07-08-周三_10-09-42

15_图结构.ipynb

@@ -0,0 +1,656 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "1801d2fc",
+   "metadata": {},
+   "source": [
+    "# 15 图结构\n",
+    "\n",
+    "## 学习目标\n",
+    "1. 理解为什么要用图结构来组织复杂流程\n",
+    "2. 掌握 LangGraph 中节点、边、状态的配合方式\n",
+    "3. 学会构建顺序执行、条件分支和循环三种常见图结构\n",
+    "4. 能够看懂图结构的执行路径，并根据需求修改流程\n",
+    "5. 为后续多智能体、工作流和复杂 Agent 打下基础"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1f4e8fef",
+   "metadata": {},
+   "source": [
+    "## 1. 为什么需要图结构\n",
+    "\n",
+    "在前面的课程中，我们已经学过链（Chain）：\n",
+    "\n",
+    "```\n",
+    "输入 -> Prompt -> 模型 -> 输出\n",
+    "```\n",
+    "\n",
+    "这种结构很适合**线性流程**，也就是每一步都按固定顺序往下走。\n",
+    "\n",
+    "但是在真实应用中，流程往往没有这么简单。例如：\n",
+    "\n",
+    "- 如果用户的问题不完整，要先追问再继续\n",
+    "- 如果模型判断需要调用工具，要进入工具节点\n",
+    "- 如果工具返回的信息还不够，要再循环一次\n",
+    "- 如果满足某个条件，要走 A 路线，否则走 B 路线\n",
+    "\n",
+    "这时，单纯的链就不够用了。我们需要一种更灵活的结构，让流程能够：\n",
+    "\n",
+    "- **分支**：根据条件走不同路线\n",
+    "- **回路**：重复执行直到满足条件\n",
+    "- **汇合**：多条路径最后回到同一个结果\n",
+    "\n",
+    "这就是**图结构（Graph）**的价值。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "761e8840",
+   "metadata": {},
+   "source": [
+    "## 2. 图结构的核心组成\n",
+    "\n",
+    "在 LangGraph 中，一个图通常由三部分组成：\n",
+    "\n",
+    "### 2.1 State（状态）\n",
+    "状态可以理解为‘流程执行时共享的一份数据’。\n",
+    "\n",
+    "例如一个状态可能长这样：\n",
+    "\n",
+    "```python\n",
+    "{'value': 3, 'result': '通过', 'step': 2}\n",
+    "```\n",
+    "\n",
+    "图中的每个节点都可以读取这份状态，也可以返回对状态的更新。\n",
+    "\n",
+    "### 2.2 Node（节点）\n",
+    "节点就是图中的一个处理步骤。\n",
+    "\n",
+    "你可以把节点理解成‘流水线上的一个工位’。每个工位做一件事，例如：\n",
+    "\n",
+    "- 给数字加 1\n",
+    "- 判断成绩是否及格\n",
+    "- 调用大模型生成回复\n",
+    "- 调用搜索工具获取资料\n",
+    "\n",
+    "### 2.3 Edge（边）\n",
+    "边表示‘从哪个节点走到哪个节点’。\n",
+    "\n",
+    "边决定了执行顺序。常见有两种：\n",
+    "\n",
+    "- **普通边**：固定走向下一个节点\n",
+    "- **条件边**：根据状态判断下一步走哪条路\n",
+    "\n",
+    "可以把图结构想象成一张地铁路线图：\n",
+    "\n",
+    "- 状态 = 你当前身上的信息\n",
+    "- 节点 = 你经过的站点\n",
+    "- 边 = 站点之间的路线"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "bf0b4be0",
+   "metadata": {},
+   "source": [
+    "## 3. 第一个图：顺序执行\n",
+    "\n",
+    "先从最简单的情况开始：\n",
+    "\n",
+    "```\n",
+    "START -> 加一 -> 乘二 -> END\n",
+    "```\n",
+    "\n",
+    "也就是说：\n",
+    "\n",
+    "1. 初始值是 `value=3`\n",
+    "2. 第一个节点先把它加 1，变成 4\n",
+    "3. 第二个节点再把它乘 2，变成 8\n",
+    "4. 最后输出结果"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "527c7a13",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from typing_extensions import TypedDict\n",
+    "from langgraph.graph import StateGraph, START, END\n",
