xyaz1313/xyskill / skills/xy-ontology
xy-ontology
Verified【本体论与FDE入门】把"本体论/Ontology 是什么、FDE 是什么岗位、为什么企业 AI 项目八成栽在这一步"讲成人话——先问你处境再答,答完给下一步,用 197 条本体论原子当依据。管的是"消费方"视角——我自己/我们公司要不要用这套方法论、要不要给自己建本体、值不值得投入("我自己想学/想接单去做FDE"这种供给方问法走xy-fde)。用户说「本体论到底是什么」「Ontology 跟知识图谱有什么区别」「FDE 是干嘛的」「为什么我们上了 RAG 还是不好用」「Palantir 那套是怎么回事」「要不要给我们公司建本体」「我们要不要做AI变革/AI改造,从哪开始」「值不值得投入本体论」时使用——判据是"建/用/投入/值不值得"这类消费方动词,不是"做/学/接"这类供给方动词。|小爷出品
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.6 KB |
| agents/openai.yaml | config | 353 B |
| evals/evals.json | config | 1.4 KB |
| references/atoms.jsonl | other | 168.4 KB |
| scripts/atoms-search.pyexecutable | script | 11.5 KB |
references/atoms.jsonl
View on GitHub{"id": "XY-ONT-001", "type": "principle", "knowledge": "Palantir 平台通过 \"AI Mesh\" 协同,形成从数据整合(Foundry)到决策分析(Gotham),再到调度部署(Apollo)与智能赋能(AIP)的 \"数据-决策-执行\" 闭环。", "original": "平台通过\"AI Mesh\"协同:Foundry整合数据 → Gotham决策分析 → Apollo调度部署 → AIP智能赋能,形成\"数据-决策-执行\"闭环。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-002", "type": "definition", "knowledge": "Object Types 是对真实世界实体或事件的 Schema 定义。", "original": "Object Types(对象类型)\\n\\n**定义**:真实世界实体或事件的Schema定义。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-003", "type": "definition", "knowledge": "Object 是对特定实体的实例化,类似于数据集中的一行;Object Set 是经过过滤后的多个对象集合。", "original": "Object Type | 数据集Schema\\nObject(对象/实例) | 数据集的一行(如\"某员工\"作为一个Employee)\\nObject Set(对象集) | 过滤后的多行(如\"所有资深员工\")", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-004", "type": "method", "knowledge": "在 Ontology Manager 中通过为对象类型添加支持的数据源来创建对象。", "original": "通过在Ontology Manager中为对象类型添加支持数据源来创建对象。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-005", "type": "method", "knowledge": "创建对象类型时应确保其代表现实实体(如员工、飞机),保持类型聚焦以代表独立实体,并使用具体、描述性的命名。", "original": "代表现实实体(员工、飞机、采购订单)\\n保持对象类型聚焦,每个代表一个独立实体\\n使用具体、描述性的命名", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-006", "type": "definition", "knowledge": "Properties 是对象的特征,分为直接映射数据源字段的基础属性、通过计算获得的派生属性以及支持历史数据追踪的时间序列属性。", "original": "Properties(属性)\\n\\n**定义**:对象的特征(firstName、salary、latitude等)。\\n\\n类型:\\n- 基础属性:直接映射数据源字段\\n- 派生属性(Derived Properties):通过计算获得(Beta功能,仅Python OSDK 2.x支持)\\n- 时间序列属性:支持历史数据追踪", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-007", "type": "definition", "knowledge": "Link Types 是对象之间关系的 Schema 定义,用于建立实体间的关联。", "original": "Link Types(链接类型)\\n\\n**定义**:对象之间关系的Schema定义。\\n\\n用途:建立实体间的关联(如Employee链接到Department、Order链接到Customer)。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-008", "type": "definition", "knowledge": "Action Types 定义了用户可以一次性执行的一组对象、属性值和链接的变更,以及提交时的副作用行为。", "original": "Action Types(动作类型)\\n\\n**定义**:用户可以一次性执行的一组对象、属性值和链接的变更定义,以及提交时的副作用行为。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-009", "type": "insight", "knowledge": "对对象、属性值和链接的任何变更都会提交到 Ontology 并反映在所有用户应用中,变更会被捕获在对象类型的\"回写数据集\"中,从而创建可审计的用户决策历史。", "original": "\"对对象、属性值和链接的任何变更将在用户执行动作时提交到Ontology,并反映在所有用户应用中\"\\n\\n变更被捕获在对象类型的\"回写数据集\"中,创建可审计的用户决策历史。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-010", "type": "method", "knowledge": "针对不同场景应选择合适的工具:人类决策或用户编辑使用 Action Types;数据转换/聚合/预计算使用批处理管道;连续低延迟处理使用流式管道;事件驱动变更使用自动化;复杂实时计算/验证使用函数。", "original": "人类决策、用户发起的编辑 | Action Types\\n数据转换、聚合、预计算派生值 | 批处理管道\\n连续低延迟数据处理 | 流式管道\\n事件驱动的Ontology变更反应 | 自动化\\n复杂实时计算、验证逻辑 | 函数", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-011", "type": "definition", "knowledge": "Functions 是可复用的计算逻辑,支持 Python 和 TypeScript v2 对 OSDK 的一流支持,并支持属性子选择。", "original": "Functions(函数)\\n\\n**定义**:可复用的计算逻辑。\\n\\n2025年更新:\\n- Python和TypeScript v2 Functions对OSDK提供一流支持\\n- 属性子选择:只加载需要的数据\\n- 可在Marketplace产品中作为内容包含", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-012", "type": "definition", "knowledge": "Interfaces 是用于建模共享特征的多态抽象,可以支持不同对象类型的统一处理。", "original": "Interfaces(接口)\\n\\n**定义**:多态抽象,用于建模共享特征(较新功能)。\\n\\n用途:\\n- 用接口建模共享特征\\n- 避免\"God Object\"反模式\\n- 支持不同对象类型的统一处理", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-013", "type": "anti-pattern", "knowledge": "应避免以下反模式:系统孤岛(同一实体按源系统拆分)、厨房水槽(包含无关技术列)、部门孤岛(各部门创建自己的版本)、上帝对象(一个对象类型代表多个实体)、金锤(单一工具过度使用)、动作蔓延(大量单属性动作)、时间机器(历史版本作为独立对象)、命名不当(命名模糊误导)。", "original": "System Silos(系统孤岛)... The Kitchen Sink(厨房水槽)... Department Silos(部门孤岛)... The God Object(上帝对象)... The Golden Hammer(金锤)... Action Sprawl(动作蔓延)... The Time Machine(时间机器)... The Misnomer(命名不当)", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-014", "type": "anti-pattern", "knowledge": "针对 System Silos(系统孤岛),应通过管道合并数据来创建统一的对象类型。", "original": "System Silos(系统孤岛) | 同一实体按源系统拆分 | 管道合并数据,创建统一对象类型", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-015", "type": "anti-pattern", "knowledge": "针对 The Kitchen Sink(厨房水槽),应有意筛选属性,排除 ETL 元数据。", "original": "The Kitchen Sink(厨房水槽) | 包含无关技术列 | 有意筛选属性,排除ETL元数据", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-016", "type": "anti-pattern", "knowledge": "针对 Department Silos(部门孤岛),应创建共享对象类型,并用属性或链接进行区分。", "original": "Department Silos(部门孤岛) | 各部门创建自己的版本 | 创建共享对象类型,用属性/链接区分", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-017", "type": "anti-pattern", "knowledge": "针对 The God Object(上帝对象),应创建独立对象类型,并使用接口来共享特征。", "original": "The God Object(上帝对象) | 一个对象类型代表多个实体 | 创建独立对象类型,用接口共享特征", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-018", "type": "anti-pattern", "knowledge": "针对 Action Sprawl(动作蔓延),应围绕业务操作进行设计,将相关变更进行捆绑。", "original": "Action Sprawl(动作蔓延) | 大量单属性动作 | 围绕业务操作设计,捆绑相关变更", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-019", "type": "anti-pattern", "knowledge": "针对 The Time Machine(时间机器),应采用单对象结合链接历史对象及时间序列属性的设计。", "original": "The Time Machine(时间机器) | 历史版本作为独立对象 | 单对象+链接历史对象+时间序列属性", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-020", "type": "principle", "knowledge": "AIP Agent Studio 不是独立产品,而是 Palantir AIP 平台的一个功能模块,用于在 Foundry 平台内构建 AI 代理。", "original": "AIP Agent Studio不是独立产品,它是Palantir AIP平台的一个功能模块,用于在Foundry平台内构建AI代理。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-021", "type": "definition", "knowledge": "Agent Studio 是 Foundry/AIP 平台中的一个配置界面,允许用户通过可视化方式搭建 AI 助手。", "original": "Agent Studio是Foundry/AIP平台的一个配置界面,让你用可视化方式搭建AI助手", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-022", "type": "principle", "knowledge": "在 Agent Studio 的架构中,Agent Studio 作为配置界面,而 Ontology 是 Agent 的大脑知识库。", "original": "Agent Studio是\"配置界面\",Ontchantology是Agent的\"大脑知识库\"。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-023", "type": "method", "knowledge": "Agent 通过读取 Object Types 获取结构化知识,使用 Object Query Tool 查询数据,使用 Action Tool 执行编辑操作,并通过 Links 理解实体间的关系。", "original": "Agent通过以下方式使用Ontology:\n• 读取Object Types作为结构化知识\n• 使用Object Query Tool查询对象数据\n• 使用Action Tool执行Ontology编辑(写操作)\n• 通过Links理解实体间关系", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-024", "type": "method", "knowledge": "在 Agent Studio 配置中,可以通过为 Agent 配备 Object Query Tool(查询对象)、Action Tool(执行编辑)和 Function Tool(调用函数)来实现与 Ontology 的交互。", "original": "当用户在Agent Studio配置Agent时,可以给Agent配备以下与Ontology相关的工具:\n• Object Query Tool: 查询Ontology中的对象\n• Action Tool: 执行Ontology编辑操作\n• Function Tool: 调用Foundry函数", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-025", "type": "principle", "knowledge": "Agent 使用 Ontology 的核心机制包括:利用其作为上下文回答问题、通过读取和更新应用状态来管理状态,以及利用平台安全模型控制 LLM 的访问权限。", "original": "Agent使用Ontology作为上下文(Context)回答用户问题\nAgent可以读取和更新应用状态(通过Ontology)\n平台安全模型控制LLM只能访问必要的数据", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-026", "type": "insight", "knowledge": "与传统 RAG 从文档中检索文本片段不同,Palantir 的 OAG (Ontology Augmented Generation) 模式是基于从 Ontology 中检索结构化数据(如对象属性值和对象间关系)来增强生成过程。", "original": "传统RAG:LLM从文档中检索文本片段 → 生成回答\nPalantir的OAG(Ontology Augmented Generation):从Ontology检索结构化数据(对象属性值、对象间关系)", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-027", "type": "method", "knowledge": "OAG(Ontology-Augmented Generation)的工作流程包括:用户提问 -> Agent理解意图 -> 从Ontology检索结构化数据(对象属性值、对象间关系、相关对象集) -> 基于结构化业务数据生成回答 -> (可选)通过Action Tool执行Ontology编辑。", "original": "用户提问 -> Agent理解意图 -> 从Ontology检索结构化数据(对象属性值、对象间关系、相关对象集) -> 基于结构化业务数据生成回答 -> (可选)通过Action Tool执行Ontology编辑。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-028", "type": "insight", "knowledge": "相比普通RAG,OAG的优势在于:1. 语义准确性:理解业务实体而非仅匹配文本字符串;2. 关系推理:通过Links理解实体间的业务关系;3. 可执行操作:能通过Actions修改数据。", "original": "OAG vs 普通R가的优势:语义准确性:不是匹配文本字符串,而是理解业务实体;关系推理:通过Links理解实体间的业务关系;可执行操作:不仅能回答,还能通过Actions修改数据。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-029", "type": "case", "knowledge": "在销售场景中,Agent通过Object Query Tool查询具有特定属性(如合同价值、最后联系日期)的Customer对象,生成高价值客户列表;用户通过指令让Agent使用Action Tool更新对象属性或创建新的Link(如创建跟进任务),实现从查询到业务执行的闭环。", "original": "场景:销售经理问Agent\"帮我看看有哪些高价值客户最近没有跟进\"... Agent使用Action Tool执行Ontology编辑:更新Customer_A对象的status属性,创建与FollowUp_Task对象的Link,触发通知给相关销售。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-030", "type": "insight", "knowledge": "AI与Ontology的协同价值体现在:1. 语义一致性:AI理解业务概念而非仅是数据库表名;2. 结构化RAG:检索的是结构化业务数据而非文本片段;3. 可执行AI:Agent能通过Actions执行业务操作;4. 治理安全:利用Ontology权限模型控制AI访问范围;5. 双向增强:AI发现洞察,用户通过Actions完善Ont义。", "original": "AI与Ontology的协同价值:语义一致性:AI理解\"客户\"、\"订单\"等业务概念,而非只是数据库表名;结构化RAG:从Ontology检索的是结构化业务数据,而非文本片段;可执行AI:Agent不仅能回答,还能通过Actions执行业务操作;治理安全:通过Ontology的权限模型控制AI的数据访问范围;双向增强:AI帮助用户发现Ontology中的洞察,用户使用Actions完善Ontolog", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-031", "type": "principle", "knowledge": "Palantir Ontology与数据仓库的区别在于:建模面向业务操作而非分析,数据为实时运营数据而非清洗后的历史数据,用户为业务人员,查询方式为自然语言,支持Actions实时回写。", "original": "vs 数据仓库(Data Warehouse):建模:面向业务操作;数据:实时/近实时的运营数据;用户:业务人员+AI Agent;查询:自然语言+对象导航;写入:Actions实时回写。