[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:zh":3,"public-menus:all":38,"post:source-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from:zh":205,"related:post:source-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from:zh:1":2781},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","zh","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,181,193],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-folder","custom",[],{"id":176,"title":177,"url":179,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":180},"item-28",{"de":178,"en":178,"es":178,"fr":178,"it":178,"ru":178,"sr":178,"zh":178},"Solr Suggester","\u002Fportfolio\u002Fsolr-fuzzy-suggester-und-solr-infix-suggester-abfrage-ueber-ajax-und-filterung",[],{"id":182,"title":183,"url":191,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":192},"item-27",{"de":184,"en":185,"es":186,"fr":187,"it":188,"ru":189,"sr":190,"zh":185},"Firmenwebseite SEO","Company Website SEO","Sitio web corporativo SEO","Site web d’entreprise SEO","Sito web aziendale SEO","Корпоративный сайт SEO","Пословна веб-страница SEO","\u002Fportfolio\u002Fseo-sem-branding-mobile-webseite-muenchen",[],{"id":194,"title":195,"url":203,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":204},"item-31",{"de":196,"en":197,"es":198,"fr":199,"it":200,"ru":201,"sr":202,"zh":197},"Digitalisierungsportal","Digitalization Portal","Portal de digitalización","Portail de numérisation","Portale di digitalizzazione","Портал цифровизации","Портал за дигитализацију","\u002Fportfolio\u002Fdigitalisierungsportal-archiv-museum-bibliothek-ead-lido-mets-mods",[],{"statusCode":4,"data":206,"message":2780},{"id":207,"title":208,"slug":209,"content":210,"contentJson":211,"excerpt":1296,"featuredImage":1297,"featuredImageAlt":1298,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":1299,"publishedAt":1300,"createdAt":1301,"updatedAt":1302,"seoLocalePaths":1303,"categories":1312,"author":1325,"translations":1330},"486","AI系统中的真相来源：可靠知识究竟从何而来","source-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u003Cp>在AI系统中，真相来源是权威来源，被允许定义特定事实、状态或规则在特定范围、版本和时间下是否应被视为真实。它不自动是语言模型、向量数据库、排名最高的检索文档、代理记忆或上下文中的最新消息。可靠的AI架构必须保留哪个来源对哪个声明具有权威性，然后保持来源、检索和验证与该权威性相连。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--info my-6 rounded-xl border p-5 border-blue-300 bg-blue-50 dark:border-blue-900 dark:bg-blue-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">直接回答\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">\u003Cstrong>真相来源是一种权威规则，而不是AI组件。\u003C\u002Fstrong> AI系统可以搜索许多来源，但只有部分来源对给定问题具有权威性。正确的来源可能是用于当前账户状态的应用程序数据库、用于规则的已批准政策文档、用于已实现行为的源代码、用于协议的官方规范，或用于观察到的声明的主要研究工件。良好的架构使该权威性明确，而不是要求模型从相关性中推断它。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">术语边界\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">“真相来源”在软件、数据和组织实践中被广泛使用，但它并不是一个具有单一通用正式定义的标准化AI组件。本文将其用作架构概念：\u003Cstrong>确定哪些证据被允许建立特定事实或状态的来源或权威映射\u003C\u002Fstrong>。诸如W3C PROV之类的来源标准描述了起源和派生；它们不会自动决定您的应用程序应将哪个来源视为权威。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">当前来源说明 — 2026年10月8日\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">权威性、来源、检索和模型上下文之间的核心区别是稳定的。NIST AI RMF 1.0目前正在修订中，但当前的NIST Playbook仍明确建议记录数据来源，包括来源、起源、转换、依赖关系、约束和元数据。W3C PROV仍然是W3C关于可互操作来源描述的推荐标准。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"目录\">\u003Cstrong class=\"editorjs-toc__title\">目录\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-6\" class=\"editorjs-toc__link\">“真相来源”的真正含义\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-11\" class=\"editorjs-toc__link\">最简单的例子\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-16\" class=\"editorjs-toc__link\">简单例子止步之处\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">权威性受主张、版本和时间范围的限定\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-22\" class=\"editorjs-toc__link\">真相来源不是什么\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-23\" class=\"editorjs-toc__link\">真相来源与记录系统\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-26\" class=\"editorjs-toc__link\">真相来源与溯源\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-29\" class=\"editorjs-toc__link\">真相来源与证据\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-32\" class=\"editorjs-toc__link\">真相来源与 RAG\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-35\" class=\"editorjs-toc__link\">真相来源与向量数据库\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">事实来源与记忆\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-41\" class=\"editorjs-toc__link\">事实来源与上下文\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-44\" class=\"editorjs-toc__link\">事实来源与评估基准真值\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-47\" class=\"editorjs-toc__link\">事实来源与数据质量\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-50\" class=\"editorjs-toc__link\">一种实用的事实来源架构模型\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-53\" class=\"editorjs-toc__link\">权威应显式声明，而非从相似性推断\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">检索应将权威性作为排序约束\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-60\" class=\"editorjs-toc__link\">新鲜度是权威性的一部分\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-63\" class=\"editorjs-toc__link\">派生值需要追溯到权威输入\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-67\" class=\"editorjs-toc__link\">当权威来源不一致时会发生什么？\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-71\" class=\"editorjs-toc__link\">网络搜索是发现，不自动是证据\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-74\" class=\"editorjs-toc__link\">语言模型不应自行决定权威性\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-77\" class=\"editorjs-toc__link\">权威性必须在执行追踪中得以保留\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-80\" class=\"editorjs-toc__link\">原始实现证据：真相来源研究引擎\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-87\" class=\"editorjs-toc__link\">Aaasaasa 文档与知识引擎：将相同的边界应用于企业文档\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-90\" class=\"editorjs-toc__link\">常见的真相来源失败模式\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-92\" class=\"editorjs-toc__link\">实用的真相来源决策框架\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-94\" class=\"editorjs-toc__link\">真相来源架构检查清单\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-96\" class=\"editorjs-toc__link\">常见误解\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-98\" class=\"editorjs-toc__link\">边缘情况\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-103\" class=\"editorjs-toc__link\">局限性\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-107\" class=\"editorjs-toc__link\">什么会改变这个答案？\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-110\" class=\"editorjs-toc__link\">相关规范知识\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-117\" class=\"editorjs-toc__link\">常见问题\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-119\" class=\"editorjs-toc__link\">术语表\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-121\" class=\"editorjs-toc__link\">结论\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-125\" class=\"editorjs-toc__link\">主要来源和实现证据\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Ch2 id=\"section-6\">“真相来源”的真正含义\u003C\u002Fh2>\n\u003Cp>这个短语常被误解为“包含一切的单一数据库”。在狭窄的系统中这可能是真的，但对于AI来说通常过于简单化。一个真实的AI应用程序可以结合操作数据库、文档、API、向量索引、用户输入、模型记忆、外部网络来源和生成的摘要。\u003C\u002Fp>\n\u003Cp>这些来源并不具有同等的权威性。客户支持手册可能定义政策，但不定义客户的当前余额。CRM可能定义当前账户所有者，但不定义法规的法律含义。源代码仓库可能定义已实现的行为，而产品规范定义预期的行为。因此，架构必须回答一个更精确的问题：哪个来源对这个特定声明具有权威性？\u003C\u002Fp>\n\u003Cp>这使得真相来源成为声明与权威之间的关系，而不仅仅是存储技术的属性。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--success my-6 rounded-xl border p-5 border-emerald-300 bg-emerald-50 dark:border-emerald-900 dark:bg-emerald-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">核心原则\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">\u003Cstrong>一个系统可以有许多真相来源，因为权威性是特定于事实的。\u003C\u002Fstrong> 重要的架构属性不是某个数据库在全球范围内胜出，而是每个重要的事实或状态都有一个明确的权威所有者。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-11\">最简单的例子\u003C\u002Fh2>\n\u003Cp>用户问AI助手：“我当前的订阅计划是什么？”助手有三个可能的输入：上个月的支持记录、描述计划类型的索引帮助中心文档，以及实时计费数据库。\u003C\u002Fp>\n\u003Cp>支持记录可能提到用户有Pro计划。帮助中心文档解释了Pro的含义。但实时计费记录是用户当前订阅状态的权威来源。\u003C\u002Fp>\n\u003Cp>语义搜索引擎可能将支持记录排在计费记录之上，因为它包含更接近问题的语言。这种排名仍然不会使记录具有权威性。相关性和权威性是不同的维度。\u003C\u002Fp>\n\u003Csection class=\"editorjs-comparison my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">同一个问题可能涉及不同的来源角色\u003C\u002Fh3>\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left dark:border-gray-700 dark:bg-gray-900\">\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">来源\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">角色\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">对当前计划的权威性？\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">计费数据库\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">帮助中心文档\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">旧支持记录\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">模型记忆\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-16\">简单例子止步之处\u003C\u002Fh2>\n\u003Cp>并非每个领域都有一个无可争议的权威。历史研究可能包含相互矛盾的主要来源。科学声明可能随着新研究的出现而演变。法律解释可能取决于司法管辖区、日期和法院权威。产品行为可能在文档和部署代码之间有所不同。\u003C\u002Fp>\n\u003Cp>在这些情况下，正确的架构不是发明一个单一的赢家。而是保留相互竞争的来源、它们的来源、它们的权威类别、它们的适用范围以及未解决的矛盾。一个可靠的真相来源系统必须能够表示不确定性和分歧。\u003C\u002Fp>\n\u003Ch2 id=\"section-19\">权威性受主张、版本和时间范围的限定\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">问题\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">可能的权威来源\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">为什么范围很重要\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">用户当前的账户余额是多少？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">账本\u002F会计记录系统\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">历史导出数据可能对较早时间点是准确的，但不代表当前状态。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">公司政策目前允许什么？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">已批准的当前政策版本\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">旧政策可能仍然是过去规则的有效证据，但不代表当前规则。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">实际部署的是哪个代码？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">部署产物\u002F提交\u002F发布记录\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">主分支可能与生产环境不同。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">合同签署时规定了什么？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">已签署的合同版本\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">草稿或后续模板对已签署的协议不具有权威性。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">技术协议规定了什么？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">相关版本的当前官方规范\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">博客解释可能有用，但属于次要证据。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">历史事件中发生了什么？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">相关的一手证据加上明确的史料批判\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">可能不存在单一的权威来源；相互矛盾的证据必须保持可见。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">用户偏好什么？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">当前明确的用户设置或已确认的偏好\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">旧的对话记忆可能已过时或被取代。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Cp>因此，“真相”一词如果不说明其边界，可能会产生误导。在架构中，真相来源通常更好地理解为：在既定条件下，被授权确定某一特定命题的来源。\u003C\u002Fp>\n\u003Ch2 id=\"section-22\">真相来源不是什么\u003C\u002Fh2>\n\u003Ch3 id=\"section-23\">真相来源与记录系统\u003C\u002Fh3>\n\u003Cp>记录系统通常是某一类记录的权威运营系统：例如，计费账本、人力资源主记录或订单数据库。它是真相来源权威性的一种常见实现方式。\u003C\u002Fp>\n\u003Cp>但真相来源的范围更广。一份已签署的 PDF 合同、一项官方标准、一个部署产物或一份原始档案文件，都可能具有权威性，而不必是一个事务性记录系统。\u003C\u002Fp>\n\u003Ch3 id=\"section-26\">真相来源与溯源\u003C\u002Fh3>\n\u003Cp>溯源回答的是这样的问题：这些数据来自哪里？是谁或什么产生了它？哪个转换创建了这个衍生结果？使用了哪个先前的实体？W3C PROV 对实体、活动、代理和衍生关系进行建模，从而可以表示来源和责任。\u003C\u002Fp>\n\u003Cp>溯源本身并不能确立权威性。知道某个值来自某位特定员工编写的电子表格，有助于评估它，但应用程序仍然需要一条规则来说明该电子表格对该主张是否具有权威性。\u003C\u002Fp>\n\u003Ch3 id=\"section-29\">真相来源与证据\u003C\u002Fh3>\n\u003Cp>证据支持或反驳某一主张。真相来源则定义了在当前应用上下文中，哪个来源有权裁定或强力约束该主张。\u003C\u002Fp>\n\u003Cp>一个来源可以是有价值的证据，但不具有权威性。五封客户电子邮件可能是用户不喜欢某个工作流的证据，但它们不是当前产品配置的记录系统。\u003C\u002Fp>\n\u003Ch3 id=\"section-32\">真相来源与 RAG\u003C\u002Fh3>\n\u003Cp>RAG 是一种检索模式。它查找信息并向模型提供选定的内容。RAG 不会自动知道哪个来源应当具有权威性。\u003C\u002Fp>\n\u003Cp>RAG 流水线可能检索到过时的文档、次要摘要或高度相似但不具有权威性的来源。来源权威性必须通过语料库设计、元数据、过滤器、排序策略、验证或检索后检查来编码。\u003C\u002Fp>\n\u003Ch3 id=\"section-35\">真相来源与向量数据库\u003C\u002Fh3>\n\u003Cp>向量数据库存储或索引用于语义检索的表示。它是一个访问层，并不自动是一个真相层。\u003C\u002Fp>\n\u003Cp>同一份权威文档可能被分块、嵌入、复制和重新索引多次。向量记录应保留对权威来源和版本的引用，而不是成为一个无法追溯的新权威。\u003C\u002Fp>\n\u003Ch3 id=\"section-38\">事实来源与记忆\u003C\u002Fh3>\n\u003Cp>智能体或应用程序的记忆存储以后可能有用的信息。记忆可以保留先前的决策、偏好或观察，但它可能变得过时。\u003C\u002Fp>\n\u003Cp>对于易变或后果重大的状态，可靠的智能体通常应重新读取权威的当前来源，而不是假设记忆中的状态仍然为真。\u003C\u002Fp>\n\u003Ch3 id=\"section-41\">事实来源与上下文\u003C\u002Fh3>\n\u003Cp>上下文是模型在当前推理过程中接收到的内容。权威信息可能不在上下文中，而非权威信息可能存在于上下文中。\u003C\u002Fp>\n\u003Cp>因此，上下文构建需要一种感知权威的策略：检索或读取被允许定义该主张的来源，然后保留足够的元数据，以便模型或验证器理解其适用范围。\u003C\u002Fp>\n\u003Ch3 id=\"section-44\">事实来源与评估基准真值\u003C\u002Fh3>\n\u003Cp>评估基准真值是用于对系统进行评分的参考答案、标签或结果。它可以来自权威来源、专家裁定或精心整理的测试数据。\u003C\u002Fp>\n\u003Cp>因此，基准真值是一种评估构造。事实来源是一种应用\u002F领域权威构造。它们可以重叠，但不可互换。\u003C\u002Fp>\n\u003Ch3 id=\"section-47\">事实来源与数据质量\u003C\u002Fh3>\n\u003Cp>权威来源仍可能包含错误。权威说明哪个来源正式管辖该事实；数据质量则询问该来源是否准确、完整、及时、一致且适合用途。\u003C\u002Fp>\n\u003Cp>当已知权威系统有误时，架构应记录缺陷、纠正过程或例外情况，而不是静默地用非官方来源替代并隐藏差异。\u003C\u002Fp>\n\u003Ch2 id=\"section-50\">一种实用的事实来源架构模型\u003C\u002Fh2>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">建议的架构模型\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">以下模型是针对 AI 系统的实用综合方案。它不是 W3C、NIST 或 ISO 标准。它将权威、来源、检索和面向模型的上下文分开，因为这些职责经常被错误地合并。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">感知权威的 AI 回答路径\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">1. 定义主张类型\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">确定用户实际在询问什么：当前状态、政策、历史事实、技术规范、用户偏好、计算或解释。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">2. 解析权威\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">确定在所需范围、版本和时间下，哪个来源或权威类别被允许定义该类型的主张。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">3. 获取证据\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">读取或检索权威来源以及任何必要的支持性或冲突性证据。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">4. 保留来源\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">携带来源标识、版本、时间戳、定位符、转换历史和责任元数据。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">5. 构建模型上下文\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">向模型提供相关证据，同时不丢弃权威性和适用性元数据。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">6\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">6. 生成或计算\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">模型可以总结、比较、推理或转换证据，但不会仅仅因为处理了证据就继承来源的权威性。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">7\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">7. 验证主张\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">检查答案是否由正确的来源支持，并且仍处于其范围和有效边界之内。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">8\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">8. 保留矛盾\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">如果权威或相关来源不一致，应暴露冲突，而不是捏造虚假的确定性。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-53\">权威应显式声明，而非从相似性推断\u003C\u002Fh2>\n\u003Cp>一种稳健的实现模式是权威注册表或等效的策略层，将主张类别映射到权威来源类别。实现可以是代码、元数据、配置或领域规则；重要的特性是权威是刻意确定的。\u003C\u002Fp>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">声明类别\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">权威规则\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">回退行为\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">当前账户状态\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">读取实时账户服务\u002F记录系统\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">如果不可用，则报告无法验证当前状态。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">产品文档\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">当前批准的文档版本\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">仅可在带有版本警告的情况下显示旧版本。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">已实现的软件行为\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">相关已部署版本\u002F源工件\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">仅凭文档无法证明已部署的行为。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">内部政策\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">已批准的政策存储库和有效版本\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">草稿是支持材料，不是当前权威。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">外部技术标准\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">相关版本的官方标准机构出版物\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">二手解释可以澄清但不能覆盖规范。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">研究声明\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">适合该领域的证据政策\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">保留相互冲突的证据和置信度，而不是强行采用单一来源。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-56\">检索应将权威性作为排序约束\u003C\u002Fh2>\n\u003Cp>语义相关性回答的是“哪个候选看起来与这个查询相关？”权威性回答的是“哪个候选被允许确立这个事实？”生产检索系统通常两者都需要。\u003C\u002Fp>\n\u003Cp>一个有用的顺序是：先通过身份、租户、来源类别、状态、版本或日期约束候选空间，然后在允许的空间内对相关证据进行排序。如果在关键授权或权威过滤之前计算相关性，管道可能会返回一个看似有说服力但无效的结果。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--warning my-6 rounded-xl border p-5 border-amber-300 bg-amber-50 dark:border-amber-900 dark:bg-amber-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">最高相似度不等于最高权威\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">搜索引擎可能将过时的摘要排在当前主要来源之上。语言模型可能偏好写得更好的解释，而不是官方记录。这两种行为都不会改变哪个来源拥有该事实。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-60\">新鲜度是权威性的一部分\u003C\u002Fh2>\n\u003Cp>许多真相来源的失败实际上是时间失败。正确的来源是已知的，但系统使用了旧快照、过时嵌入、缓存 API 响应或被取代的文档。\u003C\u002Fp>\n\u003Cp>因此，在事实可能发生变化的地方，权威规则应包含失效或刷新语义。当应用程序读取一周前的复制导出时，“CRM 是权威的”是不完整的。\u003C\u002Fp>\n\u003Ch2 id=\"section-63\">派生值需要追溯到权威输入\u003C\u002Fh2>\n\u003Cp>一些重要事实并非直接存储。它们是从权威输入计算得出的：风险评分、账户总额、资格状态或聚合指标。\u003C\u002Fp>\n\u003Cp>对于派生值，真相来源架构应保留输入权威、转换或计算版本以及执行时间。W3C PROV 对实体、活动和派生之间的区分在这里很有用，因为它建模了一个实体是如何从其他实体产生的。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--success my-6 rounded-xl border p-5 border-emerald-300 bg-emerald-50 dark:border-emerald-900 dark:bg-emerald-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">派生真相是有条件的\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">计算值的当前性和权威性仅取决于其输入、转换和有效性条件。存储重新计算或审计它所需的血统。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-67\">当权威来源不一致时会发生什么？\u003C\u002Fh2>\n\u003Cp>在严肃的知识系统中，冲突不是边缘情况。签署的合同可能与 CRM 字段不一致。生产行为可能与文档不一致。两个主要历史来源可能相互矛盾。当前政策可能与过时的本地副本冲突。\u003C\u002Fp>\n\u003Cp>系统需要适合该领域的解决政策。有时一个权威明显高于另一个。有时新版本取代旧版本。有时专家或业务所有者必须裁决。而有时正确的结果就是：证据尚未解决。\u003C\u002Fp>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">冲突类型\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">典型处理\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">当前版本与被取代版本\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">对当前状态使用当前版本；保留旧版本作为历史证据。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">记录系统与过时副本\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">使用记录系统；标记复制新鲜度问题。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">合同与 CRM 转录\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">已签署合同管辖合同措辞；CRM 差异成为更正任务。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">文档与已部署行为\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">区分预期行为与观察\u002F已部署行为；不要静默合并它们。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">两个可信的主要来源\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">保留两者，评估来源和范围，如果不存在管辖权威，则表示未解决的分歧。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">用户记忆与当前用户设置\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">使用当前显式设置；在适当情况下将记忆标记为被取代。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-71\">网络搜索是发现，不自动是证据\u003C\u002Fh2>\n\u003Cp>搜索引擎是出色的发现系统。搜索片段、结果排名和生成的摘要不自动是主要证据。\u003C\u002Fp>\n\u003Cp>对于需要权威性的主张，搜索结果应指向原始出版物、官方记录、源文件、数据集或其他适当的工件。结果页面帮助定位来源；它并不继承来源的权威性。\u003C\u002Fp>\n\u003Ch2 id=\"section-74\">语言模型不应自行决定权威性\u003C\u002Fh2>\n\u003Cp>模型可以帮助分类问题、提取主张或比较证据，但权威性不应仅取决于模型的偏好。模型优化的是基于上下文的生成；它们并不拥有一个保证的领域特定注册表，来说明每个事实归属于哪个数据库、文档或组织。\u003C\u002Fp>\n\u003Cp>这就是为什么应用架构应在实际可行的情况下以确定性方式编码关键权威规则。模型可以在边界内推理，但边界本身不应为每个提示从头重新创建。\u003C\u002Fp>\n\u003Ch2 id=\"section-77\">权威性必须在执行追踪中得以保留\u003C\u002Fh2>\n\u003Cp>如果生产答案重要到需要审计，追踪应使得能够重建以下内容：咨询了哪些来源、使用了哪个版本、哪个段落或记录支持了该主张、发生了哪些转换，以及是否存在相互矛盾的证据。\u003C\u002Fp>\n\u003Cp>这与 W3C PROV 中更广泛的溯源原则以及 NIST AI RMF Playbook 中关于记录来源、起源、转换、依赖关系、约束和元数据的指导相一致。\u003C\u002Fp>\n\u003Ch2 id=\"section-80\">原始实现证据：真相来源研究引擎\u003C\u002Fh2>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">原始实现证据\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">真相来源研究引擎是我自己对证据优先架构的实现证据。它展示了一种具体方式，将发现、来源获取、溯源、主张、证据类别、矛盾和结论分离开来。它是一种实现模式，而非通用的行业标准。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Cp>该引擎围绕可追踪的流水线设计，而非直接进行 AI 摘要：研究任务 → 搜索 → 原始来源或数字工件 → 本地快照 → SHA-256 → 来源 ID → 主张 → 证据类别 → 关系或矛盾 → 解释 → 结论。\u003C\u002Fp>\n\u003Cp>其共享证据核心存储来源、工件、溯源、主张、关系、矛盾、参考模型和审计追踪。不同的研究模式可以共享该核心，同时应用不同的领域方法论。\u003C\u002Fp>\n\u003Cp>该架构有意将发现与证据分离。搜索片段不被视为证据，文件名不被视为内容，AI 摘要不被视为主要来源，语义相似性仅是发现信号，直到结果被追溯到具体来源和定位符。\u003C\u002Fp>\n\u003Cp>原始文件被保留，本地字节获得 SHA-256 标识符。矛盾和被拒绝的假设不会被静默删除。新证据被允许改变当前参考模型，同时先前的证据路径仍可审计。\u003C\u002Fp>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">已实现的规则\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">为何对真相来源架构重要\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">搜索 ≠ 证据\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">发现排名不能静默地变成权威性。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">文件名 ≠ 内容\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">元数据线索不能替代阅读实际工件。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">本地快照 + SHA-256\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">证据可以绑定到确切的字节，而非可变的远程标签。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">来源 ID + 精确定位符\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">主张可以追溯到具体的证据位置。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">主张\u002F证据分离\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">断言不会与支持它的材料混淆。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">矛盾被保留\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">系统可以表示未解决的分歧，而非覆盖历史。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">语义相似性仅用于发现\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">检索相关性被明确地与证据权威性分离。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">新证据可以更新模型\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">真相来源状态是版本化的且可修订的，而非被视为不可变的教条。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-87\">Aaasaasa 文档与知识引擎：将相同的边界应用于企业文档\u003C\u002Fh2>\n\u003Cp>Aaasaasa 文档与知识引擎概念将相同的设计原则扩展到企业文档：用户应能够搜索文档、提出有来源依据的问题，并根据明确的标准审查集合，同时保留文档所述内容与系统推断内容之间的区别。\u003C\u002Fp>\n\u003Cp>重要的架构规则是，通用检索核心并不会使每个集合都具有同等权威性。合同文档、维护记录、财务文档和研究材料需要不同的权威性、验证和覆盖规则，即使它们共享摄取和搜索基础设施。\u003C\u002Fp>\n\u003Ch2 id=\"section-90\">常见的真相来源失败模式\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">失败模式\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">出了什么问题\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">将模型视为事实来源\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">参数化知识可能过时、不完整、无法验证，或超出应用的权威范围。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">检索排名最高的结果自动胜出\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">将相似度误认为权威性。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">向量数据库成为权威来源\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">派生索引记录丢失了原始来源的身份和版本。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">所有内容都复制到一个知识库中\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">副本模糊了所有权、新鲜度和更正路径。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">将记忆当作当前状态复用\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">旧观察结果悄然覆盖当前的记录系统。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">没有版本元数据\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">正确的文档被用于错误的时间段。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">没有来源定位符\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">引文存在，但支撑段落或记录无法验证。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">冲突被覆盖\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">系统通过销毁分歧证据来显得一致。