+    "\n",
+    "# 1. 定义状态结构\n",
+    "class NumberState(TypedDict):\n",
+    "    value: int\n",
+    "\n",
+    "# 2. 定义节点函数\n",
+    "def add_one(state: NumberState):\n",
+    "    print(f'进入 add_one 前，value = {state[\"value\"]}')\n",
+    "    return {'value': state['value'] + 1}\n",
+    "\n",
+    "def multiply_two(state: NumberState):\n",
+    "    print(f'进入 multiply_two 前，value = {state[\"value\"]}')\n",
+    "    return {'value': state['value'] * 2}\n",
+    "\n",
+    "# 3. 创建图\n",
+    "builder = StateGraph(NumberState)\n",
+    "\n",
+    "# 4. 添加节点\n",
+    "builder.add_node('add_one', add_one)\n",
+    "builder.add_node('multiply_two', multiply_two)\n",
+    "\n",
+    "# 5. 添加边\n",
+    "builder.add_edge(START, 'add_one')\n",
+    "builder.add_edge('add_one', 'multiply_two')\n",
+    "builder.add_edge('multiply_two', END)\n",
+    "\n",
+    "# 6. 编译图\n",
+    "graph = builder.compile()\n",
+    "\n",
+    "# 7. 执行图\n",
+    "result = graph.invoke({'value': 3})\n",
+    "print('最终结果：', result)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "62406ccb",
+   "metadata": {},
+   "source": [
+    "### 代码解释\n",
+    "\n",
+    "这段代码虽然不长，但非常关键，值得逐行理解。\n",
+    "\n",
+    "#### 第 1 步：定义状态\n",
+    "`NumberState` 规定了图里共享的数据结构，这里只有一个字段：`value`。\n",
+    "\n",
+    "也就是说，整张图传来传去的核心数据就是一个整数。\n",
+    "\n",
+    "#### 第 2 步：定义节点函数\n",
+    "`add_one(state)` 和 `multiply_two(state)` 都接收当前状态。\n",
+    "\n",
+    "例如当状态是：\n",
+    "\n",
+    "```python\n",
+    "{'value': 3}\n",
+    "```\n",
+    "\n",
+    "那么 `add_one` 返回：\n",
+    "\n",
+    "```python\n",
+    "{'value': 4}\n",
+    "```\n",
+    "\n",
+    "节点函数**不需要返回完整状态**，只要返回要更新的字段即可。LangGraph 会把更新结果合并回状态中。\n",
+    "\n",
+    "#### 第 3 步：创建图\n",
+    "`StateGraph(NumberState)` 表示：我们要创建一张图，并且这张图中的状态结构由 `NumberState` 定义。\n",
+    "\n",
+    "#### 第 4 步：添加节点\n",
+    "`builder.add_node('add_one', add_one)` 的意思是：\n",
+    "\n",
+    "- 节点名字叫 `add_one`\n",
+    "- 这个节点真正执行的函数是 Python 函数 `add_one`\n",
+    "\n",
+    "节点名字更像是图里的‘标签’，函数才是‘真正干活的人’。\n",
+    "\n",
+    "#### 第 5 步：添加边\n",
+    "这三行定义了执行路径：\n",
+    "\n",
+    "- 从 `START` 进入 `add_one`\n",
+    "- 从 `add_one` 进入 `multiply_two`\n",
+    "- 从 `multiply_two` 进入 `END`\n",
+    "\n",
+    "因此执行顺序是固定的，没有分支，也没有循环。\n",
+    "\n",
+    "#### 第 6 步：编译图\n",
+    "`compile()` 会把‘图的定义’变成一个真正可运行的对象。\n",
+    "\n",
+    "你可以把它理解为：前面是在画流程图，这一步是把流程图变成程序。\n",
+    "\n",
+    "#### 第 7 步：运行图\n",
+    "`graph.invoke({'value': 3})` 表示用初始状态 `{'value': 3}` 启动整张图。\n",
+    "\n",
+    "执行过程是：\n",
+    "\n",