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-032", "type": "principle", "knowledge": "Palantir Ontology与知识图谱的区别在于:建模采用支持继承的类面向对象模型而非RDF三元组,数据直接映射源头实现实时同步,定位为决策执行中台而非仅作为知识展示层,且具备驱动工作流执行的能力。", "original": "vs 知识图谱(Knowledge Graph):建模:类面向对象,支持继承;数据:直接映射数据源,实时同步;执行:可驱动工作流执行;定位:决策执行中台。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-033", "type": "principle", "knowledge": "Palentir Ontology与ORM的区别在于:范围是跨系统的企业级数据层而非应用内部层,用户涵盖业务人员,且具备AI原生设计支持。", "original": "vs ORM(对象关系映射):范围:跨系统企业级数据层;用户:业务人员+开发人员;AI:AI原生设计。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-034", "type": "definition", "knowledge": "Palantir Ontology 的定义是让业务人员和 AI 能够理解、操作企业数据,并能够自动执行决策的平台。", "original": "Palantir Ontology:\"让业务和AI能理解和操作企业数据,并自动执行决策\"", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-035", "type": "definition", "knowledge": "数据仓库的核心功能是存储历史数据用于分析报表。", "original": "数据仓库:\"存储历史数据用于分析报表\"", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-036", "type": "definition", "knowledge": "知识图谱的核心功能是展示实体关系用于知识发现。", "original": "知识图谱:\"展示实体关系用于知识发现\"", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-037", "type": "definition", "knowledge": "ORM(对象关系映射)的作用是让代码操作数据库更方便。", "original": "ORM:\"让代码操作数据库更方便\"", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-038", "type": "case", "knowledge": "在供应链优化场景中,通过使用 Foundry ERP 套件连接 7 个以上 ERP 系统,将 ERP 数据映射为工厂、SKU、客户等实体对象,在 5 天内生成了物料清单工作流本体模型,实现了决策时间从数周缩短至分钟级的效果,预估年度节省达 1 亿美元。", "original": "场景:整合7个以上ERP系统(含SAP)的数据... 效果:年度节省预估达1亿美元(生产效率提升1-2%)... 决策时间从数周缩短至分钟级", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-039", "type": "case", "knowledge": "某航空制造商的某机型 通过基于 Ontology 的全保真数字孪生技术,将工厂数据映射为机器遥测数据进行实时运营监控,实现了产能提升 33% 的效果。", "original": "案例:某航空制造商A350数字孪生... 产能提升33%... 技术栈:基于Ontology的全保진数字孪生... 实时运营监控", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-040", "type": "case", "knowledge": "某大型医院通过应用合规工作流、患者卡片(Patient Card)及 AI 算法整合临床数据,使患者住院时长减少了 30%。", "original": "案例:某大型医院... 患者住院时长减少30%... 应用:合规工作流、患者卡片(Patient Card)利用AI算法、临床数据整合", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-041", "type": "method", "knowledge": "Python OSDK 2.x 提供了属性子选择功能,允许开发者只加载需要的数据以优化性能。", "original": "属性子选择:只加载需要的数据", "confidence": "high", "topics": ["AI与工具", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-042", "type": "method", "knowledge": "在 Functions 与 OSDK 的集成中,Python 和 TypeScript v2 Functions 对 OSDK 提供了一流支持,使得开发者无需学习不同的 SDK,且安装时 OSDK 会自动添加,本体实体可重新配置。", "original": "Python和TypeScript v2 Functions对OSDK提供一流支持... 无需学习不同SDK... 安装时OSDK自动添加,本体实体可重新配置。", "confidence": "high", "topics": ["AI与工具", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-043", "type": "definition", "knowledge": "GraphRAG 是微软开源的一种检索增强生成(RAG)改进方案,通过利用图结构来增强 AI 的知识检索能力。", "original": "GraphRAG是微软开源的检索增强生成(RAG)改进方案,核心思想是利用图结构增强AI的知识检索能力", "confidence": "high", "topics": ["AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-044", "type": "method", "knowledge": "GraphRAG 的索引阶段流程包括:文档分块、建立文档-文本块映射、利用 LLM 提取实体与关系、构建图、进行社区检测以及生成报告。", "original": "索引阶段(离线):文档 → 分块 → LLM提取实体/关系 → 构建图 → 社区检测 → 生成报告", "confidence": "high", "topics": ["AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-045", "type": "anti-pattern", "knowledge": "在以下场景中采用Ontology属于过度工程:1. 仅需简单查询(如查询订单表)时,应使用SQL或NL2SQL;2. 纯文档问答场景,应使用向量RAG或GraphRAG;3. 一次性分析任务,应使用数据仓库与BI工具;4. 技术团队内部使用,应直接写SQL。", "original": "以下场景不需要Ontology:简单查询(\"查询订单表\")... 纯文档问答... 一次性分析... 技术团队内部使用...", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-046", "type": "case", "knowledge": "在某航空制造商的某机型交付延迟问题中,通过构建Ontology实时追踪零件状态、预测瓶颈并自动调整排期,解决了数百万零件与数千供应商带来的复杂交付问题,使产能提升了33%。", "original": "案例:某航空制造商的某机型交付延迟问题\n- 痛点:数百万零件、数千供应商,延迟成本极高\n- Ontology价值:实时追踪零件状态、预测瓶颈、自动调整排期\n- 结果:产能提升33%", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-047", "type": "case", "knowledge": "针对大型制造企业存在多个ERP系统导致无法全局优化库存的痛点,通过Ontology统一建模“工厂-物料-供应商”关系,实现了年度节省1亿美元,并将决策时间从数周缩短至分钟级。", "original": "案例:大型制造企业有7个ERP系统\n- 痛点:每个工厂用不同系统,无法全局优化库存\n- Ontology价值:统一建模\"工厂-物料-供应商\"关系\n- 结果:年度节省$100M,决策时间从数周缩短至分钟级", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-048", "type": "insight", "knowledge": "在AI落地场景中,Ontology可作为“语义墙”解决LLM的安全与合规问题:通过限制AI只能访问定义好的对象和属性,结合Action Log记录操作与字段级权限控制,可防止AI直接访问原始数据库带来的幻觉与数据泄露风险。", "original": "场景:金融机构想用LLM,但担心幻觉和数据泄露\n- 痛点:不能让AI直接访问原始数据库,无法审计AI行为\n- Ontology价值:\n • \"语义墙\"限制AI只能访问定义好的对象和属性\n • Action Log记录所有AI操作,可审计\n • 精确到字段级的权限控制", "confidence": "high", "topics": ["AI与工具", "合规与风控", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-049", "type": "insight", "knowledge": "在监管合规(如制药行业GxP)场景下,Ontology通过Action记录数据变更、提供自动审计、影响分析(追踪配方变更影响的批次)以及时间旅行(恢复历史状态)功能,满足数据血缘与可追溯性要求。", "original": "场景:制药公司GxP合规\n- 痛点:FDA要求完整数据血缘、变更审计、可追溯性\n- Ontology价值:\n • 每次数据变更通过Action记录,自动审计\n • 影响分析:变更一个配方能自动追踪影响哪些批次\n • 时间旅行:任意历史时刻的数据状态可恢复", "confidence": "high", "topics": ["合规与风控", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-050", "type": "insight", "knowledge": "在运营效率瓶颈场景下,Ontology通过让AI Agent自动执行常规操作,可实现人类专注审批,从而使AI能够“安全地操作企业数据”。", "original": "场景:SOMPO保险公司8,000人日常作业\n- 痛点:大量重复性数据操作,人工录入错误率高\n- Ontology价值:AI Agent自动执行常规操作,人类专注审批\n- 关键:Ontology让AI能\"安全地操作企业数据\"", "confidence": "high", "topics": ["商业案例与实战复盘", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-051", "type": "anti-pattern", "knowledge": "企业采用Ontology的主要障碍包括:高昂的实施成本(POC需100-300万,实施需1000-2000万)、组织内部权力分配(所有者不愿放弃控制权)、技能缺口(缺乏专家且混淆taxonomy与ontology)、心理壁垒(风险厌恶)、短期思维(追求孤立用例而非长期投资)以及政治因素(管理层利益与架构师偏好)。", "original": "成本... 组织阻力... 技能缺口... 心理壁垒... 短期思维... 政治因素...", "confidence": "high", "topics": ["团队与模式设计", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-052", "type": "principle", "knowledge": "企业是否值得投入Ontology取决于业务复杂度、数据孤岛程度与AI驱动需求的平衡:高业务复杂度与高数据孤岛程度应考虑Ontology;若业务复杂度低或AI驱动需求不高,则可能更适合传统方案(SQL/BI)或轻量级KG/GraphRAG。", "original": "决策框架\n 业务复杂度\n 高 ↑\n │\n 值得投入 │ 考虑轻量KG\n Ontology │ 或GraphRAG\n │\n 数据孤岛程度 ←────┼────→ AI驱", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-053", "type": "insight", "knowledge": "国内大型金融机构不直接使用Palantir Ontology的原因包括:地缘政治风险(与美方深度绑定)、数据主权安全(金融数据敏感)、技术可替代性(国内知识图谱技术成熟)以及成本因素(自研更经济)。", "original": "为什么国内不直接用Palantizer?\n1. 地缘政治... 2. 数据主权... 3. 技术可替代... 4. 成本...", "confidence": "high", "topics": ["中国市场与下沉", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-054", "type": "insight", "knowledge": "国内大型银行在风控领域主要使用传统知识图谱(多为与第三方风控科技公司合作的图平台)进行反欺诈和风控决策,这类应用侧重于图计算,而非Palantir式的Ontology。", "original": "注:国内大型银行与第三方风控科技公司合作,使用其图平台进行反欺诈和风控决策,但这是**传统知识图谱**(侧重图计算),非Palantir式Ontology。", "confidence": "high", "topics": ["中国市场与下沉", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-055", "type": "definition", "knowledge": "传统知识图谱与Palantir式Ontology的区别在于:前者侧重于图计算,而后者具备更强的业务运营属性。", "original": "国内大型银行与第三方风控科技公司合作,使用其图平台进行反欺诈和风控决策,但这是**传统知识图谱**(侧重图计算),非Palantir式Ontology。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-056", "type": "case", "knowledge": "国内某大型金融科技公司是少数明确关注本体论(Ontology)的金融机构之一,旨在通过本体论解决数据“烟囱式”割裂问题,面临的挑战包括语义统一难和长链路一致性难。", "original": "国内某大型金融科技公司...目标:解决\"烟囱式\"数据割裂问题;挑战:语义统一难、长链路一致性难。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-057", "type": "case", "knowledge": "国内某头部支付平台通过自研图数据库构建了全图风控能力,结合图计算与图神经网络,实现了毫秒级极速风控,资金损失率连续三年低于亿分之一。该架构是国内最接近Palantir Ontology理念的系统,但其应用范围聚焦于风控领域而非全业务运营。", "original": "国内某头部支付平台...资金损失率连续三年低于亿分之一。该平台的\"全图风控\"是国内最接近Palantir Ontology理念的系统——但它聚焦在**风控领域**,而非全业务运营。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-058", "type": "case", "knowledge": "在证券行业,知识图谱的应用主要集中在风控合规和投研领域,具体包括产业链智能挖掘、投行知识库、智能投顾及财报解析等场景。", "original": "24家券商入局大模型,知识图谱主要应用于风控合规和投研。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-059", "type": "insight", "knowledge": "国内制造业(如车企)对知识图谱的应用普遍停留在“数据展示和分析”层面,尚未达到Palantir式“运营执行”的高度。", "original": "国内车企也在探索类似Ontology理念的数字孪生和知识图谱应用,但**普遍停留在\"数据展示和分析\"层面**,尚未达到Palantir式\"运营执行\"的高度。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-060", "type": "anti-pattern", "knowledge": "传统的供应链管理(如某头部车企的供应链)若仅使用传统知识图谱进行ERP整合与零部件追溯,会因为缺少Actions回写能力而无法实现自动调整采购计划,无法达到Ontology的运营高度。", "original": "某头部车企供应链...传统知识图谱...❌ 缺少Actions回写,无法自动调整采购计划。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-061", "type": "anti-pattern", "knowledge": "单点式的用户运营(如某新势力车企的用户运营)若仅基于图数据库与推荐算法进行车主与车辆关联,会因为缺少全业务Ontology而无法实现跨场景的业务闭环。", "original": "某新势力车企用户运营...图数据库+推荐算法...❌ 缺少全业务Ontology,多为单点应用。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-062", "type": "anti-pattern", "knowledge": "工业物联网与时序数据库结合的数字孪生(如某车企的智能工厂)若语义层薄弱,将无法实现从数据监控到智能决策的跨越。", "original": "某车企智能工厂...接近数字孪生,但语义层薄弱。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-063", "type": "anti-pattern", "knowledge": "具备语义建模意识但未形成企业级Ontology的场景(如某车企的自动驾驶业务),无法实现从传感器数据到企业级决策的统一。", "original": "某车企自动驾驶...有语义建模意识,但未形成企业级Ontology。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-064", "type": "anti-pattern", "knowledge": "强追溯能力但缺少AI Agent操作能力的场景(如某电池厂商的电池全生命周期追溯),无法实现从数据追溯到自动化的闭环。", "original": "某电池厂商电池...追溯能力强,但缺少AI Agent操作能力。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-065", "type": "definition", "knowledge": "Palantir Ontology与传统知识图谱的核心差异在于:传统知识图谱多为只读的查询工具,而Palantir Ontology是具备'Writeback'(写回)能力的业务操作系统,能够通过数据变更自动触发业务动作。", "original": "国内车企的\"知识图谱\"大多是只读的查询工具,而Palantir Ontology是可读可写的业务操作系统——数据变更能自动触发业务动作。