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">生成的摘要取代原始内容\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">有损转换成为表面上的权威。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">权威性是全局的，而非针对具体主张\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">某个来源被信任的范围超出了它实际拥有的领域或事实类别。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">将网页摘要视为证据\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">发现元数据取代了原始出版物。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">权威数据有误，但异常被隐藏\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">运营缺陷变得不可见，无法透明地更正。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-92\">实用的真相来源决策框架\u003C\u002Fh2>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">如何决定什么应该定义一项主张\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">1. 精确陈述主张\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">区分当前状态、历史状态、政策、解释、预测和派生计算。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">2. 确定权威所有者\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">确定负责该主张类型的系统、文档、机构、人员或证据类别。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">3. 定义范围\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">指定租户、司法管辖区、产品、环境、用户、文档集或其他适用边界。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">4. 定义时间和版本\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">确定该主张需要当前状态、历史快照，还是特定的标准\u002F发布版本。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">5. 保留来源\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">记录来源身份、出处、定位符、转换过程以及负责的代理或流程。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">6\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">6. 定义检索\u002F访问路径\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">确保应用确实能够在正确的身份和权限下获取权威信息。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">7\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">7. 定义冲突策略\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">决定优先级、取代、裁决或明确未解决状态的行为。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">8\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">8. 定义失效机制\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">指定缓存、索引、记忆或派生表示何时必须刷新。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">9\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">9. 验证答案路径\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">验证重要的生成主张能够追溯到预期的权威来源，而不仅仅是看似合理的来源。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-94\">真相来源架构检查清单\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">问题\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">预期答案\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">正在确立的确切事实或状态是什么？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">主张足够精确，可以分配权威性。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">谁或什么拥有该事实？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">具名的权威系统、来源类别或裁决规则。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">权威来源在此范围内是最新的吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">租户、司法管辖区、环境、用户或领域边界是明确的。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">版本\u002F时间正确吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">当前、历史或特定版本的适用性是已知的。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">来源可以验证吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">存在稳定的标识符、定位符或记录引用。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">来源被保留了吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">出处、转换和责任元数据在摄取和检索过程中得以保留。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">检索可能返回非权威材料吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">如果可能，过滤器或验证会区分相关性与权威性。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">来源可能变化吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">存在刷新、失效或取代规则。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">来源可能不一致吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">冲突和裁决行为是明确的。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">记忆可能过时吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">在产生后果的使用之前，从当前权威来源重新读取易变状态。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">派生答案可以复现吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">输入、转换版本和执行条件可追溯。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">审计员可以重建答案吗？\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">执行证据为重要主张保留了来源路径。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-96\">常见误解\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">误解\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">更正\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“真相来源意味着一个数据库。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">一个数据库可以是一个领域的权威来源；复杂系统通常有多个针对具体事实的权威来源。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“最新的文档自动具有权威性。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">只有在较新的制品获得批准并实际取代较旧制品时，时效性才有帮助。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“RAG 解决了真相问题。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">RAG 解决的是检索问题。权威性、出处、证据质量和有效性仍然是独立的问题。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“引文证明了答案。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">被引用的来源必须实际支持该主张，具有正确的权威性，并适用于当前范围。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“出处告诉我们什么是真的。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">出处告诉我们来源和派生过程；权威性和正确性仍然需要领域规则和评估。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“记录系统总是正确的。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">它对运营记录具有权威性，但数据质量缺陷仍然可能存在，需要可见的更正。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“如果多个来源一致，该主张就是权威的。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">一致性增加了证据，但不一定确立所有权或适用性。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“AI 记忆可以取代重复读取。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">仅适用于过时风险可接受的信息；易变或会产生后果的状态应从权威来源刷新。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-98\">边缘情况\u003C\u002Fh2>\n\u003Cp>有些问题是解释性的，而非事实性的。“哪种架构最好？”没有单一的真相来源。系统可以检索权威约束和证据，但最终判断是一种推断，应当暴露假设和权衡。\u003C\u002Fp>\n\u003Cp>有些领域使用分布式权威。科学结论可能依赖于多项研究、数据集和重复实验。历史结论可能依赖于相互冲突的一手和二手证据。架构应当表示证据结构，而不是发明一个据称拥有真相的中央数据库。\u003C\u002Fp>\n\u003Cp>用户也可以是主观个人信息的权威：偏好、目标、选定设置或明确指令。即便如此，较新的明确输入也可以取代较旧的记忆。\u003C\u002Fp>\n\u003Cp>外部事件可能使先前具有权威性的数据失效。价格源、库存系统或安全策略在捕获时可能是正确的，但不再有效。快照出处保留了当时为真的内容；它不会使快照永远保持最新。\u003C\u002Fp>\n\u003Ch2 id=\"section-103\">局限性\u003C\u002Fh2>\n\u003Cp>真相来源架构无法保证权威来源在事实上是正确的。它提供了问责、出处和确定性的所有权边界；数据质量和领域验证流程仍然是必要的。\u003C\u002Fp>\n\u003Cp>权威性也可能存在争议。不同的机构可以在不同的司法管辖区或方法论中合法地主张权威。在这些情况下，系统应当暴露权威模型和分歧，而不是将其隐藏在通用的“真相分数”背后。\u003C\u002Fp>\n\u003Cp>最后，权威规则需要维护。系统、所有者、政策、版本和法规都会变化。过时的权威注册表可能比完全没有注册表更危险。\u003C\u002Fp>\n\u003Ch2 id=\"section-107\">什么会改变这个答案？\u003C\u002Fh2>\n\u003Cp>具体的权威映射随领域而变化。银行、医疗、软件交付、科学研究和历史分析有不同的记录系统、证据规则和监管义务。\u003C\u002Fp>\n\u003Cp>实现方式也会随架构而变化。小型应用可以直接在服务调用中编码权威性。较大的平台可能需要注册表、源元数据、策略引擎、血缘系统或数据契约。核心原则保持不变：不要让检索顺序或模型偏好悄然决定什么算作权威。\u003C\u002Fp>\n\u003Ch2 id=\"section-110\">相关规范知识\u003C\u002Fh2>\n\u003Cp>真相源架构是后续检索和治理概念的前提，因为仅靠检索质量无法确定证据是否被允许定义答案。\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works\" class=\"flex flex-col sm:flex-row gap-4 rounded-xl border border-gray-200 dark:border-gray-700 p-4 transition hover:border-primary-500\">\u003Cdiv class=\"min-w-0 flex-1\">\u003Cstrong class=\"block text-lg text-gray-900 dark:text-gray-100\">什么是RAG？对其工作原理的最简单解释\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">RAG为模型检索外部知识。这一基础解释了为什么检索和权威真相是相互独立的职责。\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">阅读RAG基础 →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Cp>记忆是另一个相邻概念。可靠的智能体将记住的信息与当前权威的应用状态分开。\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fai-agent-memory-is-not-rag-how-to-separate-memory-retrieval-state-and-context\" class=\"flex flex-col sm:flex-row gap-4 rounded-xl border border-gray-200 dark:border-gray-700 p-4 transition hover:border-primary-500\">\u003Cdiv class=\"min-w-0 flex-1\">\u003Cstrong class=\"block text-lg text-gray-900 dark:text-gray-100\">AI智能体记忆不是RAG：如何分离记忆、检索、状态和上下文\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">一种四层架构，分离持久记忆、权威状态、检索以及实际提供给模型的上下文。\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">阅读记忆架构文章 →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Cp>权威性也直接关系到答案有效性。即使是权威来源，也只支持其版本、日期、范围和证据边界内的主张。\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers\" class=\"flex flex-col sm:flex-row gap-4 rounded-xl border border-gray-200 dark:border-gray-700 p-4 transition hover:border-primary-500\">\u003Cdiv class=\"min-w-0 flex-1\">\u003Cstrong class=\"block text-lg text-gray-900 dark:text-gray-100\">答案有效性边界：相关性与可靠AI答案之间缺失的层\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">一个框架，用于明确AI答案在何处仍受证据支持，以及哪些条件变化会使该支持失效。\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">阅读答案有效性边界 →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Ch2 id=\"section-117\">常见问题\u003C\u002Fh2>\n\u003Csection class=\"editorjs-faq my-6 rounded-xl border border-gray-200 p-5 dark:border-gray-700\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">AI系统中的真相源\u003C\u002Fh3>\u003Cdiv id=\"faq1\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">AI系统中的真相源是什么？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">它是权威来源或权威规则，决定在定义的范围、版本和时间下，哪个来源被允许确立特定事实、状态或规则。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq2\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">语言模型是真相源吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">通常不是。语言模型可以生成、总结和推理，但其参数化知识对于当前应用状态、公司政策、特定文档版本或受监管领域的事实并不自动具有权威性。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq3\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">向量数据库是RAG的真相源吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">不自动是。向量数据库通常是索引或检索存储。它应保留对权威原始来源和版本的引用，而不是悄然替换它们。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq4\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">来源和真相源有什么区别？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">来源描述数据来自哪里、如何产生或转换以及涉及谁或什么。真相源规则决定该来源是否对特定主张具有权威性。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq5\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">AI系统可以有多个真相源吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">可以。在复杂系统中这是正常的，因为不同事实属于不同的权威系统或来源类别。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq6\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">当两个权威来源不一致时会发生什么？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">系统需要特定领域的冲突规则：优先级、版本取代、专家裁决或明确未解决状态。它不应悄然选择模型偏好的来源。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq7\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">RAG能保证AI答案使用真相源吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">不能。RAG检索候选。需要权威感知元数据、过滤器、来源策略和验证，以确保重要主张使用正确的来源。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq8\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">真相源可能是错的吗？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">可能。权威性和正确性是不同的属性。权威系统可能包含数据质量缺陷，应以透明方式纠正，而不是通过替换非官方来源来隐藏。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-119\">术语表\u003C\u002Fh2>\n\u003Csection class=\"editorjs-glossary my-6 rounded-xl border border-gray-200 dark:border-gray-700 p-5\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">关键真相源术语\u003C\u002Fh3>\u003Cdl>\u003Cdiv id=\"source-of-truth\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">真相源\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">被允许在定义的范围、版本和时间下确立特定事实、状态或规则的权威来源或规则。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"system-of-record\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">记录系统\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">负责定义类别记录或当前业务状态的权威运营系统。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"provenance\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">来源\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">描述数据或其他实体的起源、派生、转换、负责主体和历史的信息。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"evidence\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">证据\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">支持、反驳或约束某一主张的信息或人工制品。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"authority\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">权威性\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">决定哪个来源有权定义特定主张的应用或领域规则。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"freshness\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">新鲜度\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">某一表示对于其被使用的主张或操作是否仍足够当前。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"supersession\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">取代\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">较新的权威版本对较旧版本的明确替换，同时保留历史可追溯性。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"ground-truth\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">基准真相\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">用于评估系统的参考答案、标签或结果；它是评估构造，并不自动是应用的真相源。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"lineage\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">血缘\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">数据或派生值在来源和处理步骤之间流动和转换的轨迹。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"validity-boundary\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">有效性边界\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">某一主张仍受支持的范围、时间、版本、证据和假设条件。\u003C\u002Fdd>\u003C\u002Fdiv>\u003C\u002Fdl>\u003C\u002Fsection>\n\u003Ch2 id=\"section-121\">结论\u003C\u002Fh2>\n\u003Cp>可靠的AI并非来自给模型更多信息。它来自知道哪些信息被允许定义主张、保留该信息的来源、检索正确版本，并将最终答案保持在来源的范围内。\u003C\u002Fp>\n\u003Cp>这就是为什么真相源、来源、检索、记忆和上下文必须保持为独立概念。真相源定义权威性。来源解释起源。检索找到候选。记忆保留选定的过去信息。上下文是模型看到的内容。生成将这些输入转化为输出。\u003C\u002Fp>\n\u003Cp>当这些层保持明确时，AI系统不仅能听起来合理：重要主张可以追溯到实际有权确立它们的来源。\u003C\u002Fp>\n\u003Ch2 id=\"section-125\">主要来源和实现证据\u003C\u002Fh2>\n\u003Cp>以下外部来源支持来源和AI风险管理的说法。真相源研究引擎部分是原创实现证据，并明确呈现为一种实现模式，而非通用标准。\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-dm\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">W3C PROV-DM — PROV 数据模型\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">W3C 推荐标准，定义了一个与领域无关的溯源模型，围绕实体、活动、代理、派生和责任。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002Fgroups\u002Fwg\u002Fprov\u002Fpublications\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">W3C 溯源工作组 — 出版物\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">W3C PROV 推荐标准及相关规范的官方索引，用于溯源交换和约束。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.nist.gov\u002Fitl\u002Fai-risk-management-framework\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">NIST AI 风险管理框架\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">NIST 的自愿性框架，用于在 AI 生命周期中纳入可信度和风险管理考量；AI RMF 1.0 目前正在修订中。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fairc.nist.gov\u002Fairmf-resources\u002Fplaybook\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">NIST AI RMF 操作手册\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">与 AI RMF 一致的操作指南，包括数据溯源、来源、起源、转换、依赖关系、约束和元数据的文档实践。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fairc.nist.gov\u002Fairmf-resources\u002Fplaybook\u002Fmeasure\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">NIST AI RMF 操作手册 — 测量\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">关于记录测量、数据溯源以及 AI 系统输出上下文解释的指南。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.nist.gov\u002Fpublications\u002Fartificial-intelligence-risk-management-framework-generative-artificial-intelligence\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">NIST AI 600-1 — 生成式 AI 概况\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">NIST 生成式 AI 概况，包括生成式 AI 系统的溯源和信息完整性考量。\u003C\u002Fp>\u003C\u002Fa>",{"time":212,"blocks":213,"version":1295},1791479501696,[214,220,228,235,241,249,254,259,264,269,276,281,286,291,296,331,336,341,346,351,388,393,398,403,408,413,418,423,428,433,438,443,448,453,458,463,468,473,478,483,488,493,498,503,508,513,518,523,528,533,538,544,576,581,586,619,624,629,634,641,646,651,656,661,666,671,677,682,687,692,718,723,728,733,738,743,748,753,758,763,768,774,779,784,789,794,826,831,836,841,846,890,895,928,933,976,981,1013,1018,1023,1028,1033,1038,1043,1048,1053,1058,1063,1068,1073,1078,1083,1092,1097,1105,1110,1118,1123,1161,1166,1210,1215,1220,1225,1230,1235,1240,1250,1259,1268,1277,1286],{"id":215,"data":216,"type":218,"tunes":219},"intro",{"text":217},"在AI系统中，真相来源是权威来源，被允许定义特定事实、状态或规则在特定范围、版本和时间下是否应被视为真实。它不自动是语言模型、向量数据库、排名最高的检索文档、代理记忆或上下文中的最新消息。可靠的AI架构必须保留哪个来源对哪个声明具有权威性，然后保持来源、检索和验证与该权威性相连。","paragraph",{},{"id":221,"data":222,"type":226,"tunes":227},"direct",{"body":223,"title":224,"variant":225},"\u003Cstrong>真相来源是一种权威规则，而不是AI组件。\u003C\u002Fstrong> AI系统可以搜索许多来源，但只有部分来源对给定问题具有权威性。正确的来源可能是用于当前账户状态的应用程序数据库、用于规则的已批准政策文档、用于已实现行为的源代码、用于协议的官方规范，或用于观察到的声明的主要研究工件。良好的架构使该权威性明确，而不是要求模型从相关性中推断它。","直接回答","info","callout",{},{"id":229,"data":230,"type":226,"tunes":234},"terminology",{"body":231,"title":232,"variant":233},"“真相来源”在软件、数据和组织实践中被广泛使用，但它并不是一个具有单一通用正式定义的标准化AI组件。本文将其用作架构概念：\u003Cstrong>确定哪些证据被允许建立特定事实或状态的来源或权威映射\u003C\u002Fstrong>。诸如W3C PROV之类的来源标准描述了起源和派生；它们不会自动决定您的应用程序应将哪个来源视为权威。","术语边界","note",{},{"id":236,"data":237,"type":226,"tunes":240},"current",{"body":238,"title":239,"variant":233},"权威性、来源、检索和模型上下文之间的核心区别是稳定的。NIST AI RMF 1.0目前正在修订中，但当前的NIST Playbook仍明确建议记录数据来源，包括来源、起源、转换、依赖关系、约束和元数据。W3C PROV仍然是W3C关于可互操作来源描述的推荐标准。","当前来源说明 — 2026年10月8日",{},{"id":242,"data":243,"type":247,"tunes":248},"toc",{"title":244,"maxLevel":245,"minLevel":246},"目录",3,2,"tableOfContents",{},{"id":250,"data":251,"type":42,"tunes":253},"h-meaning",{"text":252,"level":246},"“真相来源”的真正含义",{},{"id":255,"data":256,"type":218,"tunes":258},"p-meaning-1",{"text":257},"这个短语常被误解为“包含一切的单一数据库”。在狭窄的系统中这可能是真的，但对于AI来说通常过于简单化。一个真实的AI应用程序可以结合操作数据库、文档、API、向量索引、用户输入、模型记忆、外部网络来源和生成的摘要。",{},{"id":260,"data":261,"type":218,"tunes":263},"p-meaning-2",{"text":262},"这些来源并不具有同等的权威性。客户支持手册可能定义政策，但不定义客户的当前余额。CRM可能定义当前账户所有者，但不定义法规的法律含义。源代码仓库可能定义已实现的行为，而产品规范定义预期的行为。因此，架构必须回答一个更精确的问题：哪个来源对这个特定声明具有权威性？",{},{"id":265,"data":266,"type":218,"tunes":268},"p-meaning-3",{"text":267},"这使得真相来源成为声明与权威之间的关系，而不仅仅是存储技术的属性。",{},{"id":270,"data":271,"type":226,"tunes":275},"principle",{"body":272,"title":273,"variant":274},"\u003Cstrong>一个系统可以有许多真相来源，因为权威性是特定于事实的。\u003C\u002Fstrong> 重要的架构属性不是某个数据库在全球范围内胜出，而是每个重要的事实或状态都有一个明确的权威所有者。","核心原则","success",{},{"id":277,"data":278,"type":42,"tunes":280},"h-simple",{"text":279,"level":246},"最简单的例子",{},{"id":282,"data":283,"type":218,"tunes":285},"p-simple-1",{"text":284},"用户问AI助手：“我当前的订阅计划是什么？”助手有三个可能的输入：上个月的支持记录、描述计划类型的索引帮助中心文档，以及实时计费数据库。",{},{"id":287,"data":288,"type":218,"tunes":290},"p-simple-2",{"text":289},"支持记录可能提到用户有Pro计划。帮助中心文档解释了Pro的含义。但实时计费记录是用户当前订阅状态的权威来源。",{},{"id":292,"data":293,"type":218,"tunes":295},"p-simple-3",{"text":294},"语义搜索引擎可能将支持记录排在计费记录之上，因为它包含更接近问题的语言。这种排名仍然不会使记录具有权威性。相关性和权威性是不同的维度。",{},{"id":297,"data":298,"type":329,"tunes":330},"simple-comparison",{"rows":299,"title":317,"layout":318,"columns":319},[300,305,309,313],{"id":301,"label":302,"values":303},"billing","计费数据库",[304,304,304],"",{"id":306,"label":307,"values":308},"help","帮助中心文档",[304,304,304],{"id":310,"label":311,"values":312},"transcript","旧支持记录",[304,304,304],{"id":314,"label":315,"values":316},"model","模型记忆",[304,304,304],"同一个问题可能涉及不同的来源角色","table",[320,323,326],{"id":321,"label":322},"source","来源",{"id":324,"label":325},"role","角色",{"id":327,"label":328},"authority","对当前计划的权威性？","comparison",{},{"id":332,"data":333,"type":42,"tunes":335},"h-stops",{"text":334,"level":246},"简单例子止步之处",{},{"id":337,"data":338,"type":218,"tunes":340},"p-stops-1",{"text":339},"并非每个领域都有一个无可争议的权威。历史研究可能包含相互矛盾的主要来源。科学声明可能随着新研究的出现而演变。法律解释可能取决于司法管辖区、日期和法院权威。产品行为可能在文档和部署代码之间有所不同。",{},{"id":342,"data":343,"type":218,"tunes":345},"p-stops-2",{"text":344},"在这些情况下，正确的架构不是发明一个单一的赢家。而是保留相互竞争的来源、它们的来源、它们的权威类别、它们的适用范围以及未解决的矛盾。一个可靠的真相来源系统必须能够表示不确定性和分歧。",{},{"id":347,"data":348,"type":42,"tunes":350},"h-authority",{"text":349,"level":246},"权威性受主张、版本和时间范围的限定",{},{"id":352,"data":353,"type":318,"tunes":387},"authority-table",{"content":354,"stretched":43,"withHeadings":14},[355,359,363,367,371,375,379,383],[356,357,358],"问题","可能的权威来源","为什么范围很重要",[360,361,362],"用户当前的账户余额是多少？","账本\u002F会计记录系统","历史导出数据可能对较早时间点是准确的，但不代表当前状态。",[364,365,366],"公司政策目前允许什么？","已批准的当前政策版本","旧政策可能仍然是过去规则的有效证据，但不代表当前规则。",[368,369,370],"实际部署的是哪个代码？","部署产物\u002F提交\u002F发布记录","主分支可能与生产环境不同。",[372,373,374],"合同签署时规定了什么？","已签署的合同版本","草稿或后续模板对已签署的协议不具有权威性。",[376,377,378],"技术协议规定了什么？","相关版本的当前官方规范","博客解释可能有用，但属于次要证据。",[380,381,382],"历史事件中发生了什么？","相关的一手证据加上明确的史料批判","可能不存在单一的权威来源；相互矛盾的证据必须保持可见。",[384,385,386],"用户偏好什么？","当前明确的用户设置或已确认的偏好","旧的对话记忆可能已过时或被取代。",{},{"id":389,"data":390,"type":218,"tunes":392},"p-authority-1",{"text":391},"因此，“真相”一词如果不说明其边界，可能会产生误导。在架构中，真相来源通常更好地理解为：在既定条件下，被授权确定某一特定命题的来源。",{},{"id":394,"data":395,"type":42,"tunes":397},"h-what-is-not",{"text":396,"level":246},"真相来源不是什么",{},{"id":399,"data":400,"type":42,"tunes":402},"h-vs-sor",{"text":401,"level":245},"真相来源与记录系统",{},{"id":404,"data":405,"type":218,"tunes":407},"p-vs-sor-1",{"text":406},"记录系统通常是某一类记录的权威运营系统：例如，计费账本、人力资源主记录或订单数据库。它是真相来源权威性的一种常见实现方式。",{},{"id":409,"data":410,"type":218,"tunes":412},"p-vs-sor-2",{"text":411},"但真相来源的范围更广。一份已签署的 PDF 合同、一项官方标准、一个部署产物或一份原始档案文件，都可能具有权威性，而不必是一个事务性记录系统。",{},{"id":414,"data":415,"type":42,"tunes":417},"h-vs-prov",{"text":416,"level":245},"真相来源与溯源",{},{"id":419,"data":420,"type":218,"tunes":422},"p-vs-prov-1",{"text":421},"溯源回答的是这样的问题：这些数据来自哪里？是谁或什么产生了它？哪个转换创建了这个衍生结果？使用了哪个先前的实体？W3C PROV 对实体、活动、代理和衍生关系进行建模，从而可以表示来源和责任。",{},{"id":424,"data":425,"type":218,"tunes":427},"p-vs-prov-2",{"text":426},"溯源本身并不能确立权威性。知道某个值来自某位特定员工编写的电子表格，有助于评估它，但应用程序仍然需要一条规则来说明该电子表格对该主张是否具有权威性。",{},{"id":429,"data":430,"type":42,"tunes":432},"h-vs-evidence",{"text":431,"level":245},"真相来源与证据",{},{"id":434,"data":435,"type":218,"tunes":437},"p-vs-evidence-1",{"text":436},"证据支持或反驳某一主张。真相来源则定义了在当前应用上下文中，哪个来源有权裁定或强力约束该主张。",{},{"id":439,"data":440,"type":218,"tunes":442},"p-vs-evidence-2",{"text":441},"一个来源可以是有价值的证据，但不具有权威性。五封客户电子邮件可能是用户不喜欢某个工作流的证据，但它们不是当前产品配置的记录系统。",{},{"id":444,"data":445,"type":42,"tunes":447},"h-vs-rag",{"text":446,"level":245},"真相来源与 RAG",{},{"id":449,"data":450,"type":218,"tunes":452},"p-vs-rag-1",{"text":451},"RAG 是一种检索模式。它查找信息并向模型提供选定的内容。RAG 不会自动知道哪个来源应当具有权威性。",{},{"id":454,"data":455,"type":218,"tunes":457},"p-vs-rag-2",{"text":456},"RAG 流水线可能检索到过时的文档、次要摘要或高度相似但不具有权威性的来源。来源权威性必须通过语料库设计、元数据、过滤器、排序策略、验证或检索后检查来编码。",{},{"id":459,"data":460,"type":42,"tunes":462},"h-vs-vector",{"text":461,"level":245},"真相来源与向量数据库",{},{"id":464,"data":465,"type":218,"tunes":467},"p-vs-vector-1",{"text":466},"向量数据库存储或索引用于语义检索的表示。它是一个访问层，并不自动是一个真相层。",{},{"id":469,"data":470,"type":218,"tunes":472},"p-vs-vector-2",{"text":471},"同一份权威文档可能被分块、嵌入、复制和重新索引多次。向量记录应保留对权威来源和版本的引用，而不是成为一个无法追溯的新权威。",{},{"id":474,"data":475,"type":42,"tunes":477},"h-vs-memory",{"text":476,"level":245},"事实来源与记忆",{},{"id":479,"data":480,"type":218,"tunes":482},"p-vs-memory-1",{"text":481},"智能体或应用程序的记忆存储以后可能有用的信息。记忆可以保留先前的决策、偏好或观察，但它可能变得过时。",{},{"id":484,"data":485,"type":218,"tunes":487},"p-vs-memory-2",{"text":486},"对于易变或后果重大的状态，可靠的智能体通常应重新读取权威的当前来源，而不是假设记忆中的状态仍然为真。",{},{"id":489,"data":490,"type":42,"tunes":492},"h-vs-context",{"text":491,"level":245},"事实来源与上下文",{},{"id":494,"data":495,"type":218,"tunes":497},"p-vs-context-1",{"text":496},"上下文是模型在当前推理过程中接收到的内容。权威信息可能不在上下文中，而非权威信息可能存在于上下文中。",{},{"id":499,"data":500,"type":218,"tunes":502},"p-vs-context-2",{"text":501},"因此，上下文构建需要一种感知权威的策略：检索或读取被允许定义该主张的来源，然后保留足够的元数据，以便模型或验证器理解其适用范围。",{},{"id":504,"data":505,"type":42,"tunes":507},"h-vs-ground-truth",{"text":506,"level":245},"事实来源与评估基准真值",{},{"id":509,"data":510,"type":218,"tunes":512},"p-vs-ground-1",{"text":511},"评估基准真值是用于对系统进行评分的参考答案、标签或结果。它可以来自权威来源、专家裁定或精心整理的测试数据。",{},{"id":514,"data":515,"type":218,"tunes":517},"p-vs-ground-2",{"text":516},"因此，基准真值是一种评估构造。事实来源是一种应用\u002F领域权威构造。