepting legacy field names."""
+        self.chunk_id = chunk_id
+        self.doc_id = doc_id
+        self.doc_title = doc_title if doc_title is not None else (doc_name or "")
+        self.text = text if text is not None else (content or "")
+        self.score = float(score)
+        self.chunk_type = chunk_type or (block_type or "")
+        self.section_title = section_title
+        self.page_start = int(page_start or page_number or 0)
+        self.page_end = int(page_end or self.page_start)
+        self.section_level = int(section_level or 0)
+        self.chunk_index = int(chunk_index or 0)
+        self.piece_index = int(piece_index or 0)
+        self.metadata = dict(metadata or {})
+
+    @property
+    def doc_name(self) -> str:
+        """Return the legacy document name alias."""
+        return self.doc_title
+
+    @doc_name.setter
+    def doc_name(self, value: str) -> None:
+        """Update the legacy document name alias."""
+        self.doc_title = value
+
+    @property
+    def content(self) -> str:
+        """Return the legacy content alias."""
+        return self.text
+
+    @content.setter
+    def content(self, value: str) -> None:
+        """Update the legacy content alias."""
+        self.text = value
+
+    @property
+    def page_number(self) -> int:
+        """Return the legacy page number alias."""
+        return self.page_start
+
+    @page_number.setter
+    def page_number(self, value: int) -> None:
+        """Update the legacy page number alias."""
+        self.page_start = value
+        self.page_end = max(self.page_end, value)

backend/app/infrastructure/llm/openai_compatible_answer_generator.py

@@ -45,10 +45,10 @@ class OpenAICompatibleAnswerGenerator(AnswerGenerator):
         context_tokens = 0
         for idx, chunk in enumerate(retrieved_chunks, start=1):
             block = (
-                f"[{idx}] 文档: {chunk.doc_name}\n"
+                f"[{idx}] 文档: {chunk.doc_title}\n"
                 f"章节: {chunk.section_title or '未标注'}\n"
-                f"页码: {chunk.page_number}\n"
-                f"内容: {chunk.content}"
+                f"页码: {chunk.page_start}" + (f"-{chunk.page_end}" if chunk.page_end and chunk.page_end != chunk.page_start else "") + "\n"
+                f"内容: {chunk.text}"
             )
             block_tokens = self._estimate_tokens(block)
             if context_tokens + block_tokens > settings.rag_max_context_tokens:
@@ -67,17 +67,37 @@ class OpenAICompatibleAnswerGenerator(AnswerGenerator):
         )
         return messages, context_tokens
 
+    def _is_context_truncated(self, *, retrieved_chunks: list[RetrievedChunk], context_tokens: int) -> bool:
+        """Return whether the prompt context had to omit retrieved chunks to fit the token budget."""
+        if not retrieved_chunks:
+            return False
+        estimated_total_tokens = sum(
+            self._estimate_tokens(
+                f"[{idx}] 文档: {chunk.doc_title}\n"
+                f"章节: {chunk.section_title or '未标注'}\n"
+                f"页码: {chunk.page_start}" + (f"-{chunk.page_end}" if chunk.page_end and chunk.page_end != chunk.page_start else "") + "\n"
+                f"内容: {chunk.text}"
+            )
+            for idx, chunk in enumerate(retrieved_chunks, start=1)
+        )
+        return estimated_total_tokens > context_tokens
+
     def _sources(self, chunks: list[RetrievedChunk]) -> list[AnswerSource]:
         """Handle sources for this module for the Open A I Compatible Answer Generator instance."""
         return [
             AnswerSource(
                 doc_id=chunk.doc_id,
-                doc_name=chunk.doc_name,
+                doc_title=chunk.doc_title,
                 chunk_id=chunk.chunk_id,
+                chunk_type=chunk.chunk_type,
                 section_title=chunk.section_title,
-                page_number=chunk.page_number,
+                page_start=chunk.page_start,
+                page_end=chunk.page_end,
+                section_level=chunk.section_level,
+                chunk_index=chunk.chunk_index,
+                piece_index=chunk.piece_index,
                 score=chunk.score,
-                content=chunk.content,
+                text=chunk.text,
                 metadata=chunk.metadata,
             )
             for chunk in chunks
@@ -111,7 +131,10 @@ class OpenAICompatibleAnswerGenerator(AnswerGenerator):
             latency_ms=latency_ms,
             retrieved_count=len(retrieved_chunks),
             context_tokens=context_tokens,
-            truncated=len(retrieved_chunks) > len(messages),
+            truncated=self._is_context_truncated(
+                retrieved_chunks=retrieved_chunks,
+                context_tokens=context_tokens,
+            ),
             error=response.error,
         )
 

backend/app/infrastructure/parser/local_chunk_builder.py

@@ -10,6 +10,7 @@ class LocalRegulationChunkBuilder(ChunkBuilder):
     """Adapt the existing markdown chunker to the new chunk builder port."""
 
     def __init__(self, *, chunk_size: int = 512, chunk_overlap: int = 50) -> None:
+        """Initialize the local markdown chunk builder."""
         self.chunker = RegulationChunker(
             chunk_size=chunk_size,
             chunk_overlap=chunk_overlap,
@@ -22,6 +23,7 @@ class LocalRegulationChunkBuilder(ChunkBuilder):
         regulation_type: str,
         version: str,
     ) -> list[Chunk]:
+        """Build migrated chunk objects from the legacy markdown chunker output."""
         markdown_text = parsed_document.raw_text.strip()
         if not markdown_text:
             return []
@@ -50,16 +52,18 @@ class LocalRegulationChunkBuilder(ChunkBuilder):
                 Chunk(
                     chunk_id=item.metadata.chunk_id,
                     doc_id=parsed_document.doc_id,
-                    doc_name=parsed_document.doc_name,
-                    content=item.content,
+                    doc_title=parsed_document.doc_name,
+                    text=item.content,
                     embedding_text=item.content,
+                    chunk_type="local_markdown_chunk",
                     section_title=item.metadata.section_title or item.metadata.section_number,
                     section_path=section_path,
-                    page_number=item.metadata.page_number,
+                    page_start=item.metadata.page_number,
+                    page_end=item.metadata.page_number,
+                    section_level=len(section_path),
                     regulation_type=regulation_type,
                     version=version,
                     semantic_id=item.metadata.clause_number,
-                    block_type="local_markdown_chunk",
                     metadata=metadata,
                 )
             )

backend/app/infrastructure/parser/vector_chunk_builder.py

@@ -19,29 +19,35 @@ class AliyunVectorChunkBuilder(ChunkBuilder):
         """Handle build for the Aliyun Vector Chunk Builder instance."""
         chunks: list[Chunk] = []
         for index, item in enumerate(parsed_document.vector_chunks):
-            content = item.get("content") or item.get("text") or ""
-            embedding_text = item.get("embedding_text") or content
+            text = item.get("text") or ""
+            embedding_text = item.get("embedding_text") or text
             if not embedding_text.strip():
                 continue
             section_path = item.get("section_path") or []
             section_title = item.get("section_title") or (section_path[-1] if section_path else "")
-            page_number = item.get("page_start") or item.get("page") or 0
             chunk_id = item.get("chunk_id") or f"{parsed_document.doc_id}-chunk-{index}"
-            metadata = {k: v for k, v in item.items() if k not in {"content", "embedding_text"}}
+            metadata = dict(item)
+            metadata["regulation_type"] = regulation_type
+            metadata["version"] = version
             chunks.append(
                 Chunk(
                     chunk_id=str(chunk_id),
                     doc_id=parsed_document.doc_id,
-                    doc_name=parsed_document.doc_name,
-                    content=content,
+                    doc_title=str(item.get("doc_title") or parsed_document.doc_name),
+                    text=text,
                     embedding_text=embedding_text,
+                    chunk_type=str(item.get("chunk_type", item.get("block_type", ""))),
+                    chunk_index=int(item.get("chunk_index") or 0),
+                    piece_index=int(item.get("piece_index") or 0),
+                    page_start=int(item.get("page_start") or 0),
+                    page_end=int(item.get("page_end") or 0),
                     section_title=section_title,
                     section_path=section_path,
-                    page_number=int(page_number or 0),
+                    section_level=int(item.get("section_level") or len(section_path)),
+                    source_ids=[str(v) for v in item.get("source_ids", [])],
                     regulation_type=regulation_type,
                     version=version,
                     semantic_id=item.get("semantic_id", ""),
-                    block_type=item.get("block_type", ""),
                     metadata=metadata,
                 )
             )

backend/app/infrastructure/perception/__init__.py

@@ -0,0 +1 @@
+"""Perception infrastructure package."""

backend/app/infrastructure/perception/mock_event_store.py

@@ -0,0 +1,421 @@
+"""Mock regulatory event store with 20 high-quality pre-seeded events."""
+
+from __future__ import annotations
+
+from typing import Any
+
+MOCK_EVENTS: list[dict[str, Any]] = [
+    # ------------------------------------------------------------------ HIGH
+    {
+        "id": "evt-001",
+        "source": "MIIT",
+        "source_label": "工业和信息化部",
+        "standard_code": "GB 18384-2025",
+        "title": "《电动汽车安全要求》国家标准第三版正式发布",
+        "summary": (
+            "新增 IP67 级别高压系统密封防护要求；热失控预警响应时间压缩至 5 分钟；"
+            "调整碰撞安全测试工况，新增侧柱碰工况。本标准于 2026 年 7 月 1 日强制实施。"
+        ),
+        "impact_level": "high",
+        "published_at": "2025-11-15",
+        "effective_at": "2026-07-01",
+        "category": "电动汽车安全",
+        "tags": ["电池安全", "高压防护", "碰撞安全", "热失控"],
+        "source_url": "https://openstd.samr.gov.cn",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-002",
+        "source": "UN-ECE",
+        "source_label": "联合国欧洲经委会",
+        "standard_code": "UN R155 Amendment 3",
+        "title": "UN-ECE R155 网络安全法规第三次修订正式生效",
+        "summary": (
+            "新增对 OTA（空中升级）全生命周期的安全审计要求；强化车辆 TARA"
+            "（威胁分析与风险评估）文档化义务；扩展 CSMS 监控范围至售后服务商。"
+        ),
+        "impact_level": "high",
+        "published_at": "2026-01-20",
+        "effective_at": "2026-07-01",
+        "category": "网络安全",
+        "tags": ["OTA", "网络安全", "CSMS", "TARA", "R155"],
+        "source_url": "https://unece.org/transport/vehicle-regulations",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-003",
+        "source": "MIIT",
+        "source_label": "工业和信息化部",
+        "standard_code": "GB/T 40429-2026（征求意见稿）",
+        "title": "《汽车整车信息安全技术要求》修订征求意见",
+        "summary": (
+            "增加基于人工智能的异常行为检测要求；新增车云通信双向认证机制规范；"
+            "提出数据最小化原则在车辆 OBD 数据收集中的应用细则。"
+        ),
+        "impact_level": "high",
+        "published_at": "2026-03-05",
+        "effective_at": None,
+        "category": "信息安全",
+        "tags": ["信息安全", "数据安全", "AI检测", "OBD"],
+        "source_url": "https://www.miit.gov.cn/",
+        "status": "draft",
+    },
+    {
+        "id": "evt-004",
+        "source": "MIIT",
+        "source_label": "工业和信息化部",
+        "standard_code": "NEV 双积分 2026",
+        "title": "2026 年度新能源汽车双积分管理办法年度调整",
+        "summary": (
+            "纯电动乘用车标准车型积分（CAFC）基准值上调 8%；"
+            "提高 A 级及以上续航里程门槛；新增氢燃料电池商用车积分计算细则。"
+        ),
+        "impact_level": "high",
+        "published_at": "2026-02-28",
+        "effective_at": "2026-04-01",
+        "category": "新能源政策",
+        "tags": ["双积分", "纯电动", "燃料电池", "碳配额"],
+        "source_url": "https://www.miit.gov.cn/",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-017",
+        "source": "MIIT",
+        "source_label": "工业和信息化部",
+        "standard_code": "智能网联汽车准入管理办法实施细则",
+        "title": "智能网联汽车准入管理实施细则正式落地",
+        "summary": (
+            "明确 L3 及以上自动驾驶功能的准入申报路径；"
+            "要求 OEM 建立数据安全管理体系并完成等保 2.0 三级认证；"
+            "道路测试数据留存期延长至 3 年。"
+        ),
+        "impact_level": "high",
+        "published_at": "2026-03-01",
+        "effective_at": "2026-09-01",
+        "category": "智能网联",
+        "tags": ["智能网联", "L3自动驾驶", "准入管理", "数据留存"],
+        "source_url": "https://www.miit.gov.cn/",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-018",
+        "source": "EUR-Lex",
+        "source_label": "欧盟官方公报",
+        "standard_code": "EU Cyber Resilience Act (CRA)",
+        "title": "《欧盟网络韧性法案》核心条款对车联网设备生效",
+        "summary": (
+            "联网汽车 ECU 须满足 CRA「重要类 II」安全要求；"
+            "强制 SBOM（软件物料清单）公开披露；"
+            "OEM 须提供至少 10 年的漏洞修复支持承诺。"
+        ),
+        "impact_level": "high",
+        "published_at": "2026-02-15",
+        "effective_at": "2027-01-01",
+        "category": "网络安全",
+        "tags": ["CRA", "SBOM", "漏洞管理", "网络韧性"],
+        "source_url": "https://eur-lex.europa.eu",
+        "status": "enacted",
+    },
+    # --------------------------------------------------------------- MEDIUM
+    {
+        "id": "evt-005",
+        "source": "UN-ECE",
+        "source_label": "联合国欧洲经委会",
+        "standard_code": "UN R156 Amendment 2",
+        "title": "UN-ECE R156 软件升级与 SUMS 法规补充修订",
+        "summary": (
+            "明确 SUMS（软件更新管理系统）对 ECU 版本追溯的最低保留年限为 15 年；"
+            "新增售后 OTA 推送的用户知情同意要求规范。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2026-01-10",
+        "effective_at": "2026-07-01",
+        "category": "软件升级",
+        "tags": ["OTA", "SUMS", "软件版本", "R156"],
+        "source_url": "https://unece.org/transport/vehicle-regulations",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-006",
+        "source": "国标委",
+        "source_label": "国家标准化管理委员会",
+        "standard_code": "GB/T 35273-2026",
+        "title": "《信息安全技术 个人信息安全规范》更新版发布",
+        "summary": (
+            "将车内人脸识别、声纹采集列为敏感个人信息；"
+            "补充自动驾驶场景下乘员行为数据的去标识化技术规范；"
+            "强化数据出境安全评估触发阈值。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2025-12-01",
+        "effective_at": "2026-06-01",
+        "category": "数据安全",
+        "tags": ["个人信息", "PIPL", "数据安全", "生物识别"],
+        "source_url": "https://openstd.samr.gov.cn",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-007",
+        "source": "EUR-Lex",
+        "source_label": "欧盟官方公报",
+        "standard_code": "EU AI Act — Art.13 & Art.14",
+        "title": "《欧盟人工智能法案》第13-14条透明度与人工监督条款正式生效",
+        "summary": (
+            "要求在汽车 ADAS 系统中植入 AI 使用记录日志；"
+            "驾驶员监控 AI 系统须披露决策逻辑；"
+            "高风险 AI 系统需提供人工干预接口。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2026-02-01",
+        "effective_at": "2026-08-01",
+        "category": "AI 法规",
+        "tags": ["AI法案", "透明度", "ADAS", "高风险AI"],
+        "source_url": "https://eur-lex.europa.eu",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-008",
+        "source": "ISO",
+        "source_label": "国际标准化组织",
+        "standard_code": "ISO 45001:2025 Amd.1",
+        "title": "ISO 45001 职业健康安全管理体系第一次修正",
+        "summary": (
+            "新增心理健康风险纳入 OHS 危害辨识范围；"
+            "明确远程办公人员安全管理职责；"
+            "更新绩效评价指标体系，新增事故未遂事件统计要求。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2025-10-20",
+        "effective_at": "2026-01-01",
+        "category": "EHS 管理",
+        "tags": ["ISO 45001", "EHS", "职业健康", "安全管理"],
+        "source_url": "https://www.iso.org",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-009",
+        "source": "MIIT",
+        "source_label": "工业和信息化部",
+        "standard_code": "GB/T 28001-2026（征求意见）",
+        "title": "《汽车产品安全召回管理规程》修订征求意见",
+        "summary": (
+            "扩展召回触发条件，将 OTA 推送导致的功能异常纳入强制报告范围；"
+            "缩短重大安全隐患召回启动时限至 15 个工作日。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2026-03-15",
+        "effective_at": None,
+        "category": "召回管理",
+        "tags": ["召回", "OTA", "安全隐患", "产品安全"],
+        "source_url": "https://www.miit.gov.cn/",
+        "status": "draft",
+    },
+    {
+        "id": "evt-010",
+        "source": "国标委",
+        "source_label": "国家标准化管理委员会",
+        "standard_code": "GB 38031-2025",
+        "title": "《电动汽车用动力蓄电池安全要求》修订版发布",
+        "summary": (
+            "新增电池系统针刺、浸水、挤压等极端工况测试程序；"
+            "热扩散防护等级要求升级；"
+            "强化 BMS（电池管理系统）状态监测数据记录要求。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2025-09-15",
+        "effective_at": "2026-03-01",
+        "category": "电池安全",
+        "tags": ["动力电池", "BMS", "热扩散", "安全测试"],
+        "source_url": "https://openstd.samr.gov.cn",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-016",
+        "source": "UN-ECE",
+        "source_label": "联合国欧洲经委会",
+        "standard_code": "UN R100 Rev.4（草案）",
+        "title": "UN R100 电动汽车安全认证法规第四次修订草案发布",
+        "summary": (
+            "拟对 400V 以上高压系统的绝缘电阻监测提出实时 CAN 总线传输要求；"
+            "新增极低温工况（-40°C）的电池性能验证程序。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2026-04-08",
+        "effective_at": None,
+        "category": "电动汽车安全",
+        "tags": ["R100", "高压安全", "绝缘监测", "低温性能"],
+        "source_url": "https://unece.org/transport/vehicle-regulations",
+        "status": "draft",
+    },
+    {
+        "id": "evt-019",
+        "source": "ISO",
+        "source_label": "国际标准化组织",
+        "standard_code": "ISO/SAE 21434:2026 Amd.1",
+        "title": "ISO/SAE 21434 汽车网络安全工程第一次修正",
+        "summary": (
+            "将 AI 推理组件纳入汽车网络安全工程范围；"
+            "补充端到端加密通信在 V2X 场景中的 TARA 建模要求；"
+            "新增第三方 ECU 供应商 CSMS 审计方法。"
+        ),
+        "impact_level": "medium",
+        "published_at": "2026-04-10",
+        "effective_at": "2026-10-01",
+        "category": "网络安全",
+        "tags": ["ISO 21434", "网络安全", "V2X", "AI安全"],
+        "source_url": "https://www.iso.org",
+        "status": "enacted",
+    },
+    # ------------------------------------------------------------------ LOW
+    {
+        "id": "evt-011",
+        "source": "ISO",
+        "source_label": "国际标准化组织",
+        "standard_code": "ISO 26262:2026 Ed.3（征求意见）",
+        "title": "ISO 26262 功能安全第三版征求意见启动",
+        "summary": (
+            "拟新增对 AI/ML 组件功能安全验证方法的指导附录；"
+            "讨论 SOTIF（预期功能安全）与 ISO 26262 的协调融合路径。"
+        ),
+        "impact_level": "low",
+        "published_at": "2026-04-01",
+        "effective_at": None,
+        "category": "功能安全",
+        "tags": ["功能安全", "ASIL", "AI安全", "SOTIF"],
+        "source_url": "https://www.iso.org",
+        "status": "consultation",
+    },
+    {
+        "id": "evt-012",
+        "source": "EUR-Lex",
+        "source_label": "欧盟官方公报",
+        "standard_code": "REACH Regulation Update 2026",
+        "title": "欧盟 REACH 法规限制物质清单更新（第 22 批）",
+        "summary": (
+            "新增 3 种 SVHCs（高度关注物质），包括特定阻燃剂和密封材料成分；"
+            "汽车零部件豁免条款调整，影响部分内饰材料供应商。"
+        ),
+        "impact_level": "low",
+        "published_at": "2026-01-30",
+        "effective_at": "2026-09-01",
+        "category": "环保法规",
+        "tags": ["REACH", "SVHCs", "环保", "化学品管理"],
+        "source_url": "https://eur-lex.europa.eu",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-013",
+        "source": "MIIT",
+        "source_label": "工业和信息化部",
+        "standard_code": "CCER 汽车碳配额 2026",
+        "title": "自愿减排（CCER）汽车行业核算方法学更新",
+        "summary": (
+            "更新纯电动汽车全生命周期碳排放核算边界；"
+            "新增动力电池回收环节碳减排量认定方法；"
+            "与全国碳市场对接的企业碳账户数据接口规范发布。"
+        ),
+        "impact_level": "low",
+        "published_at": "2026-02-10",
+        "effective_at": "2026-06-01",
+        "category": "碳排放",
+        "tags": ["CCER", "碳排放", "碳中和", "碳核算"],
+        "source_url": "https://www.miit.gov.cn/",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-014",
+        "source": "IATF",
+        "source_label": "国际汽车工作组",
+        "standard_code": "IATF 16949:2025 CSR 通告",
+        "title": "IATF 16949 质量管理体系客户特殊要求更新通告",
+        "summary": (
+            "多家主机厂（OEM）同步更新 CSR，涵盖软件定义汽车（SDV）"
+            "场景下的质量过程管控；电子电气 BOM 变更管理流程补充规范。"
+        ),
+        "impact_level": "low",
+        "published_at": "2026-03-20",
+        "effective_at": "2026-07-01",
+        "category": "质量管理",
+        "tags": ["IATF 16949", "质量管理", "SDV", "CSR"],
+        "source_url": "https://www.iatfglobaloversight.org",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-015",
+        "source": "国标委",
+        "source_label": "国家标准化管理委员会",
+        "standard_code": "GB 7258-2025 勘误",
+        "title": "《机动车运行安全技术条件》年度勘误发布",
+        "summary": (
+            "更正第 12 章灯光系统技术要求中的参数引用错误；"
+            "澄清前雾灯安装位置尺寸定义；此次为勘误性修订，不影响已认证车型。"
+        ),
+        "impact_level": "low",
+        "published_at": "2026-01-05",
+        "effective_at": "2026-01-05",
+        "category": "运行安全",
+        "tags": ["GB 7258", "灯光", "运行安全", "勘误"],
+        "source_url": "https://openstd.samr.gov.cn",
+        "status": "enacted",
+    },
+    {
+        "id": "evt-020",
+        "source": "国标委",
+        "source_label": "国家标准化管理委员会",
+        "standard_code": "GB/T 27930-2026",
+        "title": "《电动汽车非车载传导式充电通信协议》更新版发布",
+        "summary": (
+            "兼容 CHAdeMO 4.0 与 CCS2 双协议栈；"
+            "新增大功率充电（>350kW）通信握手流程；"
+            "强化充电过程 BMS 实时诊断数据上报规范。"
+        ),
+        "impact_level": "low",
+        "published_at": "2026-03-25",
+        "effective_at": "2026-12-01",
+        "category": "充电标准",
+        "tags": ["充电协议", "BMS", "大功率充电", "CHAdeMO"],
+        "source_url": "https://openstd.samr.gov.cn",
+        "status": "enacted",
+    },
+]
+
+# Index for fast lookup
+_EVENT_INDEX: dict[str, dict] = {e["id"]: e for e in MOCK_EVENTS}
+
+
+class MockEventStore:
+    """In-memory mock store for regulatory events."""
+
+    def all(self) -> list[dict]:
+        return list(MOCK_EVENTS)
+
+    def get(self, event_id: str) -> dict | None:
+        return _EVENT_INDEX.get(event_id)
+
+    def filter(
+        self,
+        *,
+        source: str | None = None,
+        impact_level: str | None = None,
+        limit: int = 50,
+    ) -> list[dict]:
+        events = list(MOCK_EVENTS)
+        if source:
+            events = [e for e in events if e["source"] == source]
+        if impact_level:
+            events = [e for e in events if e["impact_level"] == impact_level]
+        events.sort(key=lambda e: e["published_at"], reverse=True)
+        return events[:limit]
+
+    def stats(self) -> dict:
+        from datetime import date, timedelta
+
+        events = MOCK_EVENTS
+        cutoff = (date.today() - timedelta(days=90)).isoformat()
+        return {
+            "total": len(events),
+            "high_impact": sum(1 for e in events if e["impact_level"] == "high"),
+            "medium_impact": sum(1 for e in events if e["impact_level"] == "medium"),
+            "low_impact": sum(1 for e in events if e["impact_level"] == "low"),
+            "recent_90d": sum(1 for e in events if e["published_at"] >= cutoff),
+        }