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-066", "type": "case", "knowledge": "在车企场景下,Palantir Ontology可以通过'召回批次电池'动作自动关联所有搭载车辆并通知车主;通过'供应商评级下降'自动调整采购配额并激活备选供应商;通过'产线OOE低于阈值'自动排程优化并推送维修工单。", "original": "• \"召回批次电池\" → 自动关联所有搭载车辆 → 通知车主\n• \"供应商评级下降\" → 自动调整采购配额 → 激活备选供应商\n• \"产线OEE低于阈值\" → 自动排程优化 → 推送维修工单", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-067", "type": "insight", "knowledge": "Palantir的创新不在于发明了Ontology,而在于将知识图谱与'Actions'(动作)结合,使其从单纯的查询工具演变为能够执行业务操作的操作系统。", "original": "Ontology不是Palantir发明的,但Palantir把它做成了一个操作系统。知识图谱学术界搞了几十年,Palantir的创新是加了\"Actions\"——让AI不仅能查,还能改。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-068", "type": "insight", "knowledge": "国内在单点技术(如图数据库、知识抽取、语义建模)上并不落后,差距在于缺乏系统工程能力,无法将这些技术高效地组合、拼装成完整的闭环解决方案。", "original": "国内技术不落后,差的是\"拼成解决方案\"的能力。国内几家大厂的自研图数据库、OpenKG的DeepKE,单点都不差Palantls。但Palentir强在把\"图数据库+知识抽取+语义建模+权限审计+AI Agent\"拼成了一个闭环产品。", "confidence": "high", "topics": ["本体论与FDE方法论", "团队与模式设计"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-069", "type": "anti-pattern", "knowledge": "GraphRAG无法替代Ontology,因为GraphRAG本质上是'只读'的,无法像Ontology那样通过数据变更自动触发业务操作(如自动触发采购订单)。", "original": "GraphRAG不能替代Ontology,因为GraphRAG是\"只读\"的。它能回答\"这个零件短缺会影响哪些车型\",但不能自动\"触发采购订单\"。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-070", "type": "insight", "knowledge": "国内金融机构广泛应用的知识图谱多处于分析与查询层,缺乏像Palantir Ontology那样具备'Writeback'(写回)能力的业务操作层。", "original": "国内金融机构用的是\"知识图谱\",不是\"Ontology\"。国内几家大型银行与支付平台都在用图技术做风控——但他们用的是传统知识图谱(查询展示层)。Palantir Ontology的核心差异是Writeback(写回),这一点国内几乎没有。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-071", "type": "insight", "knowledge": "国内造车行业的数字化应用多停留在'数字孪生看板'阶段,即侧重于展示与报警,而缺乏数据变更自动触发业务动作的能力。", "original": "造车行业也在探索,但普遍停留在\"数字孪生看板\"。几家头部车企有供应链追溯、用户画像、产线监控——但缺少\"数据变更自动触发业务动作\"的能力。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-072", "type": "definition", "knowledge": "Ontology不是单纯的数据库,而是企业的'业务操作系统'。它定义了业务实体的含义(如'客户'、'订单'是什么)以及AI可以对这些实体执行的操作。", "original": "Ontology不是数据库,而是企业的'业务操作系统'——它定义了'客户'是什么、'订单'是什么,以及AI能对这些实体做什么操作。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-073", "type": "insight", "knowledge": "Palantir的创新在于将'知识图谱'与'业务操作'结合,使AI从单纯的查询能力进化为具备'可查、可改、可审计'的闭环能力。", "original": "Palentir没有发明知识图谱,它发明的是'知识图谱+业务操作'——让AI从'只能查'变成'能查能改还能审计'。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-074", "type": "insight", "knowledge": "GraphRAG与Ontology的区别在于:GraphRAG是更聪明的搜索引擎,侧重于回答问题;而Ontology是企业的AI操作员,侧重于执行任务。", "original": "GraphRAG是更聪明的搜索引擎,Ontology是企业的AI操作员——一个只能回答,一个还能执行。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-075", "type": "case", "knowledge": "在造车行业,传统数字孪生看板侧重于数据展示(如供应链追溯、产线监控),而Palantir Ontology通过'数据变更自动触发业务动作'实现闭环,例如当电池供应商评级下降时,系统会自动调整采购配额并激活备选供应商。", "original": "比如:\"电池供应商评级下降\"→国内系统只会报警;Palentir Ontology会自动调整采购配额、激活备选供应商。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-076", "type": "insight", "knowledge": "国内技术在单点性能(如某头部支付平台的自研图数据库)上并不落后,但与Palantir的差距在于缺乏将技术整合为闭环产品的系统工程能力。", "original": "不是技术不行,是数据主权和地缘政治不允许。某头部支付平台的自研图数据库单点性能不比Palantir差,但缺的是把技术拼成产品的系统工程能力。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-077", "type": "principle", "knowledge": "Ontology的核心理念是'Model reality, not systems'(建模现实,而非建模系统),它通过定义业务实体的属性、关系和生命周期,而非仅仅是数据表结构。", "original": "理解Ontology的核心理念(Model reality, not systems)... 传统数据库只存数据,Ontology还定义了数据的业务含义——比如'客户'不是一张表,而是一个有属性、有关系、有生命周期的实体。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-078", "type": "principle", "knowledge": "在金融等高合规场景中,安全的核心在于审计。Ontology通过Action Log(操作日志)为AI的操作提供'行车记录仪'式的审计追踪,解决AI乱改数据且无留痕的问题。", "original": "金融机构不怕AI答错,怕的是AI乱改数据还没留痕。Ontology的Action Log就是给AI戴上的'行车记录仪'。", "confidence": "high", "topics": ["合规与风控", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01-学习指南", "skills": ["xy-ontology"]}
{"id": "XY-ONT-079", "type": "method", "knowledge": "图构建的8步流程为:1. create_base_text_units(文档分块);2. create_final_documents(建立文档与文本块映射);3. extract_graph(LLM提取实体关系并去重);4. finalize_graph(NetworkX构建图并计算节点度数);5. create_communities(Leiden算法分层聚类);6. create_final_text_units(关联文本块与实体/关系ID);7. create_community_reports(LLM生成社区摘要);8. generate_embeddings(文本块、实体、社区报告向量化)。", "original": "图构建8步流程:第1步create_base_text_units做文档分块...第8步generate_embeddings把文本块、实体、社区报告做向量化。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-080", "type": "principle", "knowledge": "传统RAG与GraphRAG的核心区别在于检索逻辑:传统RAG基于向量相似性进行广度优先的单跳检索,侧重于寻找与查询最相似的文档块;GraphRAG基于实体节点与显式关系进行深度优先的多跳推理,侧重于通过图关系实现跨文档关联。", "original": "传统RAG和GraphRAG的核心区别:传统RAG靠向量相似性检索文档块...属于广度优先的单跳检索...GraphRAG靠实体节点与显式关系遍历...属于深度优先的多跳推理...", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-081", "type": "principle", "knowledge": "Ontology是GraphRAG的蓝图,通过定义实体类型、关系和层次结构来决定图谱内容;它指导数据映射(结构化与非结构化数据转化)、支持推理增强(自动推断与逻辑检查)并确保多源数据的语义一致性。", "original": "Ontology和GraphRAG的关系:Ontology是GraphRG的蓝图...结构定义上...数据映射上...推理增强上...语义一致性上...", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-082", "type": "principle", "knowledge": "在安全合规场景下,GraphRAG与Palantir Ontology存在本质差异:GraphRAG侧重于从非结构化文档自动提取关系,具有只读属性、粗粒度权限控制、离线索引特征及文档级血缘;Palantir Ontology侧重于结构化数据库与文档的结合,通过人工建模加自动映射实现,具备可写(Actions机制)、对象/属性级精细权限、完整业务血缘及实时同步能力。", "original": "GraphRAG和Palantis Ontology在安全合规场景下的对比:数据来源上...写入能力上...权限控制上...数据血缘上...", "confidence": "high", "topics": ["本体论与FDE方法论", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-083", "type": "case", "knowledge": "GraphRAG适用于多跳推理需求(如分析项目延迟对利润的影响)、合规与审计关系追踪、金融实体映射、产品层级结构以及文档密集型知识问答场景。", "original": "GraphRAG适合的场景包括:多跳推理需求...文档密集型知识问答。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-084", "type": "anti-pattern", "knowledge": "在需要AI执行业务操作(如更新客户状态)、需要完整业务操作审计(强合规行业)、需要细粒度权限控制或实时数据同步的场景下,不应仅依赖GraphRAG,因为这些场景无法满足GraphRAG的局限性。", "original": "GraphRAG无法替代Ontology的场景包括:需要AI执行业务操作...需要完整的业务操作审计...需要细粒度权限...需要实时数据同步...", "confidence": "high", "topics": ["本体论与FDE方法论", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-085", "type": "insight", "knowledge": "从安全可控与合规审计维度的综合评分来看:纯RAG为2星;GraphRAG为3星;RAG加权限层为4星/3星;传统知识图谱为4星/4星;Palantir Ontology为5星/5星,其高成本对应着极高的安全与合规能力,适合强监管行业。", "original": "从安全可控和合规审计两个维度对比几种方案...Palantir Ontology两项都是5星,实施成本很高,适合金融、制药等强监管行业。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "01b-chunk6-prose", "skills": ["xy-ontology"]}
{"id": "XY-ONT-086", "type": "definition", "knowledge": "本体论(Ontology)是构成业务的‘名词’与‘动词’,它反映了业务实际运行过程中的复杂互联关系。", "original": "So if we think about what is an ontology? It's the nouns and verbs that make em up your business, right?... So this really reflects the grown truth about how your business is actually operating. There", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-087", "type": "principle", "knowledge": "构建本体论的目标是模拟业务实际的运行方式,而不是为了迎合现有系统的运行需求而进行建模。", "original": "As we model this into the ontology, the goal is to actually model how your business is actually operating, not how these other systems need it so they can work.", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-088", "type": "principle", "knowledge": "决策中心化系统(Decision-centric system)需要具备三个核心要素:数据(Data)、逻辑(Logic)和行动(Actions)。", "original": "So if we think about that in a decision-centric system, what do I need to be able to make decisions, right? I need three things, I need the data, I need logic, and I need actions, right, so I need dat", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-089", "type": "definition", "knowledge": "在决策中心化系统中,数据代表业务的当前状态,逻辑代表对这些状态的思考方式,行动代表可以用来影响现实世界的手段。", "original": "The data represents what the current state of my business is, the logic about how do I think about those things, and the actions I can take to affect the real world.", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-090", "type": "method", "knowledge": "本体论的数据层可以通过连接各种企业系统(如自研、开箱即用软件、遗留系统或云端系统)以及通过MMDP连接Snowflake、Databricks、BigQuery等数据湖中的虚拟化数据来实现。", "original": "I'm connecting to all my different enterprise systems, whether they're homegrown, out of the box software, legacy, new stuff in the web, doesn't matter.... or with MMDP, you can connect to things like", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-091", "type": "method", "knowledge": "逻辑层可以集成多种来源,包括基于规则的Excel表格、机器学习模型、预测模型或第三方优化器,这些逻辑可以存在于平台内部或外部。", "original": "So these could be really simple. That could be an Excel spreadsheet, rules-based logic. It It could be ML models, forecast, third-party optimizers you bought. They can be living inside of the the plat", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-092", "type": "definition", "knowledge": "Palantir Ontology 是将数据、业务逻辑与可执行动作整合在一起的数字孪生,能够展示业务实际运行状态,并提供决策逻辑与行动路径。", "original": "So if we bring all the data, the logic, the action together here, then I have this ontology that is very rich, showing me my actual digital twin of how my business is operating, how to think about it ", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-093", "type": "method", "knowledge": "通过将数据源(如 SAP、遗留系统、云端或边缘端系统)与业务逻辑及行动(如向 SAP 写回以创建 STO)进行建模,可以实现从数据获取到业务执行的闭环。", "original": "So whether it's SAP and I want to write back to SAP to create an STO to move product from A to B across my Network right those should be modeled in here as the actions I can take there can be other th", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-094", "type": "insight", "knowledge": "在本体论之上构建的工作流与分析结果是业务流程的副产品,通过连接人、流程与业务系统,可以从中提炼出洞察。", "original": "So then on top of that, and we can then build these workflows and analytics and looks really are a by-product of the workflows that I'm building. So a workflow across many different business processes", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-095", "type": "insight", "knowledge": "本体论为大语言模型(LLM)提供了业务上下文与逻辑,使其能够基于业务规则进行推理,并调用确定性模型或驱动写回动作,而不仅仅是处理原始数据。", "original": "So now can we have generative AI models integrated through AIP and