它们可以重叠，但不可互换。",{},{"id":519,"data":520,"type":42,"tunes":522},"h-vs-quality",{"text":521,"level":245},"事实来源与数据质量",{},{"id":524,"data":525,"type":218,"tunes":527},"p-vs-quality-1",{"text":526},"权威来源仍可能包含错误。权威说明哪个来源正式管辖该事实；数据质量则询问该来源是否准确、完整、及时、一致且适合用途。",{},{"id":529,"data":530,"type":218,"tunes":532},"p-vs-quality-2",{"text":531},"当已知权威系统有误时，架构应记录缺陷、纠正过程或例外情况，而不是静默地用非官方来源替代并隐藏差异。",{},{"id":534,"data":535,"type":42,"tunes":537},"h-model",{"text":536,"level":246},"一种实用的事实来源架构模型",{},{"id":539,"data":540,"type":226,"tunes":543},"model-note",{"body":541,"title":542,"variant":233},"以下模型是针对 AI 系统的实用综合方案。它不是 W3C、NIST 或 ISO 标准。它将权威、来源、检索和面向模型的上下文分开，因为这些职责经常被错误地合并。","建议的架构模型",{},{"id":545,"data":546,"type":574,"tunes":575},"sot-flow",{"steps":547,"title":572,"orientation":573},[548,551,554,557,560,563,566,569],{"label":549,"description":550},"1. 定义主张类型","确定用户实际在询问什么：当前状态、政策、历史事实、技术规范、用户偏好、计算或解释。",{"label":552,"description":553},"2. 解析权威","确定在所需范围、版本和时间下，哪个来源或权威类别被允许定义该类型的主张。",{"label":555,"description":556},"3. 获取证据","读取或检索权威来源以及任何必要的支持性或冲突性证据。",{"label":558,"description":559},"4. 保留来源","携带来源标识、版本、时间戳、定位符、转换历史和责任元数据。",{"label":561,"description":562},"5. 构建模型上下文","向模型提供相关证据，同时不丢弃权威性和适用性元数据。",{"label":564,"description":565},"6. 生成或计算","模型可以总结、比较、推理或转换证据，但不会仅仅因为处理了证据就继承来源的权威性。",{"label":567,"description":568},"7. 验证主张","检查答案是否由正确的来源支持，并且仍处于其范围和有效边界之内。",{"label":570,"description":571},"8. 保留矛盾","如果权威或相关来源不一致，应暴露冲突，而不是捏造虚假的确定性。","感知权威的 AI 回答路径","auto","processFlow",{},{"id":577,"data":578,"type":42,"tunes":580},"h-registry",{"text":579,"level":246},"权威应显式声明，而非从相似性推断",{},{"id":582,"data":583,"type":218,"tunes":585},"p-registry-1",{"text":584},"一种稳健的实现模式是权威注册表或等效的策略层，将主张类别映射到权威来源类别。实现可以是代码、元数据、配置或领域规则；重要的特性是权威是刻意确定的。",{},{"id":587,"data":588,"type":318,"tunes":618},"registry-table",{"content":589,"stretched":43,"withHeadings":14},[590,594,598,602,606,610,614],[591,592,593],"声明类别","权威规则","回退行为",[595,596,597],"当前账户状态","读取实时账户服务\u002F记录系统","如果不可用，则报告无法验证当前状态。",[599,600,601],"产品文档","当前批准的文档版本","仅可在带有版本警告的情况下显示旧版本。",[603,604,605],"已实现的软件行为","相关已部署版本\u002F源工件","仅凭文档无法证明已部署的行为。",[607,608,609],"内部政策","已批准的政策存储库和有效版本","草稿是支持材料，不是当前权威。",[611,612,613],"外部技术标准","相关版本的官方标准机构出版物","二手解释可以澄清但不能覆盖规范。",[615,616,617],"研究声明","适合该领域的证据政策","保留相互冲突的证据和置信度，而不是强行采用单一来源。",{},{"id":620,"data":621,"type":42,"tunes":623},"h-retrieval",{"text":622,"level":246},"检索应将权威性作为排序约束",{},{"id":625,"data":626,"type":218,"tunes":628},"p-retrieval-1",{"text":627},"语义相关性回答的是“哪个候选看起来与这个查询相关？”权威性回答的是“哪个候选被允许确立这个事实？”生产检索系统通常两者都需要。",{},{"id":630,"data":631,"type":218,"tunes":633},"p-retrieval-2",{"text":632},"一个有用的顺序是：先通过身份、租户、来源类别、状态、版本或日期约束候选空间，然后在允许的空间内对相关证据进行排序。如果在关键授权或权威过滤之前计算相关性，管道可能会返回一个看似有说服力但无效的结果。",{},{"id":635,"data":636,"type":226,"tunes":640},"relevance-authority",{"body":637,"title":638,"variant":639},"搜索引擎可能将过时的摘要排在当前主要来源之上。语言模型可能偏好写得更好的解释，而不是官方记录。这两种行为都不会改变哪个来源拥有该事实。","最高相似度不等于最高权威","warning",{},{"id":642,"data":643,"type":42,"tunes":645},"h-freshness",{"text":644,"level":246},"新鲜度是权威性的一部分",{},{"id":647,"data":648,"type":218,"tunes":650},"p-freshness-1",{"text":649},"许多真相来源的失败实际上是时间失败。正确的来源是已知的，但系统使用了旧快照、过时嵌入、缓存 API 响应或被取代的文档。",{},{"id":652,"data":653,"type":218,"tunes":655},"p-freshness-2",{"text":654},"因此，在事实可能发生变化的地方，权威规则应包含失效或刷新语义。当应用程序读取一周前的复制导出时，“CRM 是权威的”是不完整的。",{},{"id":657,"data":658,"type":42,"tunes":660},"h-derivation",{"text":659,"level":246},"派生值需要追溯到权威输入",{},{"id":662,"data":663,"type":218,"tunes":665},"p-derivation-1",{"text":664},"一些重要事实并非直接存储。它们是从权威输入计算得出的：风险评分、账户总额、资格状态或聚合指标。",{},{"id":667,"data":668,"type":218,"tunes":670},"p-derivation-2",{"text":669},"对于派生值，真相来源架构应保留输入权威、转换或计算版本以及执行时间。W3C PROV 对实体、活动和派生之间的区分在这里很有用，因为它建模了一个实体是如何从其他实体产生的。",{},{"id":672,"data":673,"type":226,"tunes":676},"derived-rule",{"body":674,"title":675,"variant":274},"计算值的当前性和权威性仅取决于其输入、转换和有效性条件。存储重新计算或审计它所需的血统。","派生真相是有条件的",{},{"id":678,"data":679,"type":42,"tunes":681},"h-conflict",{"text":680,"level":246},"当权威来源不一致时会发生什么？",{},{"id":683,"data":684,"type":218,"tunes":686},"p-conflict-1",{"text":685},"在严肃的知识系统中，冲突不是边缘情况。签署的合同可能与 CRM 字段不一致。生产行为可能与文档不一致。两个主要历史来源可能相互矛盾。当前政策可能与过时的本地副本冲突。",{},{"id":688,"data":689,"type":218,"tunes":691},"p-conflict-2",{"text":690},"系统需要适合该领域的解决政策。有时一个权威明显高于另一个。有时新版本取代旧版本。有时专家或业务所有者必须裁决。而有时正确的结果就是：证据尚未解决。",{},{"id":693,"data":694,"type":318,"tunes":717},"conflict-table",{"content":695,"stretched":43,"withHeadings":14},[696,699,702,705,708,711,714],[697,698],"冲突类型","典型处理",[700,701],"当前版本与被取代版本","对当前状态使用当前版本；保留旧版本作为历史证据。",[703,704],"记录系统与过时副本","使用记录系统；标记复制新鲜度问题。",[706,707],"合同与 CRM 转录","已签署合同管辖合同措辞；CRM 差异成为更正任务。",[709,710],"文档与已部署行为","区分预期行为与观察\u002F已部署行为；不要静默合并它们。",[712,713],"两个可信的主要来源","保留两者，评估来源和范围，如果不存在管辖权威，则表示未解决的分歧。",[715,716],"用户记忆与当前用户设置","使用当前显式设置；在适当情况下将记忆标记为被取代。",{},{"id":719,"data":720,"type":42,"tunes":722},"h-web",{"text":721,"level":246},"网络搜索是发现，不自动是证据",{},{"id":724,"data":725,"type":218,"tunes":727},"p-web-1",{"text":726},"搜索引擎是出色的发现系统。搜索片段、结果排名和生成的摘要不自动是主要证据。",{},{"id":729,"data":730,"type":218,"tunes":732},"p-web-2",{"text":731},"对于需要权威性的主张，搜索结果应指向原始出版物、官方记录、源文件、数据集或其他适当的工件。结果页面帮助定位来源；它并不继承来源的权威性。",{},{"id":734,"data":735,"type":42,"tunes":737},"h-ai",{"text":736,"level":246},"语言模型不应自行决定权威性",{},{"id":739,"data":740,"type":218,"tunes":742},"p-ai-1",{"text":741},"模型可以帮助分类问题、提取主张或比较证据，但权威性不应仅取决于模型的偏好。模型优化的是基于上下文的生成；它们并不拥有一个保证的领域特定注册表，来说明每个事实归属于哪个数据库、文档或组织。",{},{"id":744,"data":745,"type":218,"tunes":747},"p-ai-2",{"text":746},"这就是为什么应用架构应在实际可行的情况下以确定性方式编码关键权威规则。模型可以在边界内推理，但边界本身不应为每个提示从头重新创建。",{},{"id":749,"data":750,"type":42,"tunes":752},"h-observability",{"text":751,"level":246},"权威性必须在执行追踪中得以保留",{},{"id":754,"data":755,"type":218,"tunes":757},"p-observability-1",{"text":756},"如果生产答案重要到需要审计，追踪应使得能够重建以下内容：咨询了哪些来源、使用了哪个版本、哪个段落或记录支持了该主张、发生了哪些转换，以及是否存在相互矛盾的证据。",{},{"id":759,"data":760,"type":218,"tunes":762},"p-observability-2",{"text":761},"这与 W3C PROV 中更广泛的溯源原则以及 NIST AI RMF Playbook 中关于记录来源、起源、转换、依赖关系、约束和元数据的指导相一致。",{},{"id":764,"data":765,"type":42,"tunes":767},"h-implementation",{"text":766,"level":246},"原始实现证据：真相来源研究引擎",{},{"id":769,"data":770,"type":226,"tunes":773},"implementation-note",{"body":771,"title":772,"variant":233},"真相来源研究引擎是我自己对证据优先架构的实现证据。它展示了一种具体方式，将发现、来源获取、溯源、主张、证据类别、矛盾和结论分离开来。它是一种实现模式，而非通用的行业标准。","原始实现证据",{},{"id":775,"data":776,"type":218,"tunes":778},"p-impl-1",{"text":777},"该引擎围绕可追踪的流水线设计，而非直接进行 AI 摘要：研究任务 → 搜索 → 原始来源或数字工件 → 本地快照 → SHA-256 → 来源 ID → 主张 → 证据类别 → 关系或矛盾 → 解释 → 结论。",{},{"id":780,"data":781,"type":218,"tunes":783},"p-impl-2",{"text":782},"其共享证据核心存储来源、工件、溯源、主张、关系、矛盾、参考模型和审计追踪。不同的研究模式可以共享该核心，同时应用不同的领域方法论。",{},{"id":785,"data":786,"type":218,"tunes":788},"p-impl-3",{"text":787},"该架构有意将发现与证据分离。搜索片段不被视为证据，文件名不被视为内容，AI 摘要不被视为主要来源，语义相似性仅是发现信号，直到结果被追溯到具体来源和定位符。",{},{"id":790,"data":791,"type":218,"tunes":793},"p-impl-4",{"text":792},"原始文件被保留，本地字节获得 SHA-256 标识符。矛盾和被拒绝的假设不会被静默删除。新证据被允许改变当前参考模型，同时先前的证据路径仍可审计。",{},{"id":795,"data":796,"type":318,"tunes":825},"implementation-table",{"content":797,"stretched":43,"withHeadings":14},[798,801,804,807,810,813,816,819,822],[799,800],"已实现的规则","为何对真相来源架构重要",[802,803],"搜索 ≠ 证据","发现排名不能静默地变成权威性。",[805,806],"文件名 ≠ 内容","元数据线索不能替代阅读实际工件。",[808,809],"本地快照 + SHA-256","证据可以绑定到确切的字节，而非可变的远程标签。",[811,812],"来源 ID + 精确定位符","主张可以追溯到具体的证据位置。",[814,815],"主张\u002F证据分离","断言不会与支持它的材料混淆。",[817,818],"矛盾被保留","系统可以表示未解决的分歧，而非覆盖历史。",[820,821],"语义相似性仅用于发现","检索相关性被明确地与证据权威性分离。",[823,824],"新证据可以更新模型","真相来源状态是版本化的且可修订的，而非被视为不可变的教条。",{},{"id":827,"data":828,"type":42,"tunes":830},"h-document-engine",{"text":829,"level":246},"Aaasaasa 文档与知识引擎：将相同的边界应用于企业文档",{},{"id":832,"data":833,"type":218,"tunes":835},"p-doc-1",{"text":834},"Aaasaasa 文档与知识引擎概念将相同的设计原则扩展到企业文档：用户应能够搜索文档、提出有来源依据的问题，并根据明确的标准审查集合，同时保留文档所述内容与系统推断内容之间的区别。",{},{"id":837,"data":838,"type":218,"tunes":840},"p-doc-2",{"text":839},"重要的架构规则是，通用检索核心并不会使每个集合都具有同等权威性。合同文档、维护记录、财务文档和研究材料需要不同的权威性、验证和覆盖规则，即使它们共享摄取和搜索基础设施。",{},{"id":842,"data":843,"type":42,"tunes":845},"h-failures",{"text":844,"level":246},"常见的真相来源失败模式",{},{"id":847,"data":848,"type":318,"tunes":889},"failures-table",{"content":849,"stretched":43,"withHeadings":14},[850,853,856,859,862,865,868,871,874,877,880,883,886],[851,852],"失败模式","出了什么问题",[854,855],"将模型视为事实来源","参数化知识可能过时、不完整、无法验证，或超出应用的权威范围。",[857,858],"检索排名最高的结果自动胜出","将相似度误认为权威性。",[860,861],"向量数据库成为权威来源","派生索引记录丢失了原始来源的身份和版本。",[863,864],"所有内容都复制到一个知识库中","副本模糊了所有权、新鲜度和更正路径。",[866,867],"将记忆当作当前状态复用","旧观察结果悄然覆盖当前的记录系统。",[869,870],"没有版本元数据","正确的文档被用于错误的时间段。",[872,873],"没有来源定位符","引文存在，但支撑段落或记录无法验证。",[875,876],"冲突被覆盖","系统通过销毁分歧证据来显得一致。",[878,879],"生成的摘要取代原始内容","有损转换成为表面上的权威。",[881,882],"权威性是全局的，而非针对具体主张","某个来源被信任的范围超出了它实际拥有的领域或事实类别。",[884,885],"将网页摘要视为证据","发现元数据取代了原始出版物。",[887,888],"权威数据有误，但异常被隐藏","运营缺陷变得不可见，无法透明地更正。",{},{"id":891,"data":892,"type":42,"tunes":894},"h-decision",{"text":893,"level":246},"实用的真相来源决策框架",{},{"id":896,"data":897,"type":574,"tunes":927},"decision-flow",{"steps":898,"title":926,"orientation":573},[899,902,905,908,911,914,917,920,923],{"label":900,"description":901},"1. 精确陈述主张","区分当前状态、历史状态、政策、解释、预测和派生计算。",{"label":903,"description":904},"2. 确定权威所有者","确定负责该主张类型的系统、文档、机构、人员或证据类别。",{"label":906,"description":907},"3. 定义范围","指定租户、司法管辖区、产品、环境、用户、文档集或其他适用边界。",{"label":909,"description":910},"4. 定义时间和版本","确定该主张需要当前状态、历史快照，还是特定的标准\u002F发布版本。",{"label":912,"description":913},"5. 保留来源","记录来源身份、出处、定位符、转换过程以及负责的代理或流程。",{"label":915,"description":916},"6. 定义检索\u002F访问路径","确保应用确实能够在正确的身份和权限下获取权威信息。",{"label":918,"description":919},"7. 定义冲突策略","决定优先级、取代、裁决或明确未解决状态的行为。",{"label":921,"description":922},"8. 定义失效机制","指定缓存、索引、记忆或派生表示何时必须刷新。",{"label":924,"description":925},"9. 验证答案路径","验证重要的生成主张能够追溯到预期的权威来源，而不仅仅是看似合理的来源。","如何决定什么应该定义一项主张",{},{"id":929,"data":930,"type":42,"tunes":932},"h-checklist",{"text":931,"level":246},"真相来源架构检查清单",{},{"id":934,"data":935,"type":318,"tunes":975},"checklist-table",{"content":936,"stretched":43,"withHeadings":14},[937,939,942,945,948,951,954,957,960,963,966,969,972],[356,938],"预期答案",[940,941],"正在确立的确切事实或状态是什么？","主张足够精确，可以分配权威性。",[943,944],"谁或什么拥有该事实？","具名的权威系统、来源类别或裁决规则。",[946,947],"权威来源在此范围内是最新的吗？","租户、司法管辖区、环境、用户或领域边界是明确的。",[949,950],"版本\u002F时间正确吗？","当前、历史或特定版本的适用性是已知的。",[952,953],"来源可以验证吗？","存在稳定的标识符、定位符或记录引用。",[955,956],"来源被保留了吗？","出处、转换和责任元数据在摄取和检索过程中得以保留。",[958,959],"检索可能返回非权威材料吗？","如果可能，过滤器或验证会区分相关性与权威性。",[961,962],"来源可能变化吗？","存在刷新、失效或取代规则。",[964,965],"来源可能不一致吗？","冲突和裁决行为是明确的。",[967,968],"记忆可能过时吗？","在产生后果的使用之前，从当前权威来源重新读取易变状态。",[970,971],"派生答案可以复现吗？","输入、转换版本和执行条件可追溯。",[973,974],"审计员可以重建答案吗？","执行证据为重要主张保留了来源路径。",{},{"id":977,"data":978,"type":42,"tunes":980},"h-misconceptions",{"text":979,"level":246},"常见误解",{},{"id":982,"data":983,"type":318,"tunes":1012},"misconceptions-table",{"content":984,"stretched":43,"withHeadings":14},[985,988,991,994,997,1000,1003,1006,1009],[986,987],"误解","更正",[989,990],"“真相来源意味着一个数据库。”","一个数据库可以是一个领域的权威来源；复杂系统通常有多个针对具体事实的权威来源。",[992,993],"“最新的文档自动具有权威性。”","只有在较新的制品获得批准并实际取代较旧制品时，时效性才有帮助。",[995,996],"“RAG 解决了真相问题。”","RAG 解决的是检索问题。权威性、出处、证据质量和有效性仍然是独立的问题。",[998,999],"“引文证明了答案。”","被引用的来源必须实际支持该主张，具有正确的权威性，并适用于当前范围。",[1001,1002],"“出处告诉我们什么是真的。”","出处告诉我们来源和派生过程；权威性和正确性仍然需要领域规则和评估。",[1004,1005],"“记录系统总是正确的。”","它对运营记录具有权威性，但数据质量缺陷仍然可能存在，需要可见的更正。",[1007,1008],"“如果多个来源一致，该主张就是权威的。”","一致性增加了证据，但不一定确立所有权或适用性。",[1010,1011],"“AI 记忆可以取代重复读取。”","仅适用于过时风险可接受的信息；易变或会产生后果的状态应从权威来源刷新。",{},{"id":1014,"data":1015,"type":42,"tunes":1017},"h-edge",{"text":1016,"level":246},"边缘情况",{},{"id":1019,"data":1020,"type":218,"tunes":1022},"p-edge-1",{"text":1021},"有些问题是解释性的，而非事实性的。“哪种架构最好？”没有单一的真相来源。系统可以检索权威约束和证据，但最终判断是一种推断，应当暴露假设和权衡。",{},{"id":1024,"data":1025,"type":218,"tunes":1027},"p-edge-2",{"text":1026},"有些领域使用分布式权威。科学结论可能依赖于多项研究、数据集和重复实验。历史结论可能依赖于相互冲突的一手和二手证据。架构应当表示证据结构，而不是发明一个据称拥有真相的中央数据库。",{},{"id":1029,"data":1030,"type":218,"tunes":1032},"p-edge-3",{"text":1031},"用户也可以是主观个人信息的权威：偏好、目标、选定设置或明确指令。即便如此，较新的明确输入也可以取代较旧的记忆。",{},{"id":1034,"data":1035,"type":218,"tunes":1037},"p-edge-4",{"text":1036},"外部事件可能使先前具有权威性的数据失效。价格源、库存系统或安全策略在捕获时可能是正确的，但不再有效。快照出处保留了当时为真的内容；它不会使快照永远保持最新。",{},{"id":1039,"data":1040,"type":42,"tunes":1042},"h-limit",{"text":1041,"level":246},"局限性",{},{"id":1044,"data":1045,"type":218,"tunes":1047},"p-limit-1",{"text":1046},"真相来源架构无法保证权威来源在事实上是正确的。它提供了问责、出处和确定性的所有权边界；数据质量和领域验证流程仍然是必要的。",{},{"id":1049,"data":1050,"type":218,"tunes":1052},"p-limit-2",{"text":1051},"权威性也可能存在争议。不同的机构可以在不同的司法管辖区或方法论中合法地主张权威。在这些情况下，系统应当暴露权威模型和分歧，而不是将其隐藏在通用的“真相分数”背后。",{},{"id":1054,"data":1055,"type":218,"tunes":1057},"p-limit-3",{"text":1056},"最后，权威规则需要维护。系统、所有者、政策、版本和法规都会变化。过时的权威注册表可能比完全没有注册表更危险。",{},{"id":1059,"data":1060,"type":42,"tunes":1062},"h-change",{"text":1061,"level":246},"什么会改变这个答案？",{},{"id":1064,"data":1065,"type":218,"tunes":1067},"p-change-1",{"text":1066},"具体的权威映射随领域而变化。银行、医疗、软件交付、科学研究和历史分析有不同的记录系统、证据规则和监管义务。",{},{"id":1069,"data":1070,"type":218,"tunes":1072},"p-change-2",{"text":1071},"实现方式也会随架构而变化。小型应用可以直接在服务调用中编码权威性。较大的平台可能需要注册表、源元数据、策略引擎、血缘系统或数据契约。核心原则保持不变：不要让检索顺序或模型偏好悄然决定什么算作权威。",{},{"id":1074,"data":1075,"type":42,"tunes":1077},"h-related",{"text":1076,"level":246},"相关规范知识",{},{"id":1079,"data":1080,"type":218,"tunes":1082},"p-related-1",{"text":1081},"真相源架构是后续检索和治理概念的前提，因为仅靠检索质量无法确定证据是否被允许定义答案。",{},{"id":1084,"data":1085,"type":1090,"tunes":1091},"ref-rag",{"url":1086,"title":1087,"excerpt":1088,"ctaLabel":1089},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works","什么是RAG？对其工作原理的最简单解释","RAG为模型检索外部知识。这一基础解释了为什么检索和权威真相是相互独立的职责。","阅读RAG基础","referralArticle",{},{"id":1093,"data":1094,"type":218,"tunes":1096},"p-related-2",{"text":1095},"记忆是另一个相邻概念。可靠的智能体将记住的信息与当前权威的应用状态分开。",{},{"id":1098,"data":1099,"type":1090,"tunes":1104},"ref-memory",{"url":1100,"title":1101,"excerpt":1102,"ctaLabel":1103},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fai-agent-memory-is-not-rag-how-to-separate-memory-retrieval-state-and-context","AI智能体记忆不是RAG：如何分离记忆、检索、状态和上下文","一种四层架构，分离持久记忆、权威状态、检索以及实际提供给模型的上下文。","阅读记忆架构文章",{},{"id":1106,"data":1107,"type":218,"tunes":1109},"p-related-3",{"text":1108},"权威性也直接关系到答案有效性。即使是权威来源，也只支持其版本、日期、范围和证据边界内的主张。",{},{"id":1111,"data":1112,"type":1090,"tunes":1117},"ref-avb",{"url":1113,"title":1114,"excerpt":1115,"ctaLabel":1116},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers","答案有效性边界：相关性与可靠AI答案之间缺失的层","一个框架，用于明确AI答案在何处仍受证据支持，以及哪些条件变化会使该支持失效。","阅读答案有效性边界",{},{"id":1119,"data":1120,"type":42,"tunes":1122},"h-faq",{"text":1121,"level":246},"常见问题",{},{"id":1124,"data":1125,"type":1124,"tunes":1160},"faq",{"items":1126,"title":1159},[1127,1131,1135,1139,1143,1147,1151,1155],{"id":1128,"answer":1129,"question":1130},"faq1","它是权威来源或权威规则，决定在定义的范围、版本和时间下，哪个来源被允许确立特定事实、状态或规则。","AI系统中的真相源是什么？",{"id":1132,"answer":1133,"question":1134},"faq2","通常不是。语言模型可以生成、总结和推理，但其参数化知识对于当前应用状态、公司政策、特定文档版本或受监管领域的事实并不自动具有权威性。","语言模型是真相源吗？",{"id":1136,"answer":1137,"question":1138},"faq3","不自动是。向量数据库通常是索引或检索存储。它应保留对权威原始来源和版本的引用，而不是悄然替换它们。","向量数据库是RAG的真相源吗？",{"id":1140,"answer":1141,"question":1142},"faq4","来源描述数据来自哪里、如何产生或转换以及涉及谁或什么。真相源规则决定该来源是否对特定主张具有权威性。","来源和真相源有什么区别？",{"id":1144,"answer":1145,"question":1146},"faq5","可以。在复杂系统中这是正常的，因为不同事实属于不同的权威系统或来源类别。","AI系统可以有多个真相源吗？",{"id":1148,"answer":1149,"question":1150},"faq6","系统需要特定领域的冲突规则：优先级、版本取代、专家裁决或明确未解决状态。它不应悄然选择模型偏好的来源。","当两个权威来源不一致时会发生什么？",{"id":1152,"answer":1153,"question":1154},"faq7","不能。RAG检索候选。需要权威感知元数据、过滤器、来源策略和验证，以确保重要主张使用正确的来源。","RAG能保证AI答案使用真相源吗？",{"id":1156,"answer":1157,"question":1158},"faq8","可能。权威性和正确性是不同的属性。权威系统可能包含数据质量缺陷，应以透明方式纠正，而不是通过替换非官方来源来隐藏。","真相源可能是错的吗？","AI系统中的真相源",{},{"id":1162,"data":1163,"type":42,"tunes":1165},"h-glossary",{"text":1164,"level":246},"术语表",{},{"id":1167,"data":1168,"type":1167,"tunes":1209},"glossary",{"title":1169,"entries":1170},"关键真相源术语",[1171,1175,1179,1182,1186,1189,1193,1197,1201,1205],{"term":1172,"anchor":1173,"definition":1174},"真相源","source-of-truth","被允许在定义的范围、版本和时间下确立特定事实、状态或规则的权威来源或规则。",{"term":1176,"anchor":1177,"definition":1178},"记录系统","system-of-record","负责定义类别记录或当前业务状态的权威运营系统。",{"term":322,"anchor":1180,"definition":1181},"provenance","描述数据或其他实体的起源、派生、转换、负责主体和历史的信息。",{"term":1183,"anchor":1184,"definition":1185},"证据","evidence","支持、反驳或约束某一主张的信息或人工制品。",{"term":1187,"anchor":327,"definition":1188},"权威性","决定哪个来源有权定义特定主张的应用或领域规则。",{"term":1190,"anchor":1191,"definition":1192},"新鲜度","freshness","某一表示对于其被使用的主张或操作是否仍足够当前。",{"term":1194,"anchor":1195,"definition":1196},"取代","supersession","较新的权威版本对较旧版本的明确替换，同时保留历史可追溯性。",{"term":1198,"anchor":1199,"definition":1200},"基准真相","ground-truth","用于评估系统的参考答案、标签或结果；它是评估构造，并不自动是应用的真相源。",{"term":1202,"anchor":1203,"definition":1204},"血缘","lineage","数据或派生值在来源和处理步骤之间流动和转换的轨迹。",{"term":1206,"anchor":1207,"definition":1208},"有效性边界","validity-boundary","某一主张仍受支持的范围、时间、版本、证据和假设条件。",{},{"id":1211,"data":1212,"type":42,"tunes":1214},"h-conclusion",{"text":1213,"level":246},"结论",{},{"id":1216,"data":1217,"type":218,"tunes":1219},"p-conclusion-1",{"text":1218},"可靠的AI并非来自给模型更多信息。它来自知道哪些信息被允许定义主张、保留该信息的来源、检索正确版本，并将最终答案保持在来源的范围内。",{},{"id":1221,"data":1222,"type":218,"tunes":1224},"p-conclusion-2",{"text":1223},"这就是为什么真相源、来源、检索、记忆和上下文必须保持为独立概念。真相源定义权威性。来源解释起源。检索找到候选。记忆保留选定的过去信息。上下文是模型看到的内容。生成将这些输入转化为输出。",{},{"id":1226,"data":1227,"type":218,"tunes":1229},"p-conclusion-3",{"text":1228},"当这些层保持明确时，AI系统不仅能听起来合理：重要主张可以追溯到实际有权确立它们的来源。",{},{"id":1231,"data":1232,"type":42,"tunes":1234},"h-sources",{"text":1233,"level":246},"主要来源和实现证据",{},{"id":1236,"data":1237,"type":218,"tunes":1239},"p-sources-note",{"text":1238},"以下外部来源支持来源和AI风险管理的说法。真相源研究引擎部分是原创实现证据，并明确呈现为一种实现模式，而非通用标准。",{},{"id":1241,"data":1242,"type":1248,"tunes":1249},"src-w3c-prov",{"link":1243,"meta":1244},"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-dm\u002F",{"image":1245,"title":1246,"description":1247},{"url":304},"W3C PROV-DM — PROV 数据模型","W3C 推荐标准，定义了一个与领域无关的溯源模型，围绕实体、活动、代理、派生和责任。","linkTool",{},{"id":1251,"data":1252,"type":1248,"tunes":1258},"src-w3c-publications",{"link":1253,"meta":1254},"https:\u002F\u002Fwww.w3.org\u002Fgroups\u002Fwg\u002Fprov\u002Fpublications\u002F",{"image":1255,"title":1256,"description":1257},{"url":304},"W3C 溯源工作组 — 出版物","W3C PROV 推荐标准及相关规范的官方索引，用于溯源交换和约束。",{},{"id":1260,"data":1261,"type":1248,"tunes":1267},"src-nist-rmf",{"link":1262,"meta":1263},"https:\u002F\u002Fwww.nist.gov\u002Fitl\u002Fai-risk-management-framework",{"image":1264,"title":1265,"description":1266},{"url":304},"NIST AI 风险管理框架","NIST 的自愿性框架，用于在 AI 生命周期中纳入可信度和风险管理考量；AI RMF 1.0 目前正在修订中。",{},{"id":1269,"data":1270,"type":1248,"tunes":1276},"src-nist-playbook",{"link":1271,"meta":1272},"https:\u002F\u002Fairc.nist.gov\u002Fairmf-resources\u002Fplaybook\u002F",{"image":1273,"title":1274,"description":1275},{"url":304},"NIST AI RMF 操作手册","与 AI RMF 一致的操作指南，包括数据溯源、来源、起源、转换、依赖关系、约束和元数据的文档实践。",{},{"id":1278,"data":1279,"type":1248,"tunes":1285},"src-nist-measure",{"link":1280,"meta":1281},"https:\u002F\u002Fairc.nist.gov\u002Fairmf-resources\u002Fplaybook\u002Fmeasure\u002F",{"image":1282,"title":1283,"description":1284},{"url":304},"NIST AI RMF 操作手册 — 测量","关于记录测量、数据溯源以及 AI 系统输出上下文解释的指南。",{},{"id":1287,"data":1288,"type":1248,"tunes":1294},"src-nist-genai",{"link":1289,"meta":1290},"https:\u002F\u002Fwww.nist.gov\u002Fpublications\u002Fartificial-intelligence-risk-management-framework-generative-artificial-intelligence",{"image":1291,"title":1292,"description":1293},{"url":304},"NIST AI 600-1 — 生成式 AI 概况","NIST 生成式 AI 概况，包括生成式 AI 系统的溯源和信息完整性考量。",{},"2.31","事实来源（Source of Truth）定义了对于特定事实或状态，哪个来源具有权威性。了解它与RAG、溯源、记忆、上下文、向量数据库和记录系统有何不同。","\u002Fuploads\u002F2026\u002F10\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from-1791479103235-6bq9em.webp","source-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from-1791479103235-6bq9em","PUBLISHED","2026-10-08T13:02:00.000Z","2026-10-08T17:02:49.770Z","2026-10-08T17:11:41.743Z",{"en":1304,"de":1305,"sr":1306,"es":1307,"fr":1308,"it":1309,"ru":1310,"zh":1311},"\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Fde\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Fsr\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Fes\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Ffr\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Fit\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Fru\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from","\u002Fzh\u002Fblog\u002Fsource-of-truth-in-ai-systems-where-reliable-knowledge-actually-comes-from",[1313,1317,1321],{"id":1314,"name":1315,"slug":1316},65,"主题权威体系","topical-authority",{"id":1318,"name":1319,"slug":1320},57,"数据边界","data-boundaries",{"id":1322,"name":1323,"slug":1324},84,"策略与数据边界","policy-and-data",{"id":1326,"login":1327,"email":1328,"displayName":1329},"20","rooth8233","aleksandar@stajic.de","Aleksandar Stajić",[1331,2244],{"lang":1332,"title":1333,"content":1334,"contentJson":1335,"excerpt":2243},"en","Source of Truth in AI Systems: Where Reliable Knowledge Actually Comes From","{\"time\":1791479104597,\"blocks\":[{\"id\":\"intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"A Source of Truth in an AI system is the authoritative source allowed to define whether a specific fact, state or rule should be treated as true for a particular scope, version and time. It is not automatically the language model, the vector database, the top-ranked retrieved document, agent memory or the latest message in context. Reliable AI architecture must preserve which source has authority for which claim, then keep provenance, retrieval and validation connected to that authority.\"},\"tunes\":{}},{\"id\":\"direct\",\"type\":\"callout\",\"data\":{\"variant\":\"info\",\"title\":\"Direct answer\",\"body\":\"\u003Cstrong>The Source of Truth is an authority rule, not an AI component.\u003C\u002Fstrong> An AI system can search many sources, but only some sources are authoritative for a given question. The correct source may be an application database for current account state, an approved policy document for a rule, source code for implemented behavior, an official specification for a protocol, or a primary research artifact for an observed claim. Good architecture makes that authority explicit instead of asking the model to infer it from relevance.\"},\"tunes\":{}},{\"id\":\"terminology\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"Terminology boundary\",\"body\":\"“Source of Truth” is widely used in software, data and organizational practice, but it is not a standardized AI component with one universal formal definition. This article uses it as an architecture concept: \u003Cstrong>the source or authority mapping that determines which evidence is allowed to establish a particular fact or state\u003C\u002Fstrong>. Provenance standards such as W3C PROV describe origin and derivation; they do not automatically decide which source your application should treat as authoritative.\"},\"tunes\":{}},{\"id\":\"current\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"Current-source note — 8 October 2026\",\"body\":\"The core distinction between authority, provenance, retrieval and model context is stable. NIST AI RMF 1.0 is currently being revised, but the current NIST Playbook still explicitly recommends documenting data provenance, including sources, origins, transformations, dependencies, constraints and metadata. W3C PROV remains a W3C Recommendation for interoperable provenance descriptions.\"},\"tunes\":{}},{\"id\":\"toc\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"h-meaning\",\"type\":\"header\",\"data\":{\"text\":\"What “Source of Truth” really means\",\"level\":2},\"tunes\":{}},{\"id\":\"p-meaning-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The phrase is often misunderstood as “the one database that contains everything.” That can be true in a narrow system, but it is usually too simplistic for AI. A real AI application can combine operational databases, documents, APIs, vector indexes, user input, model memory, external web sources and generated summaries.\"},\"tunes\":{}},{\"id\":\"p-meaning-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Those sources do not have equal authority. A customer-support manual may define policy but not a customer's current balance. A CRM may define the current account owner but not the legal meaning of a regulation. A source-code repository may define implemented behavior while a product specification defines intended behavior. The architecture must therefore answer a more precise question: which source is authoritative for this specific claim?\"},\"tunes\":{}},{\"id\":\"p-meaning-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"This makes Source of Truth a relationship between a claim and an authority, not merely a property of a storage technology.\"},\"tunes\":{}},{\"id\":\"principle\",\"type\":\"callout\",\"data\":{\"variant\":\"success\",\"title\":\"Core principle\",\"body\":\"\u003Cstrong>One system can have many Sources of Truth because authority is fact-specific.\u003C\u002Fstrong> The important architectural property is not that one database wins globally, but that each important fact or state has an explicit authoritative owner.\"},\"tunes\":{}},{\"id\":\"h-simple\",\"type\":\"header\",\"data\":{\"text\":\"The simplest example\",\"level\":2},\"tunes\":{}},{\"id\":\"p-simple-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A user asks an AI assistant: “What is my current subscription plan?” The assistant has three possible inputs: last month's support transcript, an indexed help-center document describing plan types, and the live billing database.\"},\"tunes\":{}},{\"id\":\"p-simple-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The support transcript may mention that the user had a Pro plan. The help-center document explains what Pro means. But the live billing record is the authoritative source for the user's current subscription state.\"},\"tunes\":{}},{\"id\":\"p-simple-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"A semantic search engine could rank the support transcript above the billing record because it contains language closer to the question. That ranking would still not make the transcript authoritative. Relevance and authority are different dimensions.\"},\"tunes\":{}},{\"id\":\"simple-comparison\",\"type\":\"comparison\",\"data\":{\"title\":\"The same question can involve different source roles\",\"layout\":\"table\",\"columns\":[{\"id\":\"source\",\"label\":\"Source\"},{\"id\":\"role\",\"label\":\"Role\"},{\"id\":\"authority\",\"label\":\"Authority for current plan?\"}],\"rows\":[{\"id\":\"billing\",\"label\":\"Billing database\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"help\",\"label\":\"Help-center documentation\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"transcript\",\"label\":\"Old support transcript\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"model\",\"label\":\"Model memory\",\"values\":[\"\",\"\",\"\"]}]},\"tunes\":{}},{\"id\":\"h-stops\",\"type\":\"header\",\"data\":{\"text\":\"Where the simple example stops\",\"level\":2},\"tunes\":{}},{\"id\":\"p-stops-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Not every domain has one unquestioned authority. Historical research can contain conflicting primary sources. Scientific claims can evolve as new studies appear. Legal interpretation can depend on jurisdiction, date and court authority. Product behavior can differ between documentation and deployed code.\"},\"tunes\":{}},{\"id\":\"p-stops-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"In these cases the correct architecture is not to invent a single winner. It is to preserve the competing sources, their provenance, their authority class, their applicable scope and the unresolved contradiction. A reliable Source-of-Truth system must be able to represent uncertainty and disagreement.\"},\"tunes\":{}},{\"id\":\"h-authority\",\"type\":\"header\",\"data\":{\"text\":\"Authority is scoped by claim, version and time\",\"level\":2},\"tunes\":{}},{\"id\":\"authority-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Question\",\"Possible authoritative source\",\"Why scope matters\"],[\"What is the user's current account balance?\",\"Ledger \u002F accounting system of record\",\"Historical exports may be accurate for an earlier time but not current state.\"],[\"What does company policy currently allow?\",\"Approved current policy version\",\"An older policy may remain valid evidence of past rules but not present rules.\"],[\"What code is actually deployed?\",\"Deployment artifact \u002F commit \u002F release record\",\"Main branch may differ from production.\"],[\"What did a contract state when signed?\",\"Executed contract version\",\"A draft or later template is not authoritative for the signed agreement.\"],[\"What does a technical protocol specify?\",\"Current official specification for the relevant version\",\"A blog explanation may be useful but is secondary evidence.\"],[\"What happened in a historical event?\",\"Relevant primary evidence plus explicit source criticism\",\"There may be no single authority; conflicting evidence must remain visible.\"],[\"What does a user prefer?\",\"Current explicit user setting or confirmed preference\",\"Old conversation memory may be stale or superseded.\"]]},\"tunes\":{}},{\"id\":\"p-authority-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The word “truth” can therefore be misleading unless its boundary is stated. In architecture, the Source of Truth is usually better understood as the source authorized to determine a specific proposition under defined conditions.\"},\"tunes\":{}},{\"id\":\"h-what-is-not\",\"type\":\"header\",\"data\":{\"text\":\"What a Source of Truth is not\",\"level\":2},\"tunes\":{}},{\"id\":\"h-vs-sor\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs system of record\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-sor-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A system of record is typically the authoritative operational system for a class of records: for example, a billing ledger, HR master record or order database. It is one common implementation of Source-of-Truth authority.\"},\"tunes\":{}},{\"id\":\"p-vs-sor-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"But Source of Truth is broader. A signed PDF contract, an official standard, a deployment artifact or a primary archival document may be authoritative without being a transactional system of record.\"},\"tunes\":{}},{\"id\":\"h-vs-prov\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs provenance\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-prov-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Provenance answers questions such as: Where did this data come from? Who or what produced it? Which transformation created this derivative? Which prior entity was used? W3C PROV models entities, activities, agents and derivations so origin and responsibility can be represented.\"},\"tunes\":{}},{\"id\":\"p-vs-prov-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Provenance does not by itself establish authority. Knowing that a value came from a spreadsheet written by a specific employee helps evaluate it, but the application still needs a rule saying whether that spreadsheet is authoritative for the claim.\"},\"tunes\":{}},{\"id\":\"h-vs-evidence\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs evidence\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-evidence-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence supports or contradicts a claim. A Source of Truth defines which source has the authority to settle or strongly constrain that claim in the current application context.\"},\"tunes\":{}},{\"id\":\"p-vs-evidence-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A source can be valuable evidence without being authoritative. Five customer emails may be evidence that users dislike a workflow, but they are not the system of record for current product configuration.\"},\"tunes\":{}},{\"id\":\"h-vs-rag\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs RAG\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-rag-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"RAG is a retrieval pattern. It finds information and supplies selected content to the model. RAG does not automatically know which source deserves authority.\"},\"tunes\":{}},{\"id\":\"p-vs-rag-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A RAG pipeline can retrieve a stale document, a secondary summary or a highly similar but non-authoritative source. Source authority must be encoded through corpus design, metadata, filters, ranking policy, validation or post-retrieval checks.\"},\"tunes\":{}},{\"id\":\"h-vs-vector\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs vector database\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-vector-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A vector database stores or indexes representations used for semantic retrieval. It is an access layer, not automatically a truth layer.\"},\"tunes\":{}},{\"id\":\"p-vs-vector-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The same authoritative document may be chunked, embedded, copied and re-indexed many times. The vector record should retain a reference back to the authoritative source and version rather than becoming an untraceable new authority.\"},\"tunes\":{}},{\"id\":\"h-vs-memory\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs memory\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-memory-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Agent or application memory stores information that may be useful later. Memory can preserve a prior decision, preference or observation, but it can become stale.\"},\"tunes\":{}},{\"id\":\"p-vs-memory-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For volatile or consequential state, a reliable agent should normally re-read the authoritative current source rather than assume that remembered state is still true.\"},\"tunes\":{}},{\"id\":\"h-vs-context\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs context\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-context-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Context is what the model receives during the current inference. Authoritative information can be absent from context, while non-authoritative information can be present.\"},\"tunes\":{}},{\"id\":\"p-vs-context-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Context construction therefore needs an authority-aware policy: retrieve or read the source that is allowed to define the claim, then preserve enough metadata for the model or validator to understand its scope.\"},\"tunes\":{}},{\"id\":\"h-vs-ground-truth\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs evaluation ground truth\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-ground-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evaluation ground truth is the reference answer, label or outcome against which a system is scored. It can be derived from authoritative sources, expert adjudication or curated test data.\"},\"tunes\":{}},{\"id\":\"p-vs-ground-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Ground truth is therefore an evaluation construct. A Source of Truth is an application\u002Fdomain authority construct. They can overlap, but they are not interchangeable.\"},\"tunes\":{}},{\"id\":\"h-vs-quality\",\"type\":\"header\",\"data\":{\"text\":\"Source of Truth vs data quality\",\"level\":3},\"tunes\":{}},{\"id\":\"p-vs-quality-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"An authoritative source can still contain errors. Authority says which source officially governs the fact; data quality asks whether that source is accurate, complete, timely, consistent and fit for purpose.\"},\"tunes\":{}},{\"id\":\"p-vs-quality-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"When an authoritative system is known to be wrong, architecture should record the defect, correction process or exception instead of silently substituting an unofficial source and hiding the discrepancy.\"},\"tunes\":{}},{\"id\":\"h-model\",\"type\":\"header\",\"data\":{\"text\":\"A practical Source-of-Truth architecture model\",\"level\":2},\"tunes\":{}},{\"id\":\"model-note\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"Proposed architecture model\",\"body\":\"The model below is a practical synthesis for AI systems. It is not a W3C, NIST or ISO standard. It separates authority, provenance, retrieval and model-facing context because those responsibilities are often incorrectly collapsed.\"},\"tunes\":{}},{\"id\":\"sot-flow\",\"type\":\"processFlow\",\"data\":{\"title\":\"Authority-aware AI answer path\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"1. Define the claim type\",\"description\":\"Identify what the user is actually asking: current state, policy, historical fact, technical specification, user preference, calculation or interpretation.\"},{\"label\":\"2. Resolve authority\",\"description\":\"Determine which source or authority class is permitted to define that type of claim for the required scope, version and time.\"},{\"label\":\"3. Acquire evidence\",\"description\":\"Read or retrieve the authoritative source and any necessary supporting or conflicting evidence.\"},{\"label\":\"4. Preserve provenance\",\"description\":\"Carry source identity, version, timestamp, locator, transformation history and responsibility metadata.\"},{\"label\":\"5. Build model context\",\"description\":\"Supply the relevant evidence to the model without discarding authority and applicability metadata.\"},{\"label\":\"6. Generate or compute\",\"description\":\"The model may summarize, compare, reason or transform the evidence, but it does not inherit source authority merely by processing it.\"},{\"label\":\"7. Validate the claim\",\"description\":\"Check that the answer is supported by the right source and remains inside its scope and validity boundary.\"},{\"label\":\"8. Preserve contradictions\",\"description\":\"If authoritative or relevant sources disagree, expose the conflict rather than fabricating false certainty.\"}]},\"tunes\":{}},{\"id\":\"h-registry\",\"type\":\"header\",\"data\":{\"text\":\"Authority should be explicit, not inferred from similarity\",\"level\":2},\"tunes\":{}},{\"id\":\"p-registry-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"One robust implementation pattern is an authority registry or equivalent policy layer that maps claim classes to authoritative source classes. The implementation can be code, metadata, configuration or domain rules; the important property is that authority is deliberate.\"},\"tunes\":{}},{\"id\":\"registry-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Claim class\",\"Authority rule\",\"Fallback behavior\"],[\"Current account state\",\"Read live account service \u002F system of record\",\"If unavailable, report that current state cannot be verified.\"],[\"Product documentation\",\"Current approved documentation version\",\"Older version may be shown only with version warning.\"],[\"Implemented software behavior\",\"Relevant deployed release \u002F source artifact\",\"Documentation alone cannot prove deployed behavior.\"],[\"Internal policy\",\"Approved policy repository and active version\",\"Drafts are supporting material, not current authority.\"],[\"External technical standard\",\"Official standards body publication for relevant version\",\"Secondary explanations can clarify but not override the specification.\"],[\"Research claim\",\"Evidence policy appropriate to the domain\",\"Preserve conflicting evidence and confidence rather than force one source.\"]]},\"tunes\":{}},{\"id\":\"h-retrieval\",\"type\":\"header\",\"data\":{\"text\":\"Retrieval should use authority as a ranking constraint\",\"level\":2},\"tunes\":{}},{\"id\":\"p-retrieval-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Semantic relevance answers “Which candidate looks related to this query?” Authority answers “Which candidate is allowed to establish this fact?” A production retrieval system often needs both.\"},\"tunes\":{}},{\"id\":\"p-retrieval-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A useful sequence is to first constrain the candidate space by identity, tenant, source class, status, version or date, then rank relevant evidence inside the permitted space. If relevance is calculated before critical authorization or authority filters, the pipeline can return a convincing but invalid result.\"},\"tunes\":{}},{\"id\":\"relevance-authority\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Highest similarity is not highest authority\",\"body\":\"A search engine can rank a stale summary above the current primary source. A language model can prefer the better-written explanation over the official record. Neither behavior changes which source owns the fact.\"},\"tunes\":{}},{\"id\":\"h-freshness\",\"type\":\"header\",\"data\":{\"text\":\"Freshness is part of authority\",\"level\":2},\"tunes\":{}},{\"id\":\"p-freshness-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Many Source-of-Truth failures are actually time failures. The correct source was known, but the system used an old snapshot, stale embedding, cached API response or superseded document.\"},\"tunes\":{}},{\"id\":\"p-freshness-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"An authority rule should therefore include invalidation or refresh semantics where the fact can change. “CRM is authoritative” is incomplete when the application reads a week-old replicated export.\"},\"tunes\":{}},{\"id\":\"h-derivation\",\"type\":\"header\",\"data\":{\"text\":\"Derived values need a trace back to authoritative inputs\",\"level\":2},\"tunes\":{}},{\"id\":\"p-derivation-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Some important facts are not stored directly. They are computed from authoritative inputs: a risk score, account total, eligibility status or aggregate metric.\"},\"tunes\":{}},{\"id\":\"p-derivation-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For derived values, Source-of-Truth architecture should preserve the input authorities, transformation or calculation version and execution time. W3C PROV's distinction between entities, activities and derivations is useful here because it models how one entity was produced from others.\"},\"tunes\":{}},{\"id\":\"derived-rule\",\"type\":\"callout\",\"data\":{\"variant\":\"success\",\"title\":\"Derived truth is conditional\",\"body\":\"A calculated value is only as current and authoritative as its inputs, transformation and validity conditions. Store the lineage needed to recompute or audit it.\"},\"tunes\":{}},{\"id\":\"h-conflict\",\"type\":\"header\",\"data\":{\"text\":\"What happens when authoritative sources disagree?\",\"level\":2},\"tunes\":{}},{\"id\":\"p-conflict-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Conflicts are not edge cases in serious knowledge systems. A signed contract may disagree with a CRM field. Production behavior may disagree with documentation. Two primary historical sources may contradict each other. A current policy can conflict with an outdated local copy.\"},\"tunes\":{}},{\"id\":\"p-conflict-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The system needs a resolution policy appropriate to the domain. Sometimes one authority clearly outranks the other. Sometimes the newer version supersedes the old one. Sometimes an expert or business owner must adjudicate. And sometimes the correct result is simply: the evidence is unresolved.\"},\"tunes\":{}},{\"id\":\"conflict-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Conflict type\",\"Typical handling\"],[\"Current vs superseded version\",\"Use current version for present state; retain older version as historical evidence.\"],[\"System of record vs stale replica\",\"Use system of record; flag replication freshness issue.\"],[\"Contract vs CRM transcription\",\"Executed contract governs contractual wording; CRM discrepancy becomes a correction task.\"],[\"Documentation vs deployed behavior\",\"Distinguish intended behavior from observed\u002Fdeployed behavior; do not silently merge them.\"],[\"Two credible primary sources\",\"Preserve both, assess provenance and scope, and represent unresolved disagreement if no governing authority exists.\"],[\"User memory vs current user setting\",\"Use current explicit setting; mark memory as superseded where appropriate.\"]]},\"tunes\":{}},{\"id\":\"h-web\",\"type\":\"header\",\"data\":{\"text\":\"Web search is discovery, not automatically evidence\",\"level\":2},\"tunes\":{}},{\"id\":\"p-web-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Search engines are excellent discovery systems. Search snippets, result ranking and generated summaries are not automatically primary evidence.\"},\"tunes\":{}},{\"id\":\"p-web-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For claims that require authority, the search result should lead to the original publication, official record, source document, dataset or other appropriate artifact. The result page helps locate the source; it does not inherit the source's authority.\"},\"tunes\":{}},{\"id\":\"h-ai\",\"type\":\"header\",\"data\":{\"text\":\"The language model should not decide authority by itself\",\"level\":2},\"tunes\":{}},{\"id\":\"p-ai-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A model can help classify a question, extract claims or compare evidence, but authority should not depend only on the model's preference. Models optimize generation from context; they do not possess a guaranteed domain-specific registry of which database, document or organization owns each fact.\"},\"tunes\":{}},{\"id\":\"p-ai-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"This is why application architecture should encode critical authority rules deterministically where practical. The model may reason within the boundary, but the boundary itself should not be recreated from scratch for every prompt.\"},\"tunes\":{}},{\"id\":\"h-observability\",\"type\":\"header\",\"data\":{\"text\":\"Authority must survive the execution trace\",\"level\":2},\"tunes\":{}},{\"id\":\"p-observability-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"If a production answer is important enough to audit, the trace should make it possible to reconstruct which sources were consulted, which version was used, which passage or record supported the claim, which transformations occurred and whether conflicting evidence was available.\"},\"tunes\":{}},{\"id\":\"p-observability-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"This aligns with the broader provenance principle in W3C PROV and with NIST AI RMF Playbook guidance to document sources, origins, transformations, dependencies, constraints and metadata.\"},\"tunes\":{}},{\"id\":\"h-implementation\",\"type\":\"header\",\"data\":{\"text\":\"Original implementation evidence: Source of Truth Research Engine\",\"level\":2},\"tunes\":{}},{\"id\":\"implementation-note\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"Original implementation evidence\",\"body\":\"The Source of Truth Research Engine is my own implementation evidence for an evidence-first architecture. It demonstrates one concrete way to separate discovery, source acquisition, provenance, claims, evidence classes, contradictions and conclusions. It is an implementation pattern, not a universal industry standard.\"},\"tunes\":{}},{\"id\":\"p-impl-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The engine is designed around a traceable pipeline rather than direct AI summarization: research task → search → original source or digital artifact → local snapshot → SHA-256 → source ID → claim → evidence class → relation or contradiction → interpretation → conclusion.\"},\"tunes\":{}},{\"id\":\"p-impl-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Its shared evidence core stores Sources, Artifacts, provenance, Claims, Relations, Contradictions, a Reference Model and an audit trail. Different research modes can share that core while applying different domain methodologies.\"},\"tunes\":{}},{\"id\":\"p-impl-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"The architecture deliberately separates discovery from evidence. Search snippets are not treated as evidence, filenames are not treated as content, AI summaries are not treated as primary sources, and semantic similarity is only a discovery signal until a result is traced back to a concrete source and locator.\"},\"tunes\":{}},{\"id\":\"p-impl-4\",\"type\":\"paragraph\",\"data\":{\"text\":\"Original files are preserved and local bytes receive SHA-256 identifiers. Contradictions and rejected hypotheses are not silently deleted. New evidence is allowed to change the current reference model while the prior evidence path remains auditable.\"},\"tunes\":{}},{\"id\":\"implementation-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Implemented rule\",\"Why it matters for Source-of-Truth architecture\"],[\"Search ≠ evidence\",\"Discovery ranking cannot silently become authority.\"],[\"Filename ≠ content\",\"Metadata clues cannot substitute for reading the actual artifact.\"],[\"Local snapshot + SHA-256\",\"Evidence can be tied to exact bytes rather than a mutable remote label.\"],[\"Source ID + exact locator\",\"Claims can be traced back to the concrete evidence location.\"],[\"Claim\u002Fevidence separation\",\"The assertion is not confused with the material supporting it.\"],[\"Contradictions preserved\",\"The system can represent unresolved disagreement rather than overwrite history.\"],[\"Semantic similarity is discovery only\",\"Retrieval relevance is explicitly separated from evidentiary authority.\"],[\"New evidence may update the model\",\"Source-of-Truth state is versioned and revisable rather than treated as immutable dogma.\"]]},\"tunes\":{}},{\"id\":\"h-document-engine\",\"type\":\"header\",\"data\":{\"text\":\"Aaasaasa Document & Knowledge Engine: applying the same boundary to enterprise documents\",\"level\":2},\"tunes\":{}},{\"id\":\"p-doc-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Aaasaasa Document & Knowledge Engine concept extends the same design principle to enterprise documentation: users should be able to search documents, ask source-grounded questions and review collections against explicit criteria while preserving the distinction between what a document states and what the system infers.\"},\"tunes\":{}},{\"id\":\"p-doc-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The important architecture rule is that a universal retrieval core does not make every collection equally authoritative. Contract documents, maintenance records, financial documents and research material need different authority, validation and coverage rules even when they share ingestion and search infrastructure.\"},\"tunes\":{}},{\"id\":\"h-failures\",\"type\":\"header\",\"data\":{\"text\":\"Common Source-of-Truth failure modes\",\"level\":2},\"tunes\":{}},{\"id\":\"failures-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Failure mode\",\"What goes wrong\"],[\"The model is treated as the source of truth\",\"Parametric knowledge can be stale, incomplete, unverifiable or outside the application's authoritative scope.\"],[\"Top retrieval result wins automatically\",\"Similarity is mistaken for authority.\"],[\"Vector database becomes authoritative\",\"Derived index records lose the identity and version of the original source.\"],[\"Everything is copied into one knowledge base\",\"Copies obscure ownership, freshness and correction paths.\"],[\"Memory is reused as current state\",\"Old observations silently override the current system of record.\"],[\"No version metadata\",\"The right document is used for the wrong time period.\"],[\"No source locator\",\"A citation exists but the supporting passage or record cannot be verified.\"],[\"Conflicts are overwritten\",\"The system appears consistent by destroying evidence of disagreement.\"],[\"Generated summaries replace originals\",\"A lossy transformation becomes the apparent authority.\"],[\"Authority is global instead of claim-specific\",\"One source is trusted beyond the domain or fact class it actually owns.\"],[\"Web snippet is treated as evidence\",\"Discovery metadata replaces the original publication.\"],[\"Authoritative data is wrong but exceptions are hidden\",\"Operational defects become invisible and cannot be corrected transparently.\"]]},\"tunes\":{}},{\"id\":\"h-decision\",\"type\":\"header\",\"data\":{\"text\":\"A practical Source-of-Truth decision framework\",\"level\":2},\"tunes\":{}},{\"id\":\"decision-flow\",\"type\":\"processFlow\",\"data\":{\"title\":\"How to decide what should define a claim\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"1. State the claim precisely\",\"description\":\"Separate current state, historical state, policy, interpretation, prediction and derived calculation.\"},{\"label\":\"2. Identify the authority owner\",\"description\":\"Determine the system, document, institution, person or evidence class responsible for that claim type.\"},{\"label\":\"3. Define scope\",\"description\":\"Specify tenant, jurisdiction, product, environment, user, document set or other applicability boundary.\"},{\"label\":\"4. Define time and version\",\"description\":\"Determine whether the claim requires current state, a historical snapshot or a specific standard\u002Frelease version.\"},{\"label\":\"5. Preserve provenance\",\"description\":\"Record source identity, origin, locator, transformations and responsible agents or processes.\"},{\"label\":\"6. Define retrieval\u002Faccess path\",\"description\":\"Make sure the application can actually obtain the authoritative information under the correct identity and permissions.\"},{\"label\":\"7. Define conflict policy\",\"description\":\"Decide precedence, supersession, adjudication or explicit unresolved-state behavior.\"},{\"label\":\"8. Define invalidation\",\"description\":\"Specify when cached, indexed, remembered or derived representations must be refreshed.\"},{\"label\":\"9. Validate the answer path\",\"description\":\"Verify that important generated claims can be traced to the intended authority, not merely to a plausible source.\"}]},\"tunes\":{}},{\"id\":\"h-checklist\",\"type\":\"header\",\"data\":{\"text\":\"Source-of-Truth architecture checklist\",\"level\":2},\"tunes\":{}},{\"id\":\"checklist-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Question\",\"Expected answer\"],[\"What exact fact or state is being established?\",\"A claim precise enough to assign authority.\"],[\"Who or what owns that fact?\",\"Named authoritative system, source class or adjudication rule.\"],[\"Is the authority current for this scope?\",\"Tenant, jurisdiction, environment, user or domain boundary is explicit.\"],[\"Is the version\u002Ftime correct?\",\"Current, historical or version-specific applicability is known.\"],[\"Can the source be verified?\",\"Stable identifier, locator or record reference exists.\"],[\"Is provenance preserved?\",\"Origin, transformation and responsibility metadata survive ingestion and retrieval.\"],[\"Can retrieval return non-authoritative material?\",\"If yes, filters or validation distinguish relevance from authority.\"],[\"Can the source change?\",\"Refresh, invalidation or supersession rules exist.\"],[\"Can sources disagree?\",\"Conflict and adjudication behavior is explicit.\"],[\"Can memory become stale?\",\"Volatile state is re-read from the current authority before consequential use.\"],[\"Can a derived answer be reproduced?\",\"Inputs, transformation version and execution conditions are traceable.\"],[\"Can an auditor reconstruct the answer?\",\"Execution evidence preserves the source path for important claims.\"]]},\"tunes\":{}},{\"id\":\"h-misconceptions\",\"type\":\"header\",\"data\":{\"text\":\"Common misconceptions\",\"level\":2},\"tunes\":{}},{\"id\":\"misconceptions-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Misconception\",\"Correction\"],[\"“Source of Truth means one database.”\",\"One database can be authoritative for one domain; complex systems usually have multiple fact-specific authorities.\"],[\"“The newest document is automatically authoritative.”\",\"Recency helps only when the newer artifact is approved and actually supersedes the older one.\"],[\"“RAG solves truth.”\",\"RAG solves retrieval. Authority, provenance, evidence quality and validity remain separate problems.\"],[\"“A citation proves the answer.”\",\"The cited source must actually support the claim, have the right authority and apply to the current scope.\"],[\"“Provenance tells us what is true.”\",\"Provenance tells us origin and derivation; authority and correctness still require domain rules and evaluation.\"],[\"“The system of record is always correct.”\",\"It is authoritative for the operational record, but data-quality defects can still exist and require visible correction.\"],[\"“If several sources agree, the claim is authoritative.”\",\"Agreement increases evidence but does not necessarily establish ownership or applicability.\"],[\"“AI memory can replace repeated reads.”\",\"Only for information whose staleness risk is acceptable; volatile or consequential state should be refreshed from authority.\"]]},\"tunes\":{}},{\"id\":\"h-edge\",\"type\":\"header\",\"data\":{\"text\":\"Edge cases\",\"level\":2},\"tunes\":{}},{\"id\":\"p-edge-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Some questions are interpretive rather than factual. “Which architecture is best?” has no single Source of Truth. The system can retrieve authoritative constraints and evidence, but the final judgment is an inference that should expose assumptions and trade-offs.\"},\"tunes\":{}},{\"id\":\"p-edge-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Some domains use distributed authority. A scientific conclusion can depend on multiple studies, datasets and replications. A historical conclusion can depend on conflicting primary and secondary evidence. The architecture should represent evidence structure rather than invent a central database that supposedly owns truth.\"},\"tunes\":{}},{\"id\":\"p-edge-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"A user can also be the authority for subjective personal information: preferences, goals, chosen settings or explicit instructions. Even then, newer explicit input can supersede older memory.\"},\"tunes\":{}},{\"id\":\"p-edge-4\",\"type\":\"paragraph\",\"data\":{\"text\":\"External events can invalidate previously authoritative data. A price feed, inventory system or security policy may have been correct when captured but no longer valid. Snapshot provenance preserves what was true then; it does not make the snapshot current forever.\"},\"tunes\":{}},{\"id\":\"h-limit\",\"type\":\"header\",\"data\":{\"text\":\"Limitations\",\"level\":2},\"tunes\":{}},{\"id\":\"p-limit-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Source-of-Truth architecture cannot guarantee that an authoritative source is factually correct. It provides accountability, provenance and deterministic ownership boundaries; data-quality and domain-verification processes remain necessary.\"},\"tunes\":{}},{\"id\":\"p-limit-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Authority can also be contested. Different institutions can legitimately claim authority in different jurisdictions or methodologies. In those situations the system should expose the authority model and disagreement rather than hide it behind a universal “truth score.”\"},\"tunes\":{}},{\"id\":\"p-limit-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"Finally, authority rules require maintenance. Systems, owners, policies, versions and regulations change. A stale authority registry can be as dangerous as no registry at all.\"},\"tunes\":{}},{\"id\":\"h-change\",\"type\":\"header\",\"data\":{\"text\":\"What would change this answer?\",\"level\":2},\"tunes\":{}},{\"id\":\"p-change-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The specific authority mapping changes with the domain. Banking, healthcare, software delivery, scientific research and historical analysis have different systems of record, evidentiary rules and regulatory obligations.\"},\"tunes\":{}},{\"id\":\"p-change-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The implementation also changes with architecture. A small application can encode authority directly in service calls. A larger platform may need registries, source metadata, policy engines, lineage systems or data contracts. The core principle remains the same: do not let retrieval order or model preference silently decide what counts as authoritative.\"},\"tunes\":{}},{\"id\":\"h-related\",\"type\":\"header\",\"data\":{\"text\":\"Related canonical knowledge\",\"level\":2},\"tunes\":{}},{\"id\":\"p-related-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Source-of-Truth architecture is a prerequisite for later retrieval and governance concepts because retrieval quality alone cannot determine whether evidence is allowed to define the answer.\"},\"tunes\":{}},{\"id\":\"ref-rag\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works\",\"title\":\"What Is RAG? The Simplest Explanation of How It Works\",\"excerpt\":\"RAG retrieves external knowledge for a model. This foundation explains why retrieval and authoritative truth are separate responsibilities.\",\"ctaLabel\":\"Read the RAG foundation\"},\"tunes\":{}},{\"id\":\"p-related-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Memory is another adjacent concept. A reliable agent separates remembered information from current authoritative application state.\"},\"tunes\":{}},{\"id\":\"ref-memory\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fai-agent-memory-is-not-rag-how-to-separate-memory-retrieval-state-and-context\",\"title\":\"AI Agent Memory Is Not RAG: How to Separate Memory, Retrieval, State and Context\",\"excerpt\":\"A four-layer architecture separating persistent memory, authoritative state, retrieval and the context actually supplied to the model.\",\"ctaLabel\":\"Read the memory architecture article\"},\"tunes\":{}},{\"id\":\"p-related-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"Authority also connects directly to answer validity. Even an authoritative source supports only claims inside its version, date, scope and evidence boundary.\"},\"tunes\":{}},{\"id\":\"ref-avb\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers\",\"title\":\"The Answer Validity Boundary: The Missing Layer Between Relevance and Reliable AI Answers\",\"excerpt\":\"A framework for making explicit where an AI answer remains supported by evidence and which changed conditions invalidate that support.\",\"ctaLabel\":\"Read the Answer Validity Boundary\"},\"tunes\":{}},{\"id\":\"h-faq\",\"type\":\"header\",\"data\":{\"text\":\"Frequently asked questions\",\"level\":2},\"tunes\":{}},{\"id\":\"faq\",\"type\":\"faq\",\"data\":{\"title\":\"Source of Truth in AI systems\",\"items\":[{\"id\":\"faq1\",\"question\":\"What is a Source of Truth in an AI system?\",\"answer\":\"It is the authoritative source or authority rule that determines which source is allowed to establish a specific fact, state or rule for a defined scope, version and time.\"},{\"id\":\"faq2\",\"question\":\"Is the language model a Source of Truth?\",\"answer\":\"Normally no. A language model can generate, summarize and reason, but its parametric knowledge is not automatically authoritative for current application state, company policy, a specific document version or a regulated domain fact.\"},{\"id\":\"faq3\",\"question\":\"Is a vector database the Source of Truth for RAG?\",\"answer\":\"Not automatically. A vector database is usually an index or retrieval store. It should preserve references to the authoritative original source and version rather than silently replacing them.\"},{\"id\":\"faq4\",\"question\":\"What is the difference between provenance and Source of Truth?\",\"answer\":\"Provenance describes where data came from, how it was produced or transformed and who or what was involved. Source-of-Truth rules determine whether that source has authority for the specific claim.\"},{\"id\":\"faq5\",\"question\":\"Can an AI system have multiple Sources of Truth?\",\"answer\":\"Yes. In complex systems this is normal because different facts belong to different authoritative systems or source classes.\"},{\"id\":\"faq6\",\"question\":\"What happens when two authoritative sources disagree?\",\"answer\":\"The system needs a domain-specific conflict rule: precedence, version supersession, expert adjudication or an explicit unresolved state. It should not silently choose the source the model prefers.\"},{\"id\":\"faq7\",\"question\":\"Does RAG guarantee that an AI answer uses the Source of Truth?\",\"answer\":\"No. RAG retrieves candidates. Authority-aware metadata, filters, source policy and validation are needed to ensure that important claims use the correct source.\"},{\"id\":\"faq8\",\"question\":\"Can the Source of Truth be wrong?\",\"answer\":\"Yes. Authority and correctness are different properties. An authoritative system can contain a data-quality defect, which should be corrected transparently rather than hidden by substituting an unofficial source.\"}]},\"tunes\":{}},{\"id\":\"h-glossary\",\"type\":\"header\",\"data\":{\"text\":\"Glossary\",\"level\":2},\"tunes\":{}},{\"id\":\"glossary\",\"type\":\"glossary\",\"data\":{\"title\":\"Key Source-of-Truth terms\",\"entries\":[{\"term\":\"Source of Truth\",\"definition\":\"The authoritative source or rule permitted to establish a particular fact, state or rule for a defined scope, version and time.\",\"anchor\":\"source-of-truth\"},{\"term\":\"System of record\",\"definition\":\"The authoritative operational system responsible for a defined class of records or current business state.\",\"anchor\":\"system-of-record\"},{\"term\":\"Provenance\",\"definition\":\"Information describing the origin, derivation, transformations, responsible agents and history of data or another entity.\",\"anchor\":\"provenance\"},{\"term\":\"Evidence\",\"definition\":\"Information or an artifact that supports, contradicts or constrains a claim.\",\"anchor\":\"evidence\"},{\"term\":\"Authority\",\"definition\":\"The application or domain rule determining which source is entitled to define a specific claim.\",\"anchor\":\"authority\"},{\"term\":\"Freshness\",\"definition\":\"Whether a representation remains current enough for the claim or operation in which it is used.\",\"anchor\":\"freshness\"},{\"term\":\"Supersession\",\"definition\":\"The explicit replacement of an older authoritative version by a newer one while preserving historical traceability.\",\"anchor\":\"supersession\"},{\"term\":\"Ground truth\",\"definition\":\"A reference answer, label or outcome used to evaluate a system; it is an evaluation construct rather than automatically the application's Source of Truth.\",\"anchor\":\"ground-truth\"},{\"term\":\"Lineage\",\"definition\":\"The trace of how data or derived values flow and transform across sources and processing steps.\",\"anchor\":\"lineage\"},{\"term\":\"Validity boundary\",\"definition\":\"The conditions of scope, time, version, evidence and assumptions inside which a claim remains supported.\",\"anchor\":\"validity-boundary\"}]},\"tunes\":{}},{\"id\":\"h-conclusion\",\"type\":\"header\",\"data\":{\"text\":\"Conclusion\",\"level\":2},\"tunes\":{}},{\"id\":\"p-conclusion-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Reliable AI does not come from giving the model more information. It comes from knowing which information is allowed to define the claim, preserving where that information came from, retrieving the correct version and keeping the final answer inside the source's scope.\"},\"tunes\":{}},{\"id\":\"p-conclusion-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That is why Source of Truth, provenance, retrieval, memory and context must remain separate concepts. The Source of Truth defines authority. Provenance explains origin. Retrieval finds candidates. Memory preserves selected past information. Context is what the model sees. Generation turns those inputs into an output.\"},\"tunes\":{}},{\"id\":\"p-conclusion-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"When those layers remain explicit, an AI system can do more than sound plausible: important claims can be traced back to the source that actually had the right to establish them.\"},\"tunes\":{}},{\"id\":\"h-sources\",\"type\":\"header\",\"data\":{\"text\":\"Primary sources and implementation evidence\",\"level\":2},\"tunes\":{}},{\"id\":\"p-sources-note\",\"type\":\"paragraph\",\"data\":{\"text\":\"The external sources below support provenance and AI risk-management claims. The Source of Truth Research Engine section is original implementation evidence and is explicitly presented as one implementation pattern rather than a universal standard.\"},\"tunes\":{}},{\"id\":\"src-w3c-prov\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-dm\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"W3C PROV-DM — The PROV Data Model\",\"description\":\"W3C Recommendation defining a domain-agnostic provenance model around entities, activities, agents, derivations and responsibility.\"}},\"tunes\":{}},{\"id\":\"src-w3c-publications\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.w3.org\u002Fgroups\u002Fwg\u002Fprov\u002Fpublications\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"W3C Provenance Working Group — Publications\",\"description\":\"Official index of W3C PROV Recommendations and related specifications for provenance interchange and constraints.\"}},\"tunes\":{}},{\"id\":\"src-nist-rmf\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.nist.gov\u002Fitl\u002Fai-risk-management-framework\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"NIST AI Risk Management Framework\",\"description\":\"NIST's voluntary framework for incorporating trustworthiness and risk-management considerations across the AI lifecycle; AI RMF 1.0 is currently being revised.\"}},\"tunes\":{}},{\"id\":\"src-nist-playbook\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fairc.nist.gov\u002Fairmf-resources\u002Fplaybook\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"NIST AI RMF Playbook\",\"description\":\"Operational guidance aligned to the AI RMF, including documentation practices for data provenance, sources, origins, transformations, dependencies, constraints and metadata.\"}},\"tunes\":{}},{\"id\":\"src-nist-measure\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fairc.nist.gov\u002Fairmf-resources\u002Fplaybook\u002Fmeasure\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"NIST AI RMF Playbook — Measure\",\"description\":\"Guidance on documenting measurement, data provenance and contextual interpretation of AI system outputs.\"}},\"tunes\":{}},{\"id\":\"src-nist-genai\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.nist.gov\u002Fpublications\u002Fartificial-intelligence-risk-management-framework-generative-artificial-intelligence\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"NIST AI 600-1 — Generative AI Profile\",\"description\":\"NIST generative-AI profile, including provenance and information-integrity considerations for generative AI systems.\"}},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":1336,"blocks":1337,"version":2242},1791479104597,[1338,1342,1347,1352,1357,1361,1365,1369,1373,1377,1382,1386,1390,1394,1398,1422,1426,1430,1434,1438,1474,1478,1482,1486,1490,1494,1498,1502,1506,1510,1514,1518,1522,1526,1530,1534,1538,1542,1546,1550,1554,1558,1562,1566,1570,1574,1578,1582,1586,1590,1594,1599,1628,1632,1636,1668,1672,1676,1680,1685,1689,1693,1697,1701,1705,1709,1714,1718,1722,1726,1751,1755,1759,1763,1767,1771,1775,1779,1783,1787,1791,1796,1800,1804,1808,1812,1843,1847,1851,1855,1859,1902,1906,1938,1942,1984,1988,2019,2023,2027,2031,2035,2039,2043,2047,2051,2055,2059,2063,2067,2071,2075,2082,2086,2093,2097,2104,2108,2137,2141,2176,2180,2184,2188,2192,2196,2200,2207,2214,2221,2228,2235],{"id":215,"data":1339,"type":218,"tunes":1341},{"text":1340},"A Source of Truth in an AI system is the authoritative source allowed to define whether a specific fact, state or rule should be treated as true for a particular scope, version and time. It is not automatically the language model, the vector database, the top-ranked retrieved document, agent memory or the latest message in context. Reliable AI architecture must preserve which source has authority for which claim, then keep provenance, retrieval and validation connected to that authority.",{},{"id":221,"data":1343,"type":226,"tunes":1346},{"body":1344,"title":1345,"variant":225},"\u003Cstrong>The Source of Truth is an authority rule, not an AI component.\u003C\u002Fstrong> An AI system can search many sources, but only some sources are authoritative for a given question. The correct source may be an application database for current account state, an approved policy document for a rule, source code for implemented behavior, an official specification for a protocol, or a primary research artifact for an observed claim. Good architecture makes that authority explicit instead of asking the model to infer it from relevance.","Direct answer",{},{"id":229,"data":1348,"type":226,"tunes":1351},{"body":1349,"title":1350,"variant":233},"“Source of Truth” is widely used in software, data and organizational practice, but it is not a standardized AI component with one universal formal definition. This article uses it as an architecture concept: \u003Cstrong>the source or authority mapping that determines which evidence is allowed to establish a particular fact or state\u003C\u002Fstrong>. Provenance standards such as W3C PROV describe origin and derivation; they do not automatically decide which source your application should treat as authoritative.","Terminology boundary",{},{"id":236,"data":1353,"type":226,"tunes":1356},{"body":1354,"title":1355,"variant":233},"The core distinction between authority, provenance, retrieval and model context is stable. NIST AI RMF 1.0 is currently being revised, but the current NIST Playbook still explicitly recommends documenting data provenance, including sources, origins, transformations, dependencies, constraints and metadata. W3C PROV remains a W3C Recommendation for interoperable provenance descriptions.","Current-source note — 8 October 2026",{},{"id":242,"data":1358,"type":247,"tunes":1360},{"title":1359,"maxLevel":245,"minLevel":246},"Contents",{},{"id":250,"data":1362,"type":42,"tunes":1364},{"text":1363,"level":246},"What “Source of Truth” really means",{},{"id":255,"data":1366,"type":218,"tunes":1368},{"text":1367},"The phrase is often misunderstood as “the one database that contains everything.” That can be true in a narrow system, but it is usually too simplistic for AI. A real AI application can combine operational databases, documents, APIs, vector indexes, user input, model memory, external web sources and generated summaries.",{},{"id":260,"data":1370,"type":218,"tunes":1372},{"text":1371},"Those sources do not have equal authority. A customer-support manual may define policy but not a customer's current balance. A CRM may define the current account owner but not the legal meaning of a regulation. A source-code repository may define implemented behavior while a product specification defines intended behavior. The architecture must therefore answer a more precise question: which source is authoritative for this specific claim?",{},{"id":265,"data":1374,"type":218,"tunes":1376},{"text":1375},"This makes Source of Truth a relationship between a claim and an authority, not merely a property of a storage technology.",{},{"id":270,"data":1378,"type":226,"tunes":1381},{"body":1379,"title":1380,"variant":274},"\u003Cstrong>One system can have many Sources of Truth because authority is fact-specific.\u003C\u002Fstrong> The important architectural property is not that one database wins globally, but that each important fact or state has an explicit authoritative owner.","Core principle",{},{"id":277,"data":1383,"type":42,"tunes":1385},{"text":1384,"level":246},"The simplest example",{},{"id":282,"data":1387,"type":218,"tunes":1389},{"text":1388},"A user asks an AI assistant: “What is my current subscription plan?” The assistant has three possible inputs: last month's support transcript, an indexed help-center document describing plan types, and the live billing database.",{},{"id":287,"data":1391,"type":218,"tunes":1393},{"text":1392},"The support transcript may mention that the user had a Pro plan. The help-center document explains what Pro means. But the live billing record is the authoritative source for the user's current subscription state.",{},{"id":292,"data":1395,"type":218,"tunes":1397},{"text":1396},"A semantic search engine could rank the support transcript above the billing record because it contains language closer to the question. That ranking would still not make the transcript authoritative. Relevance and authority are different dimensions.",{},{"id":297,"data":1399,"type":329,"tunes":1421},{"rows":1400,"title":1413,"layout":318,"columns":1414},[1401,1404,1407,1410],{"id":301,"label":1402,"values":1403},"Billing database",[304,304,304],{"id":306,"label":1405,"values":1406},"Help-center documentation",[304,304,304],{"id":310,"label":1408,"values":1409},"Old support transcript",[304,304,304],{"id":314,"label":1411,"values":1412},"Model memory",[304,304,304],"The same question can involve different source roles",[1415,1417,1419],{"id":321,"label":1416},"Source",{"id":324,"label":1418},"Role",{"id":327,"label":1420},"Authority for current plan?",{},{"id":332,"data":1423,"type":42,"tunes":1425},{"text":1424,"level":246},"Where the simple example stops",{},{"id":337,"data":1427,"type":218,"tunes":1429},{"text":1428},"Not every domain has one unquestioned authority. Historical research can contain conflicting primary sources. Scientific claims can evolve as new studies appear. Legal interpretation can depend on jurisdiction, date and court authority. Product behavior can differ between documentation and deployed code.",{},{"id":342,"data":1431,"type":218,"tunes":1433},{"text":1432},"In these cases the correct architecture is not to invent a single winner. It is to preserve the competing sources, their provenance, their authority class, their applicable scope and the unresolved contradiction. A reliable Source-of-Truth system must be able to represent uncertainty and disagreement.",{},{"id":347,"data":1435,"type":42,"tunes":1437},{"text":1436,"level":246},"Authority is scoped by claim, version and time",{},{"id":352,"data":1439,"type":318,"tunes":1473},{"content":1440,"stretched":43,"withHeadings":14},[1441,1445,1449,1453,1457,1461,1465,1469],[1442,1443,1444],"Question","Possible authoritative source","Why scope matters",[1446,1447,1448],"What is the user's current account balance?","Ledger \u002F accounting system of record","Historical exports may be accurate for an earlier time but not current state.",[1450,1451,1452],"What does company policy currently allow?","Approved current policy version","An older policy may remain valid evidence of past rules but not present rules.",[1454,1455,1456],"What code is actually deployed?","Deployment artifact \u002F commit \u002F release record","Main branch may differ from production.",[1458,1459,1460],"What did a contract state when signed?","Executed contract version","A draft or later template is not authoritative for the signed agreement.",[1462,1463,1464],"What does a technical protocol specify?","Current official specification for the relevant version","A blog explanation may be useful but is secondary evidence.",[1466,1467,1468],"What happened in a historical event?","Relevant primary evidence plus explicit source criticism","There may be no single authority; conflicting evidence must remain visible.",[1470,1471,1472],"What does a user prefer?","Current explicit user setting or confirmed preference","Old conversation memory may be stale or superseded.",{},{"id":389,"data":1475,"type":218,"tunes":1477},{"text":1476},"The word “truth” can therefore be misleading unless its boundary is stated. In architecture, the Source of Truth is usually better understood as the source authorized to determine a specific proposition under defined conditions.",{},{"id":394,"data":1479,"type":42,"tunes":1481},{"text":1480,"level":246},"What a Source of Truth is not",{},{"id":399,"data":1483,"type":42,"tunes":1485},{"text":1484,"level":245},"Source of Truth vs system of record",{},{"id":404,"data":1487,"type":218,"tunes":1489},{"text":1488},"A system of record is typically the authoritative operational system for a class of records: for example, a billing ledger, HR master record or order database. It is one common implementation of Source-of-Truth authority.",{},{"id":409,"data":1491,"type":218,"tunes":1493},{"text":1492},"But Source of Truth is broader. A signed PDF contract, an official standard, a deployment artifact or a primary archival document may be authoritative without being a transactional system of record.",{},{"id":414,"data":1495,"type":42,"tunes":1497},{"text":1496,"level":245},"Source of Truth vs provenance",{},{"id":419,"data":1499,"type":218,"tunes":1501},{"text":1500},"Provenance answers questions such as: Where did this data come from? Who or what produced it? Which transformation created this derivative? Which prior entity was used? W3C PROV models entities, activities, agents and derivations so origin and responsibility can be represented.",{},{"id":424,"data":1503,"type":218,"tunes":1505},{"text":1504},"Provenance does not by itself establish authority. Knowing that a value came from a spreadsheet written by a specific employee helps evaluate it, but the application still needs a rule saying whether that spreadsheet is authoritative for the claim.",{},{"id":429,"data":1507,"type":42,"tunes":1509},{"text":1508,"level":245},"Source of Truth vs evidence",{},{"id":434,"data":1511,"type":218,"tunes":1513},{"text":1512},"Evidence supports or contradicts a claim. A Source of Truth defines which source has the authority to settle or strongly constrain that claim in the current application context.",{},{"id":439,"data":1515,"type":218,"tunes":1517},{"text":1516},"A source can be valuable evidence without being authoritative. Five customer emails may be evidence that users dislike a workflow, but they are not the system of record for current product configuration.",{},{"id":444,"data":1519,"type":42,"tunes":1521},{"text":1520,"level":245},"Source of Truth vs RAG",{},{"id":449,"data":1523,"type":218,"tunes":1525},{"text":1524},"RAG is a retrieval pattern. It finds information and supplies selected content to the model. RAG does not automatically know which source deserves authority.",{},{"id":454,"data":1527,"type":218,"tunes":1529},{"text":1528},"A RAG pipeline can retrieve a stale document, a secondary summary or a highly similar but non-authoritative source. Source authority must be encoded through corpus design, metadata, filters, ranking policy, validation or post-retrieval checks.",{},{"id":459,"data":1531,"type":42,"tunes":1533},{"text":1532,"level":245},"Source of Truth vs vector database",{},{"id":464,"data":1535,"type":218,"tunes":1537},{"text":1536},"A vector database stores or indexes representations used for semantic retrieval. It is an access layer, not automatically a truth layer.",{},{"id":469,"data":1539,"type":218,"tunes":1541},{"text":1540},"The same authoritative document may be chunked, embedded, copied and re-indexed many times. The vector record should retain a reference back to the authoritative source and version rather than becoming an untraceable new authority.",{},{"id":474,"data":1543,"type":42,"tunes":1545},{"text":1544,"level":245},"Source of Truth vs memory",{},{"id":479,"data":1547,"type":218,"tunes":1549},{"text":1548},"Agent or application memory stores information that may be useful later. Memory can preserve a prior decision, preference or observation, but it can become stale.",{},{"id":484,"data":1551,"type":218,"tunes":1553},{"text":1552},"For volatile or consequential state, a reliable agent should normally re-read the authoritative current source rather than assume that remembered state is still true.",{},{"id":489,"data":1555,"type":42,"tunes":1557},{"text":1556,"level":245},"Source of Truth vs context",{},{"id":494,"data":1559,"type":218,"tunes":1561},{"text":1560},"Context is what the model receives during the current inference. Authoritative information can be absent from context, while non-authoritative information can be present.",{},{"id":499,"data":1563,"type":218,"tunes":1565},{"text":1564},"Context construction therefore needs an authority-aware policy: retrieve or read the source that is allowed to define the claim, then preserve enough metadata for the model or validator to understand its scope.",{},{"id":504,"data":1567,"type":42,"tunes":1569},{"text":1568,"level":245},"Source of Truth vs evaluation ground truth",{},{"id":509,"data":1571,"type":218,"tunes":1573},{"text":1572},"Evaluation ground truth is the reference answer, label or outcome against which a system is scored. It can be derived from authoritative sources, expert adjudication or curated test data.",{},{"id":514,"data":1575,"type":218,"tunes":1577},{"text":1576},"Ground truth is therefore an evaluation construct. A Source of Truth is an application\u002Fdomain authority construct. They can overlap, but they are not interchangeable.",{},{"id":519,"data":1579,"type":42,"tunes":1581},{"text":1580,"level":245},"Source of Truth vs data quality",{},{"id":524,"data":1583,"type":218,"tunes":1585},{"text":1584},"An authoritative source can still contain errors. Authority says which source officially governs the fact; data quality asks whether that source is accurate, complete, timely, consistent and fit for purpose.",{},{"id":529,"data":1587,"type":218,"tunes":1589},{"text":1588},"When an authoritative system is known to be wrong, architecture should record the defect, correction process or exception instead of silently substituting an unofficial source and hiding the discrepancy.",{},{"id":534,"data":1591,"type":42,"tunes":1593},{"text":1592,"level":246},"A practical Source-of-Truth architecture model",{},{"id":539,"data":1595,"type":226,"tunes":1598},{"body":1596,"title":1597,"variant":233},"The model below is a practical synthesis for AI systems. It is not a W3C, NIST or ISO standard. It separates authority, provenance, retrieval and model-facing context because those responsibilities are often incorrectly collapsed.","Proposed architecture model",{},{"id":545,"data":1600,"type":574,"tunes":1627},{"steps":1601,"title":1626,"orientation":573},[1602,1605,1608,1611,1614,1617,1620,1623],{"label":1603,"description":1604},"1. Define the claim type","Identify what the user is actually asking: current state, policy, historical fact, technical specification, user preference, calculation or interpretation.",{"label":1606,"description":1607},"2. Resolve authority","Determine which source or authority class is permitted to define that type of claim for the required scope, version and time.",{"label":1609,"description":1610},"3. Acquire evidence","Read or retrieve the authoritative source and any necessary supporting or conflicting evidence.",{"label":1612,"description":1613},"4. Preserve provenance","Carry source identity, version, timestamp, locator, transformation history and responsibility metadata.",{"label":1615,"description":1616},"5. Build model context","Supply the relevant evidence to the model without discarding authority and applicability metadata.",{"label":1618,"description":1619},"6. Generate or compute","The model may summarize, compare, reason or transform the evidence, but it does not inherit source authority merely by processing it.",{"label":1621,"description":1622},"7. Validate the claim","Check that the answer is supported by the right source and remains inside its scope and validity boundary.",{"label":1624,"description":1625},"8. Preserve contradictions","If authoritative or relevant sources disagree, expose the conflict rather than fabricating false certainty.","Authority-aware AI answer path",{},{"id":577,"data":1629,"type":42,"tunes":1631},{"text":1630,"level":246},"Authority should be explicit, not inferred from similarity",{},{"id":582,"data":1633,"type":218,"tunes":1635},{"text":1634},"One robust implementation pattern is an authority registry or equivalent policy layer that maps claim classes to authoritative source classes. The implementation can be code, metadata, configuration or domain rules; the important property is that authority is deliberate.",{},{"id":587,"data":1637,"type":318,"tunes":1667},{"content":1638,"stretched":43,"withHeadings":14},[1639,1643,1647,1651,1655,1659,1663],[1640,1641,1642],"Claim class","Authority rule","Fallback behavior",[1644,1645,1646],"Current account state","Read live account service \u002F system of record","If unavailable, report that current state cannot be verified.",[1648,1649,1650],"Product documentation","Current approved documentation version","Older version may be shown only with version warning.",[1652,1653,1654],"Implemented software behavior","Relevant deployed release \u002F source artifact","Documentation alone cannot prove deployed behavior.",[1656,1657,1658],"Internal policy","Approved policy repository and active version","Drafts are supporting material, not current authority.",[1660,1661,1662],"External technical standard","Official standards body publication for relevant version","Secondary explanations can clarify but not override the specification.",[1664,1665,1666],"Research claim","Evidence policy appropriate to the domain","Preserve conflicting evidence and confidence rather than force one source.",{},{"id":620,"data":1669,"type":42,"tunes":1671},{"text":1670,"level":246},"Retrieval should use authority as a ranking constraint",{},{"id":625,"data":1673,"type":218,"tunes":1675},{"text":1674},"Semantic relevance answers “Which candidate looks related to this query?” Authority answers “Which candidate is allowed to establish this fact?” A production retrieval system often needs both.",{},{"id":630,"data":1677,"type":218,"tunes":1679},{"text":1678},"A useful sequence is to first constrain the candidate space by identity, tenant, source class, status, version or date, then rank relevant evidence inside the permitted space. If relevance is calculated before critical authorization or authority filters, the pipeline can return a convincing but invalid result.",{},{"id":635,"data":1681,"type":226,"tunes":1684},{"body":1682,"title":1683,"variant":639},"A search engine can rank a stale summary above the current primary source. A language model can prefer the better-written explanation over the official record. Neither behavior changes which source owns the fact.","Highest similarity is not highest authority",{},{"id":642,"data":1686,"type":42,"tunes":1688},{"text":1687,"level":246},"Freshness is part of authority",{},{"id":647,"data":1690,"type":218,"tunes":1692},{"text":1691},"Many Source-of-Truth failures are actually time failures. The correct source was known, but the system used an old snapshot, stale embedding, cached API response or superseded document.",{},{"id":652,"data":1694,"type":218,"tunes":1696},{"text":1695},"An authority rule should therefore include invalidation or refresh semantics where the fact can change. “CRM is authoritative” is incomplete when the application reads a week-old replicated export.",{},{"id":657,"data":1698,"type":42,"tunes":1700},{"text":1699,"level":246},"Derived values need a trace back to authoritative inputs",{},{"id":662,"data":1702,"type":218,"tunes":1704},{"text":1703},"Some important facts are not stored directly. They are computed from authoritative inputs: a risk score, account total, eligibility status or aggregate metric.",{},{"id":667,"data":1706,"type":218,"tunes":1708},{"text":1707},"For derived values, Source-of-Truth architecture should preserve the input authorities, transformation or calculation version and execution time. W3C PROV's distinction between entities, activities and derivations is useful here because it models how one entity was produced from others.",{},{"id":672,"data":1710,"type":226,"tunes":1713},{"body":1711,"title":1712,"variant":274},"A calculated value is only as current and authoritative as its inputs, transformation and validity conditions. Store the lineage needed to recompute or audit it.","Derived truth is conditional",{},{"id":678,"data":1715,"type":42,"tunes":1717},{"text":1716,"level":246},"What happens when authoritative sources disagree?",{},{"id":683,"data":1719,"type":218,"tunes":1721},{"text":1720},"Conflicts are not edge cases in serious knowledge systems. A signed contract may disagree with a CRM field. Production behavior may disagree with documentation. Two primary historical sources may contradict each other. A current policy can conflict with an outdated local copy.",{},{"id":688,"data":1723,"type":218,"tunes":1725},{"text":1724},"The system needs a resolution policy appropriate to the domain. Sometimes one authority clearly outranks the other. Sometimes the newer version supersedes the old one. Sometimes an expert or business owner must adjudicate. And sometimes the correct result is simply: the evidence is unresolved.",{},{"id":693,"data":1727,"type":318,"tunes":1750},{"content":1728,"stretched":43,"withHeadings":14},[1729,1732,1735,1738,1741,1744,1747],[1730,1731],"Conflict type","Typical handling",[1733,1734],"Current vs superseded version","Use current version for present state; retain older version as historical evidence.",[1736,1737],"System of record vs stale replica","Use system of record; flag replication freshness issue.",[1739,1740],"Contract vs CRM transcription","Executed contract governs contractual wording; CRM discrepancy becomes a correction task.",[1742,1743],"Documentation vs deployed behavior","Distinguish intended behavior from observed\u002Fdeployed behavior; do not silently merge them.",[1745,1746],"Two credible primary sources","Preserve both, assess provenance and scope, and represent unresolved disagreement if no governing authority exists.",[1748,1749],"User memory vs current user setting","Use current explicit setting; mark memory as superseded where appropriate.",{},{"id":719,"data":1752,"type":42,"tunes":1754},{"text":1753,"level":246},"Web search is discovery, not automatically evidence",{},{"id":724,"data":1756,"type":218,"tunes":1758},{"text":1757},"Search engines are excellent discovery systems. Search snippets, result ranking and generated summaries are not automatically primary evidence.",{},{"id":729,"data":1760,"type":218,"tunes":1762},{"text":1761},"For claims that require authority, the search result should lead to the original publication, official record, source document, dataset or other appropriate artifact. The result page helps locate the source; it does not inherit the source's authority.",{},{"id":734,"data":1764,"type":42,"tunes":1766},{"text":1765,"level":246},"The language model should not decide authority by itself",{},{"id":739,"data":1768,"type":218,"tunes":1770},{"text":1769},"A model can help classify a question, extract claims or compare evidence, but authority should not depend only on the model's preference. Models optimize generation from context; they do not possess a guaranteed domain-specific registry of which database, document or organization owns each fact.",{},{"id":744,"data":1772,"type":218,"tunes":1774},{"text":1773},"This is why application architecture should encode critical authority rules deterministically where practical. The model may reason within the boundary, but the boundary itself should not be recreated from scratch for every prompt.",{},{"id":749,"data":1776,"type":42,"tunes":1778},{"text":1777,"level":246},"Authority must survive the execution trace",{},{"id":754,"data":1780,"type":218,"tunes":1782},{"text":1781},"If a production answer is important enough to audit, the trace should make it possible to reconstruct which sources were consulted, which version was used, which passage or record supported the claim, which transformations occurred and whether conflicting evidence was available.",{},{"id":759,"data":1784,"type":218,"tunes":1786},{"text":1785},"This aligns with the broader provenance principle in W3C PROV and with NIST AI RMF Playbook guidance to document sources, origins, transformations, dependencies, constraints and metadata.",{},{"id":764,"data":1788,"type":42,"tunes":1790},{"text":1789,"level":246},"Original implementation evidence: Source of Truth Research Engine",{},{"id":769,"data":1792,"type":226,"tunes":1795},{"body":1793,"title":1794,"variant":233},"The Source of Truth Research Engine is my own implementation evidence for an evidence-first architecture. It demonstrates one concrete way to separate discovery, source acquisition, provenance, claims, evidence classes, contradictions and conclusions. It is an implementation pattern, not a universal industry standard.","Original implementation evidence",{},{"id":775,"data":1797,"type":218,"tunes":1799},{"text":1798},"The engine is designed around a traceable pipeline rather than direct AI summarization: research task → search → original source or digital artifact → local snapshot → SHA-256 → source ID → claim → evidence class → relation or contradiction → interpretation → conclusion.",{},{"id":780,"data":1801,"type":218,"tunes":1803},{"text":1802},"Its shared evidence core stores Sources, Artifacts, provenance, Claims, Relations, Contradictions, a Reference Model and an audit trail. Different research modes can share that core while applying different domain methodologies.",{},{"id":785,"data":1805,"type":218,"tunes":1807},{"text":1806},"The architecture deliberately separates discovery from evidence. Search snippets are not treated as evidence, filenames are not treated as content, AI summaries are not treated as primary sources, and semantic similarity is only a discovery signal until a result is traced back to a concrete source and locator.",{},{"id":790,"data":1809,"type":218,"tunes":1811},{"text":1810},"Original files are preserved and local bytes receive SHA-256 identifiers. Contradictions and rejected hypotheses are not silently deleted. New evidence is allowed to change the current reference model while the prior evidence path remains auditable.",{},{"id":795,"data":1813,"type":318,"tunes":1842},{"content":1814,"stretched":43,"withHeadings":14},[1815,1818,1821,1824,1827,1830,1833,1836,1839],[1816,1817],"Implemented rule","Why it matters for Source-of-Truth architecture",[1819,1820],"Search ≠ evidence","Discovery ranking cannot silently become authority.",[1822,1823],"Filename ≠ content","Metadata clues cannot substitute for reading the actual artifact.",[1825,1826],"Local snapshot + SHA-256","Evidence can be tied to exact bytes rather than a mutable remote label.",[1828,1829],"Source ID + exact locator","Claims can be traced back to the concrete evidence location.",[1831,1832],"Claim\u002Fevidence separation","The assertion is not confused with the material supporting it.",[1834,1835],"Contradictions preserved","The system can represent unresolved disagreement rather than overwrite history.",[1837,1838],"Semantic similarity is discovery only","Retrieval relevance is explicitly separated from evidentiary authority.",[1840,1841],"New evidence may update the model","Source-of-Truth state is versioned and revisable rather than treated as immutable dogma.",{},{"id":827,"data":1844,"type":42,"tunes":1846},{"text":1845,"level":246},"Aaasaasa Document & Knowledge Engine: applying the same boundary to enterprise documents",{},{"id":832,"data":1848,"type":218,"tunes":1850},{"text":1849},"The Aaasaasa Document & Knowledge Engine concept extends the same design principle to enterprise documentation: users should be able to search documents, ask source-grounded questions and review collections against explicit criteria while preserving the distinction between what a document states and what the system infers.",{},{"id":837,"data":1852,"type":218,"tunes":1854},{"text":1853},"The important architecture rule is that a universal retrieval core does not make every collection equally authoritative. Contract documents, maintenance records, financial documents and research material need different authority, validation and coverage rules even when they share ingestion and search infrastructure.",{},{"id":842,"data":1856,"type":42,"tunes":1858},{"text":1857,"level":246},"Common Source-of-Truth failure modes",{},{"id":847,"data":1860,"type":318,"tunes":1901},{"content":1861,"stretched":43,"withHeadings":14},[1862,1865,1868,1871,1874,1877,1880,1883,1886,1889,1892,1895,1898],[1863,1864],"Failure mode","What goes wrong",[1866,1867],"The model is treated as the source of truth","Parametric knowledge can be stale, incomplete, unverifiable or outside the application's authoritative scope.",[1869,1870],"Top retrieval result wins automatically","Similarity is mistaken for authority.",[1872,1873],"Vector database becomes authoritative","Derived index records lose the identity and version of the original source.",[1875,1876],"Everything is copied into one knowledge base","Copies obscure ownership, freshness and correction paths.",[1878,1879],"Memory is reused as current state","Old observations silently override the current system of record.",[1881,1882],"No version metadata","The right document is used for the wrong time period.",[1884,1885],"No source locator","A citation exists but the supporting passage or record cannot be verified.",[1887,1888],"Conflicts are overwritten","The system appears consistent by destroying evidence of disagreement.",[1890,1891],"Generated summaries replace originals","A lossy transformation becomes the apparent authority.",[1893,1894],"Authority is global instead of claim-specific","One source is trusted beyond the domain or fact class it actually owns.",[1896,1897],"Web snippet is treated as evidence","Discovery metadata replaces the original publication.",[1899,1900],"Authoritative data is wrong but exceptions are hidden","Operational defects become invisible and cannot be corrected transparently.",{},{"id":891,"data":1903,"type":42,"tunes":1905},{"text":1904,"level":246},"A practical Source-of-Truth decision framework",{},{"id":896,"data":1907,"type":574,"tunes":1937},{"steps":1908,"title":1936,"orientation":573},[1909,1912,1915,1918,1921,1924,1927,1930,1933],{"label":1910,"description":1911},"1. State the claim precisely","Separate current state, historical state, policy, interpretation, prediction and derived calculation.",{"label":1913,"description":1914},"2. Identify the authority owner","Determine the system, document, institution, person or evidence class responsible for that claim type.",{"label":1916,"description":1917},"3. Define scope","Specify tenant, jurisdiction, product, environment, user, document set or other applicability boundary.",{"label":1919,"description":1920},"4. Define time and version","Determine whether the claim requires current state, a historical snapshot or a specific standard\u002Frelease version.",{"label":1922,"description":1923},"5. Preserve provenance","Record source identity, origin, locator, transformations and responsible agents or processes.",{"label":1925,"description":1926},"6. Define retrieval\u002Faccess path","Make sure the application can actually obtain the authoritative information under the correct identity and permissions.",{"label":1928,"description":1929},"7. Define conflict policy","Decide precedence, supersession, adjudication or explicit unresolved-state behavior.",{"label":1931,"description":1932},"8. Define invalidation","Specify when cached, indexed, remembered or derived representations must be refreshed.",{"label":1934,"description":1935},"9. Validate the answer path","Verify that important generated claims can be traced to the intended authority, not merely to a plausible source.","How to decide what should define a claim",{},{"id":929,"data":1939,"type":42,"tunes":1941},{"text":1940,"level":246},"Source-of-Truth architecture checklist",{},{"id":934,"data":1943,"type":318,"tunes":1983},{"content":1944,"stretched":43,"withHeadings":14},[1945,1947,1950,1953,1956,1959,1962,1965,1968,1971,1974,1977,1980],[1442,1946],"Expected answer",[1948,1949],"What exact fact or state is being established?","A claim precise enough to assign authority.",[1951,1952],"Who or what owns that fact?","Named authoritative system, source class or adjudication rule.",[1954,1955],"Is the authority current for this scope?","Tenant, jurisdiction, environment, user or domain boundary is explicit.",[1957,1958],"Is the version\u002Ftime correct?","Current, historical or version-specific applicability is known.",[1960,1961],"Can the source be verified?","Stable identifier, locator or record reference exists.",[1963,1964],"Is provenance preserved?","Origin, transformation and responsibility metadata survive ingestion and retrieval.",[1966,1967],"Can retrieval return non-authoritative material?","If yes, filters or validation distinguish relevance from authority.",[1969,1970],"Can the source change?","Refresh, invalidation or supersession rules exist.",[1972,1973],"Can sources disagree?","Conflict and adjudication behavior is explicit.",[1975,1976],"Can memory become stale?","Volatile state is re-read from the current authority before consequential use.",[1978,1979],"Can a derived answer be reproduced?","Inputs, transformation version and execution conditions are traceable.",[1981,1982],"Can an auditor reconstruct the answer?","Execution evidence preserves the source path for important claims.",{},{"id":977,"data":1985,"type":42,"tunes":1987},{"text":1986,"level":246},"Common misconceptions",{},{"id":982,"data":1989,"type":318,"tunes":2018},{"content":1990,"stretched":43,"withHeadings":14},[1991,1994,1997,2000,2003,2006,2009,2012,2015],[1992,1993],"Misconception","Correction",[1995,1996],"“Source of Truth means one database.”","One database can be authoritative for one domain; complex systems usually have multiple fact-specific authorities.",[1998,1999],"“The newest document is automatically authoritative.”","Recency helps only when the newer artifact is approved and actually supersedes the older one.",[2001,2002],"“RAG solves truth.”","RAG solves retrieval. Authority, provenance, evidence quality and validity remain separate problems.",[2004,2005],"“A citation proves the answer.”","The cited source must actually support the claim, have the right authority and apply to the current scope.",[2007,2008],"“Provenance tells us what is true.”","Provenance tells us origin and derivation; authority and correctness still require domain rules and evaluation.",[2010,2011],"“The system of record is always correct.”","It is authoritative for the operational record, but data-quality defects can still exist and require visible correction.",[2013,2014],"“If several sources agree, the claim is authoritative.”","Agreement increases evidence but does not necessarily establish ownership or applicability.",[2016,2017],"“AI memory can replace repeated reads.”","Only for information whose staleness risk is acceptable; volatile or consequential state should be refreshed from authority.",{},{"id":1014,"data":2020,"type":42,"tunes":2022},{"text":2021,"level":246},"Edge cases",{},{"id":1019,"data":2024,"type":218,"tunes":2026},{"text":2025},"Some questions are interpretive rather than factual. “Which architecture is best?” has no single Source of Truth. The system can retrieve authoritative constraints and evidence, but the final judgment is an inference that should expose assumptions and trade-offs.",{},{"id":1024,"data":2028,"type":218,"tunes":2030},{"text":2029},"Some domains use distributed authority. A scientific conclusion can depend on multiple studies, datasets and replications. A historical conclusion can depend on conflicting primary and secondary evidence. The architecture should represent evidence structure rather than invent a central database that supposedly owns truth.",{},{"id":1029,"data":2032,"type":218,"tunes":2034},{"text":2033},"A user can also be the authority for subjective personal information: preferences, goals, chosen settings or explicit instructions. Even then, newer explicit input can supersede older memory.",{},{"id":1034,"data":2036,"type":218,"tunes":2038},{"text":2037},"External events can invalidate previously authoritative data. A price feed, inventory system or security policy may have been correct when captured but no longer valid. Snapshot provenance preserves what was true then; it does not make the snapshot current forever.",{},{"id":1039,"data":2040,"type":42,"tunes":2042},{"text":2041,"level":246},"Limitations",{},{"id":1044,"data":2044,"type":218,"tunes":2046},{"text":2045},"Source-of-Truth architecture cannot guarantee that an authoritative source is factually correct. It provides accountability, provenance and deterministic ownership boundaries; data-quality and domain-verification processes remain necessary.",{},{"id":1049,"data":2048,"type":218,"tunes":2050},{"text":2049},"Authority can also be contested. Different institutions can legitimately claim authority in different jurisdictions or methodologies. In those situations the system should expose the authority model and disagreement rather than hide it behind a universal “truth score.”",{},{"id":1054,"data":2052,"type":218,"tunes":2054},{"text":2053},"Finally, authority rules require maintenance. Systems, owners, policies, versions and regulations change. A stale authority registry can be as dangerous as no registry at all.",{},{"id":1059,"data":2056,"type":42,"tunes":2058},{"text":2057,"level":246},"What would change this answer?",{},{"id":1064,"data":2060,"type":218,"tunes":2062},{"text":2061},"The specific authority mapping changes with the domain. Banking, healthcare, software delivery, scientific research and historical analysis have different systems of record, evidentiary rules and regulatory obligations.",{},{"id":1069,"data":2064,"type":218,"tunes":2066},{"text":2065},"The implementation also changes with architecture. A small application can encode authority directly in service calls. A larger platform may need registries, source metadata, policy engines, lineage systems or data contracts. The core principle remains the same: do not let retrieval order or model preference silently decide what counts as authoritative.",{},{"id":1074,"data":2068,"type":42,"tunes":2070},{"text":2069,"level":246},"Related canonical knowledge",{},{"id":1079,"data":2072,"type":218,"tunes":2074},{"text":2073},"Source-of-Truth architecture is a prerequisite for later retrieval and governance concepts because retrieval quality alone cannot determine whether evidence is allowed to define the answer.",{},{"id":1084,"data":2076,"type":1090,"tunes":2081},{"url":2077,"title":2078,"excerpt":2079,"ctaLabel":2080},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works","What Is RAG? The Simplest Explanation of How It Works","RAG retrieves external knowledge for a model. This foundation explains why retrieval and authoritative truth are separate responsibilities.","Read the RAG foundation",{},{"id":1093,"data":2083,"type":218,"tunes":2085},{"text":2084},"Memory is another adjacent concept. A reliable agent separates remembered information from current authoritative application state.",{},{"id":1098,"data":2087,"type":1090,"tunes":2092},{"url":2088,"title":2089,"excerpt":2090,"ctaLabel":2091},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fai-agent-memory-is-not-rag-how-to-separate-memory-retrieval-state-and-context","AI Agent Memory Is Not RAG: How to Separate Memory, Retrieval, State and Context","A four-layer architecture separating persistent memory, authoritative state, retrieval and the context actually supplied to the model.","Read the memory architecture article",{},{"id":1106,"data":2094,"type":218,"tunes":2096},{"text":2095},"Authority also connects directly to answer validity. Even an authoritative source supports only claims inside its version, date, scope and evidence boundary.",{},{"id":1111,"data":2098,"type":1090,"tunes":2103},{"url":2099,"title":2100,"excerpt":2101,"ctaLabel":2102},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers","The Answer Validity Boundary: The Missing Layer Between Relevance and Reliable AI Answers","A framework for making explicit where an AI answer remains supported by evidence and which changed conditions invalidate that support.","Read the Answer Validity Boundary",{},{"id":1119,"data":2105,"type":42,"tunes":2107},{"text":2106,"level":246},"Frequently asked questions",{},{"id":1124,"data":2109,"type":1124,"tunes":2136},{"items":2110,"title":2135},[2111,2114,2117,2120,2123,2126,2129,2132],{"id":1128,"answer":2112,"question":2113},"It is the authoritative source or authority rule that determines which source is allowed to establish a specific fact, state or rule for a defined scope, version and time.","What is a Source of Truth in an AI system?",{"id":1132,"answer":2115,"question":2116},"Normally no. A language model can generate, summarize and reason, but its parametric knowledge is not automatically authoritative for current application state, company policy, a specific document version or a regulated domain fact.","Is the language model a Source of Truth?",{"id":1136,"answer":2118,"question":2119},"Not automatically. A vector database is usually an index or retrieval store. It should preserve references to the authoritative original source and version rather than silently replacing them.","Is a vector database the Source of Truth for RAG?",{"id":1140,"answer":2121,"question":2122},"Provenance describes where data came from, how it was produced or transformed and who or what was involved. Source-of-Truth rules determine whether that source has authority for the specific claim.","What is the difference between provenance and Source of Truth?",{"id":1144,"answer":2124,"question":2125},"Yes. In complex systems this is normal because different facts belong to different authoritative systems or source classes.","Can an AI system have multiple Sources of Truth?",{"id":1148,"answer":2127,"question":2128},"The system needs a domain-specific conflict rule: precedence, version supersession, expert adjudication or an explicit unresolved state. It should not silently choose the source the model prefers.","What happens when two authoritative sources disagree?",{"id":1152,"answer":2130,"question":2131},"No. RAG retrieves candidates. Authority-aware metadata, filters, source policy and validation are needed to ensure that important claims use the correct source.","Does RAG guarantee that an AI answer uses the Source of Truth?",{"id":1156,"answer":2133,"question":2134},"Yes. Authority and correctness are different properties. An authoritative system can contain a data-quality defect, which should be corrected transparently rather than hidden by substituting an unofficial source.","Can the Source of Truth be wrong?","Source of Truth in AI systems",{},{"id":1162,"data":2138,"type":42,"tunes":2140},{"text":2139,"level":246},"Glossary",{},{"id":1167,"data":2142,"type":1167,"tunes":2175},{"title":2143,"entries":2144},"Key Source-of-Truth terms",[2145,2148,2151,2154,2157,2160,2163,2166,2169,2172],{"term":2146,"anchor":1173,"definition":2147},"Source of Truth","The authoritative source or rule permitted to establish a particular fact, state or rule for a defined scope, version and time.",{"term":2149,"anchor":1177,"definition":2150},"System of record","The authoritative operational system responsible for a defined class of records or current business state.",{"term":2152,"anchor":1180,"definition":2153},"Provenance","Information describing the origin, derivation, transformations, responsible agents and history of data or another entity.",{"term":2155,"anchor":1184,"definition":2156},"Evidence","Information or an artifact that supports, contradicts or constrains a claim.",{"term":2158,"anchor":327,"definition":2159},"Authority","The application or domain rule determining which source is entitled to define a specific claim.",{"term":2161,"anchor":1191,"definition":2162},"Freshness","Whether a representation remains current enough for the claim or operation in which it is used.",{"term":2164,"anchor":1195,"definition":2165},"Supersession","The explicit replacement of an older authoritative version by a newer one while preserving historical traceability.",{"term":2167,"anchor":1199,"definition":2168},"Ground truth","A reference answer, label or outcome used to evaluate a system; it is an evaluation construct rather than automatically the application's Source of Truth.",{"term":2170,"anchor":1203,"definition":2171},"Lineage","The trace of how data or derived values flow and transform across sources and processing steps.",{"term":2173,"anchor":1207,"definition":2174},"Validity boundary","The conditions of scope, time, version, evidence and assumptions inside which a claim remains supported.",{},{"id":1211,"data":2177,"type":42,"tunes":2179},{"text":2178,"level":246},"Conclusion",{},{"id":1216,"data":2181,"type":218,"tunes":2183},{"text":2182},"Reliable AI does not come from giving the model more information. It comes from knowing which information is allowed to define the claim, preserving where that information came from, retrieving the correct version and keeping the final answer inside the source's scope.",{},{"id":1221,"data":2185,"type":218,"tunes":2187},{"text":2186},"That is why Source of Truth, provenance, retrieval, memory and context must remain separate concepts. The Source of Truth defines authority. Provenance explains origin. Retrieval finds candidates. Memory preserves selected past information. Context is what the model sees. Generation turns those inputs into an output.",{},{"id":1226,"data":2189,"type":218,"tunes":2191},{"text":2190},"When those layers remain explicit, an AI system can do more than sound plausible: important claims can be traced back to the source that actually had the right to establish them.",{},{"id":1231,"data":2193,"type":42,"tunes":2195},{"text":2194,"level":246},"Primary sources and implementation evidence",{},{"id":1236,"data":2197,"type":218,"tunes":2199},{"text":2198},"The external sources below support provenance and AI risk-management claims. The Source of Truth Research Engine section is original implementation evidence and is explicitly presented as one implementation pattern rather than a universal standard.",{},{"id":1241,"data":2201,"type":1248,"tunes":2206},{"link":1243,"meta":2202},{"image":2203,"title":2204,"description":2205},{"url":304},"W3C PROV-DM — The PROV Data Model","W3C Recommendation defining a domain-agnostic provenance model around entities, activities, agents, derivations and responsibility.",{},{"id":1251,"data":2208,"type":1248,"tunes":2213},{"link":1253,"meta":2209},{"image":2210,"title":2211,"description":2212},{"url":304},"W3C Provenance Working Group — Publications","Official index of W3C PROV Recommendations and related specifications for provenance interchange and constraints.",{},{"id":1260,"data":2215,"type":1248,"tunes":2220},{"link":1262,"meta":2216},{"image":2217,"title":2218,"description":2219},{"url":304},"NIST AI Risk Management Framework","NIST's voluntary framework for incorporating trustworthiness and risk-management considerations across the AI lifecycle; AI RMF 1.0 is currently being revised.",{},{"id":1269,"data":2222,"type":1248,"tunes":2227},{"link":1271,"meta":2223},{"image":2224,"title":2225,"description":2226},{"url":304},"NIST AI RMF Playbook","Operational guidance aligned to the AI RMF, including documentation practices for data provenance, sources, origins, transformations, dependencies, constraints and metadata.",{},{"id":1278,"data":2229,"type":1248,"tunes":2234},{"link":1280,"meta":2230},{"image":2231,"title":2232,"description":2233},{"url":304},"NIST AI RMF Playbook — Measure","Guidance on documenting measurement, data provenance and contextual interpretation of AI system outputs.",{},{"id":1287,"data":2236,"type":1248,"tunes":2241},{"link":1289,"meta":2237},{"image":2238,"title":2239,"description":2240},{"url":304},"NIST AI 600-1 — Generative AI Profile","NIST generative-AI profile, including provenance and information-integrity considerations for generative AI systems.",{},"2.31.6","A Source of Truth defines which source is authoritative for a specific fact or state. Learn how it differs from RAG, provenance, memory, context, vector databases and systems of record.",{"lang":7,"title":208,"content":210,"contentJson":2245,"excerpt":1296},{"time":212,"blocks":2246,"version":1295},[2247,2250,2253,2256,2259,2262,2265,2268,2271,2274,2277,2280,2283,2286,2289,2305,2308,2311,2314,2317,2329,2332,2335,2338,2341,2344,2347,2350,2353,2356,2359,2362,2365,2368,2371,2374,2377,2380,2383,2386,2389,2392,2395,2398,2401,2404,2407,2410,2413,2416,2419,2422,2434,2437,2440,2451,2454,2457,2460,2463,2466,2469,2472,2475,2478,2481,2484,2487,2490,2493,2504,2507,2510,2513,2516,2519,2522,2525,2528,2531,2534,2537,2540,2543,2546,2549,2562,2565,2568,2571,2574,2591,2594,2607,2610,2627,2630,2643,2646,2649,2652,2655,2658,2661,2664,2667,2670,2673,2676,2679,2682,2685,2688,2691,2694,2697,2700,2703,2715,2718,2732,2735,2738,2741,2744,2747,2750,2755,2760,2765,2770,2775],{"id":215,"data":2248,"type":218,"tunes":2249},{"text":217},{},{"id":221,"data":2251,"type":226,"tunes":2252},{"body":223,"title":224,"variant":225},{},{"id":229,"data":2254,"type":226,"tunes":2255},{"body":231,"title":232,"variant":233},{},{"id":236,"data":2257,"type":226,"tunes":2258},{"body":238,"title":239,"variant":233},{},{"id":242,"data":2260,"type":247,"tunes":2261},{"title":244,"maxLevel":245,"minLevel":246},{},{"id":250,"data":2263,"type":42,"tunes":2264},{"text":252,"level":246},{},{"id":255,"data":2266,"type":218,"tunes":2267},{"text":257},{},{"id":260,"data":2269,"type":218,"tunes":2270},{"text":262},{},{"id":265,"data":2272,"type":218,"tunes":2273},{"text":267},{},{"id":270,"data":2275,"type":226,"tunes":2276},{"body":272,"title":273,"variant":274},{},{"id":277,"data":2278,"type":42,"tunes":2279},{"text":279,"level":246},{},{"id":282,"data":2281,"type":218,"tunes":2282},{"text":284},{},{"id":287,"data":2284,"type":218,"tunes":2285},{"text":289},{},{"id":292,"data":2287,"type":218,"tunes":2288},{"text":294},{},{"id":297,"data":2290,"type":329,"tunes":2304},{"rows":2291,"title":317,"layout":318,"columns":2300},[2292,2294,2296,2298],{"id":301,"label":302,"values":2293},[304,304,304],{"id":306,"label":307,"values":2295},[304,304,304],{"id":310,"label":311,"values":2297},[304,304,304],{"id":314,"label":315,"values":2299},[304,304,304],[2301,2302,2303],{"id":321,"label":322},{"id":324,"label":325},{"id":327,"label":328},{},{"id":332,"data":2306,"type":42,"tunes":2307},{"text":334,"level":246},{},{"id":337,"data":2309,"type":218,"tunes":2310},{"text":339},{},{"id":342,"data":2312,"type":218,"tunes":2313},{"text":344},{},{"id":347,"data":2315,"type":42,"tunes":2316},{"text":349,"level":246},{},{"id":352,"data":2318,"type":318,"tunes":2328},{"content":2319,"stretched":43,"withHeadings":14},[2320,2321,2322,2323,2324,2325,2326,2327],[356,357,358],[360,361,362],[364,365,366],[368,369,370],[372,373,374],[376,377,378],[380,381,382],[384,385,386],{},{"id":389,"data":2330,"type":218,"tunes":2331},{"text":391},{},{"id":394,"data":2333,"type":42,"tunes":2334},{"text":396,"level":246},{},{"id":399,"data":2336,"type":42,"tunes":2337},{"text":401,"level":245},{},{"id":404,"data":2339,"type":218,"tunes":2340},{"text":406},{},{"id":409,"data":2342,"type":218,"tunes":2343},{"text":411},{},{"id":414,"data":2345,"type":42,"tunes":2346},{"text":416,"level":245},{},{"id":419,"data":2348,"type":218,"tunes":2349},{"text":421},{},{"id":424,"data":2351,"type":218,"tunes":2352},{"text":426},{},{"id":429,"data":2354,"type":42,"tunes":2355},{"text":431,"level":245},{},{"id":434,"data":2357,"type":218,"tunes":2358},{"text":436},{},{"id":439,"data":2360,"type":218,"tunes":2361},{"text":441},{},{"id":444,"data":2363,"type":42,"tunes":2364},{"text":446,"level":245},{},{"id":449,"data":2366,"type":218,"tunes":2367},{"text":451},{},{"id":454,"data":2369,"type":218,"tunes":2370},{"text":456},{},{"id":459,"data":2372,"type":42,"tunes":2373},{"text":461,"level":245},{},{"id":464,"data":2375,"type":218,"tunes":2376},{"text":466},{},{"id":469,"data":2378,"type":218,"tunes":2379},{"text":471},{},{"id":474,"data":2381,"type":42,"tunes":2382},{"text":476,"level":245},{},{"id":479,"data":2384,"type":218,"tunes":2385},{"text":481},{},{"id":484,"data":2387,"type":218,"tunes":2388},{"text":486},{},{"id":489,"data":2390,"type":42,"tunes":2391},{"text":491,"level":245},{},{"id":494,"data":2393,"type":218,"tunes":2394},{"text":496},{},{"id":499,"data":2396,"type":218,"tunes":2397},{"text":501},{},{"id":504,"data":2399,"type":42,"tunes":2400},{"text":506,"level":245},{},{"id":509,"data":2402,"type":218,"tunes":2403},{"text":511},{},{"id":514,"data":2405,"type":218,"tunes":2406},{"text":516},{},{"id":519,"data":2408,"type":42,"tunes":2409},{"text":521,"level":245},{},{"id":524,"data":2411,"type":218,"tunes":2412},{"text":526},{},{"id":529,"data":2414,"type":218,"tunes":2415},{"text":531},{},{"id":534,"data":2417,"type":42,"tunes":2418},{"text":536,"level":246},{},{"id":539,"data":2420,"type":226,"tunes":2421},{"body":541,"title":542,"variant":233},{},{"id":545,"data":2423,"type":574,"tunes":2433},{"steps":2424,"title":572,"orientation":573},[2425,2426,2427,2428,2429,2430,2431,2432],{"label":549,"description":550},{"label":552,"description":553},{"label":555,"description":556},{"label":558,"description":559},{"label":561,"description":562},{"label":564,"description":565},{"label":567,"description":568},{"label":570,"description":571},{},{"id":577,"data":2435,"type":42,"tunes":2436},{"text":579,"level":246},{},{"id":582,"data":2438,"type":218,"tunes":2439},{"text":584},{},{"id":587,"data":2441,"type":318,"tunes":2450},{"content":2442,"stretched":43,"withHeadings":14},[2443,2444,2445,2446,2447,2448,2449],[591,592,593],[595,596,597],[599,600,601],[603,604,605],[607,608,609],[611,612,613],[615,616,617],{},{"id":620,"data":2452,"type":42,"tunes":2453},{"text":622,"level":246},{},{"id":625,"data":2455,"type":218,"tunes":2456},{"text":627},{},{"id":630,"data":2458,"type":218,"tunes":2459},{"text":632},{},{"id":635,"data":2461,"type":226,"tunes":2462},{"body":637,"title":638,"variant":639},{},{"id":642,"data":2464,"type":42,"tunes":2465},{"text":644,"level":246},{},{"id":647,"data":2467,"type":218,"tunes":2468},{"text":649},{},{"id":652,"data":2470,"type":218,"tunes":2471},{"text":654},{},{"id":657,"data":2473,"type":42,"tunes":2474},{"text":659,"level":246},{},{"id":662,"data":2476,"type":218,"tunes":2477},{"text":664},{},{"id":667,"data":2479,"type":218,"tunes":2480},{"text":669},{},{"id":672,"data":2482,"type":226,"tunes":2483},{"body":674,"title":675,"variant":274},{},{"id":678,"data":2485,"type":42,"tunes":2486},{"text":680,"level":246},{},{"id":683,"data":2488,"type":218,"tunes":2489},{"text":685},{},{"id":688,"data":2491,"type":218,"tunes":2492},{"text":690},{},{"id":693,"data":2494,"type":318,"tunes":2503},{"content":2495,"stretched":43,"withHeadings":14},[2496,2497,2498,2499,2500,2501,2502],[697,698],[700,701],[703,704],[706,707],[709,710],[712,713],[715,716],{},{"id":719,"data":2505,"type":42,"tunes":2506},{"text":721,"level":246},{},{"id":724,"data":2508,"type":218,"tunes":2509},{"text":726},{},{"id":729,"data":2511,"type":218,"tunes":2512},{"text":731},{},{"id":734,"data":2514,"type":42,"tunes":2515},{"text":736,"level":246},{},{"id":739,"data":2517,"type":218,"tunes":2518},{"text":741},{},{"id":744,"data":2520,"type":218,"tunes":2521},{"text":746},{},{"id":749,"data":2523,"type":42,"tunes":2524},{"text":751,"level":246},{},{"id":754,"data":2526,"type":218,"tunes":2527},{"text":756},{},{"id":759,"data":2529,"type":218,"tunes":2530},{"text":761},{},{"id":764,"data":2532,"type":42,"tunes":2533},{"text":766,"level":246},{},{"id":769,"data":2535,"type":226,"tunes":2536},{"body":771,"title":772,"variant":233},{},{"id":775,"data":2538,"type":218,"tunes":2539},{"text":777},{},{"id":780,"data":2541,"type":218,"tunes":2542},{"text":782},{},{"id":785,"data":2544,"type":218,"tunes":2545},{"text":787},{},{"id":790,"data":2547,"type":218,"tunes":2548},{"text":792},{},{"id":795,"data":2550,"type":318,"tunes":2561},{"content":2551,"stretched":43,"withHeadings":14},[2552,2553,2554,2555,2556,2557,2558,2559,2560],[799,800],[802,803],[805,806],[808,809],[811,812],[814,815],[817,818],[820,821],[823,824],{},{"id":827,"data":2563,"type":42,"tunes":2564},{"text":829,"level":246},{},{"id":832,"data":2566,"type":218,"tunes":2567},{"text":834},{},{"id":837,"data":2569,"type":218,"tunes":2570},{"text":839},{},{"id":842,"data":2572,"type":42,"tunes":2573},{"text":844,"level":246},{},{"id":847,"data":2575,"type":318,"tunes":2590},{"content":2576,"stretched":43,"withHeadings":14},[2577,2578,2579,2580,2581,2582,2583,2584,2585,2586,2587,2588,2589],[851,852],[854,855],[857,858],[860,861],[863,864],[866,867],[869,870],[872,873],[875,876],[878,879],[881,882],[884,885],[887,888],{},{"id":891,"data":2592,"type":42,"tunes":2593},{"text":893,"level":246},{},{"id":896,"data":2595,"type":574,"tunes":2606},{"steps":2596,"title":926,"orientation":573},[2597,2598,2599,2600,2601,2602,2603,2604,2605],{"label":900,"description":901},{"label":903,"description":904},{"label":906,"description":907},{"label":909,"description":910},{"label":912,"description":913},{"label":915,"description":916},{"label":918,"description":919},{"label":921,"description":922},{"label":924,"description":925},{},{"id":929,"data":2608,"type":42,"tunes":2609},{"text":931,"level":246},{},{"id":934,"data":2611,"type":318,"tunes":2626},{"content":2612,"stretched":43,"withHeadings":14},[2613,2614,2615,2616,2617,2618,2619,2620,2621,2622,2623,2624,2625],[356,938],[940,941],[943,944],[946,947],[949,950],[952,953],[955,956],[958,959],[961,962],[964,965],[967,968],[970,971],[973,974],{},{"id":977,"data":2628,"type":42,"tunes":2629},{"text":979,"level":246},{},{"id":982,"data":2631,"type":318,"tunes":2642},{"content":2632,"stretched":43,"withHeadings":14},[2633,2634,2635,2636,2637,2638,2639,2640,2641],[986,987],[989,990],[992,993],[995,996],[998,999],[1001,1002],[1004,1005],[1007,1008],[1010,1011],{},{"id":1014,"data":2644,"type":42,"tunes":2645},{"text":1016,"level":246},{},{"id":1019,"data":2647,"type":218,"tunes":2648},{"text":1021},{},{"id":1024,"data":2650,"type":218,"tunes":2651},{"text":1026},{},{"id":1029,"data":2653,"type":218,"tunes":2654},{"text":1031},{},{"id":1034,"data":2656,"type":218,"tunes":2657},{"text":1036},{},{"id":1039,"data":2659,"type":42,"tunes":2660},{"text":1041,"level":246},{},{"id":1044,"data":2662,"type":218,"tunes":2663},{"text":1046},{},{"id":1049,"data":2665,"type":218,"tunes":2666},{"text":1051},{},{"id":1054,"data":2668,"type":218,"tunes":2669},{"text":1056},{},{"id":1059,"data":2671,"type":42,"tunes":2672},{"text":1061,"level":246},{},{"id":1064,"data":2674,"type":218,"tunes":2675},{"text":1066},{},{"id":1069,"data":2677,"type":218,"tunes":2678},{"text":1071},{},{"id":1074,"data":2680,"type":42,"tunes":2681},{"text":1076,"level":246},{},{"id":1079,"data":2683,"type":218,"tunes":2684},{"text":1081},{},{"id":1084,"data":2686,"type":1090,"tunes":2687},{"url":1086,"title":1087,"excerpt":1088,"ctaLabel":1089},{},{"id":1093,"data":2689,"type":218,"tunes":2690},{"text":1095},{},{"id":1098,"data":2692,"type":1090,"tunes":2693},{"url":1100,"title":1101,"excerpt":1102,"ctaLabel":1103},{},{"id":1106,"data":2695,"type":218,"tunes":2696},{"text":1108},{},{"id":1111,"data":2698,"type":1090,"tunes":2699},{"url":1113,"title":1114,"excerpt":1115,"ctaLabel":1116},{},{"id":1119,"data":2701,"type":42,"tunes":2702},{"text":1121,"level":246},{},{"id":1124,"data":2704,"type":1124,"tunes":2714},{"items":2705,"title":1159},[2706,2707,2708,2709,2710,2711,2712,2713],{"id":1128,"answer":1129,"question":1130},{"id":1132,"answer":1133,"question":1134},{"id":1136,"answer":1137,"question":1138},{"id":1140,"answer":1141,"question":1142},{"id":1144,"answer":1145,"question":1146},{"id":1148,"answer":1149,"question":1150},{"id":1152,"answer":1153,"question":1154},{"id":1156,"answer":1157,"question":1158},{},{"id":1162,"data":2716,"type":42,"tunes":2717},{"text":1164,"level":246},{},{"id":1167,"data":2719,"type":1167,"tunes":2731},{"title":1169,"entries":2720},[2721,2722,2723,2724,2725,2726,2727,2728,2729,2730],{"term":1172,"anchor":1173,"definition":1174},{"term":1176,"anchor":1177,"definition":1178},{"term":322,"anchor":1180,"definition":1181},{"term":1183,"anchor":1184,"definition":1185},{"term":1187,"anchor":327,"definition":1188},{"term":1190,"anchor":1191,"definition":1192},{"term":1194,"anchor":1195,"definition":1196},{"term":1198,"anchor":1199,"definition":1200},{"term":1202,"anchor":1203,"definition":1204},{"term":1206,"anchor":1207,"definition":1208},{},{"id":1211,"data":2733,"type":42,"tunes":2734},{"text":1213,"level":246},{},{"id":1216,"data":2736,"type":218,"tunes":2737},{"text":1218},{},{"id":1221,"data":2739,"type":218,"tunes":2740},{"text":1223},{},{"id":1226,"data":2742,"type":218,"tunes":2743},{"text":1228},{},{"id":1231,"data":2745,"type":42,"tunes":2746},{"text":1233,"level":246},{},{"id":1236,"data":2748,"type":218,"tunes":2749},{"text":1238},{},{"id":1241,"data":2751,"type":1248,"tunes":2754},{"link":1243,"meta":2752},{"image":2753,"title":1246,"description":1247},{"url":304},{},{"id":1251,"data":2756,"type":1248,"tunes":2759},{"link":1253,"meta":2757},{"image":2758,"title":1256,"description":1257},{"url":304},{},{"id":1260,"data":2761,"type":1248,"tunes":2764},{"link":1262,"meta":2762},{"image":2763,"title":1265,"description":1266},{"url":304},{},{"id":1269,"data":2766,"type":1248,"tunes":2769},{"link":1271,"meta":2767},{"image":2768,"title":1274,"description":1275},{"url":304},{},{"id":1278,"data":2771,"type":1248,"tunes":2774},{"link":1280,"meta":2772},{"image":2773,"title":1283,"description":1284},{"url":304},{},{"id":1287,"data":2776,"type":1248,"tunes":2779},{"link":1289,"meta":2777},{"image":2778,"title":1292,"description":1293},{"url":304},{},"Post erfolgreich abgerufen",{"items":2782,"source":2866,"manualIds":2867,"manualMatchedIds":2868},[2783,2790,2797,2804,2811,2818,2825,2832,2839,2845,2852,2859],{"id":2784,"slug":2785,"title":2786,"excerpt":2787,"featuredImage":2788,"publishedAt":2789},"476","mcp-vs-a2a-vs-ucp-vs-ap2-vs-a2ui-the-agent-protocol-stack-explained","MCP vs A2A vs UCP vs AP2 vs A2UI：智能体协议栈详解","MCP、A2A、UCP、AP2 和 A2UI 常被描述为相互竞争的智能体标准。它们大多解决的是不同的互操作性问题。本指南将每个协议映射到其实际标准化的边界，并展示它们如何在同一个生产系统中协同工作。","\u002Fuploads\u002F2026\u002F09\u002Fmcp-vs-a2a-vs-ucp-vs-ap2-vs-a2ui-the-agent-protocol-stack-explained-1790352625869-2ezle0.webp","2026-09-25T12:09:00.000Z",{"id":2791,"slug":2792,"title":2793,"excerpt":2794,"featuredImage":2795,"publishedAt":2796},"471","how-to-know-whether-an-ai-agent-actually-used-the-right-evidence","如何判断一个AI智能体是否真正使用了正确的证据","AI代理可以引用来源，却仍然使用错误的证据。本文介绍一种实用方法，用于核查主张支持、来源权威性、适用性、出处，以及证据是否实际影响了答案。","\u002Fuploads\u002F2026\u002F09\u002Fhow-to-know-whether-an-ai-agent-actually-used-the-right-evidence-1790351317188-o5z9ve.webp","2026-09-25T11:47:00.000Z",{"id":2798,"slug":2799,"title":2800,"excerpt":2801,"featuredImage":2802,"publishedAt":2803},"483","what-is-an-ai-solution-architect-system-boundaries-responsibilities-and-trade-offs","什么是AI解决方案架构师？系统边界、职责与权衡","AI解决方案架构师将业务需求转化为生产就绪的AI系统，涵盖数据、模型、工具、安全、运行时、评估和运维。","\u002Fuploads\u002F2026\u002F10\u002Fwhat-is-an-ai-solution-architect-system-boundaries-responsibilities-and-trade-offs-1791476643267-1st5xz.webp","2026-10-08T12:23:00.000Z",{"id":2805,"slug":2806,"title":2807,"excerpt":2808,"featuredImage":2809,"publishedAt":2810},"489","agentic-ai-explained-when-an-ai-system-can-plan-use-tools-and-act","智能体AI解析：当AI系统能够规划、使用工具并采取行动","代理式AI在多步执行循环中使用模型，这些模型可以在明确的运行时和权限边界内选择工具、观察结果、更新状态并调整其下一步行动。","\u002Fuploads\u002F2026\u002F10\u002Fagentic-ai-explained-when-an-ai-system-can-plan-use-tools-and-act-1791481499084-wnji2a.webp","2026-10-08T11:43:00.000Z",{"id":2812,"slug":2813,"title":2814,"excerpt":2815,"featuredImage":2816,"publishedAt":2817},"487","vector-databases-embeddings-and-reranking-three-different-parts-of-retrieval","向量数据库、嵌入和重排序：检索的三个不同部分","嵌入表示含义，向量数据库检索候选结果，重排序器则精炼结果。了解这三个检索层在RAG中如何不同并协同工作。","\u002Fuploads\u002F2026\u002F10\u002Fvector-databases-embeddings-and-reranking-three-different-parts-of-retrieval-1791480129884-9dtasz.webp","2026-10-08T11:21:00.000Z",{"id":2819,"slug":2820,"title":2821,"excerpt":2822,"featuredImage":2823,"publishedAt":2824},"494","air-gapped-ai-how-ai-systems-work-without-internet-or-cloud-access","气隙AI：AI系统如何在没有互联网或云访问的情况下工作","气隙AI在隔离的安全域内运行模型、RAG和AI应用，无需互联网或云依赖。了解模型、数据、更新和工具如何离线运行。","\u002Fuploads\u002F2026\u002F10\u002Fair-gapped-ai-how-ai-systems-work-without-internet-or-cloud-access-1791487983978-e6xqf0.webp","2026-10-08T11:32:00.000Z",{"id":2826,"slug":2827,"title":2828,"excerpt":2829,"featuredImage":2830,"publishedAt":2831},"495","sovereign-ai-control-of-models-data-infrastructure-and-dependencies","主权人工智能：模型、数据、基础设施与依赖关系的控制","主权人工智能关乎对模型、数据、基础设施、软件、运营和战略依赖的有效控制——而不仅仅是人工智能模型托管在哪里。","\u002Fuploads\u002F2026\u002F10\u002Fsovereign-ai-control-of-models-data-infrastructure-and-dependencies-1791488833132-niy85x.webp","2026-10-08T15:45:00.000Z",{"id":2833,"slug":2834,"title":2835,"excerpt":2836,"featuredImage":2837,"publishedAt":2838},"485","enterprise-ai-architecture-what-changes-when-ai-enters-a-company","企业AI架构：当AI进入公司时会发生什么变化","企业AI架构阐释了AI如何在数据权限、身份、许可、提供商、风险、治理、评估、合规和运营方面改变公司系统。","\u002Fuploads\u002F2026\u002F10\u002Fenterprise-ai-architecture-what-changes-when-ai-enters-a-company-1791478161363-czrwaq.webp","2026-10-08T10:48:00.000Z",{"id":2840,"slug":2841,"title":1087,"excerpt":2842,"featuredImage":2843,"publishedAt":2844},"478","what-is-rag-the-simplest-explanation-of-how-it-works","RAG听起来很复杂，但想法很简单：在AI回答之前，它先从知识源查找有用的信息，并将该信息提供给语言模型。本指南使用一个简单的思维模型来解释RAG、LLM、状态、记忆和工具。","\u002Fuploads\u002F2026\u002F09\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works-1790377492124-khjagt.webp","2026-09-25T19:03:00.000Z",{"id":2846,"slug":2847,"title":2848,"excerpt":2849,"featuredImage":2850,"publishedAt":2851},"480","when-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger","人工智能何时应停止信任自身知识？——检索触发机制","AI 模型并非每个问题都需要检索。重要的问题在于知道何时其内部知识已不再足够。检索触发器是一个实用的决策边界，它决定 AI 系统何时应停止仅依赖模型知识，并在回答前获取外部证据。","\u002Fuploads\u002F2026\u002F09\u002Fwhen-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger-1790574991244-f4rpyg.webp","2026-09-28T01:49:00.000Z",{"id":2853,"slug":2854,"title":2855,"excerpt":2856,"featuredImage":2857,"publishedAt":2858},"466","the-gpu-is-not-the-product-future-proof-private-ai-architecture","GPU 不是产品：面向未来的私有 AI 架构","私有 AI 基础设施不应围绕单一 GPU 或单一模型来设计。更具韧性的做法是将快速推理 GPU、内存充裕的 AI 系统、物理 AI 节点以及可选的前沿云模型，统一置于一个具备能力感知的路由层之后。","\u002Fuploads\u002F2026\u002F09\u002Fthe-gpu-is-not-the-product-future-proof-private-ai-architecture-1790140878812-8hsl39.webp","2026-09-23T01:19:00.000Z",{"id":2860,"slug":2861,"title":2862,"excerpt":2863,"featuredImage":2864,"publishedAt":2865},"490","rbac-vs-tenant-isolation-two-different-security-boundaries","RBAC与租户隔离：两种不同的安全边界","RBAC 控制用户可以做什么；租户隔离控制该操作可以触及哪个租户的资源。了解为什么多租户 SaaS 安全需要这两道边界。","\u002Fuploads\u002F2026\u002F10\u002Frbac-vs-tenant-isolation-two-different-security-boundaries-1791485111528-qqtzby.webp","2026-10-08T14:43:00.000Z","fallback",[],[]]