backend/app/infrastructure/storage/json_document_processing_store.py

@@ -0,0 +1,373 @@
+"""Implement infrastructure support for json document processing history."""
+
+from __future__ import annotations
+
+import json
+from datetime import UTC, datetime
+from pathlib import Path
+from typing import Any
+
+from app.domain.documents import DocumentArtifact, DocumentProcessingRun, DocumentProcessingStore, DocumentStatusEvent
+# Keep JSON persistence behavior aligned with the lightweight document repository adapter.
+
+
+class JsonDocumentProcessingStore(DocumentProcessingStore):
+    """Persist processing history in a standalone JSON file."""
+
+    def __init__(self, file_path: str) -> None:
+        """Initialize the JSON processing history store."""
+        self.file_path = Path(file_path)
+        self.file_path.parent.mkdir(parents=True, exist_ok=True)
+        if not self.file_path.exists():
+            self._save(self._empty_payload())
+
+    def _empty_payload(self) -> dict[str, dict[str, dict[str, Any]]]:
+        """Return the canonical empty JSON structure for processing history."""
+        return {"runs": {}, "status_events": {}, "artifacts": {}}
+
+    def _load(self) -> dict[str, dict[str, dict[str, Any]]]:
+        """Load the full JSON payload and normalize missing sections."""
+        if not self.file_path.exists():
+            return self._empty_payload()
+        payload = json.loads(self.file_path.read_text(encoding="utf-8") or "{}")
+        normalized = self._empty_payload()
+        for key in normalized:
+            section = payload.get(key, {})
+            normalized[key] = section if isinstance(section, dict) else {}
+        return normalized
+
+    def _save(self, payload: dict[str, dict[str, dict[str, Any]]]) -> None:
+        """Persist the full JSON payload with stable formatting."""
+        self.file_path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    def _serialize_datetime(self, value: datetime | None) -> str | None:
+        """Serialize optional datetimes into ISO8601 strings."""
+        return value.isoformat() if value is not None else None
+
+    def _deserialize_datetime(self, value: str | None) -> datetime | None:
+        """Deserialize optional ISO8601 strings into datetimes."""
+        return datetime.fromisoformat(value) if value else None
+
+    def _serialize_run(self, run: DocumentProcessingRun) -> dict[str, Any]:
+        """Serialize one processing run to a JSON-compatible payload."""
+        return {
+            "run_id": run.run_id,
+            "doc_id": run.doc_id,
+            "trigger_type": run.trigger_type,
+            "run_status": run.run_status,
+            "parser_backend": run.parser_backend,
+            "chunk_backend": run.chunk_backend,
+            "embedding_model": run.embedding_model,
+            "index_name": run.index_name,
+            "started_at": self._serialize_datetime(run.started_at),
+            "stored_at": self._serialize_datetime(run.stored_at),
+            "parsed_at": self._serialize_datetime(run.parsed_at),
+            "indexed_at": self._serialize_datetime(run.indexed_at),
+            "finished_at": self._serialize_datetime(run.finished_at),
+            "layout_count": run.layout_count,
+            "structure_node_count": run.structure_node_count,
+            "semantic_block_count": run.semantic_block_count,
+            "vector_chunk_count": run.vector_chunk_count,
+            "chunk_count": run.chunk_count,
+            "failure_stage": run.failure_stage,
+            "error_message": run.error_message,
+            "metadata": run.metadata,
+        }
+
+    def _deserialize_run(self, payload: dict[str, Any]) -> DocumentProcessingRun:
+        """Deserialize one JSON payload into a processing run dataclass."""
+        return DocumentProcessingRun(
+            run_id=payload["run_id"],
+            doc_id=payload["doc_id"],
+            trigger_type=payload["trigger_type"],
+            run_status=payload["run_status"],
+            parser_backend=payload.get("parser_backend", ""),
+            chunk_backend=payload.get("chunk_backend", ""),
+            embedding_model=payload.get("embedding_model", ""),
+            index_name=payload.get("index_name", ""),
+            started_at=self._deserialize_datetime(payload.get("started_at")) or datetime.now(UTC),
+            stored_at=self._deserialize_datetime(payload.get("stored_at")),
+            parsed_at=self._deserialize_datetime(payload.get("parsed_at")),
+            indexed_at=self._deserialize_datetime(payload.get("indexed_at")),
+            finished_at=self._deserialize_datetime(payload.get("finished_at")),
+            layout_count=int(payload.get("layout_count", 0) or 0),
+            structure_node_count=int(payload.get("structure_node_count", 0) or 0),
+            semantic_block_count=int(payload.get("semantic_block_count", 0) or 0),
+            vector_chunk_count=int(payload.get("vector_chunk_count", 0) or 0),
+            chunk_count=int(payload.get("chunk_count", 0) or 0),
+            failure_stage=payload.get("failure_stage", ""),
+            error_message=payload.get("error_message", ""),
+            metadata=payload.get("metadata", {}),
+        )
+
+    def _serialize_event(self, event: DocumentStatusEvent) -> dict[str, Any]:
+        """Serialize one status event to a JSON-compatible payload."""
+        return {
+            "event_id": event.event_id,
+            "doc_id": event.doc_id,
+            "run_id": event.run_id,
+            "from_status": event.from_status,
+            "to_status": event.to_status,
+            "stage": event.stage,
+            "message": event.message,
+            "metadata": event.metadata,
+            "occurred_at": self._serialize_datetime(event.occurred_at),
+        }
+
+    def _deserialize_event(self, payload: dict[str, Any]) -> DocumentStatusEvent:
+        """Deserialize one JSON payload into a status event dataclass."""
+        return DocumentStatusEvent(
+            event_id=payload["event_id"],
+            doc_id=payload["doc_id"],
+            run_id=payload["run_id"],
+            from_status=payload.get("from_status", ""),
+            to_status=payload["to_status"],
+            stage=payload.get("stage", ""),
+            message=payload.get("message", ""),
+            metadata=payload.get("metadata", {}),
+            occurred_at=self._deserialize_datetime(payload.get("occurred_at")) or datetime.now(UTC),
+        )
+
+    def _serialize_artifact(self, artifact: DocumentArtifact) -> dict[str, Any]:
+        """Serialize one artifact reference to a JSON-compatible payload."""
+        return {
+            "artifact_id": artifact.artifact_id,
+            "doc_id": artifact.doc_id,
+            "run_id": artifact.run_id,
+            "artifact_type": artifact.artifact_type,
+            "object_name": artifact.object_name,
+            "content_type": artifact.content_type,
+            "byte_size": artifact.byte_size,
+            "checksum": artifact.checksum,
+            "metadata": artifact.metadata,
+            "created_at": self._serialize_datetime(artifact.created_at),
+        }
+
+    def _deserialize_artifact(self, payload: dict[str, Any]) -> DocumentArtifact:
+        """Deserialize one JSON payload into an artifact dataclass."""
+        return DocumentArtifact(
+            artifact_id=payload["artifact_id"],
+            doc_id=payload["doc_id"],
+            run_id=payload["run_id"],
+            artifact_type=payload["artifact_type"],
+            object_name=payload["object_name"],
+            content_type=payload.get("content_type", ""),
+            byte_size=int(payload.get("byte_size", 0) or 0),
+            checksum=payload.get("checksum", ""),
+            metadata=payload.get("metadata", {}),
+            created_at=self._deserialize_datetime(payload.get("created_at")) or datetime.now(UTC),
+        )
+
+    def _merge_metadata(self, original: dict[str, Any], update: dict | None) -> dict[str, Any]:
+        """Merge metadata updates onto an existing payload."""
+        merged = dict(original)
+        if update:
+            merged.update(update)
+        return merged
+
+    def create_run(self, run: DocumentProcessingRun) -> DocumentProcessingRun:
+        """Create a new processing run record."""
+        payload = self._load()
+        payload["runs"][run.run_id] = self._serialize_run(run)
+        self._save(payload)
+        return run
+
+    def mark_run_stored(
+        self,
+        run_id: str,
+        *,
+        stored_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as having persisted the source file."""
+        payload = self._load()
+        run_payload = payload["runs"].get(run_id)
+        if not run_payload:
+            return None
+        run = self._deserialize_run(run_payload)
+        run.stored_at = stored_at or datetime.now(UTC)
+        run.metadata = self._merge_metadata(run.metadata, metadata)
+        payload["runs"][run_id] = self._serialize_run(run)
+        self._save(payload)
+        return run
+
+    def mark_run_parsed(
+        self,
+        run_id: str,
+        *,
+        parser_backend: str,
+        layout_count: int,
+        structure_node_count: int,
+        semantic_block_count: int,
+        vector_chunk_count: int,
+        parsed_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Record parse completion details for a run."""
+        payload = self._load()
+        run_payload = payload["runs"].get(run_id)
+        if not run_payload:
+            return None
+        run = self._deserialize_run(run_payload)
+        run.parser_backend = parser_backend
+        run.layout_count = layout_count
+        run.structure_node_count = structure_node_count
+        run.semantic_block_count = semantic_block_count
+        run.vector_chunk_count = vector_chunk_count
+        run.parsed_at = parsed_at or datetime.now(UTC)
+        run.metadata = self._merge_metadata(run.metadata, metadata)
+        payload["runs"][run_id] = self._serialize_run(run)
+        self._save(payload)
+        return run
+
+    def mark_run_indexed(
+        self,
+        run_id: str,
+        *,
+        chunk_count: int,
+        index_name: str,
+        indexed_at: datetime | None = None,
+        finished_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as successfully indexed."""
+        payload = self._load()
+        run_payload = payload["runs"].get(run_id)
+        if not run_payload:
+            return None
+        run = self._deserialize_run(run_payload)
+        now = datetime.now(UTC)
+        run.run_status = "succeeded"
+        run.chunk_count = chunk_count
+        run.index_name = index_name
+        run.indexed_at = indexed_at or now
+        run.finished_at = finished_at or now
+        run.metadata = self._merge_metadata(run.metadata, metadata)
+        payload["runs"][run_id] = self._serialize_run(run)
+        self._save(payload)
+        return run
+
+    def mark_run_failed(
+        self,
+        run_id: str,
+        *,
+        failure_stage: str,
+        error_message: str,
+        finished_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as failed."""
+        payload = self._load()
+        run_payload = payload["runs"].get(run_id)
+        if not run_payload:
+            return None
+        run = self._deserialize_run(run_payload)
+        run.run_status = "failed"
+        run.failure_stage = failure_stage
+        run.error_message = error_message
+        run.finished_at = finished_at or datetime.now(UTC)
+        run.metadata = self._merge_metadata(run.metadata, metadata)
+        payload["runs"][run_id] = self._serialize_run(run)
+        self._save(payload)
+        return run
+
+    def append_status_event(self, event: DocumentStatusEvent) -> DocumentStatusEvent:
+        """Append a document status event."""
+        payload = self._load()
+        payload["status_events"][event.event_id] = self._serialize_event(event)
+        self._save(payload)
+        return event
+
+    def replace_artifacts_for_run(self, run_id: str, artifacts: list[DocumentArtifact]) -> list[DocumentArtifact]:
+        """Replace all artifacts for a run with the provided list."""
+        payload = self._load()
+        payload["artifacts"] = {
+            artifact_id: artifact_payload
+            for artifact_id, artifact_payload in payload["artifacts"].items()
+            if artifact_payload.get("run_id") != run_id
+        }
+        for artifact in artifacts:
+            payload["artifacts"][artifact.artifact_id] = self._serialize_artifact(artifact)
+        self._save(payload)
+        return artifacts
+
+    def delete_by_document(self, doc_id: str) -> None:
+        """Delete all processing data for a document."""
+        payload = self._load()
+        payload["runs"] = {
+            run_id: run_payload
+            for run_id, run_payload in payload["runs"].items()
+            if run_payload.get("doc_id") != doc_id
+        }
+        payload["status_events"] = {
+            event_id: event_payload
+            for event_id, event_payload in payload["status_events"].items()
+            if event_payload.get("doc_id") != doc_id
+        }
+        payload["artifacts"] = {
+            artifact_id: artifact_payload
+            for artifact_id, artifact_payload in payload["artifacts"].items()
+            if artifact_payload.get("doc_id") != doc_id
+        }
+        self._save(payload)
+
+    def list_runs_by_document(self, doc_id: str) -> list[DocumentProcessingRun]:
+        """List all processing runs for a document."""
+        payload = self._load()
+        runs = [
+            self._deserialize_run(run_payload)
+            for run_payload in payload["runs"].values()
+            if run_payload.get("doc_id") == doc_id
+        ]
+        runs.sort(key=lambda run: run.started_at)
+        return runs
+
+    def get_run(self, run_id: str) -> DocumentProcessingRun | None:
+        """Return one processing run by identifier."""
+        payload = self._load()
+        run_payload = payload["runs"].get(run_id)
+        return self._deserialize_run(run_payload) if run_payload else None
+
+    def list_status_events_by_document(self, doc_id: str) -> list[DocumentStatusEvent]:
+        """List status events for a document."""
+        payload = self._load()
+        events = [
+            self._deserialize_event(event_payload)
+            for event_payload in payload["status_events"].values()
+            if event_payload.get("doc_id") == doc_id
+        ]
+        events.sort(key=lambda event: event.occurred_at)
+        return events
+
+    def list_status_events_by_run(self, run_id: str) -> list[DocumentStatusEvent]:
+        """List status events for a run."""
+        payload = self._load()
+        events = [
+            self._deserialize_event(event_payload)
+            for event_payload in payload["status_events"].values()
+            if event_payload.get("run_id") == run_id
+        ]
+        events.sort(key=lambda event: event.occurred_at)
+        return events
+
+    def list_artifacts_by_document(self, doc_id: str) -> list[DocumentArtifact]:
+        """List artifact references for a document."""
+        payload = self._load()
+        artifacts = [
+            self._deserialize_artifact(artifact_payload)
+            for artifact_payload in payload["artifacts"].values()
+            if artifact_payload.get("doc_id") == doc_id
+        ]
+        artifacts.sort(key=lambda artifact: artifact.created_at)
+        return artifacts
+
+    def list_artifacts_by_run(self, run_id: str) -> list[DocumentArtifact]:
+        """List artifact references for a run."""
+        payload = self._load()
+        artifacts = [
+            self._deserialize_artifact(artifact_payload)
+            for artifact_payload in payload["artifacts"].values()
+            if artifact_payload.get("run_id") == run_id
+        ]
+        artifacts.sort(key=lambda artifact: artifact.created_at)
+        return artifacts

backend/app/infrastructure/storage/postgres_document_processing_store.py

@@ -0,0 +1,466 @@
+"""Implement infrastructure support for postgres document processing history."""
+
+from __future__ import annotations
+
+import json
+from contextlib import contextmanager
+from datetime import UTC, datetime
+from typing import Any
+
+import psycopg2
+import psycopg2.extras
+from psycopg2.pool import ThreadedConnectionPool
+
+from app.config.settings import settings
+from app.domain.documents import DocumentArtifact, DocumentProcessingRun, DocumentProcessingStore, DocumentStatusEvent
+# Keep SQL mapping local to this adapter so the domain stays storage-agnostic.
+
+_CREATE_RUNS_TABLE = """
+CREATE TABLE IF NOT EXISTS document_processing_runs (
+    run_id                 VARCHAR(128) PRIMARY KEY,
+    doc_id                 VARCHAR(128) NOT NULL,
+    trigger_type           VARCHAR(32) NOT NULL,
+    run_status             VARCHAR(32) NOT NULL DEFAULT 'running',
+    parser_backend         VARCHAR(128) NOT NULL DEFAULT '',
+    chunk_backend          VARCHAR(128) NOT NULL DEFAULT '',
+    embedding_model        VARCHAR(256) NOT NULL DEFAULT '',
+    index_name             VARCHAR(128) NOT NULL DEFAULT '',
+    started_at             TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    stored_at              TIMESTAMPTZ,
+    parsed_at              TIMESTAMPTZ,
+    indexed_at             TIMESTAMPTZ,
+    finished_at            TIMESTAMPTZ,
+    layout_count           INTEGER NOT NULL DEFAULT 0,
+    structure_node_count   INTEGER NOT NULL DEFAULT 0,
+    semantic_block_count   INTEGER NOT NULL DEFAULT 0,
+    vector_chunk_count     INTEGER NOT NULL DEFAULT 0,
+    chunk_count            INTEGER NOT NULL DEFAULT 0,
+    failure_stage          VARCHAR(64) NOT NULL DEFAULT '',
+    error_message          TEXT NOT NULL DEFAULT '',
+    metadata               JSONB NOT NULL DEFAULT '{}',
+    created_at             TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    updated_at             TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_dpr_doc FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE
+);
+CREATE INDEX IF NOT EXISTS idx_document_processing_runs_doc_id ON document_processing_runs(doc_id, started_at DESC);
+"""
+
+_CREATE_EVENTS_TABLE = """
+CREATE TABLE IF NOT EXISTS document_status_history (
+    event_id               VARCHAR(128) PRIMARY KEY,
+    doc_id                 VARCHAR(128) NOT NULL,
+    run_id                 VARCHAR(128) NOT NULL,
+    from_status            VARCHAR(32) NOT NULL DEFAULT '',
+    to_status              VARCHAR(32) NOT NULL,
+    stage                  VARCHAR(64) NOT NULL DEFAULT '',
+    message                TEXT NOT NULL DEFAULT '',
+    metadata               JSONB NOT NULL DEFAULT '{}',
+    occurred_at            TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_dsh_doc FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE,
+    CONSTRAINT fk_dsh_run FOREIGN KEY (run_id) REFERENCES document_processing_runs(run_id) ON DELETE CASCADE
+);
+CREATE INDEX IF NOT EXISTS idx_document_status_history_doc_id ON document_status_history(doc_id, occurred_at ASC);
+CREATE INDEX IF NOT EXISTS idx_document_status_history_run_id ON document_status_history(run_id, occurred_at ASC);
+"""
+
+_CREATE_ARTIFACTS_TABLE = """
+CREATE TABLE IF NOT EXISTS document_artifacts (
+    artifact_id            VARCHAR(128) PRIMARY KEY,
+    doc_id                 VARCHAR(128) NOT NULL,
+    run_id                 VARCHAR(128) NOT NULL,
+    artifact_type          VARCHAR(64) NOT NULL,
+    object_name            VARCHAR(1024) NOT NULL,
+    content_type           VARCHAR(128) NOT NULL DEFAULT '',
+    byte_size              BIGINT NOT NULL DEFAULT 0,
+    checksum               VARCHAR(256) NOT NULL DEFAULT '',
+    metadata               JSONB NOT NULL DEFAULT '{}',
+    created_at             TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_da_doc FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE,
+    CONSTRAINT fk_da_run FOREIGN KEY (run_id) REFERENCES document_processing_runs(run_id) ON DELETE CASCADE
+);
+CREATE INDEX IF NOT EXISTS idx_document_artifacts_doc_id ON document_artifacts(doc_id, created_at ASC);
+CREATE INDEX IF NOT EXISTS idx_document_artifacts_run_id ON document_artifacts(run_id, created_at ASC);
+"""
+
+
+class PostgresDocumentProcessingStore(DocumentProcessingStore):
+    """Persist processing history in PostgreSQL using handwritten SQL."""
+
+    def __init__(self) -> None:
+        """Initialize the store and ensure the required tables exist."""
+        self._pool = ThreadedConnectionPool(
+            minconn=1,
+            maxconn=5,
+            host=settings.postgres_host,
+            port=settings.postgres_port,
+            user=settings.postgres_user,
+            password=settings.postgres_password,
+            dbname=settings.postgres_db,
+        )
+        self._ensure_schema()
+
+    def _ensure_schema(self) -> None:
+        """Create processing history tables and indexes if they are missing."""
+        with self._conn() as conn:
+            with conn.cursor() as cur:
+                cur.execute(_CREATE_RUNS_TABLE)
+                cur.execute(_CREATE_EVENTS_TABLE)
+                cur.execute(_CREATE_ARTIFACTS_TABLE)
+            conn.commit()
+
+    @contextmanager
+    def _conn(self):
+        """Borrow one connection from the pool and return it afterwards."""
+        conn = self._pool.getconn()
+        try:
+            yield conn
+        finally:
+            self._pool.putconn(conn)
+
+    def _normalize_metadata(self, value: Any) -> dict[str, Any]:
+        """Return a JSON-object payload regardless of the row representation."""
+        if isinstance(value, dict):
+            return value
+        if not value:
+            return {}
+        return json.loads(value)
+
+    def _row_to_run(self, row: dict[str, Any]) -> DocumentProcessingRun:
+        """Map one run row into the domain dataclass."""
+        return DocumentProcessingRun(
+            run_id=row["run_id"],
+            doc_id=row["doc_id"],
+            trigger_type=row["trigger_type"],
+            run_status=row["run_status"],
+            parser_backend=row["parser_backend"],
+            chunk_backend=row["chunk_backend"],
+            embedding_model=row["embedding_model"],
+            index_name=row["index_name"],
+            started_at=row["started_at"],
+            stored_at=row["stored_at"],
+            parsed_at=row["parsed_at"],
+            indexed_at=row["indexed_at"],
+            finished_at=row["finished_at"],
+            layout_count=row["layout_count"],
+            structure_node_count=row["structure_node_count"],
+            semantic_block_count=row["semantic_block_count"],
+            vector_chunk_count=row["vector_chunk_count"],
+            chunk_count=row["chunk_count"],
+            failure_stage=row["failure_stage"],
+            error_message=row["error_message"],
+            metadata=self._normalize_metadata(row["metadata"]),
+        )
+
+    def _row_to_event(self, row: dict[str, Any]) -> DocumentStatusEvent:
+        """Map one event row into the domain dataclass."""
+        return DocumentStatusEvent(
+            event_id=row["event_id"],
+            doc_id=row["doc_id"],
+            run_id=row["run_id"],
+            from_status=row["from_status"],
+            to_status=row["to_status"],
+            stage=row["stage"],
+            message=row["message"],
+            metadata=self._normalize_metadata(row["metadata"]),
+            occurred_at=row["occurred_at"],
+        )
+
+    def _row_to_artifact(self, row: dict[str, Any]) -> DocumentArtifact:
+        """Map one artifact row into the domain dataclass."""
+        return DocumentArtifact(
+            artifact_id=row["artifact_id"],
+            doc_id=row["doc_id"],
+            run_id=row["run_id"],
+            artifact_type=row["artifact_type"],
+            object_name=row["object_name"],
+            content_type=row["content_type"],
+            byte_size=row["byte_size"],
+            checksum=row["checksum"],
+            metadata=self._normalize_metadata(row["metadata"]),
+            created_at=row["created_at"],
+        )
+
+    def _update_run(
+        self,
+        run_id: str,
+        *,
+        assignments: dict[str, Any],
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Update one run row and return the latest stored state."""
+        set_clauses = []
+        params: dict[str, Any] = {"run_id": run_id, "updated_at": datetime.now(UTC)}
+        for key, value in assignments.items():
+            set_clauses.append(f"{key} = %({key})s")
+            params[key] = value
+        set_clauses.append("updated_at = %(updated_at)s")
+        if metadata is not None:
+            set_clauses.append("metadata = COALESCE(metadata, '{}'::jsonb) || %(metadata)s::jsonb")
+            params["metadata"] = json.dumps(metadata, ensure_ascii=False)
+        sql = f"""
+            UPDATE document_processing_runs
+            SET {", ".join(set_clauses)}
+            WHERE run_id = %(run_id)s
+            RETURNING *
+        """
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, params)
+                row = cur.fetchone()
+            conn.commit()
+        return self._row_to_run(dict(row)) if row else None
+
+    def create_run(self, run: DocumentProcessingRun) -> DocumentProcessingRun:
+        """Create a new processing run record."""
+        sql = """
+            INSERT INTO document_processing_runs
+                (run_id, doc_id, trigger_type, run_status, parser_backend, chunk_backend,
+                 embedding_model, index_name, started_at, stored_at, parsed_at, indexed_at,
+                 finished_at, layout_count, structure_node_count, semantic_block_count,
+                 vector_chunk_count, chunk_count, failure_stage, error_message, metadata)
+            VALUES
+                (%(run_id)s, %(doc_id)s, %(trigger_type)s, %(run_status)s, %(parser_backend)s,
+                 %(chunk_backend)s, %(embedding_model)s, %(index_name)s, %(started_at)s,
+                 %(stored_at)s, %(parsed_at)s, %(indexed_at)s, %(finished_at)s, %(layout_count)s,
+                 %(structure_node_count)s, %(semantic_block_count)s, %(vector_chunk_count)s,
+                 %(chunk_count)s, %(failure_stage)s, %(error_message)s, %(metadata)s)
+        """
+        with self._conn() as conn:
+            with conn.cursor() as cur:
+                cur.execute(
+                    sql,
+                    {
+                        "run_id": run.run_id,
+                        "doc_id": run.doc_id,
+                        "trigger_type": run.trigger_type,
+                        "run_status": run.run_status,
+                        "parser_backend": run.parser_backend,
+                        "chunk_backend": run.chunk_backend,
+                        "embedding_model": run.embedding_model,
+                        "index_name": run.index_name,
+                        "started_at": run.started_at,
+                        "stored_at": run.stored_at,
+                        "parsed_at": run.parsed_at,
+                        "indexed_at": run.indexed_at,
+                        "finished_at": run.finished_at,
+                        "layout_count": run.layout_count,
+                        "structure_node_count": run.structure_node_count,
+                        "semantic_block_count": run.semantic_block_count,
+                        "vector_chunk_count": run.vector_chunk_count,
+                        "chunk_count": run.chunk_count,
+                        "failure_stage": run.failure_stage,
+                        "error_message": run.error_message,
+                        "metadata": json.dumps(run.metadata, ensure_ascii=False),
+                    },
+                )
+            conn.commit()
+        return run
+
+    def mark_run_stored(
+        self,
+        run_id: str,
+        *,
+        stored_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as having persisted its source file."""
+        return self._update_run(
+            run_id,
+            assignments={"stored_at": stored_at or datetime.now(UTC)},
+            metadata=metadata,
+        )
+
+    def mark_run_parsed(
+        self,
+        run_id: str,
+        *,
+        parser_backend: str,
+        layout_count: int,
+        structure_node_count: int,
+        semantic_block_count: int,
+        vector_chunk_count: int,
+        parsed_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Record parse completion metrics for a run."""
+        return self._update_run(
+            run_id,
+            assignments={
+                "parser_backend": parser_backend,
+                "parsed_at": parsed_at or datetime.now(UTC),
+                "layout_count": layout_count,
+                "structure_node_count": structure_node_count,
+                "semantic_block_count": semantic_block_count,
+                "vector_chunk_count": vector_chunk_count,
+            },
+            metadata=metadata,
+        )
+
+    def mark_run_indexed(
+        self,
+        run_id: str,
+        *,
+        chunk_count: int,
+        index_name: str,
+        indexed_at: datetime | None = None,
+        finished_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as successfully indexed."""
+        now = datetime.now(UTC)
+        return self._update_run(
+            run_id,
+            assignments={
+                "run_status": "succeeded",
+                "chunk_count": chunk_count,
+                "index_name": index_name,
+                "indexed_at": indexed_at or now,
+                "finished_at": finished_at or now,
+            },
+            metadata=metadata,
+        )
+
+    def mark_run_failed(
+        self,
+        run_id: str,
+        *,
+        failure_stage: str,
+        error_message: str,
+        finished_at: datetime | None = None,
+        metadata: dict | None = None,
+    ) -> DocumentProcessingRun | None:
+        """Mark a run as failed and persist the terminal error details."""
+        return self._update_run(
+            run_id,
+            assignments={
+                "run_status": "failed",
+                "failure_stage": failure_stage,
+                "error_message": error_message,
+                "finished_at": finished_at or datetime.now(UTC),
+            },
+            metadata=metadata,
+        )
+
+    def append_status_event(self, event: DocumentStatusEvent) -> DocumentStatusEvent:
+        """Append a document status event."""
+        sql = """
+            INSERT INTO document_status_history
+                (event_id, doc_id, run_id, from_status, to_status, stage, message, metadata, occurred_at)
+            VALUES
+                (%(event_id)s, %(doc_id)s, %(run_id)s, %(from_status)s, %(to_status)s,
+                 %(stage)s, %(message)s, %(metadata)s, %(occurred_at)s)
+        """
+        with self._conn() as conn:
+            with conn.cursor() as cur:
+                cur.execute(
+                    sql,
+                    {
+                        "event_id": event.event_id,
+                        "doc_id": event.doc_id,
+                        "run_id": event.run_id,
+                        "from_status": event.from_status,
+                        "to_status": event.to_status,
+                        "stage": event.stage,
+                        "message": event.message,
+                        "metadata": json.dumps(event.metadata, ensure_ascii=False),
+                        "occurred_at": event.occurred_at,
+                    },
+                )
+            conn.commit()
+        return event
+
+    def replace_artifacts_for_run(self, run_id: str, artifacts: list[DocumentArtifact]) -> list[DocumentArtifact]:
+        """Replace all artifact references for one run using a delete-then-insert strategy."""
+        with self._conn() as conn:
+            with conn.cursor() as cur:
+                cur.execute("DELETE FROM document_artifacts WHERE run_id = %s", (run_id,))
+                if artifacts:
+                    psycopg2.extras.execute_values(
+                        cur,
+                        """
+                        INSERT INTO document_artifacts
+                            (artifact_id, doc_id, run_id, artifact_type, object_name,
+                             content_type, byte_size, checksum, metadata, created_at)
+                        VALUES %s
+                        """,
+                        [
+                            (
+                                artifact.artifact_id,
+                                artifact.doc_id,
+                                artifact.run_id,
+                                artifact.artifact_type,
+                                artifact.object_name,
+                                artifact.content_type,
+                                artifact.byte_size,
+                                artifact.checksum,
+                                json.dumps(artifact.metadata, ensure_ascii=False),
+                                artifact.created_at,
+                            )
+                            for artifact in artifacts
+                        ],
+                    )
+            conn.commit()
+        return artifacts
+
+    def delete_by_document(self, doc_id: str) -> None:
+        """Delete all processing rows for a document explicitly."""
+        with self._conn() as conn:
+            with conn.cursor() as cur:
+                cur.execute("DELETE FROM document_status_history WHERE doc_id = %s", (doc_id,))
+                cur.execute("DELETE FROM document_artifacts WHERE doc_id = %s", (doc_id,))
+                cur.execute("DELETE FROM document_processing_runs WHERE doc_id = %s", (doc_id,))
+            conn.commit()
+
+    def list_runs_by_document(self, doc_id: str) -> list[DocumentProcessingRun]:
+        """List processing runs for a document in chronological order."""
+        sql = "SELECT * FROM document_processing_runs WHERE doc_id = %s ORDER BY started_at ASC"
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, (doc_id,))
+                rows = cur.fetchall()
+        return [self._row_to_run(dict(row)) for row in rows]
+
+    def get_run(self, run_id: str) -> DocumentProcessingRun | None:
+        """Return one processing run by identifier."""
+        sql = "SELECT * FROM document_processing_runs WHERE run_id = %s"
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, (run_id,))
+                row = cur.fetchone()
+        return self._row_to_run(dict(row)) if row else None
+
+    def list_status_events_by_document(self, doc_id: str) -> list[DocumentStatusEvent]:
+        """List all status events for a document."""
+        sql = "SELECT * FROM document_status_history WHERE doc_id = %s ORDER BY occurred_at ASC"
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, (doc_id,))
+                rows = cur.fetchall()
+        return [self._row_to_event(dict(row)) for row in rows]
+
+    def list_status_events_by_run(self, run_id: str) -> list[DocumentStatusEvent]:
+        """List all status events for a run."""
+        sql = "SELECT * FROM document_status_history WHERE run_id = %s ORDER BY occurred_at ASC"
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, (run_id,))
+                rows = cur.fetchall()
+        return [self._row_to_event(dict(row)) for row in rows]
+
+    def list_artifacts_by_document(self, doc_id: str) -> list[DocumentArtifact]:
+        """List all artifact references for a document."""
+        sql = "SELECT * FROM document_artifacts WHERE doc_id = %s ORDER BY created_at ASC"
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, (doc_id,))
+                rows = cur.fetchall()
+        return [self._row_to_artifact(dict(row)) for row in rows]
+
+    def list_artifacts_by_run(self, run_id: str) -> list[DocumentArtifact]:
+        """List all artifact references for a run."""
+        sql = "SELECT * FROM document_artifacts WHERE run_id = %s ORDER BY created_at ASC"
+        with self._conn() as conn:
+            with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
+                cur.execute(sql, (run_id,))
+                rows = cur.fetchall()
+        return [self._row_to_artifact(dict(row)) for row in rows]

backend/app/infrastructure/vectorstore/bm25_retriever.py

@@ -56,7 +56,21 @@ class BM25Retriever:
         try:
             rows = self._vector_index.collection.query(
                 expr='doc_id != ""',
-                output_fields=["id", "doc_id", "doc_name", "content", "section_title", "page_number"],
+                output_fields=[
+                    "id",
+                    "chunk_id",
+                    "doc_id",
+                    "doc_title",
+                    "text",
+                    "chunk_type",
+                    "section_title",
+                    "page_start",
+                    "page_end",
+                    "section_level",
+                    "chunk_index",
+                    "piece_index",
+                    "metadata_json",
+                ],
                 limit=16384,
             )
         except Exception:
@@ -64,19 +78,33 @@ class BM25Retriever:
             return []
         return [
             RetrievedChunk(
-                chunk_id=str(row.get("id", "")),
+                chunk_id=str(row.get("chunk_id") or row.get("id", "")),
                 doc_id=str(row.get("doc_id", "")),
-                doc_name=str(row.get("doc_name", "")),
-                content=str(row.get("content", "")),
+                doc_title=str(row.get("doc_title", "")),
+                text=str(row.get("text", "")),
                 score=0.0,
+                chunk_type=str(row.get("chunk_type", "")),
                 section_title=str(row.get("section_title", "")),
-                page_number=int(row.get("page_number") or 0),
-                metadata={},
+                page_start=int(row.get("page_start") or 0),
+                page_end=int(row.get("page_end") or 0),
+                section_level=int(row.get("section_level") or 0),
+                chunk_index=int(row.get("chunk_index") or 0),
+                piece_index=int(row.get("piece_index") or 0),
+                metadata=self._parse_metadata_json(row.get("metadata_json", "")),
             )
             for row in rows
-            if row.get("content")
+            if row.get("text")
         ]
 
+    def _parse_metadata_json(self, raw_metadata: str) -> dict:
+        """Parse metadata_json into a dict for BM25-side filtering."""
+        if not raw_metadata:
+            return {}
+        try:
+            return dict(__import__("json").loads(raw_metadata))
+        except Exception:
+            return {}
+
     def _ensure_built(self) -> None:
         if self._index is not None:
             return
@@ -93,7 +121,7 @@ class BM25Retriever:
             self._chunks = []
             self._index = BM25Okapi([[]])
             return
-        tokenized = [_tokenize(c.content) for c in chunks]
+        tokenized = [_tokenize(c.text) for c in chunks]
         self._chunks = chunks
         self._index = BM25Okapi(tokenized)
         logger.info("BM25Retriever: index built with %d chunks", len(chunks))
@@ -127,20 +155,26 @@ class BM25Retriever:
         for score, chunk in ranked[: top_k * 2]:
             if score <= 0:
                 break
-            # Apply simple regulation_type filter if provided
-            if filters and chunk.metadata.get("regulation_type"):
-                types = [t.strip() for t in filters.split(",")]
-                if chunk.metadata.get("regulation_type") not in types:
-                    continue
+            if filters:
+                normalized_filter = filters.replace("doc_name", "doc_title").strip()
+                if normalized_filter.startswith('doc_title == "'):
+                    expected_title = normalized_filter[len('doc_title == "'):-1]
+                    if chunk.doc_title != expected_title:
+                        continue
             results.append(
                 RetrievedChunk(
                     chunk_id=chunk.chunk_id,
                     doc_id=chunk.doc_id,
-                    doc_name=chunk.doc_name,
-                    content=chunk.content,
+                    doc_title=chunk.doc_title,
+                    text=chunk.text,
                     score=score,
+                    chunk_type=chunk.chunk_type,
                     section_title=chunk.section_title,
-                    page_number=chunk.page_number,
+                    page_start=chunk.page_start,
+                    page_end=chunk.page_end,
+                    section_level=chunk.section_level,
+                    chunk_index=chunk.chunk_index,
+                    piece_index=chunk.piece_index,
                     metadata=chunk.metadata,
                 )
             )

backend/app/infrastructure/vectorstore/cross_encoder_reranker.py

@@ -31,7 +31,7 @@ class OpenAICompatibleReranker(Reranker):
         if not chunks:
             return []
 
-        texts = [chunk.content for chunk in chunks]
+        texts = [chunk.text for chunk in chunks]
         start = time.time()
         try:
             scores = self._call_reranker(query, texts)

backend/app/infrastructure/vectorstore/milvus_vector_index.py

@@ -4,57 +4,150 @@ from __future__ import annotations
 
 import json
 import time
+from typing import Iterable
 
-from pymilvus import Collection, CollectionSchema, DataType, FieldSchema, connections, utility
+from loguru import logger
+from pymilvus import Collection, CollectionSchema, DataType, FieldSchema, MilvusException, connections, utility
 
 from app.config.settings import settings
 from app.domain.documents import Chunk
 from app.domain.retrieval import RetrievedChunk, VectorIndex
+from app.shared.errors import VectorStoreSchemaError
 # Keep adapter behavior explicit so integration details remain easy to audit.
 