AIP logic, for example. That now are helping actually go reason about it because they have access to the ontology, the LLM now has ac", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-096", "type": "anti-pattern", "knowledge": "避免人员在仪表盘与不同工具之间进行‘转椅式’切换(swivel chair)以执行任务,应通过本体论编排复杂的后端动作来实现自动化。", "original": "So instead of people having to swivel chair from a dashboard over to another tool to figure out how to make it real, we'll orchestrate those complex Actions on the back end to make it real and that's ", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-097", "type": "method", "knowledge": "可以通过本体论 SDK 构建数据产品或丰富的应用程序,例如移动端应用、自定义 React 应用或业务流程的自定义 SDK。", "original": "This is where we can build data products, rich applications, whether that's in a mobile app for the plant floor, it's a custom React app to interact with my customers, or it could be through the ontol", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-098", "type": "method", "knowledge": "可以通过构建基于本体(Ontology)的自定义 SDK,将业务逻辑与业务流程封装,从而在不同的应用程序或现有系统中集成代理化工作流(agented workflows)。", "original": "it could be through the ontology SDK, where I actually create a custom SDK of my ontology. So an SDK of business and business process that I can build different applications, back integrations, I can ", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-099", "type": "insight", "knowledge": "通过在企业内推广不同的 SDK,可以提升系统的集成能力和易用性,进而缩短交付周期并提高价值实现速度。", "original": "This really opens up the way that you integrate it and really improves the usability, which improves speed and time to value.", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-100", "type": "principle", "knowledge": "本体论的核心目标是实现 AI 与人类在统一模型上的协同工作,并随着时间推移实现业务流程的自动化。", "original": "So all of this comes together with the goal of AI and humans working together on the ontology. And the goal over time is to automate more and more of your business", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-101", "type": "definition", "knowledge": "本体论的作用是建模数据、逻辑与行动,从而驱动业务决策。", "original": "what you need is the ontolgy to model the data, the logic, the actions to drive decisions in your business", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "02-Palantir-Ontology-Overview", "skills": ["xy-ontology"]}
{"id": "XY-ONT-686", "type": "definition", "knowledge": "Foundry Ontology是一个数据后端,通过将基于文件的数据映射为以组织为中心的对象,为数据探索、分析、运营编辑及场景分析提供高速查询能力。", "original": "Foundry Ontology是一个数据后端,把基于文件的数据映射成以组织为中心的对象,为数据探索、数据分析、运营数据编辑、场景分析等提供高速查询能力。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-687", "type": "definition", "knowledge": "在Ontology中,对象类型(object type)是实体的语义表示(如名称和属性定义),而对象集(object set)是该类型对应的具体对象集合,其规模受导入行数及Actions操作影响。", "original": "对象类型是实体本身的语义表示(比如对象的名称和属性定义);对象集则是这个类型对应的具体对象集合,对象集的大小对应导入数据集的行数,以及本体操作(Actions)创建和删除的对象数量。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-688", "type": "number", "knowledge": "Foundry不同查询类型的最低算力开销(以compute-second计价)如下:基础查询为2个,关联查询、聚合查询、本体SQL查询均为5个,高级查询与派生属性查询为10个,Actions写回操作为18个起(首个对象18个,后续每个额外加1个)。", "original": "基础查询(Base query):...最低开销2个compute-second...关联查询(Search Around query):...最低考5个...聚合查询(Aggregation query):...最低开销5个...本体SQL查询(Ontology SQL query):...最低开销5个...高级查询(Advanced query):...最低开销10个...派生属性查询(Der", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-689", "type": "insight", "knowledge": "在Foundry中,算力消耗(compute-second)默认挂靠在触发查询的资源上(如Workshop或Quiver),而非用户身上;若无挂靠资源(如API调用),算力挂在对象类型上;若涉及多对象查询,算力在对象间平均分摊。", "original": "compute-second默认挂在触发查询的资源上(比如某个Workshop应用、某个Quver仪表盘),而不是挂在具体交互的用户身上。如果查询没有一个\"保存的资源\"可以挂靠(比如直接走API调用),算力就会挂到被查询的对象类型上;如果一次请求同时查询多个对象,算力会在这些对象之间平均分摊。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-690", "type": "anti-pattern", "knowledge": "应避免进行过于宽泛的多对象搜索,因为这会拆分成多个子查询并分别作用于各对象类型;应通过将搜索范围限定到具体对象类型来减少查询数量。", "original": "一次宽泛的多对象搜索会拆成多个子查询,分别打到每个对象类型上,所以把搜索范围限定到具体对象类型能减少查询数量。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-691", "type": "method", "knowledge": "为了降低算力消耗,应尽量查询小对象集而非大对象集,因为算力开销与对象集大小(命中数)成正比。", "original": "查询小对象集比查询大对象集省算力,因为算力开销跟对象集大小(命中数)成正比。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-692", "type": "method", "knowledge": "在执行聚合或关联查询之前,应先进行前置过滤,以减少后续处理的命中数,从而降低整体算力消耗。", "original": "在做聚合或关联查询之前先做好前置过滤,能大幅减少后续处理的命中数,从而降低整体算力消耗。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "05-官方-ComputeUsage", "skills": ["xy-ontology"]}
{"id": "XY-ONT-693", "type": "definition", "knowledge": "本体论(Ontology)的核心建模目标是“决策”本身,而非仅仅是数据。一个完整的决策由三个要素组成:数据(决策信息)、逻辑(评估过程)和动作(决策执行)。", "original": "核心观点:本体论要建模的是\"决策\"本身,而不只是数据。任何一个决策都由三个要素组成:数据(做决策用的信息)、逻辑(评估决策的过程)、动作(决策的执行)。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-694", "type": "insight", "knowledge": "在AI时代,企业面临的难题不仅是数据的清洗与统一,更是“相关性”问题。除了常规的结构化与非结构化数据,决策发生时的上下文、评估过的选项以及最终选择带来的下游影响等“决策数据”是规模化利用的关键。", "original": "AI时代真正的难题不是清洗和统一数据,而是\"相关性\"——除了结构化、流式、边缘、非结构化数据这些常规数据源,还有一类容易被忽略的\"决策数据\":决策发生时的上下文、评估过的选项、最终选择带来的下游影响。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-695", "type": "principle", "knowledge": "本体论通过将企业各种模态的数据整合为全量、高保真的语义表示,并安全采集运营者产生的决策数据,可以形成端到端的“决策血缘”,实现决策过程在时间、数据版本、应用层面的全程可追溯。", "original": "本体论把企业各种模态的数据整合成一份全量、高保真的语义表示...企业运营者的日常工作会持续产生\"决策数据\",本体论负责把这些决策数据安全地采集下来,形成端到端的\"决策血缘\"——一个决策在什么时候、基于哪个版本的企业数据、通过哪个应用产生的,全程可追溯,人和AI都能访问。", "confidence": "high", "topics": ["本体论与FDE方法论", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-696", "type": "method", "knowledge": "本体论通过“逻辑绑定”统一接口,将分散在本地数据中心、云环境、SaaS环境及平台自身的业务规则、预测模型和优化算法等逻辑资产连接起来,赋予其上下文,使人类推理过程能安全地引入AI推理。", "original": "本体论通过\"逻辑绑定\"这个统一接口,把分散在不同环境(本地数据中心、企业云环境、SaaS环境、Palantir平台本身)里的各种逻辑资产...连接起来、赋予上下文,让传统上只靠人来把关的推理过程,也能安全地引入AI推理。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-697", "type": "principle", "knowledge": "区分“运营系统”与“分析系统”的关键在于能否将决策实时闭环执行。在本体论模型中,数据元素是“名词”,动作是“动词”,动作的建模是决策中心化模型的完整闭环。", "original": "能不能把决策实时闭环执行,是区分\"运营系统\"和\"分析系统\"的关键。本体论原生地把动作建模进这个决策中心化模型:如果说数据元素是企业的\"名词\",动作就是\"动词\",两者结合才能构成完整的\"句子\"。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-698", "type": "method", "knowledge": "本体论通过统一的访问控制管理动作,使人和AI驱动的动作可以安全地暂存为待审场景,并按照与数据、逻辑相同的颗粒度安全地写回各类底层事务系统、边缘设备或自定义应用。", "original": "本体论让人和AI驱动的动作能被安全地暂存为待审场景,用跟数据、逻辑同一套颗粒度的访问控制来管理,并安全地写回企业各类底层系统(事务系统、边缘设备、自定义应用等)。", "confidence": "high", "topics": ["本体论与FDE方法论", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-699", "type": "case", "knowledge": "在医疗设备制造案例中,某医疗设备制造商通过统一的本体架构应对原材料中断:1. 快速评估营收影响并按优先级排序;2. 调用多重逻辑资产进行仿真评估替代方案;3. AI副驾提出新方案并在沙盒演练后交由人工终审;4. 终审通过后动作自动写回ERP、WMS和生产排程系统;5. 过程产生的决策数据沉淀为AI持续训练的原料。", "original": "案例(虚构企业某医疗设备制造商演示):...因为已经把数据源、逻辑资产、动作系统都接入了统一本体,运营团队能:1. 快速看清这次短缺对营收的全部影响范围... 2. 调用预测模型... 3. 团队自己训练的AI副驾... 4. 终审通过后,执行动作自动分别写回... 5. 危机处理完之后,整个过程沉淀的决策数据...成为持续训练/优化AI的原料...", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-700", "type": "insight", "knowledge": "本体论架构是“决策中心”而非“数据中心”。它通过整合数据、逻辑与动作,使新数据能快速接入语义表示,新算法能安全供人与AI使用,并实现动作向各类运营系统的实时同步。", "original": "要点总结:本体论是\"决策中心\"而不是\"数据中心\"的架构,把数据、逻辑、动作整合进同一套系统,让新数据能快速接入语义表示,新算法/业务逻辑能安全地提供给人和AI使用,动作能实时同步回各类运营系统。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "06-官方-CTO演讲", "skills": ["xy-ontology"]}
{"id": "XY-ONT-701", "type": "definition", "knowledge": "本体论(Ontology)是 Palantir 架构的核心系统,旨在代表企业复杂且相互关联的决策,而非仅仅是数据,使人类与人工智能智能体能够在与物理世界协同的业务工作流中开展协作。", "original": "本体论是构成 Palantir 架构核心的系统。本体论旨在代表企业复杂且相互关联的决策,而非仅仅是数据。这使得人类和人工智能智能体能够在必须与物理世界协同运作的业务工作流中开展协作。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-702", "type": "principle", "knowledge": "本体论通过数据、逻辑、行为和安全四个维度的整合来对决策进行建模。", "original": "本体论通过数据、逻辑、行为和安全四个维度的整合来对决策进行建模。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-703", "type": "principle", "knowledge": "在本体论建模中,数据对象(名词)必须辅以操作(动词)才能实现对决策的建模,即语义必须与动态相结合。", "original": "数据对象(即\"名词\")必须辅以\"动词\"才能对决策进行建模;语义必须与动态相结合。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-704", "type": "case", "knowledge": "航空公司通过将航班、航空器、机组人员、调度优化器等资产建模到本体论中,支持当日运营和长期规划。", "original": "航空公司可能会将航班、航空器、机组人员清单、调度优化器以及其他零散的企业资产建模到本体论中,以此为当日航班运营和长期规划提供支持。", "confidence": "high", "topics": ["商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-705", "type": "case", "knowledge": "医院系统通过对患者、护士排班、医疗物资、床位容量等实时变动要素进行建模,以推动患者诊疗全流程。", "original": "医院系统或许会转而对患者、护士排班、医疗物资、床位容量以及其他各类要素进行建模,这些要素往往是实时变动,且对推动患者诊疗全流程至关重要。", "confidence": "high", "topics": ["商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-706", "type": "case", "knowledge": "在军事场景中,本体论将前沿部署部队的战备信息与作战流程统一,为多国团队构建共享的作战世界。", "original": "在军事场景中,本体论能够将前沿部署部队的战备信息与支撑侦察和目标选择的作战流程相统一,为多国团队构建一个共享的作战世界。", "confidence": "high", "topics": ["商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-707", "type": "method", "knowledge": "本体论的逻辑层可以是模块化的,支持从简单业务规则、机器学习模型到大语言模型驱动函数的演进。", "original": "驱动每个操作的逻辑可以是模块化的,并会随着时间演变,以此体现推动决策制定的多样化计算与推理方式。某一特定操作(或对特定对象进行增强)背后的逻辑,可能是简单的业务规则、传统的机器学习模型,由大语言模型驱动的函数,或是涉及多个计算引擎的复杂多步骤编排。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-708", "type": "case", "knowledge": "在医疗制造场景中,通过对制造工厂、工作订单、客户详情等语义概念建模,整合数百个底层数据源,并根据生产、仓库、供应链等不同团队的需求实施精细化的权限控制。", "original": "对于供应链分析师、生产工程师、仓库工作人员以及其他与本体论进行交互的团队成员而言,不同的访问权限范围都具有重要意义... 随着这些不同的团队开发人工智能驱动的智能体,他们必须拥有从人类用户继承或从定义项目的权限结构继承的安全范围。", "confidence": "high", "topics": ["商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-709", "type": "principle", "knowledge": "本体论是一个由语言层、引擎层和工具链构成的多模态系统,无法依靠单一的语义层实现。", "original": "该本体论是一个多模态系统,由数十个底层组件构成,这些组件在概念上可归为语言层、引擎层和工具链三部分。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-710", "type": "method", "knowledge": "本体论的语言层负责对语义对象、链接、属性、动力学操作、自动化流程及交互逻辑进行建模。", "original": "语言层:对语义对象、链接和属性进行建模;同时也对动力学操作、自动化流程,以及定义这些操作运行方式、它们与其他系统交互方式的字面逻辑片段进行建模。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-711", "type": "method", "knowledge": "本体论的引擎层负责验证语言组件,提供支持大规模查询、实时订阅、原子化事务更新、批量突变及低延迟同步的读写架构。", "original": "引擎层:验证了语言的每一个组件。它提供了模块化的读取架构,可支持大规模SQL查询、状态变更的实时订阅... 它还提供了可扩展的写入架构,支持原子化且持久的事务更新、大规模批量突变、大规模流处理,以及变更数据捕获等机制。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-712", "type": "method", "knowledge": "本体论的工具链通过软件开发工具包(OSDK)和DevOps工具集,使开发者能将本体论作为后端构建各类AI赋能的应用。", "original": "工具链:涵盖了该语言的全部表达能力以及引擎的强大性能,使开发者能够将本体论用作后端。用于...无数其他应用场景的丰富人工智能赋能应用程序,均构建在本体论软件开发工具包(OSDK)以及为生产应用场景的规模化治理而设计的丰富Devops工具集之上。", "confidence": "high", "topics": ["AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-713", "type": "insight", "knowledge": "本体论是控制论企业动态的核心,通过构建高保真业务世界表征,使人类与AI智能体能共同使用反馈循环进行持续学习与演进。", "original": "本体论是控制ler企业动态且不断复合的核心。每一次数据整合都有助于构建一个高保真的业务世界表征,供人类和人工智能智能体共同使用...在工作流中收集到的每一条反馈都能安全地融入持续学习循环,并为从增强到自动化的演进提供动力。", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "07-官方-TheOntologySystem", "skills": ["xy-ontology"]}