+    "1. `value=3` 进入 `add_one`\n",
+    "2. 更新为 `value=4`\n",
+    "3. `value=4` 进入 `multiply_two`\n",
+    "4. 更新为 `value=8`\n",
+    "5. 图结束，输出 `{'value': 8}`"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e86fbc2b",
+   "metadata": {},
+   "source": [
+    "## 4. 查看图结构\n",
+    "\n",
+    "除了运行图，我们还可以把图的结构打印出来，帮助自己检查流程是否连对了。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "029c26ae",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print('节点：')\n",
+    "print(graph.get_graph().nodes)\n",
+    "\n",
+    "print('\\n边：')\n",
+    "print(graph.get_graph().edges)\n",
+    "\n",
+    "print('\\nMermaid 图：')\n",
+    "print(graph.get_graph().draw_mermaid())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "996a24d3",
+   "metadata": {},
+   "source": [
+    "### 代码解释\n",
+    "\n",
+    "这一段代码的重点不在‘计算’，而在‘观察图’。\n",
+    "\n",
+    "#### `graph.get_graph().nodes`\n",
+    "它会返回当前图中有哪些节点。你可以把它理解为‘流程图里有哪些方框’。\n",
+    "\n",
+    "#### `graph.get_graph().edges`\n",
+    "它会返回节点之间的连接关系，也就是‘箭头从哪里指向哪里’。\n",
+    "\n",
+    "#### `draw_mermaid()`\n",
+    "它会生成 Mermaid 格式的流程图文本。\n",
+    "\n",
+    "Mermaid 是一种用文本描述流程图的语法，例如：\n",
+    "\n",
+    "```\n",
+    "START --> add_one --> multiply_two --> END\n",
+    "```\n",
+    "\n",
+    "你可以把输出复制到支持 Mermaid 的工具中查看图形化效果。\n",
+    "\n",
+    "这在图结构变复杂之后尤其有用，因为：\n",
+    "\n",
+    "- 可以快速确认自己有没有连错边\n",
+    "- 可以帮助团队成员理解流程\n",
+    "- 可以作为调试流程的一部分"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "9117ea6b",
+   "metadata": {},
+   "source": [
+    "## 5. 条件分支图\n",
+    "\n",
+    "前面的图是固定顺序执行。现在我们来做一个更像真实业务的例子：**根据分数决定是否通过**。\n",
+    "\n",
+    "流程如下：\n",
+    "\n",
+    "```\n",
+    "START -> 判断分数 -> 合格 / 不合格 -> END\n",
+    "```\n",
+    "\n",
+    "如果分数大于等于 60，就走‘通过’节点；否则走‘补考’节点。\n",
+    "\n",
+    "这就是典型的**条件分支**。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ed38a5d4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from typing_extensions import TypedDict\n",
+    "from langgraph.graph import StateGraph, START, END\n",
+    "\n",
+    "class ScoreState(TypedDict):\n",
+    "    score: int\n",
+    "    result: str\n",
+    "\n",
+    "def check_score(state: ScoreState):\n",
+    "    print(f'当前分数：{state[\"score\"]}')\n",
+    "    return {}\n",
+    "\n",
+    "def pass_node(state: ScoreState):\n",
+    "    return {'result': '成绩合格，顺利通过'}\n",
+    "\n",
+    "def fail_node(state: ScoreState):\n",
+    "    return {'result': '成绩不合格，需要补考'}\n",
+    "\n",
+    "def route_by_score(state: ScoreState):\n",
+    "    if state['score'] >= 60:\n",
+    "        return 'pass_node'\n",
+    "    return 'fail_node'\n",
+    "\n",
+    "builder = StateGraph(ScoreState)\n",
+    "builder.add_node('check_score', check_score)\n",
+    "builder.add_node('pass_node', pass_node)\n",
+    "builder.add_node('fail_node', fail_node)\n",
+    "\n",
+    "builder.add_edge(START, 'check_score')\n",
+    "builder.add_conditional_edges(\n",
+    "    'check_score',\n",
+    "    route_by_score,\n",
+    "    {\n",
+    "        'pass_node': 'pass_node',\n",
+    "        'fail_node': 'fail_node'\n",
+    "    }\n",
+    ")\n",
+    "builder.add_edge('pass_node', END)\n",
+    "builder.add_edge('fail_node', END)\n",
+    "\n",
+    "score_graph = builder.compile()\n",
+    "\n",
+    "print('分数 85 的结果：')\n",
+    "print(score_graph.invoke({'score': 85, 'result': ''}))\n",
+    "\n",
+    "print('\\n分数 45 的结果：')\n",
+    "print(score_graph.invoke({'score': 45, 'result': ''}))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a8a9080b",
+   "metadata": {},
+   "source": [
+    "### 代码解释\n",
+    "\n",
+    "这个例子第一次出现了**条件边**，所以需要重点掌握。\n",
+    "\n",
+    "#### 状态结构\n",
+    "这里的状态包含两个字段：\n",
+    "\n",
+    "- `score`：输入分数\n",
+    "- `result`：最终判断结果\n",
+    "\n",
+    "也就是说，这张图的目标是：根据 `score` 计算出 `result`。\n",
+    "\n",
+    "#### `check_score` 节点\n",
+    "这个节点本身没有修改状态，只是把当前分数打印出来。\n",
+    "\n",
+    "它返回 `{}`，表示‘这个节点不更新任何字段’。\n",
+    "\n",
+    "这说明一个节点不一定非要做数据修改，它也可以只承担‘中转’或‘判断前准备’的作用。\n",
+    "\n",
+    "#### `pass_node` 和 `fail_node`\n",
+    "这两个节点分别代表两条不同路径的处理结果：\n",
+    "\n",
+    "- 如果通过，更新 `result='成绩合格，顺利通过'`\n",
+    "- 如果不通过，更新 `result='成绩不合格，需要补考'`\n",
+    "\n",
+    "这两个节点就像流程图里分叉后的两条支路。\n",
+    "\n",
+    "#### `route_by_score`：路由函数\n",
+    "这是本例最核心的部分。\n",
+    "\n",
+    "它不是真正的业务节点，而是一个‘路线选择器’。它根据当前状态判断：\n",
+    "\n",
+    "- `score >= 60` 返回 `'pass_node'`\n",
+    "- 否则返回 `'fail_node'`\n",
+    "\n",
+    "返回值必须和后面条件边映射表中的 key 对应上。\n",
+    "\n",
+    "#### `add_conditional_edges(...)`\n",
+    "这段代码的意思是：\n",
+    "\n",
+    "从 `check_score` 节点出来后，不是固定走下一步，而是调用 `route_by_score(state)`。\n",
+    "\n",
+    "如果返回 `'pass_node'`，就走到 `pass_node`；如果返回 `'fail_node'`，就走到 `fail_node`。\n",
+    "\n",
+    "你可以把它理解成：\n",
+    "\n",
+    "- 普通边 = 固定导航\n",
+    "- 条件边 = 智能导航\n",
+    "\n",
+    "#### 最终执行效果\n",
+    "当输入 `score=85` 时，流程是：\n",
+    "\n",
+    "```\n",
+    "START -> check_score -> pass_node -> END\n",
+    "```\n",
+    "\n",
+    "当输入 `score=45` 时，流程是：\n",
+    "\n",
+    "```\n",
+    "START -> check_score -> fail_node -> END\n",
+    "```\n",
+    "\n",
+    "这就是图结构中‘根据条件走不同路线’的最基本模式。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "3cf69b96",
+   "metadata": {},
+   "source": [
+    "## 6. 循环图\n",
+    "\n",
+    "除了分支，图结构还有一个非常重要的能力：**循环**。\n",
+    "\n",
+    "在很多智能体任务中，我们都需要重复执行某些步骤，比如：\n",
+    "\n",
+    "- 检索不到足够信息就继续搜索\n",
+    "- 模型输出格式不对就重新生成\n",
+    "- 工具调用结果不完整就继续补充\n",
+    "\n",
+    "下面我们用一个简单例子说明循环：\n",
+    "\n",
+    "让 `count` 从 0 开始，每次加 1，直到加到 3 为止。\n",
+    "\n",
+    "流程如下：\n",
+    "\n",
+    "```\n",
+    "START -> increment -> 判断是否结束\n",
+    "                 ↘ 未结束，回到 increment\n",
+    "                 ↘ 已结束，进入 END\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "d33d64ec",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from typing_extensions import TypedDict\n",
+    "from langgraph.graph import StateGraph, START, END\n",
+    "\n",
+    "class LoopState(TypedDict):\n",
+    "    count: int\n",
+    "\n",
+    "def increment(state: LoopState):\n",
+    "    new_count = state['count'] + 1\n",
+    "    print(f'当前 count 从 {state[\"count\"]} 变成 {new_count}')\n",
+    "    return {'count': new_count}\n",
+    "\n",
+    "def should_continue(state: LoopState):\n",
+    "    if state['count'] < 3:\n",
+    "        return 'continue'\n",
+    "    return 'stop'\n",
+    "\n",
+    "builder = StateGraph(LoopState)\n",
+    "builder.add_node('increment', increment)\n",
+    "\n",
+    "builder.add_edge(START, 'increment')\n",
+    "builder.add_conditional_edges(\n",
+    "    'increment',\n",
+    "    should_continue,\n",
+    "    {\n",
+    "        'continue': 'increment',\n",
+    "        'stop': END\n",
+    "    }\n",
+    ")\n",
+    "\n",
+    "loop_graph = builder.compile()\n",
+    "result = loop_graph.invoke({'count': 0})\n",
+    "print('最终状态：', result)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "26274c73",
+   "metadata": {},
+   "source": [
+    "### 代码解释\n",
+    "\n",
+    "这个例子看起来简单，但它已经具备了很多 Agent 工作流的核心思想。\n",
+    "\n",
+    "#### `increment` 节点\n",
+    "它的职责非常明确：每次把 `count` 加 1。\n",
+    "\n",
+    "例如：\n",
+    "\n",
+    "- 输入 `count=0`，输出 `count=1`\n",
+    "- 输入 `count=1`，输出 `count=2`\n",
+    "- 输入 `count=2`，输出 `count=3`\n",
+    "\n",
+    "#### `should_continue` 路由函数\n",
+    "它负责判断是继续循环，还是结束流程。\n",
+    "\n",
+    "规则很简单：\n",
+    "\n",
+    "- 如果 `count < 3`，返回 `'continue'`\n",
+    "- 否则返回 `'stop'`\n",
+    "\n",
+    "#### 条件边的映射\n",
+    "这里最关键的一点是：\n",
+    "\n",
+    "- `'continue'` 映射到 `'increment'`\n",
+    "- `'stop'` 映射到 `END`\n",
+    "\n",
+    "这就意味着：\n",
+    "\n",
+    "- 只要还没达到条件，就重新回到同一个节点\n",
+    "- 达到条件后，就退出图\n",
+    "\n",
+    "这就是循环结构的本质。\n",
+    "\n",
+    "#### 执行过程\n",
+    "如果初始状态是 `{'count': 0}`，执行路径如下：\n",
+    "\n",
+    "1. 进入 `increment`，变成 `count=1`\n",
+    "2. 判断 `1 < 3`，继续\n",
+    "3. 再次进入 `increment`，变成 `count=2`\n",
+    "4. 判断 `2 < 3`，继续\n",
+    "5. 再次进入 `increment`，变成 `count=3`\n",
+    "6. 判断 `3 < 3` 不成立，停止\n",
+    "7. 进入 `END`\n",
+    "\n",
+    "也就是说，图结构中的循环并不是 `for` 或 `while` 语法写出来的，而是通过‘节点 + 条件边回跳’实现的。\n",
+    "\n",
+    "这也是 LangGraph 很强大的地方：它把‘流程控制’变成了图本身的一部分。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "8c2c4ae1",
+   "metadata": {},
+   "source": [
+    "## 7. 图结构和普通 Python 流程控制有什么区别\n",
+    "\n",
+    "你可能会问：\n",
+    "\n",
+    "‘这些分支和循环，我直接用 `if`、`while` 不也能写吗？’\n",
+    "\n",
+    "当然可以。但图结构有几个明显优势：\n",
+    "\n",
+    "| 对比点 | 普通 Python 代码 | 图结构 |\n",
+    "| --- | --- | --- |\n",
+    "| 流程表达方式 | 写在代码逻辑里 | 显式画成节点和边 |\n",
+    "| 可视化能力 | 弱 | 强 |\n",
+    "| 分支/回路可读性 | 复杂时容易乱 | 更直观 |\n",
+    "| 与 Agent / 工作流结合 | 需要手动组织 | 天然适合 |\n",
+    "| 状态传递 | 自己管理变量 | 统一用 State 管理 |\n",
+    "\n",
+    "最核心的一点是：\n",
+    "\n",
+    "**图结构不是为了替代 Python，而是为了让复杂流程更容易设计、理解、调试和扩展。**"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "f82467cf",
+   "metadata": {},
+   "source": [
+    "## 8. 一个更贴近 Agent 的理解方式\n",
+    "\n",
+    "如果把图结构放到 AI Agent 场景中，可以这样理解：\n",
+    "\n",
+    "- 一个节点可以是‘调用大模型’\n",
+    "- 一个节点可以是‘调用工具’\n",
+    "- 一个节点可以是‘检查结果是否满足要求’\n",
+    "- 条件边可以决定‘是继续调用工具，还是给用户最终答案’\n",
+    "\n",
+    "也就是说，后面你学到的复杂 Agent，本质上大多都可以拆成：\n",
+    "\n",
+    "```\n",
+    "理解问题 -> 决策 -> 调工具 -> 检查结果 -> 继续/结束\n",
+    "```\n",
+    "\n",
+    "而图结构正是把这条路线显式表达出来的最好方式。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "f8ced671",
+   "metadata": {},
+   "source": [
+    "## 9. 本节小结\n",
+    "\n",
+    "本节课你需要记住三个最重要的结论：\n",
+    "\n",
+    "1. **State 是共享数据**：节点之间通过状态传递信息\n",
+    "2. **Node 是处理步骤**：每个节点负责完成一个明确任务\n",
+    "3. **Edge 决定执行路径**：普通边用于顺序执行，条件边用于分支和循环\n",
+    "\n",
+    "掌握了这三点，你就已经理解了图结构的核心。后面无论是多轮工作流、工具调用循环，还是多智能体协作，本质上都只是图结构的进一步扩展。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "3ec471d7",
+   "metadata": {},
+   "source": [
+    "## 10. 本节练习\n",
+    "\n",
+    "1. 修改顺序执行示例，在 `multiply_two` 后面再加一个节点 `minus_three`，让最终结果再减 3\n",
+    "2. 修改条件分支示例，把及格线从 60 改成 80，观察结果变化\n",
+    "3. 修改循环示例，把终止条件从 `count == 3` 改成 `count == 5`\n",
+    "4. 尝试自己画出三个示例的流程图\n",
+    "5. 思考：如果要做一个‘查询资料直到找到答案为止’的 Agent，它更像本节中的分支图，还是循环图？为什么？"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": ".venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.14.4"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}