 
+_REQUIRED_SCHEMA_FIELDS = (
+    "doc_id",
+    "doc_title",
+    "chunk_id",
+    "text",
+    "embedding",
+    "section_title",
+    "metadata_json",
+)
+_SCHEMA_RECOVERY_TOKENS = (
+    "field doc_title not exist",
+    "field text not exist",
+    "field embedding not exist",
+    "collection not loaded",
+    "can't find collection",
+    "not found[collection",
+)
+
+
 
 class MilvusVectorIndex(VectorIndex):
     """Provide the Milvus Vector Index index implementation."""
+
     def __init__(self) -> None:
         """Initialize the Milvus Vector Index instance."""
         self.collection_name = settings.milvus_collection
         self.db_name = settings.milvus_db_name
+        self.host = settings.milvus_host
+        self.port = settings.milvus_port
+        # Use an adapter-specific alias so this index never reuses unrelated global Milvus state.
+        self.alias = f"vector-index::{self.host}:{self.port}/{self.db_name}/{self.collection_name}"
+        self._connect()
+        self.collection = self._bind_collection()
+
+    def _connect(self, *, refresh: bool = False) -> None:
+        """Establish the Milvus connection for this adapter."""
+        if refresh:
+            try:
+                connections.disconnect(self.alias)
+            except Exception:
+                # Best-effort disconnect keeps refresh idempotent when no alias is active yet.
+                pass
         connections.connect(
-            alias="default",
-            host=settings.milvus_host,
-            port=settings.milvus_port,
+            alias=self.alias,
+            host=self.host,
+            port=self.port,
             db_name=self.db_name,
         )
-        self.collection = self._ensure_collection()
+
+    def _schema_field_names(self, collection: Collection) -> list[str]:
+        """Return the field names exposed by the bound Milvus collection."""
+        return [field.name for field in collection.schema.fields]
+
+    def _raise_schema_error(self, *, message: str, actual_fields: Iterable[str]) -> None:
+        """Raise a typed schema error for the active collection."""
+        raise VectorStoreSchemaError(
+            message=message,
+            host=self.host,
+            db_name=self.db_name,
+            collection_name=self.collection_name,
+            expected_fields=list(_REQUIRED_SCHEMA_FIELDS),
+            actual_fields=list(actual_fields),
+        )
+
+    def _validate_schema(self, collection: Collection) -> None:
+        """Ensure the collection schema matches the dense-only adapter contract."""
+        actual_fields = self._schema_field_names(collection)
+        missing_fields = [field_name for field_name in _REQUIRED_SCHEMA_FIELDS if field_name not in actual_fields]
+        if missing_fields:
+            self._raise_schema_error(
+                message=f"Milvus collection schema mismatch; missing required fields: {missing_fields}",
+                actual_fields=actual_fields,
+            )
+
+    def _log_collection_binding(self, collection: Collection, *, event: str) -> None:
+        """Record the bound collection details for runtime diagnostics."""
+        try:
+            num_entities = collection.num_entities
+        except Exception:
+            num_entities = "unknown"
+        logger.info(
+            "Milvus binding {} alias={} host={} db={} collection={} fields={} num_entities={}",
+            event,
+            self.alias,
+            self.host,
+            self.db_name,
+            self.collection_name,
+            self._schema_field_names(collection),
+            num_entities,
+        )
+
+    def _bind_collection(self, *, force_refresh: bool = False) -> Collection:
+        """Bind and validate the configured Milvus collection."""
+        if force_refresh:
+            self._connect(refresh=True)
+        collection = self._ensure_collection()
+        self._validate_schema(collection)
+        self._log_collection_binding(collection, event="refreshed" if force_refresh else "initialized")
+        return collection
 
     def _ensure_collection(self) -> Collection:
         """Handle ensure collection for this module for the Milvus Vector Index instance."""
-        if utility.has_collection(self.collection_name):
-            collection = Collection(self.collection_name)
+        if utility.has_collection(self.collection_name, using=self.alias):
+            collection = Collection(self.collection_name, using=self.alias)
             collection.load()
             return collection
         schema = CollectionSchema(
             fields=[
                 FieldSchema(name="id", dtype=DataType.VARCHAR, max_length=128, is_primary=True, auto_id=False),
                 FieldSchema(name="doc_id", dtype=DataType.VARCHAR, max_length=64),
-                FieldSchema(name="doc_name", dtype=DataType.VARCHAR, max_length=256),
-                FieldSchema(name="content", dtype=DataType.VARCHAR, max_length=65535),
+                FieldSchema(name="doc_title", dtype=DataType.VARCHAR, max_length=256),
+                FieldSchema(name="chunk_id", dtype=DataType.VARCHAR, max_length=128),
+                FieldSchema(name="chunk_index", dtype=DataType.INT64),
+                FieldSchema(name="piece_index", dtype=DataType.INT64),
+                FieldSchema(name="text", dtype=DataType.VARCHAR, max_length=65535),
+                FieldSchema(name="embedding_text", dtype=DataType.VARCHAR, max_length=65535),
                 FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=settings.embedding_dim),
-                FieldSchema(name="section_title", dtype=DataType.VARCHAR, max_length=512),
-                FieldSchema(name="section_path", dtype=DataType.VARCHAR, max_length=4096),
-                FieldSchema(name="page_number", dtype=DataType.INT64),
-                FieldSchema(name="regulation_type", dtype=DataType.VARCHAR, max_length=128),
-                FieldSchema(name="version", dtype=DataType.VARCHAR, max_length=64),
                 FieldSchema(name="semantic_id", dtype=DataType.VARCHAR, max_length=128),
-                FieldSchema(name="block_type", dtype=DataType.VARCHAR, max_length=64),
+                FieldSchema(name="chunk_type", dtype=DataType.VARCHAR, max_length=64),
+                FieldSchema(name="page_start", dtype=DataType.INT64),
+                FieldSchema(name="page_end", dtype=DataType.INT64),
+                FieldSchema(name="section_level", dtype=DataType.INT64),
+                FieldSchema(name="source_ids", dtype=DataType.VARCHAR, max_length=4096),
+                FieldSchema(name="section_path", dtype=DataType.VARCHAR, max_length=4096),
+                FieldSchema(name="section_title", dtype=DataType.VARCHAR, max_length=512),
                 FieldSchema(name="metadata_json", dtype=DataType.VARCHAR, max_length=65535),
                 FieldSchema(name="created_at", dtype=DataType.INT64),
             ],
             description="Dense-only regulations index",
             enable_dynamic_field=False,
         )
-        collection = Collection(name=self.collection_name, schema=schema)
+        collection = Collection(name=self.collection_name, schema=schema, using=self.alias)
         collection.create_index(
             field_name="embedding",
             index_params={
@@ -73,21 +166,34 @@ class MilvusVectorIndex(VectorIndex):
         data = []
         now = int(time.time())
         for chunk, vector in zip(chunks, vectors):
+            metadata = dict(chunk.metadata)
+            doc_title = str(metadata.get("doc_title", chunk.doc_title))
+            text = str(metadata.get("text", chunk.text))
+            embedding_text = str(metadata.get("embedding_text", chunk.embedding_text))
+            page_start = int(metadata.get("page_start", 0) or 0)
+            page_end = int(metadata.get("page_end", 0) or 0)
+            section_path = metadata.get("section_path", chunk.section_path)
+            source_ids = metadata.get("source_ids", [])
             data.append(
                 {
                     "id": chunk.chunk_id,
                     "doc_id": chunk.doc_id,
-                    "doc_name": chunk.doc_name,
-                    "content": chunk.content[:65535],
+                    "doc_title": doc_title[:256],
+                    "chunk_id": chunk.chunk_id[:128],
+                    "chunk_index": int(metadata.get("chunk_index", chunk.chunk_index) or 0),
+                    "piece_index": int(metadata.get("piece_index", chunk.piece_index) or 0),
+                    "text": text[:65535],
+                    "embedding_text": embedding_text[:65535],
                     "embedding": vector,
-                    "section_title": chunk.section_title[:512],
-                    "section_path": json.dumps(chunk.section_path, ensure_ascii=False)[:4096],
-                    "page_number": chunk.page_number,
-                    "regulation_type": chunk.regulation_type[:128],
-                    "version": chunk.version[:64],
-                    "semantic_id": chunk.semantic_id[:128],
-                    "block_type": chunk.block_type[:64],
-                    "metadata_json": json.dumps(chunk.metadata, ensure_ascii=False)[:65535],
+                    "semantic_id": str(metadata.get("semantic_id", chunk.semantic_id))[:128],
+                    "chunk_type": str(metadata.get("chunk_type", chunk.chunk_type))[:64],
+                    "page_start": page_start,
+                    "page_end": page_end,
+                    "section_level": int(metadata.get("section_level", chunk.section_level) or 0),
+                    "source_ids": json.dumps(source_ids, ensure_ascii=False)[:4096],
+                    "section_path": json.dumps(section_path, ensure_ascii=False)[:4096],
+                    "section_title": str(metadata.get("section_title", chunk.section_title))[:512],
+                    "metadata_json": json.dumps(metadata, ensure_ascii=False)[:65535],
                     "created_at": now,
                 }
             )
@@ -107,47 +213,97 @@ class MilvusVectorIndex(VectorIndex):
 
         filters = filters.strip()
 
+        # Normalize legacy field names so callers can keep older filter payloads.
+        replacements = {
+            "doc_name": "doc_title",
+            "content": "text",
+            "page_number": "page_start",
+            "block_type": "chunk_type",
+        }
+        for legacy_name, new_name in replacements.items():
+            filters = filters.replace(legacy_name, new_name)
+
         # Check if already a Milvus expression (contains operators)
         if any(op in filters for op in ["==", "!=", "in", "not in", ">", "<", ">=", "<=", "and", "or"]):
             return filters
 
-        # Parse simple regulation_type filter
-        # Support: "GB" or "GB,UN-ECE" or "GB, UN-ECE"
-        types = [t.strip() for t in filters.split(",") if t.strip()]
+        # Parse simple document-title filter.
+        titles = [title.strip() for title in filters.split(",") if title.strip()]
 
-        if not types:
+        if not titles:
             return None
 
-        if len(types) == 1:
-            # Single value: regulation_type == "GB"
-            return f'regulation_type == "{types[0]}"'
-        else:
-            # Multiple values: regulation_type in ["GB", "UN-ECE"]
-            quoted_types = [f'"{t}"' for t in types]
-            return f'regulation_type in [{", ".join(quoted_types)}]'
+        if len(titles) == 1:
+            return f'doc_title == "{titles[0]}"'
+
+        quoted_titles = [f'"{title}"' for title in titles]
+        return f'doc_title in [{", ".join(quoted_titles)}]'
+
+    def _should_refresh_after_exception(self, exc: Exception) -> bool:
+        """Return whether the Milvus error suggests stale connection or collection state."""
+        if not isinstance(exc, MilvusException):
+            return False
+        normalized = str(exc).lower()
+        return any(token in normalized for token in _SCHEMA_RECOVERY_TOKENS)
+
+    def _run_with_refresh(self, operation):
+        """Run a Milvus operation and retry once after a forced reconnect when appropriate."""
+        try:
+            return operation()
+        except VectorStoreSchemaError:
+            raise
+        except Exception as exc:
+            if not self._should_refresh_after_exception(exc):
+                raise
+            logger.warning(
+                "Milvus operation failed for alias={} collection={}; forcing reconnect and retry: {}",
+                self.alias,
+                self.collection_name,
+                exc,
+            )
+            self.collection = self._bind_collection(force_refresh=True)
+            try:
+                return operation()
+            except VectorStoreSchemaError:
+                raise
+            except Exception as retry_exc:
+                if isinstance(retry_exc, MilvusException):
+                    self._raise_schema_error(
+                        message=f"Milvus operation failed after refresh: {retry_exc}",
+                        actual_fields=self._schema_field_names(self.collection),
+                    )
+                raise
 
     def search(self, query_vector: list[float], top_k: int, filters: str | None = None) -> list[RetrievedChunk]:
         """Handle search for the Milvus Vector Index instance."""
         milvus_expr = self._parse_filters(filters)
 
-        results = self.collection.search(
-            data=[query_vector],
-            anns_field="embedding",
-            param={"metric_type": "COSINE", "params": {"nprobe": settings.milvus_nprobe}},
-            limit=top_k,
-            expr=milvus_expr,
-            output_fields=[
-                "doc_id",
-                "doc_name",
-                "content",
-                "section_title",
-                "page_number",
-                "regulation_type",
-                "version",
-                "semantic_id",
-                "block_type",
-                "metadata_json",
-            ],
+        results = self._run_with_refresh(
+            lambda: self.collection.search(
+                data=[query_vector],
+                anns_field="embedding",
+                param={"metric_type": "COSINE", "params": {"nprobe": settings.milvus_nprobe}},
+                limit=top_k,
+                expr=milvus_expr,
+                output_fields=[
+                    "doc_id",
+                    "doc_title",
+                    "chunk_id",
+                    "chunk_index",
+                    "piece_index",
+                    "text",
+                    "embedding_text",
+                    "section_title",
+                    "semantic_id",
+                    "chunk_type",
+                    "page_start",
+                    "page_end",
+                    "section_level",
+                    "source_ids",
+                    "section_path",
+                    "metadata_json",
+                ],
+            )
         )
         payload: list[RetrievedChunk] = []
         for hits in results:
@@ -161,13 +317,18 @@ class MilvusVectorIndex(VectorIndex):
                         metadata = {"raw_metadata": raw_metadata}
                 payload.append(
                     RetrievedChunk(
-                        chunk_id=str(hit.id),
+                        chunk_id=str(hit.entity.get("chunk_id", hit.id)),
                         doc_id=hit.entity.get("doc_id", ""),
-                        doc_name=hit.entity.get("doc_name", ""),
-                        content=hit.entity.get("content", ""),
+                        doc_title=hit.entity.get("doc_title", ""),
+                        text=hit.entity.get("text", ""),
                         score=float(hit.score),
+                        chunk_type=hit.entity.get("chunk_type", ""),
                         section_title=hit.entity.get("section_title", ""),
-                        page_number=int(hit.entity.get("page_number", 0) or 0),
+                        page_start=int(hit.entity.get("page_start", 0) or 0),
+                        page_end=int(hit.entity.get("page_end", 0) or 0),
+                        section_level=int(hit.entity.get("section_level", 0) or 0),
+                        chunk_index=int(hit.entity.get("chunk_index", 0) or 0),
+                        piece_index=int(hit.entity.get("piece_index", 0) or 0),
                         metadata=metadata,
                     )
                 )
@@ -176,7 +337,9 @@ class MilvusVectorIndex(VectorIndex):
     def count_by_document(self) -> dict[str, int]:
         """Return doc_id -> chunk count from Milvus."""
         try:
-            rows = self.collection.query(expr="doc_id != \"\"", output_fields=["doc_id"])
+            rows = self._run_with_refresh(
+                lambda: self.collection.query(expr="doc_id != \"\"", output_fields=["doc_id", "doc_title"])
+            )
         except Exception:
             return {}
         counts: dict[str, int] = {}
@@ -189,9 +352,11 @@ class MilvusVectorIndex(VectorIndex):
     def list_document_metadata(self) -> list[dict]:
         """Return one metadata row per document from Milvus (single query, no embeddings)."""
         try:
-            rows = self.collection.query(
-                expr="doc_id != \"\"",
-                output_fields=["doc_id", "doc_name", "regulation_type", "version"],
+            rows = self._run_with_refresh(
+                lambda: self.collection.query(
+                    expr="doc_id != \"\"",
+                    output_fields=["doc_id", "doc_title", "metadata_json"],
+                )
             )
         except Exception:
             return []
@@ -204,15 +369,26 @@ class MilvusVectorIndex(VectorIndex):
                 continue
             counts[doc_id] = counts.get(doc_id, 0) + 1
             if doc_id not in seen:
+                metadata: dict[str, object] = {}
+                raw_metadata = row.get("metadata_json", "")
+                if raw_metadata:
+                    try:
+                        metadata = json.loads(raw_metadata)
+                    except json.JSONDecodeError:
+                        metadata = {}
                 seen[doc_id] = {
                     "doc_id": doc_id,
-                    "doc_name": row.get("doc_name", ""),
-                    "regulation_type": row.get("regulation_type", ""),
-                    "version": row.get("version", ""),
+                    "doc_title": row.get("doc_title", ""),
+                    "regulation_type": str(metadata.get("regulation_type", "")),
+                    "version": str(metadata.get("version", "")),
                 }
 
         return [
-            {**meta, "chunk_count": counts[meta["doc_id"]]}
+            {
+                **meta,
+                "doc_name": meta.get("doc_title", ""),
+                "chunk_count": counts[meta["doc_id"]],
+            }
             for meta in seen.values()
         ]
 

backend/app/services/document_processor.py

@@ -67,14 +67,14 @@ class DocumentProcessor:
         return [
             {
                 "id": item.chunk_id,
-                "content": item.content,
+                "content": item.text,
                 "score": item.score,
                 "metadata": {
                     "doc_id": item.doc_id,
-                    "doc_name": item.doc_name,
+                    "doc_name": item.doc_title,
                     "chunk_id": item.chunk_id,
                     "section_title": item.section_title,
-                    "page_number": item.page_number,
+                    "page_number": item.page_start,
                     **item.metadata,
                 },
             }

backend/app/shared/bootstrap.py

@@ -3,29 +3,136 @@
 from __future__ import annotations
 
 from functools import lru_cache
+from typing import Callable
 
-from app.application.agent import AgentConversationService
+from app.application.agent import AgentConversationService, AgentSessionService
 from app.application.documents import DocumentCommandService, DocumentQueryService
 from app.application.knowledge import KnowledgeRetrievalService
+from app.application.perception.services import PerceptionService
 from app.config.settings import settings
+from app.domain.documents import DocumentBinaryStore
+from app.domain.retrieval import VectorIndex
 from app.infrastructure.embedding.openai_compatible_embedding_provider import OpenAICompatibleEmbeddingProvider
 from app.infrastructure.llm.openai_compatible_answer_generator import OpenAICompatibleAnswerGenerator
 from app.infrastructure.parser.aliyun_document_parser import AliyunDocumentParser
 from app.infrastructure.parser.local_chunk_builder import LocalRegulationChunkBuilder
 from app.infrastructure.parser.local_document_parser import LocalDocumentParser
 from app.infrastructure.parser.vector_chunk_builder import AliyunVectorChunkBuilder
+from app.infrastructure.perception.mock_event_store import MockEventStore
 from app.infrastructure.session.in_memory_conversation_store import InMemoryConversationStore
+from app.infrastructure.storage.json_document_processing_store import JsonDocumentProcessingStore
 from app.infrastructure.storage.json_document_repository import JsonDocumentRepository
 from app.infrastructure.storage.minio_binary_store import MinioDocumentBinaryStore
+from app.infrastructure.storage.postgres_document_processing_store import PostgresDocumentProcessingStore
 from app.infrastructure.storage.postgres_document_repository import PostgresDocumentRepository
 from app.infrastructure.storage.postgres_parse_artifact_store import PostgresParseArtifactStore
 from app.infrastructure.vectorstore.bm25_retriever import BM25Retriever
+from app.infrastructure.vectorstore.cross_encoder_reranker import OpenAICompatibleReranker
 from app.infrastructure.vectorstore.dense_retriever import DenseRetriever
 from app.infrastructure.vectorstore.milvus_vector_index import MilvusVectorIndex
-from app.infrastructure.vectorstore.cross_encoder_reranker import OpenAICompatibleReranker
+from app.services.llm.llm_factory import LLMFactory
 # Keep shared wiring centralized so dependency construction remains consistent.
 
 
+class LazyBinaryStore(DocumentBinaryStore):
+    """Delay MinIO connection work until binary storage is actually needed."""
+
+    def __init__(self, factory: Callable[[], DocumentBinaryStore]) -> None:
+        """Initialize the lazy binary store wrapper."""
+        self._factory = factory
+        self._store: DocumentBinaryStore | None = None
+
+    def _get_store(self) -> DocumentBinaryStore:
+        """Create the underlying store on first use and reuse it afterwards."""
+        if self._store is None:
+            self._store = self._factory()
+        return self._store
+
+    @property
+    def client(self):
+        """Expose the underlying client for compatibility with health endpoints."""
+        return self._get_store().client
+
+    def save(
+        self,
+        *,
+        object_name: str,
+        data: bytes,
+        content_type: str,
+        metadata: dict[str, str] | None = None,
+    ) -> None:
+        """Save data through the underlying binary store implementation."""
+        self._get_store().save(
+            object_name=object_name,
+            data=data,
+            content_type=content_type,
+            metadata=metadata,
+        )
+
+    def read(self, object_name: str) -> bytes:
+        """Read data through the underlying binary store implementation."""
+        return self._get_store().read(object_name)
+
+    def delete(self, object_name: str) -> None:
+        """Delete data through the underlying binary store implementation."""
+        self._get_store().delete(object_name)
+
+
+class LazyVectorIndex(VectorIndex):
+    """Delay Milvus connection work until vector operations are actually needed."""
+
+    def __init__(self, factory: Callable[[], VectorIndex]) -> None:
+        """Initialize the lazy vector index wrapper."""
+        self._factory = factory
+        self._index: VectorIndex | None = None
+
+    def _get_index(self) -> VectorIndex:
+        """Create the underlying index on first use and reuse it afterwards."""
+        if self._index is None:
+            self._index = self._factory()
+        return self._index
+
+    @property
+    def collection(self):
+        """Expose the underlying Milvus collection for compatibility adapters."""
+        return self._get_index().collection
+
+    def upsert(self, chunks, vectors) -> int:
+        """Insert or update vectors through the underlying vector index implementation."""
+        return self._get_index().upsert(chunks, vectors)
+
+    def delete_by_document(self, doc_id: str) -> int:
+        """Delete vectors through the underlying vector index implementation."""
+        return self._get_index().delete_by_document(doc_id)
+
+    def search(self, query_vector: list[float], top_k: int, filters: str | None = None):
+        """Search vectors through the underlying vector index implementation."""
+        return self._get_index().search(query_vector, top_k, filters)
+
+    def count_by_document(self) -> dict[str, int]:
+        """Count document vectors through the underlying vector index implementation."""
+        return self._get_index().count_by_document()
+
+    def list_document_metadata(self) -> list[dict]:
+        """List document metadata through the underlying vector index implementation."""
+        return self._get_index().list_document_metadata()
+
+    def health(self) -> dict:
+        """Return vector index health through the underlying vector index implementation."""
+        return self._get_index().health()
+
+
+@lru_cache
+def _build_binary_store() -> MinioDocumentBinaryStore:
+    """Return the concrete binary store implementation."""
+    return MinioDocumentBinaryStore()
+
+
+@lru_cache
+def _build_vector_index() -> MilvusVectorIndex:
+    """Return the concrete vector index implementation."""
+    return MilvusVectorIndex()
+
 
 @lru_cache
 def get_document_repository():
@@ -44,9 +151,17 @@ def get_parse_artifact_store():
 
 
 @lru_cache
-def get_binary_store() -> MinioDocumentBinaryStore:
+def get_document_processing_store():
+    """Return document processing store for the active repository backend."""
+    if settings.document_repository_backend == "postgres":
+        return PostgresDocumentProcessingStore()
+    return JsonDocumentProcessingStore(settings.document_processing_metadata_path)
+
+
+@lru_cache
+def get_binary_store() -> DocumentBinaryStore:
     """Return binary store."""
-    return MinioDocumentBinaryStore()
+    return LazyBinaryStore(_build_binary_store)
 
 
 @lru_cache
@@ -75,9 +190,9 @@ def get_embedding_provider() -> OpenAICompatibleEmbeddingProvider:
 
 
 @lru_cache
-def get_vector_index() -> MilvusVectorIndex:
+def get_vector_index() -> VectorIndex:
     """Return vector index."""
-    return MilvusVectorIndex()
+    return LazyVectorIndex(_build_vector_index)
 
 
 @lru_cache
@@ -121,6 +236,7 @@ def get_document_command_service() -> DocumentCommandService:
         embedding_provider=get_embedding_provider(),
         vector_index=get_vector_index(),
         parse_artifact_store=get_parse_artifact_store(),
+        document_processing_store=get_document_processing_store(),
     )
 
 
@@ -151,3 +267,28 @@ def get_agent_conversation_service() -> AgentConversationService:
         answer_generator=OpenAICompatibleAnswerGenerator(),
         conversation_store=get_conversation_store(),
     )
+
+
+@lru_cache
+def get_perception_service() -> PerceptionService:
+    """Return perception service for regulatory intelligence."""
+    return PerceptionService(
+        event_store=MockEventStore(),
+        retrieval_service=get_retrieval_service(),
+    )
+
+
+@lru_cache
+def get_agent_session_service() -> AgentSessionService:
+    """Return agent session service."""
+    return AgentSessionService(conversation_store=get_conversation_store())
+
+
+def preload_runtime_dependencies() -> None:
+    """Warm dependencies that are safe and useful to preload during startup."""
+    LLMFactory.preload_clients(["qwen", "deepseek"])
+
+
+def cleanup_runtime_dependencies() -> None:
+    """Release runtime dependencies that expose explicit cleanup hooks."""
+    LLMFactory.cleanup()

backend/app/shared/errors.py

@@ -0,0 +1,30 @@
+"""Define shared backend exception types."""
+
+from __future__ import annotations
+
+
+class VectorStoreSchemaError(RuntimeError):
+    """Signal that the active vector store schema does not match backend expectations."""
+
+    def __init__(
+        self,
+        *,
+        message: str,
+        host: str,
+        db_name: str,
+        collection_name: str,
+        expected_fields: list[str],
+        actual_fields: list[str],
+    ) -> None:
+        """Initialize the vector store schema error details."""
+        self.host = host
+        self.db_name = db_name
+        self.collection_name = collection_name
+        self.expected_fields = expected_fields
+        self.actual_fields = actual_fields
+        # Keep the message self-contained so runtime logs show the full mismatch context.
+        details = (
+            f"{message} | host={host} db={db_name} collection={collection_name} "
+            f"expected_fields={expected_fields} actual_fields={actual_fields}"
+        )
+        super().__init__(details)

backend/backend/data/documents.json

@@ -1 +0,0 @@
-{}

backend/data/document_processing.json

@@ -0,0 +1,131 @@
+{
+  "runs": {
+    "8e722053-5009-40fe-a483-535b40ebbb16": {
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "doc_id": "7cbdfe3c",
+      "trigger_type": "upload",
+      "run_status": "succeeded",
+      "parser_backend": "aliyun_docmind",
+      "chunk_backend": "aliyun",
+      "embedding_model": "text-embedding-v3",
+      "index_name": "regulations_dense_1024_v2",
+      "started_at": "2026-05-26T12:18:27.208692+00:00",
+      "stored_at": "2026-05-26T12:18:27.712855+00:00",
+      "parsed_at": "2026-05-26T12:18:42.989238+00:00",
+      "indexed_at": "2026-05-26T12:18:51.172418+00:00",
+      "finished_at": "2026-05-26T12:18:51.172418+00:00",
+      "layout_count": 48,
+      "structure_node_count": 6,
+      "semantic_block_count": 33,
+      "vector_chunk_count": 34,
+      "chunk_count": 34,
+      "failure_stage": "",
+      "error_message": "",
+      "metadata": {
+        "generate_summary": true,
+        "parse_task_id": "docmind-20260526-10b94713ccb348498b12180a5dcf32ff"
+      }
+    }
+  },
+  "status_events": {
+    "d0532baf-0d65-4130-b282-ec51f04132fd": {
+      "event_id": "d0532baf-0d65-4130-b282-ec51f04132fd",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "from_status": "",
+      "to_status": "pending",
+      "stage": "document_created",
+      "message": "Document record created",
+      "metadata": {},
+      "occurred_at": "2026-05-26T12:18:27.235921+00:00"
+    },
+    "a5e32db5-25c3-4c73-a987-7311f0e72a31": {
+      "event_id": "a5e32db5-25c3-4c73-a987-7311f0e72a31",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "from_status": "pending",
+      "to_status": "stored",
+      "stage": "store",
+      "message": "Source file stored",
+      "metadata": {},
+      "occurred_at": "2026-05-26T12:18:27.741462+00:00"
+    },
+    "18e04ce7-9d7a-4008-8600-e2590100bd85": {
+      "event_id": "18e04ce7-9d7a-4008-8600-e2590100bd85",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "from_status": "stored",
+      "to_status": "parsed",
+      "stage": "parse",
+      "message": "Document parsed",
+      "metadata": {
+        "artifact_count": 4
+      },
+      "occurred_at": "2026-05-26T12:18:43.218026+00:00"
+    },
+    "d3b06025-5c91-4a42-9e5f-dce1c5312b96": {
+      "event_id": "d3b06025-5c91-4a42-9e5f-dce1c5312b96",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "from_status": "parsed",
+      "to_status": "indexed",
+      "stage": "index",
+      "message": "Document indexed",
+      "metadata": {
+        "chunk_count": 34,
+        "index_name": "regulations_dense_1024_v2"
+      },
+      "occurred_at": "2026-05-26T12:18:51.195442+00:00"
+    }
+  },
+  "artifacts": {
+    "47fe2877-a8f5-4e1d-901b-80cd0194ba96": {
+      "artifact_id": "47fe2877-a8f5-4e1d-901b-80cd0194ba96",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "artifact_type": "layouts",
+      "object_name": "artifacts/7cbdfe3c/layouts.json",
+      "content_type": "application/json",
+      "byte_size": 0,
+      "checksum": "",
+      "metadata": {},
+      "created_at": "2026-05-26T12:18:43.188467+00:00"
+    },
+    "44aa075b-86b2-48a7-9d14-a2453bd53863": {
+      "artifact_id": "44aa075b-86b2-48a7-9d14-a2453bd53863",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "artifact_type": "structure_nodes",
+      "object_name": "artifacts/7cbdfe3c/structure_nodes.json",
+      "content_type": "application/json",
+      "byte_size": 0,
+      "checksum": "",
+      "metadata": {},
+      "created_at": "2026-05-26T12:18:43.188494+00:00"
+    },
+    "dedcc8fe-fa58-4de6-984d-f44332af5204": {
+      "artifact_id": "dedcc8fe-fa58-4de6-984d-f44332af5204",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "artifact_type": "semantic_blocks",
+      "object_name": "artifacts/7cbdfe3c/semantic_blocks.json",
+      "content_type": "application/json",
+      "byte_size": 0,
+      "checksum": "",
+      "metadata": {},
+      "created_at": "2026-05-26T12:18:43.188511+00:00"
+    },
+    "9b0d8bda-e69e-4a4e-ae06-a308afe43109": {
+      "artifact_id": "9b0d8bda-e69e-4a4e-ae06-a308afe43109",
+      "doc_id": "7cbdfe3c",
+      "run_id": "8e722053-5009-40fe-a483-535b40ebbb16",
+      "artifact_type": "vector_chunks",
+      "object_name": "artifacts/7cbdfe3c/vector_chunks.json",
+      "content_type": "application/json",
+      "byte_size": 0,
+      "checksum": "",
+      "metadata": {},
+      "created_at": "2026-05-26T12:18:43.188526+00:00"
+    }
+  }
+}