{"id": "XY-ONT-721", "type": "insight", "knowledge": "数字孪生的本质可以被理解为物理产线的知识图谱化镜像,即通过本体定义概念骨架,将原始信号灌入骨架,利用图查询语言进行语义检索,并利用图路径推理算法补全隐藏关系。", "original": "它非常早地示范了\"数字孪生 = 物理产线的知识图谱化镜像\"这个思路的最小可行版本:用本体定义概念骨架→用特征工程把原始信号灌入骨架→用图查询语言(等价于合取查询)做语义检索→用图路径推理算法补全隐藏关系。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-论文-KnowledgeGraph生成数字孪生模型-产线场景", "skills": ["xy-ontology"]}
{"id": "XY-ONT-722", "type": "method", "knowledge": "构建数字孪生模型的四阶段流水线包括:1. 特征提取(将异构数据归一化为有业务含义的特征);2. 本体构建(定义核心类);3. 知识图谱生成(将特征映射进本体类与关系);4. 语义关系抽取(利用算法挖掘隐藏关系)。", "original": "论文提出一个四阶段流水线:1. 特征提取... 2. 本体构建... 3. 知识图谱生成... 4. 语义关系抽取...", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-论文-KnowledgeGraph生成数字孪生模型-产线场景", "skills": ["xy-ontology"]}
{"id": "XY-ONT-723", "type": "principle", "knowledge": "本体设计在冷启动阶段可以采用“先跑通最小闭环,再逐步细化”的策略,即不需要一开始就设计复杂的本体,通过少量核心类即可跑通从原始数据到语义查询的全链路。", "original": "本体不需要一开始就做得很复杂——论文只用5个类就跑通了从原始数据到语义查询的全链路,说明本体设计可以\"先跑通最小闭环,再逐步细化\",这对FDE知识板块的冷启动阶段是个可参考的节奏;", "confidence": "high", "topics": ["本体论与FDE方法论", "新人起步方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-论文-KnowledgeGraph生成数字孪生模型-产线场景", "skills": ["xy-ontology"]}
{"id": "XY-ONT-724", "type": "method", "knowledge": "在进行本体建模前,必须先将结构复杂的原始记录归一化为具有业务含义的特征(原子事实),而不是让本体直接对接原始文本。", "original": "论文强调\"把结构复杂的原始记录归一化为有业务含义的特征\"是本体建模前必须做的一步——对应到FDE场景,就是先把访谈记录/案例材料等非结构化素材提炼成有命名的\"原子事实\",再往本体骨架里灌,而不是让本体直接对接原始文本;", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-论文-KnowledgeGraph生成数字孪生模型-产线场景", "skills": ["xy-ontology"]}
{"id": "XY-ONT-725", "type": "insight", "knowledge": "基于路径遍历的关系推断方法(如PRA)可以实现知识图谱中隐藏关联的自动挖掘,即通过图路径搜索自动“推”出关联,而不需要人工逐条标注。", "original": "PRA这类基于路径遍历的关系推断方法提示:知识图谱建好之后,很多有价值的隐藏关联(比如两个客户案例背后共享的同一套打法)并不需要人工逐条标注,而是可以通过图路径搜索自动\"推\"出来,这对后续做FDE案例库的关联挖掘有直接参考意义。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-论文-KnowledgeGraph生成数字孪生模型-产线场景", "skills": ["xy-ontology"]}
{"id": "XY-ONT-726", "type": "insight", "knowledge": "Node-RED等低代码工具擅长数据流转与搬运,但缺乏语义建模能力,导致基于此类工具构建的数字孪生模型之间难以实现互操作、自动发现与推理。", "original": "Node-RED擅长搬运数据,却没有语义建模能力,这导致基于它搭出来的数字孪生模型互相之间难以互操作,也难以被自动发现和推理。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-727", "type": "method", "knowledge": "通过开发自定义节点将物理资产与知识图谱语义世界连接,构建包含结构类(资产、属性、模型)、行为类(动作、事件)以及关系与引擎类(关系定义、图谱自动生成)的语义层,作为低代码工具的‘翻译中间件’。", "original": "论文的做法是自行开发了7个自定义Node-RED节点,把\"资产的物理世界\"和\"知识图谱的语义世界\"焊接在一起...语义层可以作为一层\"翻译中间件\"叠加在原有工具链之上,模型和实现互相解耦,逻辑和落地分离,更容易维护。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-728", "type": "method", "knowledge": "语义数字孪生建模五步法:1. 识别可观测资产;2. 搭建DT模型;3. 添加语义;4. 配置通信接口;5. 部署。支持在步骤间进行循环迭代。", "original": "给出完整的建模五步法:识别可观测资产→搭建DT模型→添加语义→配置通信接口→部署,且支持在中间步骤之间循环迭代。", "confidence": "high", "topics": ["本体论与FDE方法论", "新人起步方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-729", "type": "case", "knowledge": "在工业电动阀门场景中,通过dt-relation定义资产间的层级关系(如阀门属于炼油厂,炼油厂属于母公司),并利用dt-graph节点实时监听模型变化,自动将模型翻译为Cypher语句写入Neo4j,实现了物理状态同步与语义关系查询的自动化。", "original": "论文用一个\"工业电动阀门\"场景做了验证:阀门有4个传感器属性(温度、液位、压力、振动)和2个动作(开、关),通过dt-relation把阀门(Electric Valve)挂到炼油厂(Refinery,partOf关系),再挂到母公司Petrobras(belongsTo关系)。dt-graph节点全程监听并自动把这套模型翻译成Cypher查询语句写入Neo4j,无需人工手写图数据库语句,最终形成", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-730", "type": "insight", "knowledge": "将“资产-属性-关系-动作-事件”作为通用的最小本体骨架,可以跨领域复用。在企业知识管理中,可将客户视为资产,行业/规模为属性,合作项目为关系,签约/流失为事件。", "original": "论文验证的\"资产-属性-关系-动作-事件\"五元结构,本质上是一个通用的最小本体骨架,同样的骨架完全可以套用在企业知识/客户案例领域:把\"客户\"当资产、\"行业/规模/合作阶段\"当属性、\"合作过哪些项目\"\"同属哪个渠道\"当关系、\"签约/流失/复购\"当事件;", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-731", "type": "insight", "knowledge": "FDE知识库的技术实现应追求“业务人员只维护结构化条目,图谱层的同步和查询语句生成完全自动化”,以降低非技术人员的使用门槛。", "original": "dt-graph节点\"监听模型变化,自动同步写图数据库,无需人工写查询语句\"这个设计思路,提示FDE知识库的技术实现也应该追求\"业务人员只维护结构化条目,图谱层的同步和查询语句生成完全自动化\",降低对非技术人员的使用门槛;", "confidence": "medium", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-732", "type": "anti-pattern", "knowledge": "缺乏统一的属性与关系命名规范会导致语义标注失败,使得模型无法被自动发现,即便数据量巨大也难以被检索和复用。", "original": "论文强调\"语义标注失败会导致模型难以被自动发现\"——对应到FDE知识板块,没有统一属性/关系命名规范的案例记录,即使数量再多也很难被检索和复用,这提示知识板块上线前必须先把本体/字段规范定死。", "confidence": "high", "topics": ["本体论与FDE方法论", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "09-论文-NodeRED语义数字孪生建模方法论", "skills": ["xy-ontology"]}
{"id": "XY-ONT-733", "type": "insight", "knowledge": "工业从4.0向5.0演进的过程,本质是从单纯的万物互联(IoT)向具备上下文感知能力的语义理解(Semantic Understanding)转变,核心目标是实现以人为本的数据理解。", "original": "工业5.0的转变是'从单纯互联走向以人为本',系统必须具备上下文感知能力才能理解海量数据背后的真正含义。", "confidence": "high", "topics": ["本体论与FDE方法论", "认知与心态"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "10-论文-NodeRED语义数字孪生方法论-讲解版幻灯片", "skills": ["xy-ontology"]}
{"id": "XY-ONT-734", "type": "principle", "knowledge": "原始数据(Raw Data)与语义数据(Semantic Data)的区别在于上下文。语义标注通过赋予实体、属性、数值与单位的关联,使数据具备互操作性,解决数据丰富但信息匮乏的问题。", "original": "Raw Data(45、101.2、TRUE、FALSE等一堆没有上下文的数字)vs \"Semantic Data\"(Entity: Valve A / Property: Temperature / Value: value / Unit: Celsius)——数据丰富但信息匮乏,语义标注才能赋予上下文,实现互操作性。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "10-论文-NodeRED语义数字孪生方法论-讲解版幻灯片", "skills": ["xy-ontology"]}
{"id": "XY-ONT-735", "type": "method", "knowledge": "构建语义数字孪生(Semantic Digital Twin)的五步方法论流程为:识别资产 → DT建模 → 添加语义 → 通信接口 → 部署。该流程包含两个迭代优化回路:在添加语义后可退回DT建模阶段修正,在通信接口配置完后也可退回DT建模阶段修正。", "original": "五步方法论被重新画成一条清晰的流程带:识别资产→DT建模→添加语义→通信接口→部署,且标注了两处迭代优化回路(添加语义之后可以退回DT建模阶段修正,通信接口配置完之后也可以退回DT建模阶段修正)。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "10-论文-NodeRED语义数字孪生方法论-讲解版幻灯片", "skills": ["xy-ontology"]}
{"id": "XY-ONT-736", "type": "method", "knowledge": "利用知识图谱(KG)可以解决异构性问题,为机器与人类建立通用的语言,通过实体、关系、节点与边的表达实现对物理世界的建模。", "original": "利用知识图谱(KG)解决异构性问题,建立机器与人类的通用语言,并用经典的dog-is-animal三元组图示解释实体/关系/节点/边的基本概念。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "10-论文-NodeRED语义数字孪生方法论-讲解版幻灯片", "skills": ["xy-ontology"]}
{"id": "XY-ONT-737", "type": "case", "knowledge": "在工业场景(如阀门/炼油厂)中,通过Node-RED编排流程可以实现从物理信号到知识图谱的自动化转换:通过dt-asset节点结合iot.schema.org等标准将原始信号转为带单位的语义值,并自动生成Cypher查询语句(MER句法)在Neo4j中构建知识图谱,实现结构与含义的存储。", "original": "用阀门-炼油厂-Petrobras的案例演示Node-RED编排流程如何自动生成Cypher查询语句(MERGE语法),并配图展示Neo4j里最终生成的知识图谱视图,强调\"不仅存储数据,还存储了结构和含义\"。", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "10-论文-NodeRED语义数字孪生方法论-讲解版幻灯片", "skills": ["xy-ontology"]}
{"id": "XY-ONT-738", "type": "insight", "knowledge": "语义数字孪生具备四个核心价值:1. 可发现性(支持复杂语义查询检索);2. 模块化(逻辑与实现分离,易于维护);3. 实时同步(物理状态与数据库实时一致);4. 以人为本(提供可理解的上下文,打破数据孤岛)。", "original": "收尾归纳四点核心价值:可发现性(支持复杂语义查询检索)、模块化(逻辑与实现分离,易于维护)、实时同步(物理状态与数据库实时一致)、以人为本(提供可理解的上下文,打破数据孤岛)。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "10-论文-NodeRED语义数字孪生方法论-讲解版幻灯片", "skills": ["xy-ontology"]}
{"id": "XY-ONT-739", "type": "insight", "knowledge": "数字孪生在管理物理世界对象时面临三重异构性挑战:物理异构(不同类型的物理资产)、时间异构(时序变化的数据)以及结构异构(数据的层级与组织方式不同)。", "original": "数字孪生要准确管理来自物理世界对象的数据,面临三重异构性——物理异构(不同类型的物理资产)、时间异构(时序变化的数据)、结构异构(数据的层级/组织方式不同)。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-740", "type": "principle", "knowledge": "数字孪生应用中存在两个易混淆的概念:一是描述“实际发生了什么”的物理孪生,二是描述“设计者期望发生什么”的虚拟模型。", "original": "物理空间实体和其数字模型很容易混为一谈,难以区分描述\"实际发生了什么\"和\"设计者期望发生什么\"。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-741", "type": "method", "knowledge": "Digital Twin Ontology (DTO) 采用三元架构,将物理孪生-虚拟模型二元对立扩展为:物理世界(External world)、设计者预期世界(Expected world)与虚拟世界(Interpreted world)。", "original": "Digital Twin Ontology(DTO)——核心创新是把传统\"物理孪生-虚拟模型\"二元对立,扩展成物理世界(External world)、设计者预期世界(Expected world)、虚拟世界(Interpreted world)三元架构。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-742", "type": "definition", "knowledge": "在DTO架构中,Hyle(物理孪生)是Entity在物理世界中的精确投影,代表实际发生的物理状态;Eidos(虚拟模型)是设计者预期世界对Entity的投影,代表设计意图。", "original": "Hyle(物理孪生,源自亚里士多德\"质料\"概念,是Entity在物理世界里的精确投影)、Eidos(虚拟模型,源自亚里士多德\"形式\"概念,是设计者预期世界对Entity的投影)", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-743", "type": "method", "knowledge": "Data Conversion Ontology (DCO) 通过将数据转换建模为有限自动机的状态转移图,定义了System、Conversion和Transaction等类,将数据转换过程本身进行语义化描述。", "original": "Data Conversion Ontology(DCO)——专门用来形式化\"从一个系统到另一个系统的数据转换过程\",把数据转换建模为有限自动机的状态转移图,定义System(单位和数据类型)、Conversion(转换的状态转移函数)、Transaction(转换过程中实际执行的操作,用RPC远程过程调用表示)等类", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-744", "type": "case", "knowledge": "在300平方米的仿真办公室环境验证中,研究者利用DTO/DCO构建了包含建筑管理数据和AGV日志的RDF格式链接图数据集,并通过SPARQL查询实现了对物理孪生(Hyle)与虚拟模型(Eidos)属性及时序动作的同步检索。", "original": "论文在一个约300平方米的仿真办公室环境(NTI Communications建立的CROSS LAB)里实测...构建了符合DTO/DCO的链接图数据集(RDF格式),并开发了一个可视化查看器应用,通过SPARQL查询同时检索物理孪生(Hyle)和虚拟模型(Eidos)的属性、坐标、时序动作等信息。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-745", "type": "insight", "knowledge": "在对比BIM/IFC、Brick、BOT等6种既有本体后发现,DTO/DCO组合是唯一能同时实现“物理孪生与虚拟模型解耦”、“时序动态数据描述”以及“数据转换过程描述”的方案。", "original": "结果显示DTO/DCO组合是唯一能同时描述\"物理孪生与虚拟模型解耦\"、\"时序动态数据\"、\"数据转换过程\"三项的方案,其余6种既有本体各有欠缺。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-746", "type": "insight", "knowledge": "在FDE知识管理中,区分“实际执行结果”与“方案设计意图”能避免案例库信息混淆;同时,将数据加工过程(如原始访谈到知识提炼)进行语义化记录,可以保留知识的可追溯性。", "original": "1. \"Hyle(实际发生的)vs Eidos(预期/设计的)\"这组区分对FDE知识板块极具参考价值...2. DCO把\"数据转换/编辑过程\"也当作需要被语义化记录的一等公民,提示FDE知识板块不仅要记录\"案例内容\"本身,还应该记录\"这条知识是怎么从原始访谈/录音提炼出来的\"这一加工过程,保留可追溯性。", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "11-论文-语义数字孪生互操作性与数据资产管理", "skills": ["xy-ontology"]}
{"id": "XY-ONT-747", "type": "insight", "knowledge": "制造业数字孪生落地面临两个核心挑战:一是异构数据源整合难,需要整合来自多传感器、控制器与设备的复杂数据;二是实时性与模型准确性的平衡问题,即模型需在吸收实时上下文信息的同时,随生产条件和设备磨损动态调整。", "original": "论文聚焦制造业数字孪生落地中的两个老大难问题:异构数据源整合难、实时数据流管理难。作者指出,数字孪生要精确复刻物理设备/系统/流程,必须整合来自多传感器、多控制器、多设备的异构数据,构建准确的模型本身就极其复杂;同时数字孪生模型不能只是静态快照,必须能持续吸收上下文信息并随生产条件、设备磨损、工艺更新实时调整,在\"实时性\"和\"模型准确性\"之间找到平衡是一个尚待解决的迫切问题。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-748", "type": "method", "knowledge": "一种“概念层-模型层-决策层”的三层知识图谱架构:概念层提供标准化、逻辑一致的基础知识框架(语法);模型层将概念层知识实例化到具体对象上,包含生产、对象、特征、工序、规制、过程、事件、关系8类要素;决策层接收模型层输出,驱动监督、管理、调整、预测与优化的闭环。", "original": "论文提出一个\"概念层-模型层-决策层\"三层知识图谱架构:1. 概念层(Concept Layer)——整合本体库、规则约束、领域专家知识,为整个制造领域提供标准化、逻辑一致的基础知识框架,相当于给模型层提供\"语法\";2. 模型层(Model Layer)——把概念层的知识实例化到具体产品对象上,生成可应用的数字化建模,包括Production(名称/类型/几何模型)、Object(机器/工具/人员", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-749", "type": "principle", "knowledge": "在三层知识图谱架构中,各层通过数据层(专家库、工艺库、知识库等)实现双向更新:概念层保障模型层的逻辑一致性,模型层为决策层提供可操作输入,生产过程的实时反馈则持续更新知识与数据,形成自我迭代闭环。", "original": "三层之间通过数据层(专家库、工艺数据库、综合数据库、知识库、案例库)双向更新:概念层为模型层提供逻辑一致性保障,模型层为决策层提供可操作输入,制造过程的实时反馈又持续更新知识和数据,形成自我迭代闭环。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-750", "type": "case", "knowledge": "在航空发动机叶片产线的验证中,采用三层知识图谱架构后,在5个月内(200批次产品)实现了:叶片轮廓最大误差精度从0.073mm下降至0.062mm,产品合格率从81.3%上升至85.2%。", "original": "论文在企业真实的航空发动机叶片产线上做了为期5个月(2024年,200批次叶片产品)的验证,用两个关键指标衡量效果:- 叶片轮廓最大误差精度:从1月的0.073mm逐月下降到5月的0.062mm;- 产品合格率:从1月的81.3%逐月上升到5月的85.2%。", "confidence": "high", "topics": ["商业案例与实战复盘", "本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-751", "type": "number", "knowledge": "在知识三元组抽取对比实验中,提出的三层知识模型在Precision (77.38%)、Recall (83.14%)、F1 (80.15%) 指标上全面超越了传统的OpenIE、Bi-LSTM及BERT-Bi-LSTM-CRF方法。", "original": "另外,论文还做了一个知识三元组抽取的对比实验,把提出的三层知识模型与三种既有方法(传统NLP方法OpenIE、深度学习方法Bi-LSTM、混合方法BERT-Bi-LSTM-CRF)比较,本文方法在Precision(77.38%)、Recall(83.14%)、F1(80.15%)三项指标上全面超越对比方法(BERT-Bi-LSTM-CRF次之,F1为72.29%),证明领域知识和三层认知推理机制", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-752", "type": "anti-pattern", "knowledge": "过度依赖领域专家的人工输入会限制系统的通用性和灵活性,导致知识库更新速度成为瓶颈,应提前规划如何结合机器学习或NLP技术实现自动化知识抽取。", "original": "对专家知识的依赖(过度依赖领域专家输入会限制系统通用性和灵活性,未来应结合机器学习/自然语言处理降低对人工标注的依赖)。...对于FDE知识板块是个提醒——本体设计初期依赖人工整理是必要的,但要提前规划如何逐步引入自动化知识抽取(比如用LLM从访谈逐字稿里自动识别案例要素),否则知识库更新速度会成为瓶颈。", "confidence": "medium", "topics": ["本体论与FDE方法论", "团队与模式设计"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-753", "type": "method", "knowledge": "在构建知识体系时,应采用“概念层-模型层-决策层”的分离设计思路:概念层定义本体与字段规范,模型层进行案例/实例的结构化录入,决策层驱动下游应用。这种解耦设计可确保本体修改不影响已有数据,反之亦然。", "original": "\"概念层-模型层-决策层\"三层分离的设计思路可以直接迁移:概念层对应FDE知识板块的本体/字段规范定义,模型层对应具体案例/客户实例的结构化录入,决策层对应\"根据知识库内容给出建议/话术/选题推荐\"这类下游应用,三者解耦能让本体修改不影响已有案例数据,反之亦然。", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "13-论文-多层知识图谱驱动的制造业数字孪生系统", "skills": ["xy-ontology"]}
{"id": "XY-ONT-754", "type": "insight", "knowledge": "传统数字孪生系统主要聚焦实时数据和当前操作状态,侧重于回顾性分析,缺乏对历史生命周期信息和制造知识的深度整合与利用,无法利用积累的经验进行决策。", "original": "论文指出传统数字孪生系统主要聚焦实时数据和当前操作状态,偏重\"回顾性分析\",缺乏对历史生命周期信息和制造知识的深度整合与利用——换句话说,传统DT只是一面\"实时镜子\",看不到\"过去发生过什么、积累了什么经验\"。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-755", "type": "principle", "knowledge": "知识驱动的数字孪生系统应构建一个既能利用实时数据,又能将历史生命周期数据作为决策支撑核心组件的架构。", "original": "论文要解决的问题是:如何构建一个既能利用实时数据、又能把历史生命周期数据当作决策支撑核心组件的知识驱动数字孪生系统。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-756", "type": "method", "knowledge": "知识驱动的数字孪生系统采用三层框架(数据层-本体层-服务层),其中本体层是核心,承载知识模型、几何模型、决策模型的\"三位一体\"。", "original": "论文采用Models for Manufacturing(MfM)方法论搭建整体架构,本体层是核心,承载知识模型、几何模型、决策模型的\"三位一体\"。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-757", "type": "definition", "knowledge": "制造知识被形式化为包含七类元素的集合:K_O(实体对象)、K_F(物理/环境/化学特征)、K_P(工艺相关信息)、K_E(运行活动)、K_R(规范规则/标准/协议)、K_D(决策模型)、K_G(几何模型)以及R(实体间关系)。", "original": "论文把制造知识形式化为Knowledge = {K_O, K_F, K_P, K_E, K_R, K_D, K_G, R}七类元素——K_O实体对象(材料/工具/机器/产品/人员)、K_F物理/环境/化学特征、K_P工艺相关信息(传感器实时监测+历史记录)、K_E运行活动(机器操作/人员操作/异常/维护)、K_R规范规则/标准/协议、K_D决策模型、K_G几何模型、R各类实体间关系。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-758", "type": "method", "knowledge": "知识驱动的建模方法采用自下而上的路径:从多源异构数据(手册/图纸/经验/设备/传感器/几何模型)出发,通过实体识别、关系抽取、属性分析、本体设计(如使用BERT-BiLSTM-CR夫模型),经过质量检查及知识匹配消歧与实体对齐融合,最终构建制造过程知识图谱。", "original": "从操作手册/设计图纸/专家经验/设备资源/传感器数据/几何模型等多源异构数据出发,经过实体识别/关系抽取/属性分析/本体设计(用BERT-BiLSTM-CRF等序列标注模型做知识抽取)、唯一性/一致性/有效性/完整性四项质量检查、知识匹配消歧与实体对齐融合,最终构建成制造过程知识图谱;", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-759", "type": "method", "knowledge": "为了保证广泛的互操作性,通过将自建的制造过程本体(MPO)对齐到国际通用的BFO(Basic Formal Ontology)顶层本体,划分为连续实体(Continuant)和发生实体(Occurrent)两大类,实现与外部本体生态的兼容。", "original": "为了保证更广泛的互操作性,论文把自建的\"制造过程本体(Manufacturing Process Ontology, MPO)\"对齐到国际通用的顶层本体BFO之下,划分出Continuant(连续实体,如物料/几何/特征/规则)和Occurrent(发生实体,如工序/事件/决策)两大top-level类,各自再细分子类,形成一个既贴合制造场景又能与外部本体生态兼容的结构。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-760", "type": "case", "knowledge": "在航空发动机叶片产线验证中,通过知识驱动的几何模型自动更新和决策模型支持,实现了:1. 几何模型设计的快速构建;2. 结合实时数据的工艺仿真;3. 基于知识图谱嵌入的工艺知识问答与工具推荐;4. 针对下一制造阶段的预测分析。结果显示:在5个月200批次验证期内,叶片轮廓误差从0.073mm降至0.062mm,产品合格率从81.3%提升至85.2%。", "original": "论文同样在航空发动机叶片产线做了验证,通过知识驱动的几何模型自动更新和决策模型支持,展示了知识图谱如何驱动\"几何模型设计的快速构建\"(拉动模型库相关几何类作为驱动力)、\"工艺仿真\"(结合实时数据实现制造过程可视化)、\"工艺知识问答\"(基于知识图谱嵌入模型做知识补全和预测,推荐适配的加工工具)、\"预测与分析\"(针对下一制造阶段给出潜在结果预测)四大功能,系统在5个月200批次的验证周期内使叶片轮廓误", "confidence": "high", "topics": ["本体论与FDE方法论", "商业案例与实战复盘"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-761", "type": "insight", "knowledge": "将制造知识的七元素分类法(对象、特征、工艺信息、活动、规则、决策、几何、关系)转化为企业FDE知识模板,可将对象替换为客户/项目,特征替换为客户画像,工艺信息替换为合作记录,活动替换为关键动作,规则替换为合规要求,决策模型替换为打法/话术逻辑。", "original": "\"Knowledge = {K_O, K_F, K_P, K_E, K_R, K_D, K_G, R}\"这套七元素分类法可以直接套用改造:把K_O(对象)换成客户/项目主体,K_F(特征)换成客户画像属性,K_P(工艺信息)换成合作过程记录,K_E(活动)换成关键动作节点,K_R(规则)换成行业规范/合规要求,K_D(决策模型)换成打法/话术推荐逻辑,形成一套结构一致、可复用的FDE知识分类模板;", "confidence": "medium", "topics": ["本体论与FDE方法论", "团队与模式设计"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "14-论文-知识驱动的制造工艺数字孪生系统KBDTS", "skills": ["xy-ontology"]}
{"id": "XY-ONT-804", "type": "method", "knowledge": "将FDE知识板块的“给客户的建议”从自由文本升级为包含目标对象、前提条件、影响范围和可追溯依据的结构化判断,利用Action设计实现完整、受控且可审计的输出。", "original": "Action设计(完整、受控、可审计)可用来重新设计FDE知识板块\"给客户的建议\"——从自由文本升级为带目标对象、前提条件、影响范围、可追溯依据的结构化判断。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "21-解析-Ontology4篇到底是什么", "skills": ["xy-ontology"]}
{"id": "XY-ONT-805", "type": "principle", "knowledge": "在现有技术架构中,原子库往往表现为事实层的知识图谱,而缺乏定义原子间组织方式与推理逻辑的本体层,本体化改造的核心任务是填补这一缺口。", "original": "原子库更像知识图谱(事实层),缺少定义\"原子间怎么组织、怎么推理\"的本体层,是本体化改造要补的缺口。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "21-解析-Ontology4篇到底是什么", "skills": ["xy-ontology"]}
{"id": "XY-ONT-806", "type": "method", "knowledge": "在利用本地大模型进行知识蒸馏时,可参考RDF→RDFS→OWL的递进关系进行分阶段实施:首先将数据结构化为事实三元组,随后补充类型与关系定义,最后根据需求引入复杂推理规则,避免一步到位。", "original": "RDF→RDFS→OWL的递进关系,可作为喂给本地26B模型做知识蒸馏的分阶段参考:先结构化成事实三元组,再补类型/关系定义,最后视需要再上复杂推理规则,不必一步到位。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "21-解析-Ontology4篇到底是什么", "skills": ["xy-ontology"]}
{"id": "XY-ONT-807", "type": "definition", "knowledge": "数据集成分析平台的核心特征是:通过数据集成分析生产决策赋能信息,并通过预分析式可视化支撑并引导操作者的知识实践。", "original": "其核心特征是\"通过数据集成集成分析生产决策赋能信息\"和\"通过预分析式可视化支撑(并引导)操作者的知识实践\"。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-808", "type": "insight", "knowledge": "数据整合过程中的“认知不透明性”指:数据标注、结构化和组织等关键决策过程对客户组织和数据主体是不可见的,这种去语境化的过程可能导致松散的关联被误读。", "original": "认知不透明性(epistemic opacity)——数据整合分析过程中大量关键决策(如何标注、如何结构化、如何组织数据)对客户组织和被处理数据的公民都是不可见的,数据整合天然带来\"去语境化\",可能导致原本松散的关联网络被误读成(比如)恐怖主义关系网;", "confidence": "high", "topics": ["本体论与FDE方法论", "合规与风控"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-809", "type": "insight", "knowledge": "数据平台的“认知控制权”指:平台的标签与可视化设计会引导用户关注特定范式,通过预先塑造分析师的视角,决定了“什么可以被看见和知道”,使平台成为折射世界的“数字棱镜”。", "original": "认知控制权(epistemic control)——Palantir的标签/可视化设计倾向于鼓励用户关注\"网络关系\"这一特定分析范式,这种设计选择本身就在预先塑造分析师能够生产出什么样的知识...平台的可见性设计本身就在预先决定\"什么可以被看见和知道\"。", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-810", "type": "method", "knowledge": "在FDE知识板块设计中,应学习Palantir的Dynamic Ontology范式,采用“对象-属性-关系”三元结构,并确保每条属性/关系都能强制溯源回原始文档,以避免脱离原始素材的抽象总结。", "original": "Palantir的Dynamic Ontology示范了一个真实、经过大规模商业验证的本体设计范式:对象-属性-关系三元结构,且每条属性/关系强制可溯源回原始文档,这个\"强制溯源\"设计对FDE知识板块极具参考价值——案例库里的每一条结论都应该能点回到具体的访谈记录/录音/文档,而不是脱离原始素材的抽象总结;", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-811", "type": "method", "knowledge": "FDE知识录入可采用轻量化交互设计:允许业务人员手动标注案例中的关键实体,由系统自动建立跨案例的动态链接,以替代繁重的结构化表单填写。", "original": "\"标签功能让用户手动标注文档中的实体、系统据此自动建立动态链接\"这一交互设计,提示FDE知识板块的知识录入环节可以采用类似的轻量交互(业务人员标注案例里的关键实体,系统自动建立跨案例关联),而不需要每次都做繁重的结构化表单填写;", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-812", "type": "anti-pattern", "knowledge": "在本体设计阶段应避免让本体结构本身带偏业务判断,例如过度强调某一类特征而遮蔽其他重要维度,因为本体设计者掌握着预先决定知识可见性的权力。", "original": "需要有意识地避免让本体结构本身带偏业务判断(比如过度强调某类案例特征而遮蔽其他同样重要的维度);", "confidence": "high", "topics": ["本体论与FDE方法论", "认知与心态"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-813", "type": "principle", "knowledge": "本体驱动知识库的价值在于规模效应与关联丰富度:应优先保证案例条目的持续积累和跨案例关联质量,而非过早追求单条案例内容的极致精细化。", "original": "本体驱动知识库的价值不在于单条数据本身,而在于接入的数据/案例越多、跨案例关联越丰富,每条已有知识对使用者的价值就越大,这为FDE知识板块的建设节奏提供参考:应优先保证案例条目的持续积累和跨案例关联质量,而非过早追求单条案例内容的极致精细化。", "confidence": "high", "topics": ["本体论与FDE方法论", "团队与模式设计"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "22-论文-数据集成分析平台作为数字平台-Palantir案例研究", "skills": ["xy-ontology"]}