backend/data/documents.json

@@ -1,385 +1,38 @@
 {
-  "69280841": {
-    "doc_id": "69280841",
-    "doc_name": "TCT算法接口.pdf",
-    "file_name": "TCT算法接口.pdf",
-    "object_name": "69280841/TCT算法接口.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 165557,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "local_markdown_parser",
-    "index_name": "",
-    "error_message": "embedding 维度不匹配，期望 1536",
-    "created_at": "2026-05-18T07:12:16.668306+00:00",
-    "updated_at": "2026-05-18T07:12:19.417142+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "structure_nodes": 0
-    }
-  },
-  "44121fbb": {
-    "doc_id": "44121fbb",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "44121fbb/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "",
-    "index_name": "",
-    "error_message": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a5cb9d0>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-    "created_at": "2026-05-18T09:53:47.996183+00:00",
-    "updated_at": "2026-05-18T09:53:50.825868+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "failure_reason": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a5cb9d0>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-      "processing_stage": "failed"
-    }
-  },
-  "77debb4a": {
-    "doc_id": "77debb4a",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "77debb4a/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "",
-    "index_name": "",
-    "error_message": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a6dd480>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-    "created_at": "2026-05-18T10:05:46.104259+00:00",
-    "updated_at": "2026-05-18T10:05:48.704061+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "failure_reason": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a6dd480>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-      "processing_stage": "failed"
-    }
-  },
-  "d12bdcc8": {
-    "doc_id": "d12bdcc8",
-    "doc_name": "TCT算法接口.pdf",
-    "file_name": "TCT算法接口.pdf",
-    "object_name": "d12bdcc8/TCT算法接口.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 165557,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "",
-    "index_name": "",
-    "error_message": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a5bf570>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-    "created_at": "2026-05-18T10:07:22.199824+00:00",
-    "updated_at": "2026-05-18T10:07:24.653751+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "failure_reason": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a5bf570>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-      "processing_stage": "failed"
-    }
-  },
-  "3c2e8c9c": {
-    "doc_id": "3c2e8c9c",
-    "doc_name": "20260415_Continental tire mobile app solution.pdf",
-    "file_name": "20260415_Continental tire mobile app solution.pdf",
-    "object_name": "3c2e8c9c/20260415_Continental tire mobile app solution.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 2178074,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "",
-    "index_name": "",
-    "error_message": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a5bc8d0>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-    "created_at": "2026-05-18T10:09:58.338274+00:00",
-    "updated_at": "2026-05-18T10:10:01.295502+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "failure_reason": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614a5bc8d0>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-      "processing_stage": "failed"
-    }
-  },
-  "d22d21a0": {
-    "doc_id": "d22d21a0",
-    "doc_name": "20260415_Continental tire mobile app solution.pdf",
-    "file_name": "20260415_Continental tire mobile app solution.pdf",
-    "object_name": "d22d21a0/20260415_Continental tire mobile app solution.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 2178074,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "",
-    "index_name": "",
-    "error_message": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614b994160>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-    "created_at": "2026-05-18T10:12:20.078027+00:00",
-    "updated_at": "2026-05-18T10:12:22.999843+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "failure_reason": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614b994160>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-      "processing_stage": "failed"
-    }
-  },
-  "35f129d3": {
-    "doc_id": "35f129d3",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "35f129d3/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "",
-    "index_name": "",
-    "error_message": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614b995370>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-    "created_at": "2026-05-18T10:13:24.706512+00:00",
-    "updated_at": "2026-05-18T10:13:27.180509+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "failure_reason": "unable to load credentials from any of the providers in the chain: ['EnvironmentVariableCredentialsProvider: Environment variable accessKeyId cannot be empty', 'CLIProfileCredentialsProvider: unable to open credentials file: C:\\\\Users\\\\A200477427\\\\.aliyun/config.json', 'ProfileCredentialsProvider: failed to get credential from credentials file: $C:\\\\Users\\\\A200477427\\\\.alibabacloud/credentials.ini', \"EcsRamRoleCredentialsProvider: HTTPConnectionPool(host='100.100.100.200', port=80): Max retries exceeded with url: /latest/meta-data/ram/security-credentials/ (Caused by ConnectTimeoutError(<HTTPConnection(host='100.100.100.200', port=80) at 0x2614b995370>, 'Connection to 100.100.100.200 timed out. (connect timeout=1.0)'))\"]",
-      "processing_stage": "failed"
-    }
-  },
-  "efc21515": {
-    "doc_id": "efc21515",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "efc21515/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "aliyun_docmind",
-    "index_name": "",
-    "error_message": "Client error '400 Bad Request' for url 'http://6.86.80.4:30080/v1/embeddings'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400",
-    "created_at": "2026-05-18T13:47:32.076786+00:00",
-    "updated_at": "2026-05-18T13:47:57.998073+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "parser_backend": "aliyun_docmind",
-      "parse_task_id": "docmind-20260518-a6e84447457f43cb85f95225cfc6495b",
-      "layout_count": 87,
-      "structure_node_count": 20,
-      "semantic_block_count": 27,
-      "vector_chunk_count": 27,
-      "artifact_keys": {
-        "layouts": "artifacts/efc21515/layouts.json",
-        "structure_nodes": "artifacts/efc21515/structure_nodes.json",
-        "semantic_blocks": "artifacts/efc21515/semantic_blocks.json",
-        "vector_chunks": "artifacts/efc21515/vector_chunks.json"
-      },
-      "processing_stage": "failed",
-      "failure_reason": "Client error '400 Bad Request' for url 'http://6.86.80.4:30080/v1/embeddings'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400"
-    }
-  },
-  "0d4b08bc": {
-    "doc_id": "0d4b08bc",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "0d4b08bc/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "aliyun_docmind",
-    "index_name": "",
-    "error_message": "Client error '404 Not Found' for url 'http://6.86.80.4:30080/v1/embeddings'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/404",
-    "created_at": "2026-05-18T14:03:15.134344+00:00",
-    "updated_at": "2026-05-18T14:03:34.843448+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "parser_backend": "aliyun_docmind",
-      "parse_task_id": "docmind-20260518-78353d85daa24147b68d8fb71895179f",
-      "layout_count": 87,
-      "structure_node_count": 20,
-      "semantic_block_count": 27,
-      "vector_chunk_count": 27,
-      "artifact_keys": {
-        "layouts": "artifacts/0d4b08bc/layouts.json",
-        "structure_nodes": "artifacts/0d4b08bc/structure_nodes.json",
-        "semantic_blocks": "artifacts/0d4b08bc/semantic_blocks.json",
-        "vector_chunks": "artifacts/0d4b08bc/vector_chunks.json"
-      },
-      "processing_stage": "failed",
-      "failure_reason": "Client error '404 Not Found' for url 'http://6.86.80.4:30080/v1/embeddings'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/404"
-    }
-  },
-  "4302f314": {
-    "doc_id": "4302f314",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "4302f314/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "aliyun_docmind",
-    "index_name": "",
-    "error_message": "embedding 维度不匹配，期望 1536",
-    "created_at": "2026-05-18T14:11:29.943973+00:00",
-    "updated_at": "2026-05-18T14:11:48.554500+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "parser_backend": "aliyun_docmind",
-      "parse_task_id": "docmind-20260518-23935ee455ac4b26ac4201ac4781ee52",
-      "layout_count": 87,
-      "structure_node_count": 20,
-      "semantic_block_count": 27,
-      "vector_chunk_count": 27,
-      "artifact_keys": {
-        "layouts": "artifacts/4302f314/layouts.json",
-        "structure_nodes": "artifacts/4302f314/structure_nodes.json",
-        "semantic_blocks": "artifacts/4302f314/semantic_blocks.json",
-        "vector_chunks": "artifacts/4302f314/vector_chunks.json"
-      },
-      "processing_stage": "failed",
-      "failure_reason": "embedding 维度不匹配，期望 1536"
-    }
-  },
-  "765ed1ee": {
-    "doc_id": "765ed1ee",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "765ed1ee/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "aliyun_docmind",
-    "index_name": "",
-    "error_message": "<MilvusException: (code=1100, message=the dim (1024) of field data(embedding) is not equal to schema dim (1536): invalid parameter[expected=1536][actual=1024])>",
-    "created_at": "2026-05-18T14:18:28.875138+00:00",
-    "updated_at": "2026-05-18T14:18:57.389110+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "parser_backend": "aliyun_docmind",
-      "parse_task_id": "docmind-20260518-f116856bc29245baa2531b245078a701",
-      "layout_count": 87,
-      "structure_node_count": 20,
-      "semantic_block_count": 27,
-      "vector_chunk_count": 27,
-      "artifact_keys": {
-        "layouts": "artifacts/765ed1ee/layouts.json",
-        "structure_nodes": "artifacts/765ed1ee/structure_nodes.json",
-        "semantic_blocks": "artifacts/765ed1ee/semantic_blocks.json",
-        "vector_chunks": "artifacts/765ed1ee/vector_chunks.json"
-      },
-      "processing_stage": "failed",
-      "failure_reason": "<MilvusException: (code=1100, message=the dim (1024) of field data(embedding) is not equal to schema dim (1536): invalid parameter[expected=1536][actual=1024])>"
-    }
-  },
-  "05cabe09": {
-    "doc_id": "05cabe09",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "05cabe09/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
-    "status": "failed",
-    "regulation_type": "",
-    "version": "",
-    "summary": "",
-    "summary_latency_ms": 0,
-    "chunk_count": 0,
-    "parser_name": "aliyun_docmind",
-    "index_name": "",
-    "error_message": "embedding 维度不匹配，期望 1536",
-    "created_at": "2026-05-18T14:24:32.156500+00:00",
-    "updated_at": "2026-05-18T14:24:50.114138+00:00",
-    "metadata": {
-      "generate_summary": true,
-      "parser_backend": "aliyun_docmind",
-      "parse_task_id": "docmind-20260518-897d858983df48e28e9819e563d46208",
-      "layout_count": 87,
-      "structure_node_count": 20,
-      "semantic_block_count": 27,
-      "vector_chunk_count": 27,
-      "artifact_keys": {
-        "layouts": "artifacts/05cabe09/layouts.json",
-        "structure_nodes": "artifacts/05cabe09/structure_nodes.json",
-        "semantic_blocks": "artifacts/05cabe09/semantic_blocks.json",
-        "vector_chunks": "artifacts/05cabe09/vector_chunks.json"
-      },
-      "processing_stage": "failed",
-      "failure_reason": "embedding 维度不匹配，期望 1536"
-    }
-  },
-  "9acb2ba0": {
-    "doc_id": "9acb2ba0",
-    "doc_name": "大众汽车手册.pdf",
-    "file_name": "大众汽车手册.pdf",
-    "object_name": "9acb2ba0/大众汽车手册.pdf",
-    "content_type": "application/pdf",
-    "size_bytes": 766565,
+  "7cbdfe3c": {
+    "doc_id": "7cbdfe3c",
+    "doc_name": "使用RSA Token连接CheckPoint VPN及PIN码设置_220.181.114.93 or 10.25.134.3.docx",
+    "file_name": "使用RSA Token连接CheckPoint VPN及PIN码设置_220.181.114.93 or 10.25.134.3.docx",
+    "object_name": "7cbdfe3c/使用RSA Token连接CheckPoint VPN及PIN码设置_220.181.114.93 or 10.25.134.3.docx",
+    "content_type": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+    "size_bytes": 1199920,
     "status": "indexed",
     "regulation_type": "",
     "version": "",
     "summary": "",
     "summary_latency_ms": 0,
-    "chunk_count": 27,
+    "chunk_count": 34,
     "parser_name": "aliyun_docmind",
-    "index_name": "regulations_dense_1024_v1",
+    "index_name": "regulations_dense_1024_v2",
     "error_message": "",
-    "created_at": "2026-05-18T14:29:01.368719+00:00",
-    "updated_at": "2026-05-18T14:29:23.699068+00:00",
+    "created_at": "2026-05-26T12:18:27.206125+00:00",
+    "updated_at": "2026-05-26T12:18:51.171308+00:00",
     "metadata": {
       "generate_summary": true,
       "parser_backend": "aliyun_docmind",
-      "parse_task_id": "docmind-20260518-e5fd4a5419e74d569c562e389e6ae72c",
-      "layout_count": 87,
-      "structure_node_count": 20,
-      "semantic_block_count": 27,
-      "vector_chunk_count": 27,
+      "parse_task_id": "docmind-20260526-10b94713ccb348498b12180a5dcf32ff",
+      "layout_count": 48,
+      "structure_node_count": 6,
+      "semantic_block_count": 33,
+      "vector_chunk_count": 34,
       "artifact_keys": {
-        "layouts": "artifacts/9acb2ba0/layouts.json",
-        "structure_nodes": "artifacts/9acb2ba0/structure_nodes.json",
-        "semantic_blocks": "artifacts/9acb2ba0/semantic_blocks.json",
-        "vector_chunks": "artifacts/9acb2ba0/vector_chunks.json"
+        "layouts": "artifacts/7cbdfe3c/layouts.json",
+        "structure_nodes": "artifacts/7cbdfe3c/structure_nodes.json",
+        "semantic_blocks": "artifacts/7cbdfe3c/semantic_blocks.json",
+        "vector_chunks": "artifacts/7cbdfe3c/vector_chunks.json"
       },
       "processing_stage": "indexed",
-      "index_collection": "regulations_dense_1024_v1"
+      "index_collection": "regulations_dense_1024_v2"
     }
   }
 }

docs/architecture/aliyun-ingest-implementation.md

@@ -7,7 +7,7 @@
 - 上传入口保持为 `/api/v1/documents/upload`
 - 默认 `PARSER_BACKEND=aliyun`
 - 默认 `CHUNK_BACKEND=aliyun`
-- 默认 Milvus collection 为 `regulations_dense_1536_v2`
+- 默认 Milvus collection 为 `regulations_dense_1024_v2`
 - 解析产物落到 MinIO `artifacts/{doc_id}/`
 
 完整主链路如下：
@@ -19,7 +19,7 @@
 5. 转换为 `structure_nodes / semantic_blocks / vector_chunks`
 6. 三层结构 JSON 回写 MinIO
 7. 使用 `vector_chunks[*].embedding_text` 调 embedding API
-8. 写入 `regulations_dense_1536_v2`
+8. 写入 `regulations_dense_1024_v2`
 9. 文档状态更新为 `indexed`
 
 运行时转换逻辑位于 `backend/app/infrastructure/parser/aliyun_layout_normalizer.py`。

docs/architecture/backend-project-architecture.md

@@ -10,6 +10,31 @@
 - 本文档负责冻结目标模块边界、依赖规则和实现组织方式。
 - 后续任何代码重构、能力替换或底座升级，都应同时满足 RFC 与本文档。
 
+## 1.1 Document Status And Authority
+
+本文档不是仅供参考的“目标态草案”，而是当前 backend 持续开发的强制架构基线。
+
+- 新增 backend 功能默认必须遵守本文档定义的模块边界与依赖方向。
+- 历史实现、迁移中代码和兼容 façade 的存在，不构成继续偏离本文档的理由。
+- 当现状与本文档冲突时，新增代码按本文档落位；旧代码按迁移计划逐步收口，但不允许继续扩大 legacy 边界。
+- 评审、重构验收和后续架构讨论，均以本文档作为 backend 内部结构的 authority。
+
+## 1.2 Authoritative Scope
+
+本文档约束的 backend 范围包括：
+
+- `backend/app/api/*`
+- `backend/app/application/*`
+- `backend/app/domain/*`
+- `backend/app/infrastructure/*`
+- `backend/app/shared/*`
+
+说明：
+
+- `backend/app/services/*` 与 `backend/app/workflows/*` 当前属于迁移期 legacy 目录，不是新增业务逻辑的默认落点。
+- `backend/app/api/routes/docs.py` 与 `backend/app/api/routes/rag.py` 视为遗留或非主入口，除迁移、兼容或下线动作外，不应继续扩展。
+- `backend/app/api/routes/compliance.py` 当前仍对外暴露，但尚未完全满足本文档约束；在迁移到 application service 之前，应视为受控 legacy 入口，而不是新的架构样板。
+
 ## 2. Current-State Problems
 
 基于当前代码，后端已经具备以下能力：
@@ -22,6 +47,18 @@
 
 但这些能力当前主要是“可运行”，还不是“结构清晰、便于替换、便于演进”的状态。核心问题如下。
 
+### 2.0 Current-State Verdict
+
+基于当前仓库，现状裁决如下：
+
+- 已基本符合：`documents` 上传/查询主链路已经通过 `DocumentCommandService` 与 `DocumentQueryService` 收口。
+- 已基本符合：`knowledge` 检索已经通过 `KnowledgeRetrievalService` 统一对外暴露。
+- 已基本符合：`agent` 问答主链路已经通过 `AgentConversationService` 收口，`shared/bootstrap.py` 已承担 composition root 角色。
+- 部分符合：Agent session 详情、历史、删除、反馈等接口曾经直接访问 `ConversationStore`，需要继续收口到 application service。
+- 未完全符合：`compliance` 路由仍直接处理文件落盘、任务状态和 mock 结果，不符合 `api -> application -> domain ports -> infrastructure`。
+- 未完全符合：部分 `infrastructure` adapter 仍依赖 `services/*` 内的 legacy 实现，说明迁移尚未彻底完成。
+- 未完全符合：`api/main.py` 的生命周期预热逻辑仍直接依赖旧 LLM factory，尚未完全回到统一 wiring 边界。
+
 ### 2.1 `DocumentProcessor` 责任过载
 
 现状判断：
@@ -603,6 +640,7 @@ infrastructure -> external systems
 - `application` 只能依赖 `domain`、端口接口，以及通过 composition root 注入进来的实现实例
 - `domain` 不能依赖 `api` 或 `infrastructure`
 - `infrastructure` 可以依赖 `domain` 定义的端口和数据模型，但不能反向驱动 application 逻辑
+- `api/main.py` 这类应用入口可以保留轻量 startup/shutdown 生命周期代码，但不应长期直接依赖 legacy service factory；预热与装配逻辑应逐步收口到明确的 wiring 边界
 
 说明：
 
@@ -739,6 +777,54 @@ infrastructure -> external systems
 - 内部 DTO / VO / domain object 收敛到 `application` 或 `domain`
 - 不允许 API model 直接渗透到 domain
 
+### 10.10 应用入口与启动生命周期
+
+当前：
+
+- `backend/app/api/main.py`
+
+目标：
+
+- 保留 FastAPI app、middleware 和 lifespan 入口职责
+- 逐步去除对 legacy LLM factory 的直接依赖
+- 预热、清理和依赖装配应保持在明确的 wiring / bootstrap 边界内，而不是继续把旧 service factory 固化为应用入口依赖
+
+### 10.11 Compliance 路由
+
+当前：
+
+- `backend/app/api/routes/compliance.py`
+
+目标：
+
+- 如继续保留该能力，应迁移到独立的 application service 与稳定端口
+- 在迁移完成前，该路由视为受控 legacy 入口，可修 bug，但不应继续扩展业务编排职责
+
+### 10.12 遗留路由入口
+
+当前：
+
+- `backend/app/api/routes/docs.py`
+- `backend/app/api/routes/rag.py`
+
+目标：
+
+- 作为遗留或演示入口逐步归档、下线或迁移
+- 不再作为新增 backend 能力的开发入口
+
+### 10.13 Legacy Workflow 与 Service 目录
+
+当前：
+
+- `backend/app/workflows/*`
+- `backend/app/services/*`
+
+目标：
+
+- 保留迁移期兼容价值，但不再承载新的长期业务编排
+- 若某个 legacy 实现仍被 `infrastructure` adapter 间接复用，应视为过渡依赖，后续逐步迁入 `infrastructure` 或更稳定的底层支撑模块
+- 任何新增 backend 业务能力，都不应再以这些目录作为默认落点
+
 ## 11. Technology Replacement Boundaries
 
 ### 11.1 本地解析 / MinerU -> 阿里云文档解析
@@ -790,6 +876,10 @@ infrastructure -> external systems
 - 禁止新建第二个“大一统流程类”替代 `DocumentProcessor`
 - 禁止 `knowledge` 和 `agent` 各自维护独立检索实现
 - 禁止 parser、embedding、vector index、llm provider 的替换穿透到 API 层
+- 禁止新增 route 直接访问 `ConversationStore`
+- 禁止新增代码把 `backend/app/services/*` 或 `backend/app/workflows/*` 作为默认业务落点
+- 禁止新增 `infrastructure -> services/*` 的过渡依赖；已有依赖只允许在迁移窗口内逐步消除，不允许继续扩散
+- 禁止在 README、开发说明或评审结论中把 legacy 目录描述为当前 backend 的主结构
 
 ## 13. Architecture Review Checklist
 
@@ -807,3 +897,7 @@ infrastructure -> external systems
 10. 是否明确 `knowledge` 与 `agent` 共用同一 retrieval 底座。
 11. 是否明确 API 层只负责 transport concerns，不再直接承担业务编排。
 12. 是否保证后续替换方案时，上层 application service 与外部 API 契约不被迫变化。
+13. 是否仍存在 route 直接访问 `ConversationStore`、文件系统、对象存储或任务状态存储。
+14. 是否新增了 `infrastructure -> services/*` 依赖。
+15. 是否把新的 backend 业务逻辑写进了 `services/*` 或 `workflows/*`。
+16. README、backend README 与协作说明是否仍与当前 authoritative architecture 保持一致。

docs/architecture/document-core-processing-flow.md

@@ -0,0 +1,623 @@
+# 核心文档处理主链路说明
+
+本文件说明当前默认生产链路中的核心文档处理流程，也就是：
+
+- `AliyunDocumentParser`
+- `AliyunVectorChunkBuilder`
+- `OpenAICompatibleEmbeddingProvider`
+- `MilvusVectorIndex`
+
+目标是回答四个核心问题：
+
+1. `ParsedDocument` 为什么是多层结构
+2. 这些结构分别保存到哪里
+3. 哪一步才真正做了向量化
+4. Milvus 里最后到底存的是什么
+
+数据库表设计、关系模型、DDL 和 PostgreSQL 职责边界已经单独整理到 [document-processing-database-design.md](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/docs/architecture/document-processing-database-design.md:1)。本文件保留流程视角，只在必要处给出与存储分工相关的摘要，不再作为数据库设计 authority。
+
+## 1. 主链路总览
+
+当前默认实现由 `DocumentCommandService.upload_and_process()` 统一编排。它不是“parse 完直接进向量库”，而是先生成结构化解析产物，再把其中适合检索的一层送去 embedding 和 Milvus。
+
+```mermaid
+sequenceDiagram
+    participant API as API / Service
+    participant MinIO as BinaryStore
+    participant Parser as AliyunDocumentParser
+    participant PG as DocumentRepository / ParseArtifactStore
+    participant Embed as EmbeddingProvider
+    participant Milvus as VectorIndex
+
+    API->>MinIO: 保存原始文件
+    API->>Parser: parse(file_path, doc_id, doc_name)
+    Parser-->>API: ParsedDocument
+    API->>MinIO: 保存 layouts/structure_nodes/semantic_blocks/vector_chunks JSON
+    API->>PG: 更新 documents.status=parsed
+    API->>PG: 保存 structure_nodes / semantic_blocks
+    API->>API: chunk_builder.build(parsed_document)
+    API->>Embed: embed_texts([chunk.embedding_text])
+    Embed-->>API: vectors
+    API->>Milvus: upsert(chunks, vectors)
+    API->>PG: 更新 documents.status=indexed
+```
+
+主链路编排代码在 [services.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/application/documents/services.py:83)：
+
+```python
+def upload_and_process(
+    self,
+    *,
+    doc_id: str | None = None,
+    file_name: str,
+    content: bytes,
+    content_type: str,
+    doc_name: str | None,
+    regulation_type: str,
+    version: str,
+    generate_summary: bool,
+) -> DocumentProcessResult:
+    doc_id = doc_id or str(uuid.uuid4())[:8]
+    final_doc_name = doc_name or file_name
+    object_name = f"{doc_id}/{file_name}"
+
+    self.document_repository.create(document)
+    self.binary_store.save(object_name=object_name, data=content, content_type=content_type, metadata={"doc_id": doc_id})
+    self.document_repository.update_status(doc_id, DocumentStatus.STORED)
+
+    parsed_document = self.parser.parse(file_path=temp_path, doc_id=doc_id, doc_name=final_doc_name)
+    artifact_keys = self._save_parse_artifacts(doc_id=doc_id, parsed_document=parsed_document)
+    self.document_repository.update_status(doc_id, DocumentStatus.PARSED, parser_name=parsed_document.parser_name, metadata={...})
+
+    if self.parse_artifact_store:
+        self.parse_artifact_store.save(doc_id, parsed_document.structure_nodes, parsed_document.semantic_blocks)
+
+    chunks = self.chunk_builder.build(parsed_document=parsed_document, regulation_type=regulation_type, version=version)
+    vectors = self.embedding_provider.embed_texts([chunk.embedding_text for chunk in chunks])
+    inserted = self.vector_index.upsert(chunks, vectors)
+
+    self.document_repository.update_status(doc_id, DocumentStatus.INDEXED, chunk_count=len(chunks), index_name=health.get("collection_name", ""), metadata={...})
+```
+
+默认绑定关系在 [bootstrap.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/shared/bootstrap.py:157)：
+
+```python
+def get_parser():
+    if settings.parser_backend == "aliyun":
+        return AliyunDocumentParser()
+    return LocalDocumentParser()
+
+
+def get_chunk_builder():
+    if settings.chunk_backend == "aliyun":
+        return AliyunVectorChunkBuilder()
+    return LocalRegulationChunkBuilder(...)
+
+
+def get_embedding_provider() -> OpenAICompatibleEmbeddingProvider:
+    return OpenAICompatibleEmbeddingProvider()
+
+
+def get_vector_index() -> VectorIndex:
+    return LazyVectorIndex(_build_vector_index)
+```
+
+也就是说，当前默认主链路是：
+
+- parser: `AliyunDocumentParser`
+- chunk builder: `AliyunVectorChunkBuilder`
+- embedding provider: `OpenAICompatibleEmbeddingProvider`
+- vector index: `MilvusVectorIndex`
+
+## 2. `ParsedDocument` 为什么是三层结构
+
+`ParsedDocument` 不是最终入库格式，而是 parser 输出给后续处理步骤的统一中间结构。它把“结构理解”和“向量检索准备”拆成了三层。
+
+定义在 [models.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/domain/documents/models.py:49)：
+
+```python
+@dataclass
+class ParsedDocument:
+    doc_id: str
+    doc_name: str
+    structure_nodes: list[dict[str, Any]]
+    semantic_blocks: list[dict[str, Any]]
+    vector_chunks: list[dict[str, Any]]
+    parser_name: str
+    raw_text: str = ""
+    raw_layouts: list[dict[str, Any]] = field(default_factory=list)
+    metadata: dict[str, Any] = field(default_factory=dict)
+```
+
+这三层的职责不同：
+
+- `structure_nodes`
+  - 标题层级骨架
+  - 描述“文档有哪些章、节、条”
+  - 用于保留结构，不直接做 embedding
+
+- `semantic_blocks`
+  - 语义块层
+  - 把正文、表格、图片说明整理成连续的语义单元
+  - 是从原始 layout 到检索 chunk 之间的中间层
+
+- `vector_chunks`
+  - 检索和向量化层
+  - 已经是适合送给 embedding 模型的 chunk 视图
+  - 后续 `ChunkBuilder` 基本就是把这层映射成统一 `Chunk`
+
+### 2.1 这三层是怎么从 parser 结果生成的
+
+`AliyunDocumentParser.parse()` 先通过网关拿到阿里云返回的 `layouts`，再把 `layouts` 转成三层结构。
+
+代码在 [aliyun_document_parser.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/parser/aliyun_document_parser.py:28)：
+
+```python
+def parse(self, *, file_path: str, doc_id: str, doc_name: str) -> ParsedDocument:
+    payload = self.gateway.parse_document(file_path=file_path)
+    layouts = payload.layouts
+    structure_nodes = build_structure_nodes(layouts)
+    semantic_blocks = build_semantic_blocks(layouts)
+    vector_chunks = build_vector_chunks(
+        semantic_blocks,
+        doc_id=doc_id,
+        doc_title=doc_name,
+        max_chars=MAX_CHARS,
+        overlap_chars=OVERLAP_CHARS,
+    )
+    raw_text = "\n\n".join(
+        block.get("text", "")
+        for block in semantic_blocks
+        if block.get("text")
+    )
+    return ParsedDocument(
+        doc_id=doc_id,
+        doc_name=doc_name,
+        structure_nodes=structure_nodes,
+        semantic_blocks=semantic_blocks,
+        vector_chunks=vector_chunks,
+        parser_name=self.parser_name,
+        raw_text=raw_text,
+        raw_layouts=layouts,
+        metadata={...},
+    )
+```
+
+也就是说：
+
+- parser 原始输出是 `layouts`
+- 当前系统真正消费的是 `ParsedDocument`
+- `ParsedDocument` 是由 normalizer 从 `layouts` 规整出来的
+
+### 2.2 第一层：`structure_nodes`
+
+这一层只保留标题和层级。
+
+代码在 [aliyun_layout_normalizer.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/parser/aliyun_layout_normalizer.py:85)：
+
+```python
+def build_structure_nodes(layouts: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    nodes: list[dict[str, Any]] = []
+    for layout in layouts:
+        if not is_title(layout):
+            continue
+        text = get_text(layout)
+        if not text or text in TOC_TITLES:
+            continue
+        nodes.append(
+            {
+                "unique_id": layout.get("uniqueId"),
+                "page": get_page(layout),
+                "index": layout.get("index", 0),
+                "level": layout.get("level", 0),
+                "title": text,
+                "type": layout.get("type"),
+                "sub_type": layout.get("subType"),
+            }
+        )
+    return nodes
+```
+
+示例：
+
+```json
+[
+  {
+    "unique_id": "l-title-001",
+    "page": 2,
+    "index": 11,
+    "level": 1,
+    "title": "1 范围",
+    "type": "title",
+    "sub_type": "para_title"
+  },
+  {
+    "unique_id": "l-title-002",
+    "page": 3,
+    "index": 18,
+    "level": 2,
+    "title": "1.1 适用对象",
+    "type": "title",
+    "sub_type": "para_title"
+  }
+]
+```
+
+这层的意义是“保留目录树”，不是直接拿来检索。
+
+### 2.3 第二层：`semantic_blocks`
+
+这一层会把连续正文合并成一个语义块，也会单独处理表格和图片说明。
+
+代码在 [aliyun_layout_normalizer.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/parser/aliyun_layout_normalizer.py:163)：
+
+```python
+def build_semantic_blocks(layouts: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    semantic_blocks: list[dict[str, Any]] = []
+    section_stack: list[dict[str, Any]] = []
+    pending_text_blocks: list[dict[str, Any]] = []
+    block_id = 1
+
+    for layout in layouts:
+        text = get_text(layout)
+        page = get_page(layout)
+
+        if is_title(layout):
+            block_id = flush_text_block(pending_text_blocks, semantic_blocks, block_id)
+            pending_text_blocks = []
+            section_stack = update_section_path(section_stack, layout)
+            continue
+
+        section_path = section_path_titles(section_stack)
+        section_title = section_path[-1] if section_path else "未分类"
+        section_level = len(section_path)
+
+        if is_table(layout):
+            ...
+            semantic_blocks.append(
+                {
+                    "semantic_id": f"semantic-{block_id}",
+                    "block_type": "table",
+                    "page_start": page,
+                    "page_end": page,
+                    "section_path": section_path,
+                    "section_level": section_level,
+                    "section_title": section_title,
+                    "source_ids": [layout.get("uniqueId")],
+                    "text": table_text,
+                }
+            )
+            continue
+
+        if is_text(layout) and text:
+            pending_text_blocks.append(
+                {
+                    "page": page,
+                    "text": text,
+                    "unique_id": layout.get("uniqueId"),
+                    "section_path": section_path,
+                    "section_level": section_level,
+                    "section_title": section_title,
+                }
+            )
+```
+
+正文合并后会形成类似这样的语义块：
+
+```json
+[
+  {
+    "semantic_id": "semantic-1",
+    "block_type": "section_text",
+    "page_start": 2,
+    "page_end": 2,
+    "section_path": ["1 范围", "1.1 适用对象"],
+    "section_level": 2,
+    "section_title": "1.1 适用对象",
+    "source_ids": ["l-text-001", "l-text-002"],
+    "text": "本标准适用于道路车辆动力电池系统的安全要求。企业应建立一致的测试和验证方法。"
+  }
+]
+```
+
+## 3. 这些结构分别保存到哪里
+
+### 3.1 原始文件和中间 artifacts 先落 MinIO
+
+当前链路在上传阶段会先把原始文件保存到对象存储；解析完成后，又会把结构化中间产物保存为 JSON。
+
+保存 artifacts 的代码在 [services.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/application/documents/services.py:62)：
+
+```python
+def _save_parse_artifacts(self, *, doc_id: str, parsed_document: ParsedDocument) -> dict[str, str]:
+    prefix = f"{parsed_document.metadata.get('artifact_prefix', 'artifacts').strip('/')}/{doc_id}"
+    artifact_payloads = {
+        "layouts": parsed_document.raw_layouts,
+        "structure_nodes": parsed_document.structure_nodes,
+        "semantic_blocks": parsed_document.semantic_blocks,
+        "vector_chunks": parsed_document.vector_chunks,
+    }
+    artifact_keys: dict[str, str] = {}
+    for name, payload in artifact_payloads.items():
+        object_name = f"{prefix}/{name}.json"
+        self.binary_store.save(
+            object_name=object_name,
+            data=json.dumps(payload, ensure_ascii=False, indent=2).encode("utf-8"),
+            content_type="application/json",
+            metadata={"doc_id": doc_id, "artifact_type": name},
+        )
+        artifact_keys[name] = object_name
+    return artifact_keys
+```
+
+`DocumentBinaryStore` 的当前默认实现是 `MinioDocumentBinaryStore`，也就是：
+
+- 原始上传文件进 MinIO
+- `layouts.json` 进 MinIO
+- `structure_nodes.json` 进 MinIO
+- `semantic_blocks.json` 进 MinIO
+- `vector_chunks.json` 进 MinIO
+
+### 3.2 PostgreSQL 在流程中的职责摘要
+
+当前流程中，PostgreSQL 承担的是“文档元数据 + 结构化快照”的职责，而不是向量或大对象存储：
+
+- `documents` 保存当前文档主记录、状态、统计和索引信息
+- `structure_nodes` 保存当前最新解析快照的目录结构
+- `semantic_blocks` 保存当前最新解析快照的语义块结构
+
+更完整的 PostgreSQL 设计，包括：
+
+- `documents`
+- `document_processing_runs`
+- `document_status_history`
+- `document_artifacts`
+- `structure_nodes`
+- `semantic_blocks`
+
+见 [document-processing-database-design.md](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/docs/architecture/document-processing-database-design.md:1)。
+
+### 3.3 存储分工一览
+
+| 数据层 | 保存位置 | 是否直接用于 embedding | 是否最终进入 Milvus | 主要用途 |
+| --- | --- | --- | --- | --- |
+| 原始文件 | MinIO | 否 | 否 | 保留原始上传文档 |
+| `raw_layouts` | MinIO `layouts.json` | 否 | 否 | 保留 parser 原始返回 |
+| `structure_nodes` | MinIO + PostgreSQL | 否 | 否 | 目录树、层级结构 |
+| `semantic_blocks` | MinIO + PostgreSQL | 否 | 间接 | 语义单元、中间层 |
+| `vector_chunks` | MinIO | 是 | 间接 | embedding 前的检索块 |
+| `Chunk` | 内存态 + Milvus | 是 | 是 | 统一向量入库模型 |
+| `documents` 元数据 | PostgreSQL | 否 | 否 | 处理状态、统计、索引信息 |
+
+## 4. 哪一步才真正“变成向量”
+
+这是整个流程最关键的点。
+
+结论先说清楚：
+
+- parse 不做向量化
+- 保存 artifacts 不做向量化
+- `ChunkBuilder.build()` 也不做向量化
+- 只有 `EmbeddingProvider.embed_texts()` 才真正调用 embedding 模型
+- 只有 `VectorIndex.upsert()` 才真正把向量写入向量库
+
+### 4.1 `vector_chunks` 先被映射成统一 `Chunk`
+
+`AliyunVectorChunkBuilder` 并不做 embedding，它只负责把 `ParsedDocument.vector_chunks` 转成领域层统一 `Chunk` 模型。
+
+代码在 [vector_chunk_builder.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/parser/vector_chunk_builder.py:12)：
+
+```python
+def build(
+    self,
+    *,
+    parsed_document: ParsedDocument,
+    regulation_type: str,
+    version: str,
+) -> list[Chunk]:
+    chunks: list[Chunk] = []
+    for index, item in enumerate(parsed_document.vector_chunks):
+        content = item.get("content") or item.get("text") or ""
+        embedding_text = item.get("embedding_text") or content
+        if not embedding_text.strip():
+            continue
+        section_path = item.get("section_path") or []
+        section_title = item.get("section_title") or (section_path[-1] if section_path else "")
+        page_number = item.get("page_start") or item.get("page") or 0
+        chunk_id = item.get("chunk_id") or f"{parsed_document.doc_id}-chunk-{index}"
+        metadata = {k: v for k, v in item.items() if k not in {"content", "embedding_text"}}
+        chunks.append(
+            Chunk(
+                chunk_id=str(chunk_id),
+                doc_id=parsed_document.doc_id,
+                doc_name=parsed_document.doc_name,
+                content=content,
+                embedding_text=embedding_text,
+                section_title=section_title,
+                section_path=section_path,
+                page_number=int(page_number or 0),
+                regulation_type=regulation_type,
+                version=version,
+                semantic_id=item.get("semantic_id", ""),
+                block_type=item.get("block_type", ""),
+                metadata=metadata,
+            )
+        )
+    return chunks
+```
+
+`Chunk` 的定义在 [models.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/domain/documents/models.py:63)：
+
+```python
+@dataclass
+class Chunk:
+    chunk_id: str
+    doc_id: str
+    doc_name: str
+    content: str
+    embedding_text: str
+    section_title: str = ""
+    section_path: list[str] = field(default_factory=list)
+    page_number: int = 0
+    regulation_type: str = ""
+    version: str = ""
+    semantic_id: str = ""
+    block_type: str = ""
+    metadata: dict[str, Any] = field(default_factory=dict)
+```
+
+一个 `Chunk` 的典型样子如下：
+
+```json
+{
+  "chunk_id": "doc-001-chunk-1",
+  "doc_id": "doc-001",
+  "doc_name": "动力电池安全规范",
+  "content": "本标准适用于道路车辆动力电池系统的安全要求。企业应建立一致的测试和验证方法。",
+  "embedding_text": "标准：动力电池安全规范\n章节：1 范围 > 1.1 适用对象\n\n本标准适用于道路车辆动力电池系统的安全要求。企业应建立一致的测试和验证方法。",
+  "section_title": "1.1 适用对象",
+  "section_path": ["1 范围", "1.1 适用对象"],
+  "page_number": 2,
+  "regulation_type": "GB",
+  "version": "2025",
+  "semantic_id": "semantic-1",
+  "block_type": "section_text",
+  "metadata": {
+    "chunk_index": 1,
+    "piece_index": 1,
+    "source_ids": ["l-text-001", "l-text-002"]
+  }
+}
+```
+
+这里最关键的是要区分两个字段：
+
+- `content`
+  - 用于检索命中后的展示内容
+  - 更接近用户最终看到的正文片段
+
+- `embedding_text`
+  - 用于送给 embedding 模型
+  - 比 `content` 多了“标准名 + 章节路径”的上下文
+
+所以“向量化输入”不是纯正文，而是增强后的上下文文本。
+
+### 4.2 真正调用 embedding API 的地方
+
+真正把文本变成向量的是 `OpenAICompatibleEmbeddingProvider.embed_texts()`。
+
+代码在 [openai_compatible_embedding_provider.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/embedding/openai_compatible_embedding_provider.py:64)：
+
+```python
+def embed_texts(self, texts: list[str]) -> list[list[float]]:
+    if not texts:
+        return []
+```
+
+也就是说，只有在这一步：
+
+- 输入：`list[str]` 的 `embedding_text`
+- 输出：`list[list[float]]` 的 dense vectors
+
+前面的 parse、normalizer、chunk builder 都只是准备文本，没有任何向量值产生。
+
+### 4.3 真正把向量写进 Milvus 的地方
+
+向量值生成之后，`MilvusVectorIndex.upsert()` 才会把 `Chunk + vector` 写入向量库。
+
+代码在 [milvus_vector_index.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/vectorstore/milvus_vector_index.py:69)：
+
+```python
+def upsert(self, chunks: list[Chunk], vectors: list[list[float]]) -> int:
+    if len(chunks) != len(vectors):
+        raise ValueError("chunks 与 vectors 数量不一致")
+    data = []
+    now = int(time.time())
+    for chunk, vector in zip(chunks, vectors):
+        data.append(
+            {
+                "id": chunk.chunk_id,
+                "doc_id": chunk.doc_id,
+                "doc_name": chunk.doc_name,
+                "content": chunk.content[:65535],
+                "embedding": vector,
+                "section_title": chunk.section_title[:512],
+                "section_path": json.dumps(chunk.section_path, ensure_ascii=False)[:4096],
+                "page_number": chunk.page_number,
+                "regulation_type": chunk.regulation_type[:128],
+                "version": chunk.version[:64],
+                "semantic_id": chunk.semantic_id[:128],
+                "block_type": chunk.block_type[:64],
+                "metadata_json": json.dumps(chunk.metadata, ensure_ascii=False)[:65535],
+                "created_at": now,
+            }
+        )
+    self.collection.insert(data)
+    self.collection.flush()
+    return len(data)
+```
+
+也就是说，Milvus 最终存进去的是：
+
+- 主键：`chunk_id`
+- 文档维度字段：`doc_id`、`doc_name`
+- 检索展示字段：`content`
+- 向量字段：`embedding`
+- 过滤/回溯字段：`section_title`、`section_path`、`page_number`、`regulation_type`、`version`、`semantic_id`、`block_type`
+- 附加元数据：`metadata_json`
+
+## 5. Milvus 里最后到底存的是什么
+
+### 5.1 Collection schema
+
+当前 `MilvusVectorIndex` 初始化 collection 时定义的 schema 在 [milvus_vector_index.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/infrastructure/vectorstore/milvus_vector_index.py:37)：
+
+```python
+schema = CollectionSchema(
+    fields=[
+        FieldSchema(name="id", dtype=DataType.VARCHAR, max_length=128, is_primary=True, auto_id=False),
+        FieldSchema(name="doc_id", dtype=DataType.VARCHAR, max_length=64),
+        FieldSchema(name="doc_name", dtype=DataType.VARCHAR, max_length=256),
+        FieldSchema(name="content", dtype=DataType.VARCHAR, max_length=65535),
+        FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=settings.embedding_dim),
+        FieldSchema(name="section_title", dtype=DataType.VARCHAR, max_length=512),
+        FieldSchema(name="section_path", dtype=DataType.VARCHAR, max_length=4096),
+        FieldSchema(name="page_number", dtype=DataType.INT64),
+        FieldSchema(name="regulation_type", dtype=DataType.VARCHAR, max_length=128),
+        FieldSchema(name="version", dtype=DataType.VARCHAR, max_length=64),
+        FieldSchema(name="semantic_id", dtype=DataType.VARCHAR, max_length=128),
+        FieldSchema(name="block_type", dtype=DataType.VARCHAR, max_length=64),
+        FieldSchema(name="metadata_json", dtype=DataType.VARCHAR, max_length=65535),
+        FieldSchema(name="created_at", dtype=DataType.INT64),
+    ],
+    description="Dense-only regulations index",
+    enable_dynamic_field=False,
+)
+```
+
+这说明 Milvus 存的不是“只有 embedding 的极简向量表”，而是：
+
+- 一个 dense vector
+- 一组检索时要返回或过滤的结构化字段
+
+但要注意：这并不意味着 Milvus 是业务主记录库。它仍然主要服务于检索，而不是替代 PostgreSQL 的文档管理职责。
+
+### 5.2 `list_documents()` 为什么会先看 Milvus
+
+文档列表查询在 [services.py](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/backend/app/application/documents/services.py:271) 中实现，它会：
+
+1. 从 Milvus 查询当前真的存在向量的文档
+2. 从文档元数据仓储加载文档记录
+3. 以 Milvus 为索引状态真相源进行 merge
+
+原因不是“Milvus 替代 PostgreSQL”，而是：
+
+- `indexed` 这个状态最终是否真实成立，要看 Milvus 里有没有对应 chunk
+- 但下载、删除、重试、文件定位、错误信息仍然要依赖文档元数据仓储
+
+所以：
+
+- Milvus 是“索引真相源”
+- PostgreSQL/JSON 是“文档元数据真相源”
+
+这两者职责不同，不能互相替代。

docs/architecture/document-processing-database-design.md

@@ -0,0 +1,508 @@
+# 文档处理链路数据库设计
+
+## 1. Purpose
+
+本文档定义当前文档处理主链路的 PostgreSQL 数据库设计，覆盖上传、解析、索引、状态查询、重试、删除这条核心链路，以及围绕该链路的常用运维与审计需求。
+
+本文档的目标不是替代 [document-core-processing-flow.md](/abs/path/C:/Users/A200477427/Developers/AIRegulation/AIRegulation-DocAnalysis/docs/architecture/document-core-processing-flow.md:1) 的流程说明，而是补齐关系型存储的 authority，使后续从 JSON 元数据切换到 PostgreSQL 时有清晰、稳定、可实施的数据库设计基线。
+
+## 1.1 Scope And Design Target
+
+本文档只覆盖以下范围：
+
+- 文档主记录
+- 文档处理运行记录
+- 文档状态历史
+- 解析产物引用
+- 当前最新结构化解析快照
+
+本文档不覆盖以下范围：
+
+- Agent 会话
+- 反馈和人工审核
+- 合规分析任务
+- Milvus collection schema 的详细实现
+
+设计原则采用 `Compat First`：
+
+- 保持与当前 `DocumentRepository` / `ParseArtifactStore` 主流程兼容
+- 新增关系表以补足运维与审计能力
+- 不为了理想化模型而反推大规模接口重写
+
+## 2. Storage Responsibilities
+
+当前系统采用三类存储，各自职责必须清晰分离：
+
+| 存储 | 保存内容 | 是否业务主记录 | 说明 |
+| --- | --- | --- | --- |
+| MinIO | 原始文件、`layouts.json`、`structure_nodes.json`、`semantic_blocks.json`、`vector_chunks.json` | 否 | 负责大对象与产物归档，不承担关系查询 |
+| Milvus | chunk 级向量和检索辅助字段 | 否 | 负责向量检索，不承担文档生命周期管理 |
+| PostgreSQL | 文档元数据、处理状态、结构化快照、处理历史、artifact 引用 | 是 | 负责文档管理、运维可观测性和关系查询 |
+
+约束说明：
+
+- PostgreSQL 不保存 embedding 向量。
+- PostgreSQL 不新增 `vector_chunks` 内容表。
+- Milvus 可以保存 `doc_id`、`doc_name`、`regulation_type`、`version` 等检索辅助字段，但不是业务真相源。
+- 文档下载、删除、重试仍以 PostgreSQL 中的文档主记录为入口。
+
+## 3. Design Overview
+
+### 3.1 Entity Responsibilities
+
+数据库采用“当前态主记录 + 当前快照 + 历史过程”的分层模型：
+
+- `documents`
+  - 当前文档主记录
+  - 保存供管理、下载、重试、删除直接使用的元数据和当前状态
+- `document_processing_runs`
+  - 每次上传或重试对应一次处理运行
+  - 保存运行级统计、阶段时间点和失败信息
+- `document_status_history`
+  - 追加式状态事件流
+  - 保存每次状态变更的上下文
+- `document_artifacts`
+  - 保存 MinIO artifact 的引用信息
+  - 不保存 artifact 内容本体
+- `structure_nodes`
+  - 当前最新解析快照中的目录结构
+- `semantic_blocks`
+  - 当前最新解析快照中的语义块结构
+
+### 3.2 Current Snapshot Vs Historical Records
+
+本设计显式区分两类数据：
+
+- 当前快照
+  - `documents`
+  - `structure_nodes`
+  - `semantic_blocks`
+- 历史过程
+  - `document_processing_runs`
+  - `document_status_history`
+  - `document_artifacts`
+
+其中：
+
+- `structure_nodes` 和 `semantic_blocks` 只保存“最新一次成功解析后”的当前快照
+- 历史版本回溯依赖 `document_processing_runs`、`document_artifacts` 和 MinIO 中对应 run 的 artifact 文件
+
+## 4. Table Design
+
+### 4.1 `documents`
+
+用途：
+
+- 作为文档生命周期的主记录表
+- 为下载、删除、重试、管理列表、状态查询提供当前态真相
+
+字段设计：
+
+```sql
+CREATE TABLE IF NOT EXISTS documents (
+    doc_id              VARCHAR(128) PRIMARY KEY,
+    doc_name            VARCHAR(512)  NOT NULL DEFAULT '',
+    file_name           VARCHAR(512)  NOT NULL DEFAULT '',
+    object_name         VARCHAR(1024) NOT NULL DEFAULT '',
+    content_type        VARCHAR(128)  NOT NULL DEFAULT '',
+    size_bytes          BIGINT        NOT NULL DEFAULT 0,
+    status              VARCHAR(32)   NOT NULL DEFAULT 'pending',
+    regulation_type     VARCHAR(128)  NOT NULL DEFAULT '',
+    version             VARCHAR(64)   NOT NULL DEFAULT '',
+    summary             TEXT          NOT NULL DEFAULT '',
+    summary_latency_ms  INTEGER       NOT NULL DEFAULT 0,
+    chunk_count         INTEGER       NOT NULL DEFAULT 0,
+    parser_name         VARCHAR(128)  NOT NULL DEFAULT '',
+    index_name          VARCHAR(128)  NOT NULL DEFAULT '',
+    error_message       TEXT          NOT NULL DEFAULT '',
+    metadata            JSONB         NOT NULL DEFAULT '{}'::jsonb,
+    created_at          TIMESTAMPTZ   NOT NULL DEFAULT NOW(),
+    updated_at          TIMESTAMPTZ   NOT NULL DEFAULT NOW(),
+    CONSTRAINT chk_documents_status
+        CHECK (status IN ('pending', 'stored', 'parsed', 'indexed', 'failed'))
+);
+
+CREATE INDEX IF NOT EXISTS idx_documents_status_updated_at
+    ON documents(status, updated_at DESC);
+
+CREATE INDEX IF NOT EXISTS idx_documents_regulation_version
+    ON documents(regulation_type, version);
+
+CREATE INDEX IF NOT EXISTS idx_documents_updated_at
+    ON documents(updated_at DESC);
+```
+
+字段说明：
+
+- `object_name`
+  - 原始上传文件在 MinIO 中的对象路径
+  - 当前实现依赖该字段完成下载、重试和删除，v1 不拆分为独立文件表
+- `status`
+  - 当前文档处理状态
+  - 仅表示当前态，不承担历史审计职责
+- `metadata`
+  - 保存轻量、变动频率较高、暂不值得列式建模的附加信息
+  - 典型内容包括 `parse_task_id`、`processing_stage`、`artifact_keys`、统计计数等
+
+### 4.2 `document_processing_runs`
+
+用途：
+
+- 记录一次上传或一次重试的完整处理运行
+- 用于解释“这份文档本次处理为什么成功或失败”
+
+字段设计：
+
+```sql
+CREATE TABLE IF NOT EXISTS document_processing_runs (
+    run_id                 BIGSERIAL PRIMARY KEY,
+    doc_id                 VARCHAR(128) NOT NULL,
+    trigger_type           VARCHAR(16)  NOT NULL,
+    run_status             VARCHAR(16)  NOT NULL,
+    parser_backend         VARCHAR(64)  NOT NULL DEFAULT '',
+    chunk_backend          VARCHAR(64)  NOT NULL DEFAULT '',
+    embedding_model        VARCHAR(128) NOT NULL DEFAULT '',
+    index_name             VARCHAR(128) NOT NULL DEFAULT '',
+    started_at             TIMESTAMPTZ  NOT NULL DEFAULT NOW(),
+    stored_at              TIMESTAMPTZ,
+    parsed_at              TIMESTAMPTZ,
+    indexed_at             TIMESTAMPTZ,
+    finished_at            TIMESTAMPTZ,
+    layout_count           INTEGER      NOT NULL DEFAULT 0,
+    structure_node_count   INTEGER      NOT NULL DEFAULT 0,
+    semantic_block_count   INTEGER      NOT NULL DEFAULT 0,
+    vector_chunk_count     INTEGER      NOT NULL DEFAULT 0,
+    chunk_count            INTEGER      NOT NULL DEFAULT 0,
+    failure_stage          VARCHAR(32)  NOT NULL DEFAULT '',
+    error_message          TEXT         NOT NULL DEFAULT '',
+    metadata               JSONB        NOT NULL DEFAULT '{}'::jsonb,
+    CONSTRAINT fk_runs_document
+        FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE,
+    CONSTRAINT chk_runs_trigger_type
+        CHECK (trigger_type IN ('upload', 'retry')),
+    CONSTRAINT chk_runs_status
+        CHECK (run_status IN ('running', 'succeeded', 'failed'))
+);
+
+CREATE INDEX IF NOT EXISTS idx_runs_doc_started_at
+    ON document_processing_runs(doc_id, started_at DESC);
+
+CREATE INDEX IF NOT EXISTS idx_runs_status_started_at
+    ON document_processing_runs(run_status, started_at DESC);
+```
+
+字段说明：
+
+- `trigger_type`
+  - 标识该次处理由首次上传还是 retry 触发
+- `run_status`
+  - 只表示该次运行的最终结果
+- `failure_stage`
+  - 建议取值与应用层关键阶段一致，例如 `store`、`parse`、`artifact_persist`、`embed`、`index`
+- `metadata`
+  - 保存运行级附加上下文，例如配置快照、后端实现名、provider 返回信息摘要
+
+### 4.3 `document_status_history`
+
+用途：
+
+- 保存状态变化事件流
+- 用于排障、审计和运行轨迹分析
+
+字段设计：
+
+```sql
+CREATE TABLE IF NOT EXISTS document_status_history (
+    event_id         BIGSERIAL PRIMARY KEY,
+    doc_id           VARCHAR(128) NOT NULL,
+    run_id           BIGINT,
+    from_status      VARCHAR(32)  NOT NULL DEFAULT '',
+    to_status        VARCHAR(32)  NOT NULL,
+    stage            VARCHAR(32)  NOT NULL DEFAULT '',
+    message          TEXT         NOT NULL DEFAULT '',
+    metadata         JSONB        NOT NULL DEFAULT '{}'::jsonb,
+    occurred_at      TIMESTAMPTZ  NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_status_document
+        FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE,
+    CONSTRAINT fk_status_run
+        FOREIGN KEY (run_id) REFERENCES document_processing_runs(run_id) ON DELETE CASCADE,
+    CONSTRAINT chk_status_history_to_status
+        CHECK (to_status IN ('pending', 'stored', 'parsed', 'indexed', 'failed'))
+);
+
+CREATE INDEX IF NOT EXISTS idx_status_history_doc_occurred_at
+    ON document_status_history(doc_id, occurred_at DESC);
+
+CREATE INDEX IF NOT EXISTS idx_status_history_run_occurred_at
+    ON document_status_history(run_id, occurred_at DESC);
+```
+
+字段说明：
+
+- `from_status` 可以为空字符串
+  - 用于首个事件，例如文档创建时进入 `pending`
+- `stage`
+  - 用于记录状态推进对应的业务阶段
+- `message`
+  - 用于记录面向排障的人类可读说明
+
+### 4.4 `document_artifacts`
+
+用途：
+
+- 保存解析产物在 MinIO 中的位置与基本属性
+- 支持后续定位某次 run 的 artifacts，而不扫描对象存储
+
+字段设计：
+
+```sql
+CREATE TABLE IF NOT EXISTS document_artifacts (
+    artifact_id       BIGSERIAL PRIMARY KEY,
+    doc_id            VARCHAR(128)  NOT NULL,
+    run_id            BIGINT,
+    artifact_type     VARCHAR(32)   NOT NULL,
+    object_name       VARCHAR(1024) NOT NULL,
+    content_type      VARCHAR(128)  NOT NULL DEFAULT 'application/json',
+    byte_size         BIGINT        NOT NULL DEFAULT 0,
+    checksum          VARCHAR(128)  NOT NULL DEFAULT '',
+    metadata          JSONB         NOT NULL DEFAULT '{}'::jsonb,
+    created_at        TIMESTAMPTZ   NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_artifacts_document
+        FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE,
+    CONSTRAINT fk_artifacts_run
+        FOREIGN KEY (run_id) REFERENCES document_processing_runs(run_id) ON DELETE CASCADE,
+    CONSTRAINT chk_artifact_type
+        CHECK (artifact_type IN ('layouts', 'structure_nodes', 'semantic_blocks', 'vector_chunks'))
+);
+
+CREATE INDEX IF NOT EXISTS idx_artifacts_doc_created_at
+    ON document_artifacts(doc_id, created_at DESC);
+
+CREATE INDEX IF NOT EXISTS idx_artifacts_run_type
+    ON document_artifacts(run_id, artifact_type);
+
+CREATE UNIQUE INDEX IF NOT EXISTS uq_artifacts_run_type_object
+    ON document_artifacts(run_id, artifact_type, object_name);
+```
+
+字段说明：
+
+- 该表只记录 artifact 引用，不记录原始文件
+- 原始文件仍由 `documents.object_name` 表达，这是为了保持当前下载和重试逻辑兼容
+
+### 4.5 `structure_nodes`
+
+用途：
+
+- 保存当前最新解析快照中的标题层级结构
+- 供目录树查询、结构化浏览、调试和审计使用
+
+字段设计：
+
+```sql
+CREATE TABLE IF NOT EXISTS structure_nodes (
+    id             BIGSERIAL PRIMARY KEY,
+    doc_id         VARCHAR(128) NOT NULL,
+    unique_id      VARCHAR(128),
+    page           INTEGER      NOT NULL DEFAULT 0,
+    idx            INTEGER      NOT NULL DEFAULT 0,
+    level          INTEGER      NOT NULL DEFAULT 0,
+    title          TEXT         NOT NULL DEFAULT '',
+    type           VARCHAR(64),
+    sub_type       VARCHAR(64),
+    created_at     TIMESTAMPTZ  NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_structure_nodes_document
+        FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE
+);
+
+CREATE INDEX IF NOT EXISTS idx_structure_nodes_doc_idx
+    ON structure_nodes(doc_id, idx);
+
+CREATE INDEX IF NOT EXISTS idx_structure_nodes_doc_level
+    ON structure_nodes(doc_id, level);
+```
+
+设计约束：
+
+- 该表表示当前快照，不做多版本建模
+- 新一轮成功解析会覆盖同一 `doc_id` 的旧快照
+
+### 4.6 `semantic_blocks`
+
+用途：
+
+- 保存当前最新解析快照中的语义块
+- 供结构回溯、调试和后续关系型查询使用
+
+字段设计：
+
+```sql
+CREATE TABLE IF NOT EXISTS semantic_blocks (
+    id              BIGSERIAL PRIMARY KEY,
+    doc_id          VARCHAR(128) NOT NULL,
+    semantic_id     VARCHAR(128) NOT NULL,
+    block_type      VARCHAR(64)  NOT NULL DEFAULT '',
+    page_start      INTEGER      NOT NULL DEFAULT 0,
+    page_end        INTEGER      NOT NULL DEFAULT 0,
+    section_path    JSONB        NOT NULL DEFAULT '[]'::jsonb,
+    section_level   INTEGER      NOT NULL DEFAULT 0,
+    section_title   VARCHAR(512) NOT NULL DEFAULT '',
+    source_ids      JSONB        NOT NULL DEFAULT '[]'::jsonb,
+    text            TEXT         NOT NULL DEFAULT '',
+    created_at      TIMESTAMPTZ  NOT NULL DEFAULT NOW(),
+    CONSTRAINT fk_semantic_blocks_document
+        FOREIGN KEY (doc_id) REFERENCES documents(doc_id) ON DELETE CASCADE,
+    CONSTRAINT uq_semantic_blocks_doc_semantic
+        UNIQUE (doc_id, semantic_id)
+);
+
+CREATE INDEX IF NOT EXISTS idx_semantic_blocks_doc_id
+    ON semantic_blocks(doc_id);
+
+CREATE INDEX IF NOT EXISTS idx_semantic_blocks_doc_section_title
+    ON semantic_blocks(doc_id, section_title);
+
+CREATE INDEX IF NOT EXISTS idx_semantic_blocks_doc_block_type
+    ON semantic_blocks(doc_id, block_type);
+```
+
+设计约束：
+
+- 该表表示当前快照，不保存历史版本
+- 历史回溯应通过 run 对应的 artifact 文件完成
+
+## 5. Relationship Model
+
+实体关系如下：
+
+```mermaid
+erDiagram
+    documents ||--o{ document_processing_runs : has
+    documents ||--o{ document_status_history : has
+    documents ||--o{ document_artifacts : has
+    documents ||--o{ structure_nodes : has
+    documents ||--o{ semantic_blocks : has
+    document_processing_runs ||--o{ document_status_history : emits
+    document_processing_runs ||--o{ document_artifacts : produces
+```
+
+关系语义：
+
+- `documents` 是聚合根
+- `document_processing_runs` 记录一次完整处理尝试
+- `document_status_history` 记录状态推进轨迹
+- `document_artifacts` 记录 MinIO 中可回放的结构化产物
+- `structure_nodes` / `semantic_blocks` 代表“当前版本”的关系型快照
+
+## 6. Flow-To-Table Mapping
+
+### 6.1 Upload
+
+上传开始时：
+
+1. 创建 `documents`
+2. 创建一条 `document_processing_runs`
+3. 写入一条 `document_status_history`，`to_status='pending'`
+
+### 6.2 Store Original File
+
+原始文件写入 MinIO 成功后：
+
+1. 更新 `documents.status='stored'`
+2. 更新当前 run 的 `stored_at`
+3. 追加 `document_status_history`
+
+### 6.3 Parse And Persist Artifacts
+
+解析成功后：
+
+1. 更新当前 run 的 `parsed_at`
+2. 更新 run 的 `layout_count`、`structure_node_count`、`semantic_block_count`、`vector_chunk_count`
+3. 更新 `documents.status='parsed'`
+4. 刷新 `structure_nodes`
+5. 刷新 `semantic_blocks`
+6. 为 `layouts`、`structure_nodes`、`semantic_blocks`、`vector_chunks` 写入 `document_artifacts`
+7. 追加 `document_status_history`
+
+### 6.4 Embed And Index
+
+向量化和入库成功后：
+
+1. 更新当前 run 的 `indexed_at`、`finished_at`
+2. 更新当前 run 的 `run_status='succeeded'`
+3. 更新 `documents.status='indexed'`
+4. 更新 `documents.chunk_count`、`index_name`
+5. 追加 `document_status_history`
+
+### 6.5 Failure
+
+任一阶段失败时：
+
+1. 更新当前 run 的 `run_status='failed'`
+2. 记录 `failure_stage` 和 `error_message`
+3. 更新 `finished_at`
+4. 更新 `documents.status='failed'`
+5. 更新 `documents.error_message`
+6. 追加 `document_status_history`
+
+### 6.6 Retry
+
+重试时：
+
+1. 保留现有 `documents.doc_id`
+2. 新建一条 `document_processing_runs`
+3. 为本次重试重新写入状态历史
+4. 本次重试成功后覆盖 `structure_nodes` / `semantic_blocks` 当前快照
+5. 历史 run 和 artifact 记录继续保留
+
+### 6.7 Delete
+
+删除文档时：
+
+1. 应用层先删除 MinIO 原始文件和 artifacts
+2. 应用层删除 Milvus 中按 `doc_id` 关联的向量
+3. 最后删除 `documents`
+4. 依赖外键 `ON DELETE CASCADE` 清理 run、status history、artifacts、structure nodes、semantic blocks
+
+## 7. Alignment With Current Backend
+
+### 7.1 Compatible Parts
+
+当前代码已天然兼容以下设计：
+
+- `documents`
+- `structure_nodes`
+- `semantic_blocks`
+- 当前快照覆盖式更新
+- `doc_id` 作为跨 MinIO / Milvus / PostgreSQL 的统一关联键
+
+### 7.2 Required Future Additions
+
+若后续正式切到 PostgreSQL 默认元数据后端，应新增以下内部 store 或 repository：
+
+- `DocumentProcessingRunStore`
+- `DocumentStatusEventStore`
+- `DocumentArtifactStore`
+
+这些新增能力属于内部增强，不要求修改现有 HTTP API。
+
+### 7.3 Migration Guidance
+
+从当前 JSON 元数据切换到 PostgreSQL 时，建议按以下顺序进行：
+
+1. 迁移 `documents.json` 中已有文档主记录到 `documents`
+2. 将 `DOCUMENT_REPOSITORY_BACKEND` 切换为 `postgres`
+3. 为新上传或重试的文档开始写入 run / status history / artifact records
+4. 历史文档若缺少 run 级数据，可允许为空，不阻塞切换
+
+## 8. Non-Goals
+
+以下能力不在本设计 v1 范围内：
+
+- 将 Milvus 替换为 PostgreSQL 向量能力
+- 在 PostgreSQL 中保存向量字段
+- 为 `vector_chunks` 建独立关系表
+- 为 `structure_nodes` / `semantic_blocks` 建历史版本仓库
+- 将原始文件抽象成独立 `document_files` 表
+
+这些能力可能在后续重构时被讨论，但不应影响当前主链路切换和现有应用层兼容性。

frontend/.env

@@ -0,0 +1,2 @@
+VITE_API_PROXY_TARGET=http://6.86.80.8:8000
+FRONTEND_PORT=5173

frontend/.env.development

@@ -0,0 +1,2 @@
+VITE_API_PROXY_TARGET=http://127.0.0.1:8000
+FRONTEND_PORT=5173

frontend/.env.example

@@ -0,0 +1,2 @@
+VITE_API_PROXY_TARGET=http://127.0.0.1:8000
+FRONTEND_PORT=5173

frontend/README.md

@@ -49,6 +49,12 @@ npm run dev
 
 启动本地开发服务器，默认访问 `http://localhost:5173`
 
+前端环境文件约定如下：
+
+- `frontend/.env.development`：本地开发，默认代理到 `http://127.0.0.1:8000`
+- `frontend/.env.production`：生产构建，默认代理到 `http://6.86.80.8:8000`
+- `frontend/.env.local`：临时覆盖本机配置，优先级高于上面两者
+
 ### 构建生产版本
 
 ```bash

frontend/components.json

@@ -0,0 +1,25 @@
+{
+  "$schema": "https://ui.shadcn.com/schema.json",
+  "style": "radix-nova",
+  "rsc": false,
+  "tsx": true,
+  "tailwind": {
+    "config": "tailwind.config.js",
+    "css": "src/styles/globals.css",
+    "baseColor": "neutral",
+    "cssVariables": true,
+    "prefix": ""
+  },
+  "iconLibrary": "lucide",
+  "rtl": false,
+  "aliases": {
+    "components": "@/components",
+    "utils": "@/lib/utils",
+    "ui": "@/components/shadcn/ui",
+    "lib": "@/lib",
+    "hooks": "@/hooks"
+  },
+  "menuColor": "default",
+  "menuAccent": "subtle",
+  "registries": {}
+}

frontend/components/ui/button.tsx

@@ -0,0 +1,67 @@
+import * as React from "react"
+import { cva, type VariantProps } from "class-variance-authority"
+import { Slot } from "radix-ui"
+
+import { cn } from "@/lib/utils"
+
+const buttonVariants = cva(
+  "group/button inline-flex shrink-0 items-center justify-center rounded-lg border border-transparent bg-clip-padding text-sm font-medium whitespace-nowrap transition-all outline-none select-none focus-visible:border-ring focus-visible:ring-3 focus-visible:ring-ring/50 active:not-aria-[haspopup]:translate-y-px disabled:pointer-events-none disabled:opacity-50 aria-invalid:border-destructive aria-invalid:ring-3 aria-invalid:ring-destructive/20 dark:aria-invalid:border-destructive/50 dark:aria-invalid:ring-destructive/40 [&_svg]:pointer-events-none [&_svg]:shrink-0 [&_svg:not([class*='size-'])]:size-4",
+  {
+    variants: {
+      variant: {
+        default: "bg-primary text-primary-foreground [a]:hover:bg-primary/80",
+        outline:
+          "border-border bg-background hover:bg-muted hover:text-foreground aria-expanded:bg-muted aria-expanded:text-foreground dark:border-input dark:bg-input/30 dark:hover:bg-input/50",
+        secondary:
+          "bg-secondary text-secondary-foreground hover:bg-secondary/80 aria-expanded:bg-secondary aria-expanded:text-secondary-foreground",
+        ghost:
+          "hover:bg-muted hover:text-foreground aria-expanded:bg-muted aria-expanded:text-foreground dark:hover:bg-muted/50",
+        destructive:
+          "bg-destructive/10 text-destructive hover:bg-destructive/20 focus-visible:border-destructive/40 focus-visible:ring-destructive/20 dark:bg-destructive/20 dark:hover:bg-destructive/30 dark:focus-visible:ring-destructive/40",
+        link: "text-primary underline-offset-4 hover:underline",
+      },
+      size: {
+        default:
+          "h-8 gap-1.5 px-2.5 has-data-[icon=inline-end]:pr-2 has-data-[icon=inline-start]:pl-2",
+        xs: "h-6 gap-1 rounded-[min(var(--radius-md),10px)] px-2 text-xs in-data-[slot=button-group]:rounded-lg has-data-[icon=inline-end]:pr-1.5 has-data-[icon=inline-start]:pl-1.5 [&_svg:not([class*='size-'])]:size-3",
+        sm: "h-7 gap-1 rounded-[min(var(--radius-md),12px)] px-2.5 text-[0.8rem] in-data-[slot=button-group]:rounded-lg has-data-[icon=inline-end]:pr-1.5 has-data-[icon=inline-start]:pl-1.5 [&_svg:not([class*='size-'])]:size-3.5",
+        lg: "h-9 gap-1.5 px-2.5 has-data-[icon=inline-end]:pr-2 has-data-[icon=inline-start]:pl-2",
+        icon: "size-8",
+        "icon-xs":
+          "size-6 rounded-[min(var(--radius-md),10px)] in-data-[slot=button-group]:rounded-lg [&_svg:not([class*='size-'])]:size-3",
+        "icon-sm":
+          "size-7 rounded-[min(var(--radius-md),12px)] in-data-[slot=button-group]:rounded-lg",
+        "icon-lg": "size-9",
+      },
+    },
+    defaultVariants: {
+      variant: "default",
+      size: "default",
+    },
+  }
+)
+
+function Button({
+  className,
+  variant = "default",
+  size = "default",
+  asChild = false,
+  ...props
+}: React.ComponentProps<"button"> &
+  VariantProps<typeof buttonVariants> & {
+    asChild?: boolean
+  }) {
+  const Comp = asChild ? Slot.Root : "button"
+
+  return (
+    <Comp
+      data-slot="button"
+      data-variant={variant}
+      data-size={size}
+      className={cn(buttonVariants({ variant, size, className }))}
+      {...props}
+    />
+  )
+}
+
+export { Button }

frontend/package-lock.json

@@ -9,7 +9,8 @@
       "version": "0.0.0",
       "dependencies": {
         "react": "^19.2.5",
-        "react-dom": "^19.2.5"
+        "react-dom": "^19.2.5",
+        "react-router-dom": "^7.9.6"
       },
       "devDependencies": {
         "@eslint/js": "^10.0.1",
@@ -1631,6 +1632,19 @@
       "dev": true,
       "license": "MIT"
     },
+    "node_modules/cookie": {
+      "version": "1.1.1",
+      "resolved": "https://registry.npmjs.org/cookie/-/cookie-1.1.1.tgz",
+      "integrity": "sha512-ei8Aos7ja0weRpFzJnEA9UHJ/7XQmqglbRwnf2ATjcB9Wq874VKH9kfjjirM6UhU2/E5fFYadylyhFldcqSidQ==",
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/express"
+      }
+    },
     "node_modules/cross-spawn": {
       "version": "7.0.6",
       "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.6.tgz",
@@ -2751,6 +2765,44 @@
         "react": "^19.2.5"
       }
     },
+    "node_modules/react-router": {
+      "version": "7.15.1",
+      "resolved": "https://registry.npmjs.org/react-router/-/react-router-7.15.1.tgz",
+      "integrity": "sha512-R8rl9HhgikFYoPJymnUtPXWbnDb3oget6lQnfIoupbt61aT9aOhRkDsY2XRhZRyX1Z/8a5sL74fXmFNm3NRK5A==",
+      "license": "MIT",
+      "dependencies": {
+        "cookie": "^1.0.1",
+        "set-cookie-parser": "^2.6.0"
+      },
+      "engines": {
+        "node": ">=20.0.0"
+      },
+      "peerDependencies": {
+        "react": ">=18",
+        "react-dom": ">=18"
+      },
+      "peerDependenciesMeta": {
+        "react-dom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/react-router-dom": {
+      "version": "7.15.1",
+      "resolved": "https://registry.npmjs.org/react-router-dom/-/react-router-dom-7.15.1.tgz",
+      "integrity": "sha512-AzF62gjY6U9rkMq4RfP/r2EVtQ7DMfNMjyOp/flLTCrtRylLiK4wT4pSq6O8rOXZ2eXdZYJPEYe+ifomiv+Igg==",
+      "license": "MIT",
+      "dependencies": {
+        "react-router": "7.15.1"
+      },
+      "engines": {
+        "node": ">=20.0.0"
+      },
+      "peerDependencies": {
+        "react": ">=18",
+        "react-dom": ">=18"
+      }
+    },
     "node_modules/rolldown": {
       "version": "1.0.0-rc.17",
       "resolved": "https://registry.npmjs.org/rolldown/-/rolldown-1.0.0-rc.17.tgz",
@@ -2808,6 +2860,12 @@
         "semver": "bin/semver.js"
       }
     },
+    "node_modules/set-cookie-parser": {
+      "version": "2.7.2",
+      "resolved": "https://registry.npmjs.org/set-cookie-parser/-/set-cookie-parser-2.7.2.tgz",
+      "integrity": "sha512-oeM1lpU/UvhTxw+g3cIfxXHyJRc/uidd3yK1P242gzHds0udQBYzs3y8j4gCCW+ZJ7ad0yctld8RYO+bdurlvw==",
+      "license": "MIT"
+    },
     "node_modules/shebang-command": {
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/shebang-command/-/shebang-command-2.0.0.tgz",

frontend/package.json

@@ -10,8 +10,17 @@
     "preview": "vite preview"
   },
   "dependencies": {
+    "@fontsource-variable/geist": "^5.2.9",
+    "class-variance-authority": "^0.7.1",
+    "clsx": "^2.1.1",
+    "lucide-react": "^1.16.0",
+    "radix-ui": "^1.4.3",
     "react": "^19.2.5",
-    "react-dom": "^19.2.5"
+    "react-dom": "^19.2.5",
+    "react-router-dom": "^7.9.6",
+    "shadcn": "^4.8.0",
+    "tailwind-merge": "^3.6.0",
+    "tw-animate-css": "^1.4.0"
   },
   "devDependencies": {
     "@eslint/js": "^10.0.1",

frontend/src/App.tsx

@@ -1,46 +1,11 @@
 import './styles/globals.css';
-import { ThemeProvider, AppProvider, useApp, useTheme } from './contexts';
-import { Header, Tabs } from './components/layout';
-import { CompliancePage } from './pages/Compliance';
-import { DocsPage } from './pages/Docs';
-import { StatusPage } from './pages/Status';
-import { RagChatPage } from './pages/RagChat';
-
-const PageContent = () => {
-  const { activeTab } = useApp();
-
-  switch (activeTab) {
-    case 'docs':
-      return <DocsPage />;
-    case 'compliance':
-      return <CompliancePage />;
-    case 'status':
-      return <StatusPage />;
-    case 'rag':
-      return <RagChatPage />;
-    default:
-      return <CompliancePage />;
-  }
-};
-
-const AppContent = () => {
-  const { theme } = useTheme();
-
-  return (
-    <div className="h-full flex flex-col min-h-screen" style={{ backgroundColor: theme.bg }}>
-      <Header />
-      <Tabs />
-      <PageContent />
-    </div>
-  );
-};
+import { ThemeProvider } from './contexts';
+import { AppRouter } from './router/AppRouter';
 
 function App() {
   return (
     <ThemeProvider>
-      <AppProvider>
-        <AppContent />
-      </AppProvider>
+      <AppRouter />
     </ThemeProvider>
   );
 }

frontend/src/api/perception.ts

@@ -0,0 +1,128 @@
+const PERCEPTION_API_BASE = '/api/v1';
+
+export type ImpactLevel = 'high' | 'medium' | 'low';
+export type EventStatus = 'enacted' | 'draft' | 'consultation';
+export type EventSource = 'MIIT' | 'UN-ECE' | 'ISO' | '国标委' | 'EUR-Lex' | 'IATF';
+
+export interface RegulationEvent {
+  id: string;
+  source: EventSource;
+  source_label: string;
+  standard_code: string;
+  title: string;
+  summary: string;
+  impact_level: ImpactLevel;
+  published_at: string;
+  effective_at: string | null;
+  category: string;
+  tags: string[];
+  source_url: string;
+  status: EventStatus;
+}
+
+export interface PerceptionStats {
+  total: number;
+  high_impact: number;
+  medium_impact: number;
+  low_impact: number;
+  recent_90d: number;
+}
+
+export interface EventListResponse {
+  events: RegulationEvent[];
+  total: number;
+}
+
+export interface AffectedDoc {
+  doc_id: string;
+  doc_name: string;
+  score: number;
+  snippet: string;
+  clause: string;
+}
+
+export interface AnalysisSSEMessage {
+  type: 'sources' | 'content' | 'done' | 'error';
+  docs?: AffectedDoc[];
+  text?: string;
+}
+
+export async function getPerceptionStats(): Promise<PerceptionStats> {
+  const res = await fetch(`${PERCEPTION_API_BASE}/perception/stats`);
+  if (!res.ok) throw new Error(`stats failed: ${res.status}`);
+  return res.json() as Promise<PerceptionStats>;
+}
+
+export async function listEvents(params?: {
+  source?: string;
+  impact_level?: string;
+  limit?: number;
+}): Promise<EventListResponse> {
+  const query = new URLSearchParams();
+  if (params?.source) query.set('source', params.source);
+  if (params?.impact_level) query.set('impact_level', params.impact_level);
+  if (params?.limit) query.set('limit', String(params.limit));
+  const res = await fetch(`${PERCEPTION_API_BASE}/perception/events?${query.toString()}`);
+  if (!res.ok) throw new Error(`list events failed: ${res.status}`);
+  return res.json() as Promise<EventListResponse>;
+}
+
+export async function analyzeEvent(
+  eventId: string,
+  onMessage: (msg: AnalysisSSEMessage) => void,
+  onComplete?: () => void,
+  signal?: AbortSignal,
+): Promise<void> {
+  try {
+    const res = await fetch(`${PERCEPTION_API_BASE}/perception/events/${eventId}/analyze`, {
+      method: 'POST',
+      headers: { Accept: 'text/event-stream' },
+      signal,
+    });
+    if (!res.ok || !res.body) throw new Error(`analyze failed: ${res.status}`);
+
+    const reader = res.body.getReader();
+    const decoder = new TextDecoder();
+    let buffer = '';
+
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      const parts = buffer.split('\n\n');
+      buffer = parts.pop() ?? '';
+      for (const block of parts) {
+        if (!block.trim()) continue;
+        let eventName = 'message';
+        const dataLines: string[] = [];
+        for (const line of block.split('\n')) {
+          if (line.startsWith('event:')) eventName = line.slice(6).trim();
+          else if (line.startsWith('data:')) dataLines.push(line.slice(5).trim());
+        }
+        const payload = dataLines.join('\n');
+        if (!payload) continue;
+
+        if (eventName === 'sources') {
+          try {
+            const docs = JSON.parse(payload) as AffectedDoc[];
+            onMessage({ type: 'sources', docs });
+          } catch { /* ignore */ }
+        } else if (eventName === 'content') {
+          onMessage({ type: 'content', text: payload });
+        } else if (eventName === 'done') {
+          onMessage({ type: 'done' });
+        } else if (eventName === 'error') {
+          onMessage({ type: 'error', text: payload });
+        }
+      }
+    }
+    if (buffer.trim()) {
+      // flush remaining
+    }
+    onComplete?.();
+  } catch (err) {
+    if (err instanceof DOMException && err.name === 'AbortError') return;
+    onMessage({ type: 'error', text: err instanceof Error ? err.message : String(err) });
+    onComplete?.();
+  }
+}

frontend/src/components/common/ThemeToggle.tsx

@@ -1,35 +1,34 @@
 import React from 'react';
+import { Moon, Sun, SunMedium } from 'lucide-react';
+
 import { useTheme } from '../../contexts';
+import { Button } from '../shadcn/ui/button';
+
+const NEXT_LABELS: Record<string, string> = {
+  dark: '过渡色模式',
+  dim: '亮色模式',
+  light: '暗色模式',
+};
 
 export const ThemeToggle: React.FC = () => {
-  const { isDark, toggleTheme, theme } = useTheme();
+  const { themeMode, toggleTheme } = useTheme();
+
+  // Shows the NEXT state's icon: dark→SunMedium(dim next), dim→Sun(light next), light→Moon(dark next)
+  const Icon =
+    themeMode === 'dark' ? SunMedium :
+    themeMode === 'dim'  ? Sun :
+    Moon;
 
   return (
-    <button
+    <Button
       onClick={toggleTheme}
-      style={{
-        width: 44,
-        height: 44,
-        borderRadius: 10,
-        background: isDark ? theme.bgHover : theme.bgCard,
-        border: `1px solid ${theme.border}`,
-        cursor: 'pointer',
-        display: 'flex',
-        alignItems: 'center',
-        justifyContent: 'center',
-        transition: 'all 0.3s ease',
-      }}
+      variant="outline"
+      size="icon-lg"
+      className="rounded-xl border-border bg-card text-muted-foreground hover:bg-muted hover:text-foreground"
+      aria-label={`切换到${NEXT_LABELS[themeMode]}`}
+      title={`切换到${NEXT_LABELS[themeMode]}`}
     >
-      {isDark ? (
-        <svg width="20" height="20" viewBox="0 0 24 24" fill="none">
-          <circle cx="12" cy="12" r="4" fill={theme.accent}/>
-          <path d="M12 2V4M12 20V22M4 12H2M22 12H20M6.34 6.34L4.93 4.93M19.07 19.07L17.66 17.66M6.34 17.66L4.93 19.07M19.07 4.93L17.66 6.34" stroke={theme.accent} strokeWidth="2" strokeLinecap="round"/>
-        </svg>
-      ) : (
-        <svg width="20" height="20" viewBox="0 0 24 24" fill="none">
-          <path d="M21 12.79A9 9 0 1 1 11.21 3 7 7 0 0 0 21 12.79z" fill={theme.accent} stroke={theme.accent} strokeWidth="1"/>
-        </svg>
-      )}
-    </button>
+      <Icon />
+    </Button>
   );
 };

frontend/src/components/layout/AppShell.tsx

@@ -0,0 +1,22 @@
+import { useLocation } from 'react-router-dom';
+
+import { FooterLayout } from './FooterLayout';
+import { HeaderLayout } from './HeaderLayout';
+import { ContentLayout } from './ContentLayout';
+import { KeepAliveViewport } from './KeepAliveViewport';
+import { getTabByPath } from '../../router/tabs';
+
+export function AppShell() {
+  const location = useLocation();
+  const activeTab = getTabByPath(location.pathname);
+
+  return (
+    <div className="flex min-h-screen flex-col bg-t-bg text-t-text">
+      <HeaderLayout activeTab={activeTab} />
+      <ContentLayout tab={activeTab}>
+        <KeepAliveViewport activeTab={activeTab} />
+      </ContentLayout>
+      <FooterLayout />
+    </div>
+  );
+}

frontend/src/components/layout/Content.tsx

@@ -1,27 +0,0 @@
-import React from 'react';
-import { useTheme } from '../../contexts';
-
-interface ContentProps {
-  children: React.ReactNode;
-  wide?: boolean;
-}
-
-export const Content: React.FC<ContentProps> = ({ children, wide = false }) => {
-  const { theme } = useTheme();
-
-  return (
-    <main
-      style={{
-        flex: 1,
-        padding: '48px 56px',
-        maxWidth: wide ? 1400 : 1100,
-        margin: '0 auto',
-        width: '100%',
-        position: 'relative',
-        backgroundColor: theme.bg,
-      }}
-    >
-      {children}
-    </main>
-  );
-};

frontend/src/components/layout/ContentLayout.tsx

@@ -0,0 +1,40 @@
+import type { ReactNode } from 'react';
+
+import type { AppTabConfig } from '../../router/tabs';
+import { shellFrameClassName } from './shell-config';
+
+interface ContentLayoutProps {
+  children: ReactNode;
+  tab: AppTabConfig;
+}
+
+const widthClassMap = {
+  default: 'mx-auto w-full max-w-[1120px]',
+  wide: 'mx-auto w-full max-w-[1440px]',
+  full: 'w-full',
+} as const;
+
+export function ContentLayout({ children, tab }: ContentLayoutProps) {
+  const widthClass = widthClassMap[tab.contentWidth];
+
+  return (
+    <main className="flex min-h-0 flex-1 bg-t-bg">
+      <div
+        className={[
+          shellFrameClassName,
+          'relative flex min-h-0 flex-1 justify-center py-8',
+        ].join(' ')}
+      >
+        <div
+          className={[
+            'relative flex min-h-0 w-full',
+            widthClass,
+            tab.fillHeight ? 'overflow-hidden' : '',
+          ].join(' ')}
+        >
+          {children}
+        </div>
+      </div>
+    </main>
+  );
+}

frontend/src/components/layout/FooterLayout.tsx

@@ -0,0 +1,38 @@
+import { Badge } from '../shadcn/ui/badge';
+import { Separator } from '../shadcn/ui/separator';
+
+import { shellFrameClassName, shellMeta } from './shell-config';
+
+export function FooterLayout() {
+  return (
+    <footer className="border-t border-t-border bg-t-bg">
+      <div
+        className={[
+          shellFrameClassName,
+          'flex items-center justify-between gap-6 py-4 text-xs text-t-text3',
+        ].join(' ')}
+      >
+        <div className="min-w-0 max-w-[360px]">
+          <div className="mono mb-1 tracking-[0.18em] text-t-text2">
+            {shellMeta.productLabel}
+          </div>
+        </div>
+
+        <div className="flex shrink-0 items-center gap-3 whitespace-nowrap rounded-xl border border-border bg-card px-3 py-2 shadow-sm">
+          <Badge variant="secondary" className="mono border-0 bg-transparent px-0 py-0 text-[11px] tracking-[0.24em] text-muted-foreground">
+            {shellMeta.version}
+          </Badge>
+          <Separator orientation="vertical" className="h-4 bg-border" />
+          <Badge variant="outline" className="mono gap-2 border-0 bg-transparent px-0 py-0 text-[11px] tracking-[0.24em] text-[var(--t-green)]">
+            <span className="size-2 rounded-full bg-[var(--t-green)]" />
+            {shellMeta.status}
+          </Badge>
+          <Separator orientation="vertical" className="h-4 bg-border" />
+          <span className="mono text-[11px] tracking-[0.18em] text-muted-foreground">
+            {shellMeta.surface}
+          </span>
+        </div>
+      </div>
+    </footer>
+  );
+}

frontend/src/components/layout/Header.tsx

@@ -1,47 +0,0 @@
-import React from 'react';
-import { useTheme } from '../../contexts';
-import { TLogo } from '../common/TLogo';
-import { ThemeToggle } from '../common/ThemeToggle';
-
-export const Header: React.FC = () => {
-  const { theme } = useTheme();
-
-  return (
-    <header
-      className="h-[72px] flex items-center justify-between sticky top-0 z-[100]"
-      style={{
-        padding: '0 48px',
-        borderBottom: `1px solid ${theme.border}`,
-        backgroundColor: theme.bg,
-      }}
-    >
-      <div className="flex items-center" style={{ gap: 20 }}>
-        <TLogo size={80} />
-        <div className="flex items-baseline" style={{ gap: 12 }}>
-          <span style={{ fontWeight: 700, fontSize: 20, letterSpacing: '-0.5px', color: theme.text }}>
-            T-Systems
-          </span>
-          <span style={{ fontWeight: 300, fontSize: 16, color: theme.text2 }}>
-            Regulation
-          </span>
-        </div>
-      </div>
-      <div className="flex items-center" style={{ gap: 16 }}>
-        <ThemeToggle />
-        <div
-          className="flex items-center rounded-lg"
-          style={{
-            padding: '8px 16px',
-            gap: 8,
-            backgroundColor: theme.bgHover,
-            borderRadius: 8,
-          }}
-        >
-          <span className="mono" style={{ fontSize: 11, color: theme.text3 }}>v1.0.0</span>
-          <div style={{ width: 1, height: 12, background: theme.border }} />
-          <span className="mono" style={{ fontSize: 12, color: theme.green }}>● ONLINE</span>
-        </div>
-      </div>
-    </header>
-  );
-};

frontend/src/components/layout/HeaderBrand.tsx

@@ -0,0 +1,19 @@
+import { TLogo } from '../common/TLogo';
+
+export function HeaderBrand() {
+  return (
+    <div className="flex min-w-[280px] shrink-0 items-center gap-4 whitespace-nowrap">
+      <div className="shrink-0">
+        <TLogo size={46} />
+      </div>
+      <div className="flex min-w-0 items-center gap-2 whitespace-nowrap">
+        <span className="text-[1.18rem] font-semibold tracking-[-0.04em] text-foreground">
+          T-Systems
+        </span>
+        <span className="text-[1.02rem] font-light text-muted-foreground">
+          Regulation
+        </span>
+      </div>
+    </div>
+  );
+}

frontend/src/components/layout/HeaderLayout.tsx

@@ -0,0 +1,38 @@
+import type { AppTabConfig } from '../../router/tabs';
+import { ThemeToggle } from '../common/ThemeToggle';
+import { Badge } from '../shadcn/ui/badge';
+import { Separator } from '../shadcn/ui/separator';
+
+import { HeaderBrand } from './HeaderBrand';
+import { shellFrameClassName, shellMeta } from './shell-config';
+import { TabNav } from './TabNav';
+
+interface HeaderLayoutProps {
+  activeTab: AppTabConfig;
+}
+
+export function HeaderLayout({ activeTab }: HeaderLayoutProps) {
+  return (
+    <header className="sticky top-0 z-[100] border-b border-border bg-background/95 backdrop-blur supports-[backdrop-filter]:bg-background/80">
+      <div className={[shellFrameClassName, 'flex h-20 items-center gap-8'].join(' ')}>
+        <HeaderBrand />
+        <div className="min-w-0 flex-1 self-stretch overflow-hidden">
+          <TabNav activeTab={activeTab} />
+        </div>
+        <div className="ml-auto flex shrink-0 items-center gap-3 self-center">
+          <ThemeToggle />
+          <div className="flex h-11 shrink-0 items-center gap-3 whitespace-nowrap rounded-xl border border-border bg-card px-3 shadow-sm">
+            <Badge variant="secondary" className="mono border-0 bg-transparent px-0 py-0 text-[11px] tracking-[0.24em] text-muted-foreground">
+              {shellMeta.version}
+            </Badge>
+            <Separator orientation="vertical" className="h-4 bg-border" />
+            <Badge variant="outline" className="mono gap-2 border-0 bg-transparent px-0 py-0 text-[11px] tracking-[0.24em] text-[var(--t-green)]">
+              <span className="size-2 rounded-full bg-[var(--t-green)]" />
+              {shellMeta.status}
+            </Badge>
+          </div>
+        </div>
+      </div>
+    </header>
+  );
+}

frontend/src/components/layout/KeepAliveViewport.tsx

@@ -0,0 +1,45 @@
+import { useEffect, useState } from 'react';
+
+import { appTabs, type AppTabConfig } from '../../router/tabs';
+
+interface KeepAliveViewportProps {
+  activeTab: AppTabConfig;
+}
+
+export function KeepAliveViewport({ activeTab }: KeepAliveViewportProps) {
+  const [mountedTabIds, setMountedTabIds] = useState<string[]>([activeTab.id]);
+
+  useEffect(() => {
+    const timerId = window.setTimeout(() => {
+      setMountedTabIds((prev) => (prev.includes(activeTab.id) ? prev : [...prev, activeTab.id]));
+    }, 0);
+    return () => window.clearTimeout(timerId);
+  }, [activeTab.id]);
+
+  return (
+    <div className="flex min-h-0 flex-1">
+      {appTabs.map((tab) => {
+        const shouldRender = tab.keepAlive ? mountedTabIds.includes(tab.id) : tab.id === activeTab.id;
+        if (!shouldRender) {
+          return null;
+        }
+
+        const TabComponent = tab.component;
+        const isActive = tab.id === activeTab.id;
+
+        return (
+          <div
+            key={tab.id}
+            aria-hidden={!isActive}
+            className={[
+              'min-h-0 flex-1',
+              isActive ? 'flex' : 'hidden',
+            ].join(' ')}
+          >
+            <TabComponent />
+          </div>
+        );
+      })}
+    </div>
+  );
+}

frontend/src/components/layout/TabNav.tsx

@@ -0,0 +1,125 @@
+import { useEffect, useLayoutEffect, useRef, useState } from 'react';
+import { useNavigate } from 'react-router-dom';
+
+import type { AppTabConfig, TabId } from '../../router/tabs';
+import { appTabs } from '../../router/tabs';
+
+interface TabNavProps {
+  activeTab: AppTabConfig;
+}
+
+interface IndicatorStyle {
+  opacity: number;
+  transform: string;
+  width: number;
+}
+
+const reducedMotionQuery = '(prefers-reduced-motion: reduce)';
+
+export function TabNav({ activeTab }: TabNavProps) {
+  const navigate = useNavigate();
+  const trackRef = useRef<HTMLDivElement | null>(null);
+  const buttonRefs = useRef<Record<TabId, HTMLButtonElement | null>>({
+    perception: null,
+    docs: null,
+    compliance: null,
+    status: null,
+    rag: null,
+  });
+  const [indicatorStyle, setIndicatorStyle] = useState<IndicatorStyle>({
+    opacity: 0,
+    transform: 'translateX(0px)',
+    width: 0,
+  });
+  const [reducedMotion, setReducedMotion] = useState(false);
+
+  const handleValueChange = (value: string) => {
+    const nextTab = appTabs.find((tab) => tab.id === value);
+    if (nextTab && nextTab.path !== activeTab.path) {
+      navigate(nextTab.path);
+    }
+  };
+
+  useEffect(() => {
+    const mediaQuery = window.matchMedia(reducedMotionQuery);
+    const updateMotionPreference = () => {
+      setReducedMotion(mediaQuery.matches);
+    };
+
+    updateMotionPreference();
+    mediaQuery.addEventListener('change', updateMotionPreference);
+
+    return () => {
+      mediaQuery.removeEventListener('change', updateMotionPreference);
+    };
+  }, []);
+
+  useLayoutEffect(() => {
+    const updateIndicator = () => {
+      const trackNode = trackRef.current;
+      const activeNode = buttonRefs.current[activeTab.id];
+      if (!trackNode || !activeNode) {
+        return;
+      }
+
+      const trackRect = trackNode.getBoundingClientRect();
+      const activeRect = activeNode.getBoundingClientRect();
+
+      setIndicatorStyle({
+        opacity: 1,
+        transform: `translateX(${activeRect.left - trackRect.left}px)`,
+        width: activeRect.width,
+      });
+    };
+
+    updateIndicator();
+    window.addEventListener('resize', updateIndicator);
+
+    return () => {
+      window.removeEventListener('resize', updateIndicator);
+    };
+  }, [activeTab.id]);
+
+  return (
+    <nav className="flex h-full min-w-0 items-stretch overflow-x-auto overflow-y-hidden">
+      <div
+        ref={trackRef}
+        className="relative flex h-full min-w-max flex-nowrap items-stretch gap-3 pr-6"
+      >
+        <div
+          aria-hidden="true"
+          className={[
+            'pointer-events-none absolute bottom-0 left-0 h-0.5 rounded-full bg-primary',
+            reducedMotion
+              ? 'transition-none'
+              : 'transition-[transform,width,opacity] duration-220 ease-[cubic-bezier(0.22,1,0.36,1)]',
+          ].join(' ')}
+          style={indicatorStyle}
+        />
+        {appTabs.map((tab) => (
+          <button
+            key={tab.id}
+            ref={(node) => {
+              buttonRefs.current[tab.id] = node;
+            }}
+            data-shell-tab="true"
+            type="button"
+            onClick={() => handleValueChange(tab.id satisfies TabId)}
+            aria-current={tab.id === activeTab.id ? 'page' : undefined}
+            className={[
+              'inline-flex h-full shrink-0 appearance-none items-center justify-center border-0 border-b-2 border-transparent bg-transparent px-5 pt-1 text-[0.95rem] font-medium tracking-[0.02em] outline-none',
+              reducedMotion
+                ? 'transition-none'
+                : 'transition-[color,opacity] duration-200 ease-out',
+              tab.id === activeTab.id
+                ? 'text-foreground'
+                : 'text-muted-foreground hover:text-foreground',
+            ].join(' ')}
+          >
+            {tab.label}
+          </button>
+        ))}
+      </div>
+    </nav>
+  );
+}

frontend/src/components/layout/Tabs.tsx

@@ -1,48 +0,0 @@
-import React from 'react';
-import { useTheme, useApp } from '../../contexts';
-import type { TabId } from '../../contexts';
-
-const tabs: Array<{ id: TabId; label: string }> = [
-  { id: 'docs', label: '文档管理' },
-  { id: 'compliance', label: '合规分析' },
-  { id: 'status', label: '系统状态' },
-  { id: 'rag', label: '法规对话' },
-];
-
-export const Tabs: React.FC = () => {
-  const { theme } = useTheme();
-  const { activeTab, setActiveTab } = useApp();
-
-  return (
-    <nav
-      className="h-[56px] flex items-center"
-      style={{
-        padding: '0 48px',
-        borderBottom: `1px solid ${theme.border}`,
-        backgroundColor: theme.bg,
-      }}
-    >
-      {tabs.map((tab) => (
-        <button
-          key={tab.id}
-          onClick={() => setActiveTab(tab.id)}
-          style={{
-            height: 56,
-            padding: '0 32px',
-            fontSize: 15,
-            fontWeight: activeTab === tab.id ? 600 : 400,
-            color: activeTab === tab.id ? theme.accent : theme.text3,
-            background: 'transparent',
-            border: 'none',
-            borderBottom: activeTab === tab.id ? `3px solid ${theme.accent}` : '3px solid transparent',
-            marginBottom: -1,
-            cursor: 'pointer',
-            transition: 'all 0.2s ease',
-          }}
-        >
-          {tab.label}
-        </button>
-      ))}
-    </nav>
-  );
-};

frontend/src/components/layout/index.ts

@@ -1,3 +1,4 @@
-export { Header } from './Header';
-export { Tabs } from './Tabs';
-export { Content } from './Content';
+export { AppShell } from './AppShell';
+export { ContentLayout } from './ContentLayout';
+export { FooterLayout } from './FooterLayout';
+export { HeaderLayout } from './HeaderLayout';

frontend/src/components/layout/shell-config.ts

@@ -0,0 +1,12 @@
+import { appTabs } from '../../router/tabs';
+
+export const shellFrameClassName = 'mx-auto w-full max-w-[1680px] px-8';
+
+export const shellMeta = {
+  productLabel: 'T-Systems Regulation',
+  version: 'v1.0.0',
+  status: 'ONLINE',
+  surface: 'Desktop Web',
+} as const;
+
+export const shellModuleSummary = appTabs.map((tab) => tab.label).join(' / ');

frontend/src/components/shadcn/ui/badge.tsx

@@ -0,0 +1,30 @@
+import { cva, type VariantProps } from 'class-variance-authority';
+import type * as React from 'react';
+
+import { cn } from '@/lib/utils';
+
+const badgeVariants = cva(
+  'inline-flex items-center rounded-md border px-2 py-1 text-[11px] font-medium tracking-[0.22em] uppercase transition-colors',
+  {
+    variants: {
+      variant: {
+        default: 'border-primary/30 bg-primary/10 text-primary',
+        secondary: 'border-border bg-muted text-muted-foreground',
+        outline: 'border-border bg-transparent text-foreground',
+      },
+    },
+    defaultVariants: {
+      variant: 'default',
+    },
+  },
+);
+
+function Badge({
+  className,
+  variant,
+  ...props
+}: React.ComponentProps<'span'> & VariantProps<typeof badgeVariants>) {
+  return <span className={cn(badgeVariants({ variant }), className)} {...props} />;
+}
+
+export { Badge };

frontend/src/components/shadcn/ui/button.tsx

@@ -0,0 +1,65 @@
+import * as React from 'react';
+import { cva, type VariantProps } from 'class-variance-authority';
+import { Slot } from 'radix-ui';
+
+import { cn } from '@/lib/utils';
+
+const buttonVariants = cva(
+  'group/button inline-flex shrink-0 items-center justify-center rounded-lg border border-transparent bg-clip-padding text-sm font-medium whitespace-nowrap transition-all outline-none select-none focus-visible:border-ring focus-visible:ring-3 focus-visible:ring-ring/50 active:not-aria-[haspopup]:translate-y-px disabled:pointer-events-none disabled:opacity-50 aria-invalid:border-destructive aria-invalid:ring-3 aria-invalid:ring-destructive/20 dark:aria-invalid:border-destructive/50 dark:aria-invalid:ring-destructive/40 [&_svg]:pointer-events-none [&_svg]:shrink-0 [&_svg:not([class*=size-])]:size-4',
+  {
+    variants: {
+      variant: {
+        default: 'bg-primary text-primary-foreground hover:bg-primary/90',
+        outline:
+          'border-border bg-background hover:bg-muted hover:text-foreground aria-expanded:bg-muted aria-expanded:text-foreground dark:bg-input/30 dark:hover:bg-input/50',
+        secondary:
+          'bg-secondary text-secondary-foreground hover:bg-secondary/80 aria-expanded:bg-secondary aria-expanded:text-secondary-foreground',
+        ghost:
+          'hover:bg-muted hover:text-foreground aria-expanded:bg-muted aria-expanded:text-foreground dark:hover:bg-muted/50',
+        destructive:
+          'bg-destructive/10 text-destructive hover:bg-destructive/20 focus-visible:border-destructive/40 focus-visible:ring-destructive/20 dark:bg-destructive/20 dark:hover:bg-destructive/30 dark:focus-visible:ring-destructive/40',
+        link: 'text-primary underline-offset-4 hover:underline',
+      },
+      size: {
+        default:
+          'h-8 gap-1.5 px-2.5 has-data-[icon=inline-end]:pr-2 has-data-[icon=inline-start]:pl-2',
+        xs: 'h-6 gap-1 rounded-[min(var(--radius-md),10px)] px-2 text-xs has-data-[icon=inline-end]:pr-1.5 has-data-[icon=inline-start]:pl-1.5 [&_svg:not([class*=size-])]:size-3',
+        sm: 'h-7 gap-1 rounded-[min(var(--radius-md),12px)] px-2.5 text-[0.8rem] has-data-[icon=inline-end]:pr-1.5 has-data-[icon=inline-start]:pl-1.5 [&_svg:not([class*=size-])]:size-3.5',
+        lg: 'h-9 gap-1.5 px-3 has-data-[icon=inline-end]:pr-2 has-data-[icon=inline-start]:pl-2',
+        icon: 'size-8',
+        'icon-xs': 'size-6 rounded-[min(var(--radius-md),10px)] [&_svg:not([class*=size-])]:size-3',
+        'icon-sm': 'size-7 rounded-[min(var(--radius-md),12px)]',
+        'icon-lg': 'size-9',
+      },
+    },
+    defaultVariants: {
+      variant: 'default',
+      size: 'default',
+    },
+  },
+);
+
+function Button({
+  className,
+  variant,
+  size,
+  asChild = false,
+  ...props
+}: React.ComponentProps<'button'> &
+  VariantProps<typeof buttonVariants> & {
+    asChild?: boolean;
+  }) {
+  const Comp = asChild ? Slot.Root : 'button';
+
+  return (
+    <Comp
+      data-slot="button"
+      data-variant={variant}
+      data-size={size}
+      className={cn(buttonVariants({ variant, size, className }))}
+      {...props}
+    />
+  );
+}
+
+export { Button };

frontend/src/components/shadcn/ui/separator.tsx

@@ -0,0 +1,27 @@
+import * as React from 'react';
+import { Separator as SeparatorPrimitive } from 'radix-ui';
+
+import { cn } from '@/lib/utils';
+
+function Separator({
+  className,
+  orientation = 'horizontal',
+  decorative = true,
+  ...props
+}: React.ComponentProps<typeof SeparatorPrimitive.Root>) {
+  return (
+    <SeparatorPrimitive.Root
+      data-slot="separator-root"
+      decorative={decorative}
+      orientation={orientation}
+      className={cn(
+        'shrink-0 bg-border',
+        orientation === 'horizontal' ? 'h-px w-full' : 'h-full w-px',
+        className,
+      )}
+      {...props}
+    />
+  );
+}
+
+export { Separator };

frontend/src/components/shadcn/ui/tabs.tsx

@@ -0,0 +1,48 @@
+import * as React from 'react';
+import { Tabs as TabsPrimitive } from 'radix-ui';
+
+import { cn } from '@/lib/utils';
+
+function Tabs({
+  className,
+  ...props
+}: React.ComponentProps<typeof TabsPrimitive.Root>) {
+  return (
+    <TabsPrimitive.Root
+      data-slot="tabs"
+      className={cn('w-full', className)}
+      {...props}
+    />
+  );
+}
+
+function TabsList({
+  className,
+  ...props
+}: React.ComponentProps<typeof TabsPrimitive.List>) {
+  return (
+    <TabsPrimitive.List
+      data-slot="tabs-list"
+      className={cn('inline-flex items-center gap-2', className)}
+      {...props}
+    />
+  );
+}
+
+function TabsTrigger({
+  className,
+  ...props
+}: React.ComponentProps<typeof TabsPrimitive.Trigger>) {
+  return (
+    <TabsPrimitive.Trigger
+      data-slot="tabs-trigger"
+      className={cn(
+        'inline-flex items-center justify-center whitespace-nowrap outline-none transition-colors focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-2 disabled:pointer-events-none disabled:opacity-50',
+        className,
+      )}
+      {...props}
+    />
+  );
+}
+
+export { Tabs, TabsList, TabsTrigger };

frontend/src/contexts/AppContext.tsx

@@ -1,17 +0,0 @@
-import { useState, type ReactNode } from 'react';
-
-import { AppContext, type TabId } from './app-context';
-
-interface AppProviderProps {
-  children: ReactNode;
-}
-
-export const AppProvider: React.FC<AppProviderProps> = ({ children }) => {
-  const [activeTab, setActiveTab] = useState<TabId>('compliance');
-
-  return (
-    <AppContext.Provider value={{ activeTab, setActiveTab }}>
-      {children}
-    </AppContext.Provider>
-  );
-};

frontend/src/contexts/ThemeContext.tsx

@@ -1,33 +1,58 @@
 import React, { useEffect, useState, type ReactNode } from 'react';
 
-import { darkTheme, lightTheme } from '../types/theme';
+import { darkTheme, dimTheme, lightTheme, type ThemeMode } from '../types/theme';
 import { ThemeContext } from './theme-context';
 
+const STORAGE_KEY = 'app-theme-mode';
+
+function getInitialMode(): ThemeMode {
+  try {
+    const stored = localStorage.getItem(STORAGE_KEY);
+    if (stored === 'dark' || stored === 'dim' || stored === 'light') return stored;
+  } catch {
+    // ignore
+  }
+  return 'dark';
+}
+
+const THEME_MAP = { dark: darkTheme, dim: dimTheme, light: lightTheme };
+const BG_MAP: Record<ThemeMode, string> = {
+  dark:  '#0a0a12',
+  dim:   '#1e1b2e',
+  light: '#ffffff',
+};
+
 interface ThemeProviderProps {
   children: ReactNode;
 }
 
 export const ThemeProvider: React.FC<ThemeProviderProps> = ({ children }) => {
-  const [isDark, setIsDark] = useState<boolean>(true);
-  const theme = isDark ? darkTheme : lightTheme;
+  const [themeMode, setThemeMode] = useState<ThemeMode>(getInitialMode);
+  const theme = THEME_MAP[themeMode];
+  const isDark = themeMode === 'dark';
 
   const toggleTheme = () => {
-    setIsDark((prev) => !prev);
+    setThemeMode((prev) =>
+      prev === 'dark' ? 'dim' : prev === 'dim' ? 'light' : 'dark'
+    );
   };
 
   useEffect(() => {
-    if (isDark) {
-      document.documentElement.classList.add('dark');
-      document.body.style.background = '#0a0a12';
-      return;
-    }
+    const root = document.documentElement;
+    root.classList.remove('dark', 'dim');
+    if (themeMode !== 'light') root.classList.add(themeMode);
 
-    document.documentElement.classList.remove('dark');
-    document.body.style.background = '#ffffff';
-  }, [isDark]);
+    document.body.style.background = BG_MAP[themeMode];
+
+    try {
+      localStorage.setItem(STORAGE_KEY, themeMode);
+    } catch {
+      // ignore
+    }
+  }, [themeMode]);
 
   return (
-    <ThemeContext.Provider value={{ isDark, theme, toggleTheme }}>
+    <ThemeContext.Provider value={{ isDark, themeMode, theme, toggleTheme }}>
       {children}
     </ThemeContext.Provider>
   );

frontend/src/contexts/app-context.ts

@@ -1,10 +0,0 @@
-import { createContext } from 'react';
-
-export type TabId = 'docs' | 'compliance' | 'status' | 'rag';
-
-export interface AppContextValue {
-  activeTab: TabId;
-  setActiveTab: (tab: TabId) => void;
-}
-
-export const AppContext = createContext<AppContextValue | undefined>(undefined);

frontend/src/contexts/index.ts

@@ -1,5 +1,2 @@
 export { ThemeProvider } from './ThemeContext';
 export { useTheme } from './useTheme';
-export { AppProvider } from './AppContext';
-export type { AppContextValue, TabId } from './app-context';
-export { useApp } from './useApp';

frontend/src/contexts/theme-context.ts

@@ -1,9 +1,10 @@
 import { createContext } from 'react';
 
-import type { ThemeColors } from '../types/theme';
+import type { ThemeColors, ThemeMode } from '../types/theme';
 
 export interface ThemeContextValue {
   isDark: boolean;
+  themeMode: ThemeMode;
   theme: ThemeColors;
   toggleTheme: () => void;
 }

frontend/src/contexts/useApp.ts

@@ -1,11 +0,0 @@
-import { useContext } from 'react';
-
-import { AppContext, type AppContextValue } from './app-context';
-
-export function useApp(): AppContextValue {
-  const context = useContext(AppContext);
-  if (!context) {
-    throw new Error('useApp must be used within an AppProvider');
-  }
-  return context;
-}

frontend/src/pages/Compliance/CompliancePage.tsx

@@ -587,7 +587,7 @@ export const CompliancePage: React.FC = () => {
       flex: 1,
       display: 'flex',
       height: '100%',
-      minHeight: 'calc(100vh - 128px)',
+      minHeight: 0,
       position: 'relative',
     }}>
       {/* Main Content Area */}

frontend/src/pages/Docs/DocsPage.tsx

@@ -1,6 +1,5 @@
 import React, { useEffect, useRef, useState } from 'react';
 import { useTheme } from '../../contexts';
-import { Content } from '../../components/layout/Content';
 import { TPattern } from '../../components/common/TPattern';
 import { getDocumentList, getDocumentStatus, searchRegulations, uploadDocument, deleteDocument, retryDocument, type RegulationSearchItem } from '../../api/docs';
 import type { Doc } from '../../types';
@@ -40,6 +39,7 @@ export const DocsPage: React.FC = () => {
   const [searchResults, setSearchResults] = useState<RegulationSearchItem[]>([]);
   const [searchLoading, setSearchLoading] = useState(false);
   const [searchError, setSearchError] = useState('');
+  const [batchQueueLength, setBatchQueueLength] = useState(0);
 
   // Upload metadata
   const [regulationType, setRegulationType] = useState('');
@@ -48,12 +48,17 @@ export const DocsPage: React.FC = () => {
   // Batch queue: files waiting to be uploaded after the current one finishes
   const batchQueueRef = useRef<File[]>([]);
 
+  const setBatchQueue = (files: File[]) => {
+    batchQueueRef.current = files;
+    setBatchQueueLength(files.length);
+  };
+
   async function loadDocuments() {
     setLoading(true);
     try {
       const response = await getDocumentList();
-      const apiDocs: Doc[] = response.docs.map((doc) => ({
-        id: parseInt(String(doc.id).replace('doc-', ''), 10) || Math.floor(Math.random() * 10000),
+      const apiDocs: Doc[] = response.docs.map((doc, index) => ({
+        id: Number.parseInt(String(doc.id).replace('doc-', ''), 10) || -(index + 1),
         name: doc.name,
         chunks: doc.chunks,
         size: doc.updated_at ? new Date(doc.updated_at).toLocaleString() : 'Indexed document',
@@ -209,6 +214,7 @@ export const DocsPage: React.FC = () => {
 
       // Process next file in batch queue
       const next = batchQueueRef.current.shift();
+      setBatchQueueLength(batchQueueRef.current.length);
       if (next) {
         const nextRunId = pipelineRunIdRef.current + 1;
         pipelineRunIdRef.current = nextRunId;
@@ -222,7 +228,7 @@ export const DocsPage: React.FC = () => {
     if (files.length === 0 || uploading) return;
 
     const [first, ...rest] = files;
-    batchQueueRef.current = rest;
+    setBatchQueue(rest);
 
     const runId = pipelineRunIdRef.current + 1;
     pipelineRunIdRef.current = runId;
@@ -262,7 +268,7 @@ export const DocsPage: React.FC = () => {
     if (files.length === 0 || uploading) return;
 
     const [first, ...rest] = files;
-    batchQueueRef.current = rest;
+    setBatchQueue(rest);
     const runId = pipelineRunIdRef.current + 1;
     pipelineRunIdRef.current = runId;
     void uploadSingleFile(first, runId);
@@ -282,8 +288,7 @@ export const DocsPage: React.FC = () => {
 
   const getPipelineHint = () => {
     if (pipelineStatus === 'running') {
-      const queueLen = batchQueueRef.current.length;
-      const suffix = queueLen > 0 ? ` (+${queueLen} 待上传)` : '';
+      const suffix = batchQueueLength > 0 ? ` (+${batchQueueLength} 待上传)` : '';
       return `${activeStep >= 0 ? PIPELINE_STEPS[activeStep].name : 'LOAD'} · ${uploadFileName}${suffix}`;
     }
     if (pipelineStatus === 'completed') return 'PIPELINE COMPLETE';
@@ -291,6 +296,11 @@ export const DocsPage: React.FC = () => {
     return 'WAITING FOR UPLOAD';
   };
 
+  const getDocKey = (doc: Doc) => {
+    // Prefer the backend document identifier because the numeric display id is not guaranteed unique.
+    return doc.docId ?? `local-${doc.id}-${doc.name}`;
+  };
+
   const inputStyle: React.CSSProperties = {
     padding: '8px 12px',
     fontSize: 13,
@@ -302,7 +312,7 @@ export const DocsPage: React.FC = () => {
   };
 
   return (
-    <Content>
+    <div className="relative w-full">
       <TPattern />
 
       <section style={{ marginBottom: 56 }}>
@@ -432,7 +442,7 @@ export const DocsPage: React.FC = () => {
           <div style={{ display: 'flex', flexDirection: 'column', gap: 12 }}>
             {docs.map((doc) => (
               <div
-                key={doc.id}
+                key={getDocKey(doc)}
                 style={{ display: 'flex', alignItems: 'center', justifyContent: 'space-between', padding: 20, background: theme.bgCard, borderRadius: 12, border: `1px solid ${doc.status === 'parsing' ? theme.accent : theme.border}`, transition: 'all 0.2s ease', boxShadow: !isDark ? '0 2px 8px rgba(226,0,116,0.04)' : 'none' }}
               >
                 <div style={{ display: 'flex', alignItems: 'flex-start', gap: 16 }}>
@@ -543,6 +553,6 @@ export const DocsPage: React.FC = () => {
           </div>
         </div>
       </section>
-    </Content>
+    </div>
   );
 };

frontend/src/pages/Perception/AnalysisPanel.tsx

@@ -0,0 +1,207 @@
+import React, { useRef } from 'react';
+import { useTheme } from '../../contexts';
+import type { RegulationEvent, AffectedDoc } from '../../api/perception';
+
+interface AnalysisPanelProps {
+  event: RegulationEvent | null;
+  analyzing: boolean;
+  analysisText: string;
+  affectedDocs: AffectedDoc[];
+  onAnalyze: () => void;
+  onAbort: () => void;
+}
+
+// Minimal markdown renderer — handles ##/### headings, **bold**, bullet lists
+function MarkdownText({ text, textColor, accent }: { text: string; textColor: string; accent: string }) {
+  const lines = text.split('\n');
+  return (
+    <div style={{ fontSize: 14, lineHeight: 1.75, color: textColor }}>
+      {lines.map((line, i) => {
+        if (line.startsWith('## ')) {
+          return <div key={i} style={{ fontSize: 15, fontWeight: 700, color: accent, marginTop: 18, marginBottom: 6 }}>{line.slice(3)}</div>;
+        }
+        if (line.startsWith('### ')) {
+          return <div key={i} style={{ fontSize: 13, fontWeight: 700, marginTop: 12, marginBottom: 4 }}>{line.slice(4)}</div>;
+        }
+        if (line.startsWith('- ') || line.startsWith('* ')) {
+          const content = line.slice(2);
+          return (
+            <div key={i} style={{ display: 'flex', gap: 8, marginBottom: 4, paddingLeft: 8 }}>
+              <span style={{ color: accent, flexShrink: 0 }}>·</span>
+              <span dangerouslySetInnerHTML={{ __html: content.replace(/\*\*(.+?)\*\*/g, '<strong>$1</strong>') }} />
+            </div>
+          );
+        }
+        if (/^\d+\./.test(line)) {
+          return (
+            <div key={i} style={{ marginBottom: 4, paddingLeft: 8 }}>
+              <span dangerouslySetInnerHTML={{ __html: line.replace(/\*\*(.+?)\*\*/g, '<strong>$1</strong>') }} />
+            </div>
+          );
+        }
+        if (!line.trim()) return <div key={i} style={{ height: 8 }} />;
+        return (
+          <div key={i} style={{ marginBottom: 4 }}>
+            <span dangerouslySetInnerHTML={{ __html: line.replace(/\*\*(.+?)\*\*/g, '<strong>$1</strong>') }} />
+          </div>
+        );
+      })}
+    </div>
+  );
+}
+
+const IMPACT_COLORS = { high: '#d64545', medium: '#ff8800', low: '#00d4aa' };
+const SOURCE_COLORS: Record<string, string> = {
+  MIIT: '#e20074', 'UN-ECE': '#4a90d9', ISO: '#7b68ee',
+  '国标委': '#00b89c', 'EUR-Lex': '#f5a623', IATF: '#9b59b6',
+};
+const STATUS_LABEL: Record<string, string> = { enacted: '已生效', draft: '征求意见', consultation: '公众咨询' };
+
+export const AnalysisPanel: React.FC<AnalysisPanelProps> = ({
+  event, analyzing, analysisText, affectedDocs, onAnalyze, onAbort,
+}) => {
+  const { theme, isDark } = useTheme();
+  const analysisRef = useRef<HTMLDivElement>(null);
+
+  if (!event) {
+    return (
+      <div style={{ height: '100%', display: 'flex', flexDirection: 'column', alignItems: 'center', justifyContent: 'center', gap: 12 }}>
+        <div style={{ fontSize: 48, opacity: 0.15 }}>◈</div>
+        <div style={{ fontSize: 14, color: theme.text3 }}>选择左侧法规动态以查看智能影响分析</div>
+      </div>
+    );
+  }
+
+  const impactColor = IMPACT_COLORS[event.impact_level];
+  const srcColor = SOURCE_COLORS[event.source] || theme.accent;
+
+  return (
+    <div style={{ display: 'flex', flexDirection: 'column', height: '100%', gap: 0 }}>
+      {/* Event header */}
+      <div style={{
+        padding: '20px 24px',
+        background: theme.bgCard,
+        borderRadius: 12,
+        border: `1px solid ${theme.border}`,
+        borderLeft: `4px solid ${impactColor}`,
+        marginBottom: 16,
+        flexShrink: 0,
+        boxShadow: !isDark ? '0 2px 8px rgba(226,0,116,0.04)' : 'none',
+      }}>
+        {/* Source + status */}
+        <div style={{ display: 'flex', alignItems: 'center', gap: 8, marginBottom: 10 }}>
+          <span style={{ fontSize: 11, fontWeight: 700, color: srcColor, background: srcColor + '18', borderRadius: 4, padding: '3px 8px' }}>{event.source}</span>
+          <span className="mono" style={{ fontSize: 11, color: theme.text3 }}>{event.standard_code}</span>
+          <span style={{ marginLeft: 'auto', fontSize: 11, color: event.status === 'enacted' ? theme.green : '#ff8800', fontWeight: 600 }}>
+            {STATUS_LABEL[event.status] ?? event.status}
+          </span>
+        </div>
+
+        {/* Title */}
+        <div style={{ fontSize: 16, fontWeight: 700, color: theme.text, lineHeight: 1.4, marginBottom: 10 }}>
+          {event.title}
+        </div>
+
+        {/* Summary */}
+        <div style={{ fontSize: 13, color: theme.text2, lineHeight: 1.6, marginBottom: 12 }}>
+          {event.summary}
+        </div>
+
+        {/* Tags */}
+        <div style={{ display: 'flex', flexWrap: 'wrap', gap: 6, marginBottom: 12 }}>
+          {event.tags.map(tag => (
+            <span key={tag} style={{ fontSize: 11, color: theme.text3, background: theme.bgHover, borderRadius: 4, padding: '2px 8px', border: `1px solid ${theme.border}` }}>
+              {tag}
+            </span>
+          ))}
+        </div>
+
+        {/* Dates + Analyze button */}
+        <div style={{ display: 'flex', alignItems: 'center', justifyContent: 'space-between' }}>
+          <div className="mono" style={{ fontSize: 11, color: theme.text3 }}>
+            发布：{event.published_at}
+            {event.effective_at && <span style={{ marginLeft: 12 }}>生效：<span style={{ color: impactColor }}>{event.effective_at}</span></span>}
+          </div>
+          {analyzing ? (
+            <button onClick={onAbort} style={{ padding: '7px 18px', borderRadius: 8, border: '1px solid #d64545', background: 'transparent', color: '#d64545', cursor: 'pointer', fontSize: 13, fontWeight: 600 }}>
+              ■ 停止
+            </button>
+          ) : (
+            <button onClick={onAnalyze} style={{ padding: '7px 18px', borderRadius: 8, border: 'none', background: theme.gradientAccent, color: '#fff', cursor: 'pointer', fontSize: 13, fontWeight: 600, boxShadow: '0 2px 8px rgba(226,0,116,0.3)' }}>
+              ⚡ 触发智能分析
+            </button>
+          )}
+        </div>
+      </div>
+
+      {/* Affected documents */}
+      {affectedDocs.length > 0 && (
+        <div style={{ marginBottom: 16, flexShrink: 0 }}>
+          <div className="mono" style={{ fontSize: 11, color: theme.text3, letterSpacing: '1px', marginBottom: 8 }}>
+            关联文档（{affectedDocs.length}）
+          </div>
+          <div style={{ display: 'flex', flexDirection: 'column', gap: 6 }}>
+            {affectedDocs.map(doc => (
+              <div key={doc.doc_id} style={{
+                padding: '10px 14px',
+                background: theme.bgCard,
+                border: `1px solid ${theme.border}`,
+                borderLeft: `3px solid ${theme.accent}`,
+                borderRadius: 8,
+                display: 'flex',
+                alignItems: 'flex-start',
+                gap: 10,
+              }}>
+                <div style={{ flex: 1, minWidth: 0 }}>
+                  <div style={{ fontSize: 13, fontWeight: 600, color: theme.text, overflow: 'hidden', textOverflow: 'ellipsis', whiteSpace: 'nowrap' }}>
+                    {doc.doc_name}
+                  </div>
+                  {doc.snippet && (
+                    <div style={{ fontSize: 12, color: theme.text3, marginTop: 3, overflow: 'hidden', textOverflow: 'ellipsis', whiteSpace: 'nowrap' }}>
+                      {doc.snippet}
+                    </div>
+                  )}
+                </div>
+                <span className="mono" style={{ fontSize: 11, color: theme.accent, flexShrink: 0 }}>
+                  {Math.round(doc.score * 100)}%
+                </span>
+              </div>
+            ))}
+          </div>
+        </div>
+      )}
+
+      {/* Streaming analysis output */}
+      {(analysisText || analyzing) && (
+        <div ref={analysisRef} style={{
+          flex: 1,
+          overflowY: 'auto',
+          padding: '20px 24px',
+          background: theme.bgCard,
+          border: `1px solid ${theme.border}`,
+          borderRadius: 12,
+          boxShadow: !isDark ? '0 2px 8px rgba(0,0,0,0.03)' : 'none',
+        }}>
+          <div className="mono" style={{ fontSize: 11, color: theme.accent, letterSpacing: '1px', marginBottom: 14 }}>
+            ANALYSIS {analyzing && <span style={{ animation: 'blink 1s step-end infinite' }}>▌</span>}
+          </div>
+          {analysisText && (
+            <MarkdownText text={analysisText} textColor={theme.text2} accent={theme.accent} />
+          )}
+          {analyzing && !analysisText && (
+            <div style={{ color: theme.text3, fontSize: 13 }}>正在分析法规影响...</div>
+          )}
+        </div>
+      )}
+
+      {/* Empty analysis state */}
+      {!analysisText && !analyzing && (
+        <div style={{ flex: 1, display: 'flex', alignItems: 'center', justifyContent: 'center' }}>
+          <div style={{ textAlign: 'center', color: theme.text3, fontSize: 13 }}>
+            点击「触发智能分析」查看 AI 影响评估
+          </div>
+        </div>
+      )}
+    </div>
+  );
+};

frontend/src/pages/Perception/EventFeed.tsx

@@ -0,0 +1,157 @@
+import React from 'react';
+import { useTheme } from '../../contexts';
+import type { RegulationEvent, ImpactLevel, EventSource } from '../../api/perception';
+
+const IMPACT_CONFIG: Record<ImpactLevel, { color: string; label: string; dot: string }> = {
+  high:   { color: '#d64545', label: '高影响', dot: '●' },
+  medium: { color: '#ff8800', label: '中影响', dot: '●' },
+  low:    { color: '#00d4aa', label: '低影响', dot: '●' },
+};
+
+const STATUS_LABEL: Record<string, string> = {
+  enacted: '已生效',
+  draft: '征求意见',
+  consultation: '公众咨询',
+};
+
+const SOURCE_COLORS: Record<string, string> = {
+  MIIT:   '#e20074',
+  'UN-ECE': '#4a90d9',
+  ISO:    '#7b68ee',
+  '国标委': '#00b89c',
+  'EUR-Lex': '#f5a623',
+  IATF:   '#9b59b6',
+};
+
+interface EventFeedProps {
+  events: RegulationEvent[];
+  selectedId: string | null;
+  onSelect: (id: string) => void;
+  filterSource: string;
+  filterImpact: string;
+  onFilterSource: (v: string) => void;
+  onFilterImpact: (v: string) => void;
+  stats: { total: number; high_impact: number; medium_impact: number; low_impact: number; recent_90d: number } | null;
+  loading: boolean;
+}
+
+export const EventFeed: React.FC<EventFeedProps> = ({
+  events, selectedId, onSelect,
+  filterSource, filterImpact, onFilterSource, onFilterImpact,
+  stats, loading,
+}) => {
+  const { theme, isDark } = useTheme();
+
+  const sources: EventSource[] = ['MIIT', 'UN-ECE', 'ISO', '国标委', 'EUR-Lex', 'IATF'];
+  const impacts: ImpactLevel[] = ['high', 'medium', 'low'];
+
+  return (
+    <div style={{ display: 'flex', flexDirection: 'column', height: '100%', gap: 16 }}>
+
+      {/* KPI mini-cards */}
+      {stats && (
+        <div style={{ display: 'grid', gridTemplateColumns: 'repeat(4, 1fr)', gap: 8 }}>
+          {[
+            { label: '总计', value: stats.total, color: theme.text },
+            { label: '高影响', value: stats.high_impact, color: '#d64545' },
+            { label: '中影响', value: stats.medium_impact, color: '#ff8800' },
+            { label: '近90天', value: stats.recent_90d, color: theme.accent },
+          ].map(({ label, value, color }) => (
+            <div key={label} style={{
+              padding: '10px 12px',
+              background: theme.bgCard,
+              border: `1px solid ${theme.border}`,
+              borderRadius: 10,
+              position: 'relative',
+              overflow: 'hidden',
+              boxShadow: 'none',
+            }}>
+              <div style={{ position: 'absolute', top: 0, left: 0, right: 0, height: 2, background: color }} />
+              <div className="mono" style={{ fontSize: 10, color: theme.text3, letterSpacing: '0.5px' }}>{label}</div>
+              <div className="mono" style={{ fontSize: 22, fontWeight: 700, color }}>{value}</div>
+            </div>
+          ))}
+        </div>
+      )}
+
+      {/* Filter row */}
+      <div style={{ display: 'flex', gap: 6, flexWrap: 'wrap' }}>
+        <button
+          onClick={() => onFilterSource('')}
+          style={{ padding: '4px 10px', borderRadius: 20, border: `1px solid ${filterSource === '' ? theme.accent : theme.border}`, background: filterSource === '' ? theme.accent + '20' : 'transparent', color: filterSource === '' ? theme.accent : theme.text3, fontSize: 11, cursor: 'pointer' }}
+        >全部来源</button>
+        {sources.map(s => (
+          <button key={s} onClick={() => onFilterSource(filterSource === s ? '' : s)}
+            style={{ padding: '4px 10px', borderRadius: 20, border: `1px solid ${filterSource === s ? (SOURCE_COLORS[s] || theme.accent) : theme.border}`, background: filterSource === s ? (SOURCE_COLORS[s] || theme.accent) + '20' : 'transparent', color: filterSource === s ? (SOURCE_COLORS[s] || theme.accent) : theme.text3, fontSize: 11, cursor: 'pointer' }}>
+            {s}
+          </button>
+        ))}
+      </div>
+      <div style={{ display: 'flex', gap: 6 }}>
+        <button onClick={() => onFilterImpact('')}
+          style={{ padding: '4px 10px', borderRadius: 20, border: `1px solid ${filterImpact === '' ? theme.accent : theme.border}`, background: filterImpact === '' ? theme.accent + '20' : 'transparent', color: filterImpact === '' ? theme.accent : theme.text3, fontSize: 11, cursor: 'pointer' }}>
+          全部等级
+        </button>
+        {impacts.map(lvl => (
+          <button key={lvl} onClick={() => onFilterImpact(filterImpact === lvl ? '' : lvl)}
+            style={{ padding: '4px 10px', borderRadius: 20, border: `1px solid ${filterImpact === lvl ? IMPACT_CONFIG[lvl].color : theme.border}`, background: filterImpact === lvl ? IMPACT_CONFIG[lvl].color + '22' : 'transparent', color: filterImpact === lvl ? IMPACT_CONFIG[lvl].color : theme.text3, fontSize: 11, cursor: 'pointer' }}>
+            {IMPACT_CONFIG[lvl].dot} {IMPACT_CONFIG[lvl].label}
+          </button>
+        ))}
+      </div>
+
+      {/* Event list */}
+      <div style={{ flex: 1, overflowY: 'auto', display: 'flex', flexDirection: 'column', gap: 8 }}>
+        {loading && (
+          <div className="mono" style={{ fontSize: 12, color: theme.text3, padding: '16px 0' }}>加载中...</div>
+        )}
+        {!loading && events.length === 0 && (
+          <div style={{ fontSize: 13, color: theme.text3, padding: '32px 0', textAlign: 'center' }}>暂无法规动态</div>
+        )}
+        {events.map(evt => {
+          const cfg = IMPACT_CONFIG[evt.impact_level];
+          const isSelected = evt.id === selectedId;
+          const srcColor = SOURCE_COLORS[evt.source] || theme.accent;
+          return (
+            <div
+              key={evt.id}
+              onClick={() => onSelect(evt.id)}
+              style={{
+                padding: '14px 16px',
+                background: isSelected ? theme.bgHover : theme.bgCard,
+                borderRadius: 10,
+                border: `1px solid ${isSelected ? theme.accent : theme.border}`,
+                borderLeft: `4px solid ${cfg.color}`,
+                cursor: 'pointer',
+                transition: 'all 0.15s ease',
+                boxShadow: isSelected ? `0 0 0 1px ${theme.accent}40` : 'none',
+              }}
+            >
+              {/* Source + Status row */}
+              <div style={{ display: 'flex', alignItems: 'center', gap: 6, marginBottom: 6 }}>
+                <span style={{ fontSize: 10, fontWeight: 700, color: srcColor, background: srcColor + '18', borderRadius: 4, padding: '2px 7px' }}>{evt.source}</span>
+                <span className="mono" style={{ fontSize: 10, color: theme.text3 }}>{evt.standard_code}</span>
+                <span style={{ marginLeft: 'auto', fontSize: 10, color: evt.status === 'enacted' ? theme.green : '#ff8800', background: evt.status === 'enacted' ? theme.green + '18' : '#ff880018', borderRadius: 4, padding: '2px 6px', fontWeight: 600 }}>
+                  {STATUS_LABEL[evt.status] ?? evt.status}
+                </span>
+              </div>
+
+              {/* Title */}
+              <div style={{ fontSize: 13, fontWeight: 600, color: theme.text, lineHeight: 1.4, marginBottom: 6, display: '-webkit-box', WebkitLineClamp: 2, WebkitBoxOrient: 'vertical', overflow: 'hidden' }}>
+                {evt.title}
+              </div>
+
+              {/* Date + impact */}
+              <div style={{ display: 'flex', alignItems: 'center', justifyContent: 'space-between' }}>
+                <span className="mono" style={{ fontSize: 11, color: theme.text3 }}>
+                  {evt.published_at}{evt.effective_at ? ` → ${evt.effective_at}` : ''}
+                </span>
+                <span style={{ fontSize: 10, color: cfg.color, fontWeight: 700 }}>{cfg.dot} {cfg.label}</span>
+              </div>
+            </div>
+          );
+        })}
+      </div>
+    </div>
+  );
+};

frontend/src/pages/Perception/PerceptionPage.tsx

@@ -0,0 +1,146 @@
+import React, { useCallback, useEffect, useRef, useState } from 'react';
+import { useTheme } from '../../contexts';
+import { TPattern } from '../../components/common/TPattern';
+import {
+  listEvents,
+  getPerceptionStats,
+  analyzeEvent,
+  type RegulationEvent,
+  type PerceptionStats,
+  type AffectedDoc,
+} from '../../api/perception';
+import { EventFeed } from './EventFeed';
+import { AnalysisPanel } from './AnalysisPanel';
+
+export const PerceptionPage: React.FC = () => {
+  const { theme } = useTheme();
+
+  // Feed state
+  const [events, setEvents] = useState<RegulationEvent[]>([]);
+  const [stats, setStats] = useState<PerceptionStats | null>(null);
+  const [feedLoading, setFeedLoading] = useState(true);
+  const [filterSource, setFilterSource] = useState('');
+  const [filterImpact, setFilterImpact] = useState('');
+
+  // Selected event
+  const [selectedId, setSelectedId] = useState<string | null>(null);
+  const selectedEvent = events.find(e => e.id === selectedId) ?? null;
+
+  // Analysis state
+  const [analyzing, setAnalyzing] = useState(false);
+  const [analysisText, setAnalysisText] = useState('');
+  const [affectedDocs, setAffectedDocs] = useState<AffectedDoc[]>([]);
+  const abortRef = useRef<AbortController | null>(null);
+
+  // Load events + stats
+  const loadFeed = useCallback(async () => {
+    setFeedLoading(true);
+    try {
+      const [evtRes, statsRes] = await Promise.all([
+        listEvents({
+          source: filterSource || undefined,
+          impact_level: filterImpact || undefined,
+        }),
+        getPerceptionStats(),
+      ]);
+      setEvents(evtRes.events);
+      setStats(statsRes);
+    } catch {
+      // silent
+    } finally {
+      setFeedLoading(false);
+    }
+  }, [filterSource, filterImpact]);
+
+  useEffect(() => {
+    const timerId = window.setTimeout(() => { void loadFeed(); }, 0);
+    return () => window.clearTimeout(timerId);
+  }, [loadFeed]);
+
+  // When selecting a new event, clear previous analysis
+  const handleSelectEvent = (id: string) => {
+    if (id === selectedId) return;
+    abortRef.current?.abort();
+    setSelectedId(id);
+    setAnalysisText('');
+    setAffectedDocs([]);
+    setAnalyzing(false);
+  };
+
+  const handleAnalyze = useCallback(() => {
+    if (!selectedId || analyzing) return;
+    abortRef.current?.abort();
+    const ctrl = new AbortController();
+    abortRef.current = ctrl;
+    setAnalysisText('');
+    setAffectedDocs([]);
+    setAnalyzing(true);
+
+    void analyzeEvent(
+      selectedId,
+      (msg) => {
+        if (msg.type === 'sources' && msg.docs) {
+          setAffectedDocs(msg.docs);
+        } else if (msg.type === 'content' && msg.text) {
+          setAnalysisText(prev => prev + msg.text);
+        } else if (msg.type === 'error') {
+          setAnalysisText(prev => prev + `\n\n⚠ 分析出错：${msg.text ?? '未知错误'}`);
+        }
+      },
+      () => setAnalyzing(false),
+      ctrl.signal,
+    );
+  }, [selectedId, analyzing]);
+
+  const handleAbort = () => {
+    abortRef.current?.abort();
+    setAnalyzing(false);
+  };
+
+  return (
+    <div className="relative flex min-h-0 flex-1 flex-col">
+      <style>{`
+        @keyframes blink { 0%,100%{opacity:1} 50%{opacity:0} }
+      `}</style>
+      <TPattern />
+
+      {/* Page header */}
+      <div style={{ display: 'flex', alignItems: 'baseline', gap: 16, marginBottom: 24 }}>
+        <h1 style={{ fontSize: 20, fontWeight: 700, color: theme.text, margin: 0 }}>智能感知</h1>
+        <span style={{ fontSize: 13, color: theme.text3 }}>法规动态实时追踪 · 知识库影响分析</span>
+      </div>
+
+      {/* Split layout */}
+      <div style={{
+        display: 'grid',
+        gridTemplateColumns: '400px 1fr',
+        gap: 24,
+        flex: 1,
+        minHeight: 560,
+      }}>
+        {/* Left: Event feed */}
+        <EventFeed
+          events={events}
+          selectedId={selectedId}
+          onSelect={handleSelectEvent}
+          filterSource={filterSource}
+          filterImpact={filterImpact}
+          onFilterSource={setFilterSource}
+          onFilterImpact={setFilterImpact}
+          stats={stats}
+          loading={feedLoading}
+        />
+
+        {/* Right: Analysis panel */}
+        <AnalysisPanel
+          event={selectedEvent}
+          analyzing={analyzing}
+          analysisText={analysisText}
+          affectedDocs={affectedDocs}
+          onAnalyze={handleAnalyze}
+          onAbort={handleAbort}
+        />
+      </div>
+    </div>
+  );
+};

frontend/src/pages/Perception/index.ts

@@ -0,0 +1 @@
+export { PerceptionPage } from './PerceptionPage';

frontend/src/pages/RagChat/CitedAnswer.tsx

@@ -1,4 +1,4 @@
-import React, { useRef } from 'react';
+import React from 'react';
 import { useTheme } from '../../contexts';
 import type { RetrievalData } from '../../types';
 

frontend/src/pages/RagChat/RagChatPage.tsx

@@ -133,7 +133,6 @@ export const RagChatPage: React.FC = () => {
       sessionId,
       abortRef.current.signal,
     );
-  // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [filterRegulationType, sessionId]);
 
   const sendMessage = (text: string) => {
@@ -173,7 +172,7 @@ export const RagChatPage: React.FC = () => {
   };
 
   return (
-    <div style={{ flex: 1, display: 'flex', height: 'calc(100vh - 128px)' }}>
+    <div style={{ flex: 1, display: 'flex', minHeight: 0, height: '100%' }}>
       {/* ── Left: chat panel ─────────────────────────────────── */}
       <div style={{
         flex: '0 0 60%',

frontend/src/pages/Status/StatusPage.tsx

@@ -1,6 +1,5 @@
 import React, { useCallback, useEffect, useState } from 'react';
 import { useTheme } from '../../contexts';
-import { Content } from '../../components/layout/Content';
 import { TPattern } from '../../components/common/TPattern';
 import { getSystemStats, getSystemConfig, getSystemHealth, type SystemStats, type SystemConfig, type SystemHealth } from '../../api/status';
 import { getDocumentList, type DocInfo } from '../../api/docs';
@@ -107,7 +106,8 @@ export const StatusPage: React.FC = () => {
 
   // Initial load
   useEffect(() => {
-    void loadData();
+    const timerId = window.setTimeout(() => { void loadData(); }, 0);
+    return () => window.clearTimeout(timerId);
   }, [loadData]);
 
   // Auto-poll every 5 s while any document is still processing
@@ -119,7 +119,7 @@ export const StatusPage: React.FC = () => {
   }, [docs, loadData]);
 
   return (
-    <Content>
+    <div className="relative w-full">
       <style>{`@keyframes spin { to { transform: rotate(360deg); } }`}</style>
       <TPattern />
 
@@ -386,6 +386,6 @@ export const StatusPage: React.FC = () => {
           </div>
         ))}
       </section>
-    </Content>
+    </div>
   );
 };

frontend/src/router/AppRouter.tsx

@@ -0,0 +1,20 @@
+import { BrowserRouter, Navigate, Route, Routes } from 'react-router-dom';
+
+import { AppShell } from '../components/layout/AppShell';
+import { appTabs, defaultTab } from './tabs';
+
+export function AppRouter() {
+  return (
+    <BrowserRouter>
+      <Routes>
+        <Route element={<AppShell />}>
+          <Route index element={<Navigate to={defaultTab.path} replace />} />
+          {appTabs.map((tab) => (
+            <Route key={tab.id} path={tab.path.slice(1)} element={null} />
+          ))}
+          <Route path="*" element={<Navigate to={defaultTab.path} replace />} />
+        </Route>
+      </Routes>
+    </BrowserRouter>
+  );
+}

frontend/src/router/tabs.tsx

@@ -0,0 +1,73 @@
+import type { ComponentType } from 'react';
+
+import { CompliancePage } from '../pages/Compliance';
+import { DocsPage } from '../pages/Docs';
+import { PerceptionPage } from '../pages/Perception';
+import { RagChatPage } from '../pages/RagChat';
+import { StatusPage } from '../pages/Status';
+
+export type TabId = 'perception' | 'docs' | 'compliance' | 'status' | 'rag';
+
+export type ContentWidth = 'default' | 'wide' | 'full';
+
+export interface AppTabConfig {
+  id: TabId;
+  path: string;
+  label: string;
+  component: ComponentType;
+  keepAlive: boolean;
+  contentWidth: ContentWidth;
+  fillHeight?: boolean;
+}
+
+export const appTabs: AppTabConfig[] = [
+  {
+    id: 'perception',
+    path: '/perception',
+    label: '智能感知',
+    component: PerceptionPage,
+    keepAlive: true,
+    contentWidth: 'wide',
+    fillHeight: true,
+  },
+  {
+    id: 'docs',
+    path: '/docs',
+    label: '文档管理',
+    component: DocsPage,
+    keepAlive: true,
+    contentWidth: 'default',
+  },
+  {
+    id: 'compliance',
+    path: '/compliance',
+    label: '合规分析',
+    component: CompliancePage,
+    keepAlive: true,
+    contentWidth: 'wide',
+    fillHeight: true,
+  },
+  {
+    id: 'status',
+    path: '/status',
+    label: '系统状态',
+    component: StatusPage,
+    keepAlive: true,
+    contentWidth: 'default',
+  },
+  {
+    id: 'rag',
+    path: '/rag',
+    label: '法规对话',
+    component: RagChatPage,
+    keepAlive: true,
+    contentWidth: 'wide',
+    fillHeight: true,
+  },
+];
+
+export const defaultTab = appTabs.find((tab) => tab.id === 'compliance') ?? appTabs[0];
+
+export function getTabByPath(pathname: string): AppTabConfig {
+  return appTabs.find((tab) => tab.path === pathname) ?? defaultTab;
+}

frontend/src/styles/globals.css

@@ -1,8 +1,9 @@
 @import url('https://fonts.googleapis.com/css2?family=TeleNeo:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500;600&display=swap');
+@import "tw-animate-css";
+@import "tailwindcss";
+
+@custom-variant dark (&:is(.dark *));
 
-@tailwind base;
-@tailwind components;
-@tailwind utilities;
 
 /* Light mode (default) */
 :root {
@@ -19,6 +20,38 @@
   --t-orange: #ff7700;
   --t-accent-glow: rgba(226,0,116,0.08);
   --t-pattern-opacity: 0.04;
+  --background: var(--t-bg);
+  --foreground: var(--t-text);
+  --card: var(--t-bg-card);
+  --card-foreground: var(--t-text);
+  --popover: var(--t-bg-card);
+  --popover-foreground: var(--t-text);
+  --primary: #e20074;
+  --primary-foreground: #ffffff;
+  --secondary: var(--t-bg-hover);
+  --secondary-foreground: var(--t-text);
+  --muted: var(--t-bg-hover);
+  --muted-foreground: var(--t-text3);
+  --accent: rgba(226, 0, 116, 0.08);
+  --accent-foreground: #e20074;
+  --destructive: #ff4444;
+  --border: var(--t-border);
+  --input: var(--t-border);
+  --ring: rgba(226, 0, 116, 0.35);
+  --chart-1: #e20074;
+  --chart-2: #be0060;
+  --chart-3: #00b89c;
+  --chart-4: #ff7700;
+  --chart-5: #4a4a5a;
+  --radius: 0.625rem;
+  --sidebar: var(--t-bg-card);
+  --sidebar-foreground: var(--t-text);
+  --sidebar-primary: #e20074;
+  --sidebar-primary-foreground: #ffffff;
+  --sidebar-accent: var(--t-bg-hover);
+  --sidebar-accent-foreground: var(--t-text);
+  --sidebar-border: var(--t-border);
+  --sidebar-ring: rgba(226, 0, 116, 0.35);
 }
 
 /* Dark mode */
@@ -36,6 +69,80 @@
   --t-orange: #ff8800;
   --t-accent-glow: rgba(226,0,116,0.12);
   --t-pattern-opacity: 0.03;
+  --background: var(--t-bg);
+  --foreground: var(--t-text);
+  --card: var(--t-bg-card);
+  --card-foreground: var(--t-text);
+  --popover: var(--t-bg-card);
+  --popover-foreground: var(--t-text);
+  --primary: #e20074;
+  --primary-foreground: #ffffff;
+  --secondary: var(--t-bg-hover);
+  --secondary-foreground: var(--t-text);
+  --muted: var(--t-bg-hover);
+  --muted-foreground: var(--t-text3);
+  --accent: rgba(226, 0, 116, 0.14);
+  --accent-foreground: #ff7abf;
+  --destructive: #ff4444;
+  --border: var(--t-border);
+  --input: var(--t-border-light);
+  --ring: rgba(226, 0, 116, 0.45);
+  --chart-1: #e20074;
+  --chart-2: #f04090;
+  --chart-3: #00d4aa;
+  --chart-4: #ff8800;
+  --chart-5: #c0c0d0;
+  --sidebar: var(--t-bg-card);
+  --sidebar-foreground: var(--t-text);
+  --sidebar-primary: #e20074;
+  --sidebar-primary-foreground: #ffffff;
+  --sidebar-accent: var(--t-bg-hover);
+  --sidebar-accent-foreground: var(--t-text);
+  --sidebar-border: var(--t-border);
+  --sidebar-ring: rgba(226, 0, 116, 0.45);
+}
+
+/* Dim mode — Indigo Dusk: deep navy-purple mid-tone between dark and light */
+.dim {
+  --t-bg: #1e1b2e;
+  --t-bg-card: #252237;
+  --t-bg-hover: #2d2945;
+  --t-bg-elevated: #292541;
+  --t-border: #3a3650;
+  --t-border-light: #504c6e;
+  --t-text: #f0eeff;
+  --t-text2: #b8b4d8;
+  --t-text3: #7a7698;
+  --t-green: #00c4a0;
+  --t-orange: #ff8820;
+  --t-accent-glow: rgba(226,0,116,0.14);
+  --t-pattern-opacity: 0.04;
+  --background: var(--t-bg);
+  --foreground: var(--t-text);
+  --card: var(--t-bg-card);
+  --card-foreground: var(--t-text);
+  --popover: var(--t-bg-card);
+  --popover-foreground: var(--t-text);
+  --primary: #e20074;
+  --primary-foreground: #ffffff;
+  --secondary: var(--t-bg-hover);
+  --secondary-foreground: var(--t-text);
+  --muted: var(--t-bg-hover);
+  --muted-foreground: var(--t-text3);
+  --accent: rgba(226, 0, 116, 0.12);
+  --accent-foreground: #f04090;
+  --destructive: #ff4444;
+  --border: var(--t-border);
+  --input: var(--t-border-light);
+  --ring: rgba(226, 0, 116, 0.40);
+  --sidebar: var(--t-bg-card);
+  --sidebar-foreground: var(--t-text);
+  --sidebar-primary: #e20074;
+  --sidebar-primary-foreground: #ffffff;
+  --sidebar-accent: var(--t-bg-hover);
+  --sidebar-accent-foreground: var(--t-text);
+  --sidebar-border: var(--t-border);
+  --sidebar-ring: rgba(226, 0, 116, 0.40);
 }
 
 /* Base styles */
@@ -66,6 +173,13 @@ body.dark-mode {
   background: #0a0a12;
 }
 
+/* Dim mode body */
+.dim body,
+body.dim-mode {
+  color: #f0eeff;
+  background: #1e1b2e;
+}
+
 /* Selection */
 ::selection {
   background: rgba(226, 0, 116, 0.3);
@@ -100,6 +214,11 @@ button, input {
   transition: none;
 }
 
+/* Shell navigation manages its own transition timing. */
+[data-shell-tab='true'] {
+  transition: color 0.2s ease-out;
+}
+
 /* T-Systems Button Style */
 .t-btn,
 .t-btn:hover {
@@ -131,17 +250,22 @@ button, input {
   background: linear-gradient(180deg, #12121f, #0a0a12);
 }
 
+/* Card gradient for dim mode */
+.dim .t-card-gradient {
+  background: linear-gradient(180deg, #e8e5f4, #f0eef8);
+}
+
 /* Card gradient for light mode */
-:not(.dark) .t-card-gradient {
+:not(.dark):not(.dim) .t-card-gradient {
   background: linear-gradient(180deg, #ffffff, #fafafa);
 }
 
 /* Light mode shadow for cards */
-:not(.dark) .t-card-shadow {
+:not(.dark):not(.dim) .t-card-shadow {
   box-shadow: 0 2px 8px rgba(226,0,116,0.04);
 }
 
-:not(.dark) .t-card-shadow-lg {
+:not(.dark):not(.dim) .t-card-shadow-lg {
   box-shadow: 0 4px 16px rgba(226,0,116,0.08);
 }
 
@@ -272,3 +396,59 @@ button, input {
     background: linear-gradient(135deg, #f0208a 0%, #d01070 100%);
   }
 }
+
+@theme inline {
+  --font-heading: 'TeleNeo', 'Segoe UI', system-ui, sans-serif;
+  --font-sans: 'TeleNeo', 'Segoe UI', system-ui, sans-serif;
+  --font-mono: 'JetBrains Mono', monospace;
+  --color-sidebar-ring: var(--sidebar-ring);
+  --color-sidebar-border: var(--sidebar-border);
+  --color-sidebar-accent-foreground: var(--sidebar-accent-foreground);
+  --color-sidebar-accent: var(--sidebar-accent);
+  --color-sidebar-primary-foreground: var(--sidebar-primary-foreground);
+  --color-sidebar-primary: var(--sidebar-primary);
+  --color-sidebar-foreground: var(--sidebar-foreground);
+  --color-sidebar: var(--sidebar);
+  --color-chart-5: var(--chart-5);
+  --color-chart-4: var(--chart-4);
+  --color-chart-3: var(--chart-3);
+  --color-chart-2: var(--chart-2);
+  --color-chart-1: var(--chart-1);
+  --color-ring: var(--ring);
+  --color-input: var(--input);
+  --color-border: var(--border);
+  --color-destructive: var(--destructive);
+  --color-accent-foreground: var(--accent-foreground);
+  --color-accent: var(--accent);
+  --color-muted-foreground: var(--muted-foreground);
+  --color-muted: var(--muted);
+  --color-secondary-foreground: var(--secondary-foreground);
+  --color-secondary: var(--secondary);
+  --color-primary-foreground: var(--primary-foreground);
+  --color-primary: var(--primary);
+  --color-popover-foreground: var(--popover-foreground);
+  --color-popover: var(--popover);
+  --color-card-foreground: var(--card-foreground);
+  --color-card: var(--card);
+  --color-foreground: var(--foreground);
+  --color-background: var(--background);
+  --radius-sm: calc(var(--radius) * 0.6);
+  --radius-md: calc(var(--radius) * 0.8);
+  --radius-lg: var(--radius);
+  --radius-xl: calc(var(--radius) * 1.4);
+  --radius-2xl: calc(var(--radius) * 1.8);
+  --radius-3xl: calc(var(--radius) * 2.2);
+  --radius-4xl: calc(var(--radius) * 2.6);
+}
+
+@layer base {
+  * {
+    @apply border-border outline-ring/50;
+  }
+  body {
+    @apply bg-background text-foreground;
+  }
+  html {
+    @apply font-sans;
+  }
+}