{"id": "XY-ONT-814", "type": "principle", "knowledge": "本体建设的可持续性过去依赖专家个人经验,专家离职或换团队会导致建设连续性中断;成熟的做法是把建模能力沉淀进平台工程体系,让本体像软件一样被持续创建、更新、发布和治理。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "related": [{"id": "XY-ONT-004", "rel": "refines"}], "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-714"]}
{"id": "XY-ONT-815", "type": "method", "knowledge": "本体对象建模应工程化:提供统一的开发框架和对象模型规范,不管谁来建模都遵循同一套工程标准和版本管理,让业务实体能像软件组件一样被标准化、复用、持续迭代,而不是依赖某个人的个人经验。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-715"]}
{"id": "XY-ONT-816", "type": "method", "knowledge": "跨系统语义映射应该自动化配置而非人工维护:让ERP、MES、CRM等系统的数据持续收敛、映射到统一业务对象上,且不需要改动原系统本身,各系统依然共享同一套企业业务语言。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-716"]}
{"id": "XY-ONT-817", "type": "principle", "knowledge": "本体要从静态知识库升级为可执行的业务现场:把真实业务动作(如调整参数、创建工单、发起审批)封装成Agent可直接调用的动作单元,Agent理解业务对象后就能直接调用动作解决问题,而不只是查询信息。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "related": [{"id": "XY-ONT-096", "rel": "supports"}], "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-717"]}
{"id": "XY-ONT-818", "type": "method", "knowledge": "企业内AI应用增多后,应设一个统一的能力接入出口,所有Agent通过标准化方式接入、经权限控制和合规审计后再调用具体业务能力,在开放能力给AI和管住安全边界之间取得平衡,而不是让每个Agent单独对接底层系统导致接口混乱。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-718"]}
{"id": "XY-ONT-819", "type": "method", "knowledge": "本体的长期运营需要治理体系化:把版本管理、分支审核、灰度发布、调用审计纳入统一流程,不再靠开会口头对齐和人工核对版本,才能像成熟软件系统一样持续升级、长期稳定运营。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-719"]}
{"id": "XY-ONT-820", "type": "insight", "knowledge": "企业真正值钱的AI核心资产不是模型也不是Agent本身,而是标准化、可迭代、能被AI直接调用的业务语义资产——本体论正在从一次性画出来的静态业务模型,变成能持续运营、持续演进的AI时代基础设施。", "original": "", "confidence": "medium", "topics": ["本体论与FDE方法论"], "source_type": "user_import", "source_label": "Palantir Ontology学习资料-2026-09", "chapter": "08-Foundry7月更新-语义工程化", "related": [{"id": "XY-ONT-814", "rel": "prerequisite_of"}], "skills": ["xy-ontology"], "merged_ids": ["XY-ONT-720"]}
{"id": "XY-ONT-821", "type": "insight", "knowledge": "本体建模六步法(定边界→抽实体→定类型→建层级→抽关系→验证)实操中曾在一个复杂真实代码库上完整跑通一遍,证明方法论不局限于知识文本库,同样适用于代码/系统类领域,不是只能纸上谈兵。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-822", "type": "principle", "knowledge": "本体核实必须先确认数据版本一致性:核实用的真实数据快照与待核实的候选草稿如果不是同一时间点,会出现草稿里有但真实数据没有的假阴性——正确做法是先比对两份材料的时间戳/版本号,对无法核实的部分如实标注版本不对齐、本轮无法核实,而不是强行给出结论。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-823", "type": "method", "knowledge": "本体验证阶段的实体命名冲突排查,先对全部同名对象做重复分组,排除脚手架/测试代码命名撞车(非领域概念冲突),再对每一组冲突逐一判断是真实歧义还是同名不同义的合理分层——多数命名撞车其实是生产代码与草稿/简化版本各自建模的分层惯例,不是设计失误,不应该被本体建模强行合并成一个实体。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-824", "type": "insight", "knowledge": "一个复杂系统里,正式流程环节字段完整、可追溯,与草稿/简化流程环节字段精简、一次性,是常见的合理分层惯例,同名不代表设计失误,本体建模时应该区分成两个不同类别,而不是强行统一成一个实体。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-826", "type": "principle", "knowledge": "本体里的关键权威数据源(如某种统一台账/清单类实体)只能有一份,且最好有强制校验机制防止遗漏——这条规矩往往不是系统设计之初就有的,而是被真实运营事故反复验证后才补上的,本体建模时应该主动排查系统里是否存在类似本该唯一但目前靠人工记得同步的清单类实体。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-827", "type": "insight", "knowledge": "本体建模的候选草稿即使认真做过一轮,仍可能系统性遗漏核心聚合实体——真实实操中,一个把多个处理阶段串起来的顶层聚合对象在扫描材料里明明存在,却因为类名太普通、又与另一处同名类撞名,被第一轮当成普通记录类漏掉了;验证阶段要专门把\"名字最普通、且有同名冲突\"的候选单独拎出来看一遍,并注意聚合的边界止于裁决环节,下游的执行环节不在聚合之内。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-828", "type": "method", "knowledge": "本体建模的验证阶段应该明确排除范围(测试代码、纯配置目录、部署配置正文等),并如实列出本轮核实到的范围与明确排除的范围——不做全覆盖的假设,能核实多少就如实报告多少,未核实到的部分标注本轮无法核实,而不是沉默略过或者编造结论。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-829", "type": "anti-pattern", "knowledge": "本体建模时看到两个模块里有同名字段/同名分类,不能想当然认为是同一个东西——不同层级恰好用了同一个名字是常见情况,若不细查会被误判为命名冗余应合并,实际上强行合并可能丢失关键的语义层级信息。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-830", "type": "method", "knowledge": "本体的Action层建模,每个动作要记录三样:触发者是谁(系统自动/人工/哪个具体环节)、涉及哪些实体(输入/输出)、前置条件是什么(哪些判断必须先通过)——实操中发现有些系统会要求即使已经判定失败也要跑完全部检查链、不许提前短路返回,这种非直觉的完整性要求是为了让事后复盘能区分不同的失败原因,本体建模时应该把这类细节也记录进前置条件里,不能想当然假设逻辑会在第一个失败点就停止。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "XY-ONT-831", "type": "insight", "knowledge": "真实系统里可能存在两条描述同一类现实世界事件、但完全独立的观测/数据管线(互相不知道对方存在),这可能是有意的设计(两条独立信源做交叉验证),也可能是遗留的架构缺口——本体建模只能记录代码事实是什么,不能替代业务方判断这是不是问题,遇到这类情况应如实标注需要问系统原始设计者的意图,不擅自下业务结论。", "original": "", "confidence": "high", "topics": ["本体论与FDE方法论"], "source_type": "case_study", "source_label": "本体建模实操经验(已脱敏,不含任何客户/项目识别信息)", "skills": ["xy-ontology"]}
{"id": "FDE3P-004", "type": "case", "knowledge": "德国斯图加特一家七八十人的零部件制造企业,多条产线、型号繁多,材料配比和调参经验大多在老师傅脑子里,新人要跟两三个月以上,老师傅又忙又临近退休。FDE 没先谈模型,而是跟着新人培训流程走,统计新人高频问题和易错点,画出知识地图;再把已有文档和老师傅的隐性经验整理进知识库,用高频问题当测试集验收;最后把问答能力接进企业原有培训平台,做成带进度和课后问答的学习流程,并陪跑近两个月。项目前后约半年,实际投入约四个月。结果:企业第一次看清哪些经验不能随人流失,老师傅从回答重复问题转为盯进度、处理难题,培训进度也能被追踪。", "confidence": "high", "topics": ["AI与工具", "本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.1", "chapter": "NO.1 从师徒带教到AI辅助培训:FDE用AI破解制造业经验传承难题", "section": "Q1-Q4", "related": [{"id": "FDE3P-005", "rel": "supports"}, {"id": "FDE3P-006", "rel": "supports"}, {"id": "FDE3P-007", "rel": "supports"}, {"id": "FDE3P-008", "rel": "supports"}, {"id": "FDE3P-009", "rel": "supports"}]}
{"id": "FDE3P-006", "type": "method", "knowledge": "一线员工往往不愿花时间接受访谈,自己也说不全日常细节。调研时除了访谈,直接跟着员工把一天的活走一遍:看他在哪一步停下、去问谁、哪些问题一再出现。这样得到的痛点,比问一句“你有什么痛点”具体得多。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.1", "chapter": "NO.1 从师徒带教到AI辅助培训:FDE用AI破解制造业经验传承难题", "section": "Q3/Q5", "related": [{"id": "FDE3P-004", "rel": "example_of"}]}
{"id": "FDE3P-018", "type": "case", "knowledge": "一支做商事合同纠纷、股权争议的诉讼律师团队,单个案件材料动辄上千页,光整理就要 3 到 5 个工作日,新老律师效率能差 3 到 5 倍,过去买的法律数据库和文档工具都是单点功能,要在五六个系统间来回切。FDE 从早上开电脑起跟班观察,旁听案件讨论,把前期准备拆出可由 AI 辅助的环节,沉淀成 7 个标准模块;架构上让大模型做理解、摘要和初步推理,专业系统管案件知识图谱和法条判例的精确检索,律师保留最终判断。上线后发现聊天框要反复交代背景,便把入口改成“先建案、再交互”的案件工作台。", "confidence": "high", "topics": ["AI与工具", "本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.3", "chapter": "NO.3 从案头工作到智能协作:FDE用AI优化诉讼律师的案件处理能力", "section": "Q1-Q4", "related": [{"id": "FDE3P-019", "rel": "supports"}, {"id": "FDE3P-020", "rel": "supports"}, {"id": "FDE3P-021", "rel": "supports"}, {"id": "FDE3P-022", "rel": "supports"}, {"id": "FDE3P-023", "rel": "supports"}, {"id": "FDE3P-024", "rel": "supports"}]}
{"id": "FDE3P-021", "type": "method", "knowledge": "把专家的隐性经验变成系统能执行的东西:花大量时间和行业里最好的从业者一起干活,观察他们做判断时关注什么、怎么把普通人的话转成专业表述、怎么把证据串起来;把他们常用的思维框架(例如请求权基础分析、证据链规则)拆成一个个流程节点,让 AI 按专家的思路组织信息,而不是只吐答案。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.3", "chapter": "NO.3 从案头工作到智能协作:FDE用AI优化诉讼律师的案件处理能力", "section": "Q5", "related": [{"id": "FDE3P-018", "rel": "example_of"}]}
{"id": "FDE3P-054", "type": "case", "knowledge": "一家海外工业企业用一套老 ERP 管理从手套螺丝到卡车机械臂的大量物料,约 15 个团队围着它录入、采购、签合同、管仓储,人人抱怨“ERP 不好用”。FDE 团队花了数月访谈、坐在业务人员旁边看操作,整理出约 15 个可处理的功能点,其中最典型的是数据脏和搜索弱。继续追下去,发现真正的损失链是:找不到已有物料、重新录入、重复数据越来越多、买错或重复采购、库存积压多年后报废。最终做的是历史数据清洗标准化加语义检索,是否采购仍由人拍板;效果以错误采购和重复采购是否下降来衡量,节省的金额要由客户业务人员确认。", "confidence": "high", "topics": ["AI与工具", "本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.8", "chapter": "NO.8 从物料难找到精准定位:FDE用AI减少工业企业采购浪费", "section": "Q1-Q4", "related": [{"id": "FDE3P-055", "rel": "supports"}, {"id": "FDE3P-056", "rel": "supports"}, {"id": "FDE3P-057", "rel": "supports"}, {"id": "FDE3P-058", "rel": "supports"}, {"id": "FDE3P-059", "rel": "supports"}, {"id": "FDE3P-060", "rel": "supports"}]}
{"id": "FDE3P-055", "type": "method", "knowledge": "用户说“系统不好用”“搜索不好用”都还不是问题定义。要一直追问:为什么、然后呢、最后影响了什么,直到连上真实的财务损失。例如物料搜不到,一路追到重复录入、数据污染、错采重采、库存报废。只有这条链成立,才知道这个 AI 应用到底值多少钱。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.8", "chapter": "NO.8 从物料难找到精准定位:FDE用AI减少工业企业采购浪费", "section": "Q2/Q6", "related": [{"id": "FDE3P-054", "rel": "example_of"}, {"id": "XY-UZB-096", "rel": "refines"}]}
{"id": "FDE3P-100", "type": "method", "knowledge": "知识萃取和传统咨询不同:咨询带着行业模板去判断客户该在哪提效,萃取是把客户自己人脑子里的做法挖出来。具体做法是让业务人员把工作过程讲出来——怎么做、为什么这么做、哪些情况例外、什么时候必须人来判断。这一步相当于先给AI准备好这家公司的业务世界。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.14", "chapter": "NO.14 从技术交付到业务自驱", "section": "Q3", "related": [{"id": "FDE3P-096", "rel": "example_of"}, {"id": "XY-UZB-144", "rel": "refines"}]}
{"id": "FDE3P-101", "type": "principle", "knowledge": "先把流程梳理清楚,再做Agent。通用大模型靠概率生成,它对A公司电商流程的理解是对的,不代表对B公司也对;而且老员工干了几年,很多步骤已习惯到察觉不到,不写出来模型只能自己编。流程必须来自这家企业本身,不能让模型'自己理解一下业务再去做'。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.14", "chapter": "NO.14 从技术交付到业务自驱", "section": "Q3", "related": [{"id": "FDE3P-096", "rel": "example_of"}, {"id": "XY-UZB-205", "rel": "refines"}]}
{"id": "FDE3P-107", "type": "method", "knowledge": "把优秀员工的经验变成可复用的业务逻辑:不同客户的内容表面各异,背后要关注的字段往往相近(服务对象、门店类型、主推品、活动、卖点、期望风格),把这些做成明确的输入参数,具体取向用参数去调。这样一套逻辑能在同类客户间复用,而不是每次靠人临场判断。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.15", "chapter": "NO.15 从经验写稿到稳定量产", "section": "Q6", "related": [{"id": "FDE3P-104", "rel": "example_of"}]}
{"id": "FDE3P-124", "type": "insight", "knowledge": "非标制造企业早年用ERP和报价模板管报价为什么没成:这类系统擅长'字段填进来就按规则往下跑',但最难的第一步——把客户五花八门的图纸和资料理解成标准数据——始终没人解决。结果系统只覆盖标准件,非标部分照样回到人手里,模板还让人多填一遍表。", "confidence": "high", "topics": ["AI与工具", "本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.17", "chapter": "NO.17 从老师傅报价到AI深度协同", "section": "Q2", "related": [{"id": "FDE3P-116", "rel": "example_of"}]}
{"id": "FDE3P-125", "type": "principle", "knowledge": "数据清洗、结构化和业务规则还原本身就是项目的一部分,不是等客户准备好再进场的前置条件。中小企业的历史报价、物料、供应商资料很少能直接给AI用,还要从数据和老师傅的经验里还原'为什么这样映射、为什么这样报价'的规则,这些不解决,模型再强也没用。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.17", "chapter": "NO.17 从老师傅报价到AI深度协同", "section": "Q6", "related": [{"id": "FDE3P-116", "rel": "example_of"}, {"id": "XY-UZB-865", "rel": "refines"}]}
{"id": "FDE3P-136", "type": "method", "knowledge": "给AI写'岗位说明书':公司招人做某项工作时HR会写岗位说明,让AI接手这项工作也应如此。借HRBP的思路,请真正做这项工作的人自己拆解:平时审什么、输入是什么、输出是什么、什么情况要判断、什么规则可以直接执行、什么需要上下文。由懂业务的人定义AI怎么干活,避免外行指导内行。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.19", "chapter": "NO.19 从人工排队到审核前置", "section": "Q3", "related": [{"id": "FDE3P-134", "rel": "example_of"}, {"id": "XY-UZB-689", "rel": "refines"}]}
{"id": "FDE3P-137", "type": "method", "knowledge": "把一项混杂的审核工作分三层分配:确定性问题(错别字、法规硬条款、营养成分计算)直接交AI自动检查;一开始要人判断但会重复出现的问题(如各渠道的规范反馈),每次记录下来,逐步沉淀成标准后交AI预审;审美和表达这类复杂判断留给人,AI只负责把历史经验和上线后的销售表现等上下文递给人。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.19", "chapter": "NO.19 从人工排队到审核前置", "section": "Q3", "related": [{"id": "FDE3P-134", "rel": "example_of"}]}
{"id": "FDE3P-140", "type": "insight", "knowledge": "传统SaaS系统记录的是结果,不是过程。比如产品为什么要做得'更脆'、配方为什么这么选,背后的调研、渠道沟通和多次实验经验大多不在系统里,系统里只剩一个结论;而且只要信息靠人主动录入,就一定有损耗。业务规模和SKU变多后,要解决的已经不是'有没有系统',而是整条链路上的上下文能不能流动。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.19", "chapter": "NO.19 从人工排队到审核前置", "section": "Q2", "related": [{"id": "FDE3P-134", "rel": "example_of"}]}
{"id": "FDE3P-141", "type": "method", "knowledge": "'点—线—面'梳理组织:面是公司的战略目标和它最终靠什么创造价值;线是把目标拆成的一条条价值链和业务流程;点是价值链上的每个岗位——承担什么角色、输入输出是什么、哪些是这个人的独门经验。三层理清后,AI该放在哪里自然就清楚了。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.19", "chapter": "NO.19 从人工排队到审核前置", "section": "Q6", "related": [{"id": "FDE3P-134", "rel": "example_of"}]}
{"id": "FDE3P-174", "type": "insight", "knowledge": "品牌出海走经销商模式时,要区分两个数:Sell In是品牌卖给经销商多少,容易从自家发货系统拿到;Sell Through是消费者最终从渠道买走多少,更能反映真实需求和营销效果,却最难获得。总部问一句'卖得怎么样',不同角色可能理解成不同的数,口径从源头就错了,所以要从数据源头把口径对齐。", "confidence": "high", "topics": ["通用商业管理", "本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "Datawhale《FDE案例100》第一期·NO.24", "chapter": "NO.24 从发货到动销", "section": "Q1", "related": [{"id": "FDE3P-172", "rel": "example_of"}]}
{"id": "FDE3P-191", "type": "definition", "knowledge": "双向蒸馏是 FDE 的核心机制。对客户蒸馏:把散在员工脑子里的业务经验、流程规则、隐性知识,变成 AI 能调用的知识库、规则、工作流和 SOP,客户得到更稳定可复制的执行能力。对厂商自己蒸馏:把大客户项目里摸到的行业门道、对接过的系统接口、用过的测试办法、场景模板,连同踩过的坑,一起变成产品能力、行业模板、内部工具和平台组件,下一个同类客户交付更快更深。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第二章", "section": "2.3 双向蒸馏:交付如何变成学习"}
{"id": "FDE3P-192", "type": "principle", "knowledge": "本体层是双向蒸馏的共同载体。同一个业务对象在不同客户系统里字段名、编码、流程写法都不同,AI 直接面对会读不懂;本体把这些差异翻成统一的业务说法:哪个是客户、订单、审批、风险事件,彼此什么关系,能触发什么动作,什么条件下要人接手。从一家客户提炼的本体能复用到同行下一家——复用的是行业层面的抽象,不是拿这家的数据去训练模型。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-ontology", "xy-fde"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第二章", "section": "2.3 双向蒸馏:交付如何变成学习", "related": [{"id": "FDE3P-191", "rel": "prerequisite_of"}, {"id": "XY-ONT-816", "rel": "refines"}]}
{"id": "FDE3P-193", "type": "principle", "knowledge": "本体沉淀是先投后收的生意。延迟收益:头几个项目除了交付还要额外做抽象、结构化和验证,成本比纯项目制更高,而且做第一单时还不知道有没有第二单。规模效应:同类客户够多时,几十上百家能复用同一套本体,成本和周期都明显下降;按报告经验,一个典型垂直领域做到七八家,本体就基本稳住了。领域里只有一两个客户,就别指望靠沉淀回本;平台型企业有耐心、客户多,更适合扛这份前期投入。", "confidence": "high", "topics": ["本体论与FDE方法论", "团队与模式设计"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第二章", "section": "2.4 延迟收益与规模效应:底层逻辑为何成立", "related": [{"id": "XY-ONT-052", "rel": "refines"}]}
{"id": "FDE3P-197", "type": "risk", "knowledge": "别以为有了 AI,本体建模就便宜了。AI 从代码或设计文档里抽出来的实体关系也许 98% 是对的,但要揪出那 2% 的错,人还得从头读到尾。AI 能加快初稿,最后的确认和治理仍要领域专家投入——这也是负责本体的团队成本省不掉的原因。报告在虚构案例里给的比例是 AI 占两成、人工审核占八成。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "skills": ["xy-ontology", "xy-fde"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第二章", "section": "2.6 产品演进与知识沉淀形式"}
{"id": "FDE3P-199", "type": "case", "knowledge": "Palantir 本体的由来:早期给情报客户建库时,团队本能地给“人”建一张表、给“资金”建一张表,结果每换一个客户或机构就得重建结构,完全不通用。后来把抽象拉高一层:底层只保留对象、属性、媒体、链接四种原语,具体语义交给本体去标注(这个实体是人、那个是船、这组关系是资金流)。FDE 和产品经理的问题从“给这个客户设计什么表”变成“在本体框架里怎么建模,才能让功能跨客户复用”。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.2 ONTOLOGY:从技术架构到业务语义"}
{"id": "FDE3P-200", "type": "definition", "knowledge": "报告把 Palantir 本体拆成三层:语义层回答“世界里有什么”,落在实体定义、属性结构和元数据映射;动势层回答“能做什么”,落在动作定义、权限模型、审批流和函数;动态层回答“正在发生什么”,落在实时数据管道、事件流和状态机。三层合起来,本体才是面向业务执行的语义层,而不只是一张静态的概念图。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.2 ONTOLOGY:从技术架构到业务语义", "related": [{"id": "XY-ONT-703", "rel": "refines"}]}
{"id": "FDE3P-201", "type": "case", "knowledge": "国内行业里本体要抽的“稳定概念”长什么样:金融是账户、客户、交易、授信、审批、风险事件;教育是学生、课程、作业、教案、评价、教务流程;传媒是选题、素材、稿件、审核、发布、分发和运营数据。客户系统里叫“客户编号”“用户ID”“主体编码”的字段,本体要认出是同一类对象;客户一句“审核不过就打回去改”,本体要知道背后牵涉哪些角色、状态、权限和通知。缺这层,AI 只能做问答和摘要。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-ontology", "xy-fde"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.2 ONTOLOGY:从技术架构到业务语义"}
{"id": "FDE3P-203", "type": "principle", "knowledge": "在 Palantir 的分工里,FDE 不负责建本体本身,那是后方产品团队的事。FDE 的活是在本体框架下给客户搭具体应用,同时把现场冒出来的新业务对象、新关系、新规则反馈回后方,推动本体继续长。产品要给在现场交付的 FDE 提供杠杆,而且杠杆应随时间越来越多:部署第一个客户很费劲,第十个同类客户就该轻松得多。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.4 FDE 在本体体系中的位置"}
{"id": "FDE3P-204", "type": "number", "knowledge": "报告给的成本曲线对比:项目制交付几乎是线性的,服务 10 个客户的累计成本约为 1 个客户的 8—9 倍,因为每家都要重新摸组织、流程、数据口径和系统边界;本体化交付第一单约是普通项目的 3 倍,但只要抽象对了,同类客户只需映射和适配,10 个客户累计约 4—4.5 倍。前者只问这单能不能验收,后者还问这单能不能让下一单更便宜。", "confidence": "high", "topics": ["本体论与FDE方法论", "团队与模式设计"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.4 FDE 在本体体系中的位置", "related": [{"id": "FDE3P-193", "rel": "supports"}]}
{"id": "FDE3P-205", "type": "risk", "knowledge": "Palantir 的几个前提在国内不成立:一是 Foundry 建立在客户多年结构化数据之上,国内多数企业数据治理薄弱,本体没法从系统里“抽”,得在项目里重新梳理;二是其本体靠十余年、数百名专职工程师打磨,国内客单价和付费习惯撑不起;三是 Gotham 客户因安全许可天然难迁走,商业客户没有这种粘性;四是约 0.3% 的录用率意味着人才密度无法批量复制。", "confidence": "high", "topics": ["本体论与FDE方法论", "团队与模式设计"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.5 模式成立的前提条件与中国适配"}
{"id": "FDE3P-206", "type": "principle", "knowledge": "国内团队别照搬 Palantir 整套技术栈,要抓它的核心逻辑:在数据和业务之间架一层统一语义,让 AI 和 FDE 都去操作这层语义,而不是直接碰杂乱的原始系统。实现上用更轻的组合——Skill + 连接器 + 行业知识库——一步步逼近本体层的效果。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第三章", "section": "3.5 模式成立的前提条件与中国适配", "related": [{"id": "FDE3P-205", "rel": "supports"}]}
{"id": "FDE3P-209", "type": "principle", "knowledge": "本体不是新概念,过去没流行开,是因为就算把领域概念和流程抽象清楚,到客户现场最后一公里还得写大量代码、接大量系统,价值被交付成本吃掉了。大模型和 AI 编程擅长把清楚的描述变成代码、接口和工作流,但它不知道客户要什么;本体恰好把业务对象、关系和操作规则讲清楚。AI 管快速实现,本体管描述稳定可复用,两者叠加,定制交付才有机会摆脱按人月卖。", "confidence": "high", "topics": ["本体论与FDE方法论", "AI与工具"], "skills": ["xy-ontology", "xy-fde"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第四章", "section": "4.2 AI 为什么放大本体层价值"}
{"id": "FDE3P-214", "type": "method", "knowledge": "国内更现实的是分层投入:任何行业先做术语统一和场景清单,把常见问题、业务对象、高频任务理出来;同类客户攒到一定数量,再把常用的 Skill 拼成行业模板;等到客户数、客单价、复用率都够高,再去系统地建完整本体。本体不是事先设计出的宏大模型,而是从一个个项目里长出来的:先有项目里的 Skill,再有跨项目模板,最后才是本体,顺序不能倒。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第四章", "section": "4.5 产品演进路径:从 SKILL 到本体"}
{"id": "FDE3P-234", "type": "method", "knowledge": "行业本体冷启动的做法(报告以虚构的 ERP 案例演示):请 2—3 位有交付经验的领域专家用约两周理出核心实体清单、实体关系、状态流转规则,以及每次状态切换的前置条件、谁有权操作、异常怎么走;让 AI 读已有项目的设计文档和表结构先出草稿,专家逐条补正确认,不能跳过人工;存成结构化文件并配一份业务语言说明。不求完美,覆盖约七成主流场景即可,同时列一份“已知缺口”清单,交给前线当观察指引。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-ontology", "xy-fde"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第九章", "section": "阶段一 冷启动——本体从零到一", "related": [{"id": "XY-ONT-723", "rel": "refines"}]}
{"id": "FDE3P-236", "type": "method", "knowledge": "本体反馈单怎么用(虚构 ERP 案例):第一个客户出现本体没覆盖的差异——大额订单要总经理审批才能发货、成品仓和原料仓发货规则不同。前线先按客户需要定制,再填反馈单:现象、当前本体状态、修改建议、预估多少客户有同类需求。本体团队 1—2 天内评审三个问题:是否超过约三成潜在客户会遇到;纳入后维护成本是否可控;是否和已有实体冲突或重复。通过的发布新版本,附变更日志和使用指南。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第九章", "section": "阶段二 第一个客户——使用本体 + 发现缺口"}
{"id": "FDE3P-237", "type": "method", "knowledge": "客户担心“你们会不会把我们的东西拿给别人”时的边界和说法:本体层归平台,只含实体、关系、行为规则的抽象定义,不含任何客户真实数据;访问客户数据的具体 Skill 代码和配置归客户;平台基于本体重新开发的通用 Skill 归平台,但不带客户数据访问和客户特定上下文。对客户可以这样讲:订单有创建、确认、发货这类概念是行业公共语言,您的订单数据、客户名单、业务参数都留在您的系统里——就像会计准则人人可用,账本是您自己的。", "confidence": "high", "topics": ["合规与风控", "本体论与FDE方法论"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第九章", "section": "机制二:数据隐私隔离"}
{"id": "FDE3P-238", "type": "number", "knowledge": "报告虚构 ERP 案例里的估算数字(报告自注为方法论演示、数字为估算):从零定制每个客户约 38—45 人天;用本体后前线工时第 1 家 8 人天、第 4 家 4 人天、第 8 家 2 人天;上线周期从 6 周缩到 1 周;项目毛利率从约 25%(含本体投入分摊)升到 55%、再到 70%。冷启动投入约 150 万元(5 人 × 6 个月),到第 8 家累计省下约 300 万元交付成本,正式回本。", "confidence": "medium", "topics": ["本体论与FDE方法论", "团队与模式设计"], "skills": ["xy-fde", "xy-ontology"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第九章", "section": "阶段四 第五到第八个客户——规模效应显现", "related": [{"id": "FDE3P-204", "rel": "refines"}]}
{"id": "FDE3P-240", "type": "insight", "knowledge": "本体沉淀的难点不在技术,而在对业务领域的规则、风控和场景有没有吃透,再把这份认知写成结构化文档;工具只能降低表达成本,认知本身绕不过去。从代码或表结构逆向抽取本体时,AI 能快速出实体关系草稿,但前提是原系统设计得好,否则垃圾进垃圾出;抽得再好也要人确认,因为本体是业务的唯一标准说法。报告举例:某支付团队早年没有本体概念时,就用结构化的“系统用例”沉淀复杂业务规则,本质就是本体的早期形态。", "confidence": "high", "topics": ["本体论与FDE方法论"], "skills": ["xy-ontology", "xy-fde"], "source_type": "adapted_other", "source_label": "腾讯研究院《FDE模式行业观察与实践》", "chapter": "第九章", "section": "补充参考 本体沉淀的技术路径", "related": [{"id": "FDE3P-197", "rel": "supports"}]}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |