[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:de":3,"public-menus:all":37,"post:can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain:de":220,"related:post:can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain:de:1":1604},{"statusCode":4,"data":5,"message":36},200,{"tenantId":6,"lang":7,"defaultLang":7,"siteUrl":8,"contactEmail":9,"brandName":10,"logoUrl":11,"siteName":10,"siteDescription":12,"ogImage":9,"robotsIndex":13,"socialLinks":9,"reservedSlugs":9,"seoPolicy":14},"stajic","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":15,"relatedContent":16,"crossDomainLinks":17},{"logoUrl":11},{"enabled":13},[18,21,24,27,30,33],{"url":19,"label":20,"isActive":13,"showInFooter":13,"includeInSameAs":13},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":22,"label":23,"isActive":13,"showInFooter":13,"includeInSameAs":13},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":25,"label":26,"isActive":13,"showInFooter":13,"includeInSameAs":13},"https:\u002F\u002Fbazify.com","bazify.com",{"url":28,"label":29,"isActive":13,"showInFooter":13,"includeInSameAs":13},"https:\u002F\u002Fbazify.de","bazify.de",{"url":31,"label":32,"isActive":13,"showInFooter":13,"includeInSameAs":13},"https:\u002F\u002Fbazify.at","bazify.at",{"url":34,"label":35,"isActive":13,"showInFooter":13,"includeInSameAs":13},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[38,44],{"id":39,"name":40,"location":41,"isActive":13,"isDefault":42,"items":43},1,"main-navigation","header",false,[],{"id":45,"name":46,"location":47,"isActive":13,"isDefault":13,"items":48},4,"main-menu","sidebar",[49,65,78,92,102,117,132],{"id":50,"title":51,"url":59,"target":60,"icon":61,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":63,"portfolioId":9,"children":64},"item-18",{"de":52,"en":53,"es":54,"fr":55,"it":53,"ru":56,"sr":57,"zh":58},"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":66,"title":67,"url":74,"target":60,"icon":75,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":76,"portfolioId":9,"children":77},"item-22",{"de":68,"en":68,"es":69,"fr":68,"it":70,"ru":71,"sr":72,"zh":73},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":79,"title":80,"url":88,"target":60,"icon":89,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":90,"portfolioId":9,"children":91},"item-19",{"de":81,"en":82,"es":83,"fr":82,"it":84,"ru":85,"sr":86,"zh":87},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":93,"title":94,"url":98,"target":60,"icon":99,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":100,"portfolioId":9,"children":101},"item-23",{"de":95,"en":95,"es":95,"fr":95,"it":95,"ru":96,"sr":96,"zh":97},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":103,"title":104,"url":113,"target":60,"icon":114,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":115,"portfolioId":9,"children":116},"item-32",{"de":105,"en":106,"es":107,"fr":108,"it":109,"ru":110,"sr":111,"zh":112},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":118,"title":119,"url":128,"target":60,"icon":129,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":130,"portfolioId":9,"children":131},"item-20",{"de":120,"en":121,"es":122,"fr":123,"it":124,"ru":125,"sr":126,"zh":127},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":133,"title":134,"url":143,"target":60,"icon":144,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":145,"portfolioId":9,"children":146},"item-21",{"de":135,"en":136,"es":137,"fr":138,"it":139,"ru":140,"sr":141,"zh":142},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[147,160,174,190,205],{"id":148,"title":149,"url":143,"target":60,"icon":158,"isActive":13,"type":62,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":145,"portfolioId":9,"children":159},"item-24",{"de":150,"en":151,"es":152,"fr":153,"it":154,"ru":155,"sr":156,"zh":157},"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":161,"title":162,"url":170,"target":60,"icon":171,"isActive":13,"type":172,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":9,"portfolioId":9,"children":173},"item-29",{"de":163,"en":164,"es":165,"fr":166,"it":167,"ru":168,"sr":169,"zh":142},"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-globe-2","custom",[],{"id":175,"title":176,"url":185,"target":60,"icon":186,"isActive":13,"type":187,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":9,"portfolioId":188,"children":189},"menu-2916d060-8681-482a-af9a-c107a8bc30d8",{"de":177,"en":178,"es":179,"fr":180,"it":181,"ru":182,"sr":183,"zh":184},"Von Rohdaten zu verifizierbarem KI-Kontext: Eine praktische Architektur für agentenbereite Informationen","From Raw Data to Verifiable AI Context:","De datos sin procesar a contexto de IA verificable: una arquitectura práctica para información lista para agentes","Des données brutes au contexte d'IA vérifiable : une architecture pratique pour des informations prêtes pour les agents","Dai dati grezzi a un contesto IA verificabile: Un'architettura pratica per informazioni pronte per gli agenti","От необработанных данных к верифицируемому контексту ИИ: практическая архитектура информации, готовой для агентов","Od sirovih podataka do proverljivog AI konteksta: Praktična arhitektura za informacije spremne za agente","从原始数据到可验证的 AI 上下文：面向智能体就绪信息的实用架构","\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information","i-lucide-database","portfolio",22,[],{"id":191,"title":192,"url":201,"target":60,"icon":202,"isActive":13,"type":187,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":9,"portfolioId":203,"children":204},"menu-e2061cd7-ff53-49d5-a0e8-3d49090577c0",{"de":193,"en":194,"es":195,"fr":196,"it":197,"ru":198,"sr":199,"zh":200},"Präsenzplattformen — Unternehmenskommunikation","Presence Platforms — Enterprise Communication","Plataformas de presencia — Comunicación empresarial","Plateformes de présence — Communication d'entreprise","Presence Platforms — Comunicazione aziendale","Платформы присутствия — Корпоративные коммуникации","Presence platforme — Poslovna komunikacija","Presence 平台 — 企业通信","\u002Fportfolio\u002Fpresence-platforms-enterprise-communication-that-wins-without-competing","i-lucide-badge-percent",21,[],{"id":206,"title":207,"url":216,"target":60,"icon":217,"isActive":13,"type":187,"productId":9,"categoryId":9,"shopCategoryId":9,"articleId":9,"pageId":9,"portfolioId":218,"children":219},"menu-a781582e-e90d-418e-b8c9-b070bc37c37e",{"de":208,"en":209,"es":210,"fr":211,"it":212,"ru":213,"sr":214,"zh":215},"Maximierung des POS-ROI durch Konsistenz","Maximizing POS ROI Through Consistent","Maximizar el ROI del POS a través de constante","Maximiser le ROI du PDV grâce à la cohérence","Massimizzare il ROI del POS attraverso costante","Максимизация ROI POS за счет последовательного","Maksimizacija POS ROI-ja kroz dosledno","通过持续最大化 POS 投资回报率","\u002Fportfolio\u002Ffrom-vision-to-value-maximizing-pos-roi-through-consistent-ci-creation-and-strategic-branding","i-lucide-briefcase-business",11,[],{"statusCode":4,"data":221,"message":1603},{"id":222,"title":223,"slug":224,"content":225,"contentJson":226,"excerpt":797,"featuredImage":798,"featuredImageAlt":799,"featuredImageCaption":9,"featuredImageTitle":9,"featuredImageCopyright":9,"featuredImageAuthor":9,"featuredImageSourceUrl":9,"featuredImageLicense":9,"featuredImageIsAiGenerated":42,"status":800,"publishedAt":801,"createdAt":802,"updatedAt":803,"seoLocalePaths":804,"categories":813,"author":822,"translations":826},"498","Kann Beweismaterial die KI-Pipeline überleben? — Die Beweiserhaltungskette","can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u003Ch2 id=\"section-1\">Frage\u003C\u002Fh2>\n\u003Cp>Was passiert, wenn die ursprüngliche Quelle korrekt ist, aber die KI-Pipeline die Informationen entfernt, die erforderlich sind, um diese Quelle korrekt zu verstehen?\u003C\u002Fp>\n\u003Cp>Ein zuverlässiges Dokument kann zu einem unzuverlässigen Kontext werden, wenn das Parsing seine Tabellenstruktur entfernt, das Chunking eine Bedingung von ihrer Ausnahme trennt, die Normalisierung Einheiten fallen lässt, das Retrieval Versionsmetadaten verliert oder die Kontextzusammenstellung einen Konflikt verbirgt.\u003C\u002Fp>\n\u003Cp>Dieser Artikel verwendet den Begriff Evidence Preservation Chain für die Anforderung, dass Beweise die für eine korrekte Interpretation notwendigen Eigenschaften behalten müssen, während sie von der Quelle zum endgültigen KI-Kontext wandern.\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">Eine Evidence Preservation Chain ist die End-to-End-Sequenz von Transformationen, durch die Beweise Identität, Herkunft, Umfang, Bedeutung und Gültigkeit behalten müssen, bevor sie sicher eine KI-generierte Behauptung stützen können.\u003Ccite class=\"block mt-2 text-sm\">— Arbeitsdefinition — ein hier eingeführtes architektonisches Konzept, kein standardisierter Forschungsbegriff.\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"Inhalt\">\u003Cstrong class=\"editorjs-toc__title\">Inhalt\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-1\" class=\"editorjs-toc__link\">Frage\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-7\" class=\"editorjs-toc__link\">Was das wirklich bedeutet\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-12\" class=\"editorjs-toc__link\">Einfachstes Beispiel\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-20\" class=\"editorjs-toc__link\">Wo das Beispiel nicht mehr funktioniert\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-24\" class=\"editorjs-toc__link\">Direkte Antwort\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-27\" class=\"editorjs-toc__link\">Warum RAG selbst mit der richtigen Quelle fehlschlagen kann\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">Die acht Erhaltungspunkte\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-32\" class=\"editorjs-toc__link\">1. Identität\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-34\" class=\"editorjs-toc__link\">2. Herkunft\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">3. Struktur\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">4. Einheiten\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-40\" class=\"editorjs-toc__link\">5. Bedingungen und Ausnahmen\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-42\" class=\"editorjs-toc__link\">6. Version und Zeit\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-44\" class=\"editorjs-toc__link\">7. Konfliktstatus\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-46\" class=\"editorjs-toc__link\">8. Zugriffsbereich\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-48\" class=\"editorjs-toc__link\">Von Rohdaten zu kontrollierten Belegen\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-52\" class=\"editorjs-toc__link\">Kontext-Engineering ist Teil der Kette\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">Verbindung zur Evidenzdistanz\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-60\" class=\"editorjs-toc__link\">Verbindung zur Retrieval-Autoritätsschicht\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-64\" class=\"editorjs-toc__link\">Praktisches Datenmodell\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-68\" class=\"editorjs-toc__link\">Diagnostische \u002F Entscheidungsmethode\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-72\" class=\"editorjs-toc__link\">Fehlermodus: Der verwaiste Chunk\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-76\" class=\"editorjs-toc__link\">Fehlermodus: Normalisierung durch Erfindung\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-80\" class=\"editorjs-toc__link\">Fehlermodus: Konfliktkollaps\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-84\" class=\"editorjs-toc__link\">Fehlermodus: Kontextkompression\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-88\" class=\"editorjs-toc__link\">Randfall: Verlustbehaftete Kompression kann dennoch korrekt sein\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-92\" class=\"editorjs-toc__link\">Einschränkungen\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-96\" class=\"editorjs-toc__link\">Was würde diese Antwort ändern?\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-99\" class=\"editorjs-toc__link\">Fazit\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-104\" class=\"editorjs-toc__link\">Primärquellen\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Ch2 id=\"section-7\">Was das wirklich bedeutet\u003C\u002Fh2>\n\u003Cp>Die meisten Diskussionen über RAG konzentrieren sich darauf, ob relevante Informationen gefunden werden können.\u003C\u002Fp>\n\u003Cp>Aber Beweise durchlaufen normalerweise mehrere Transformationen, bevor das Modell sie sieht.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Original Source\n   ↓\nParsing\n   ↓\nNormalization\n   ↓\nChunking\n   ↓\nIndexing\n   ↓\nRetrieval\n   ↓\nReranking\n   ↓\nContext Assembly\n   ↓\nModel\n   ↓\nClaim\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Die Beweise können an jedem Punkt dieser Kette verfallen, selbst wenn das Originaldokument vollkommen korrekt war.\u003C\u002Fp>\n\u003Ch2 id=\"section-12\">Einfachstes Beispiel\u003C\u002Fh2>\n\u003Cp>Stellen Sie sich vor, ein Produkthandbuch enthält diese Aussage:\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">Maximale Betriebstemperatur: 40 °C, außer wenn Zubehör X installiert ist, in diesem Fall beträgt die Grenze 35 °C.\u003Ccite class=\"block mt-2 text-sm\">— Beispiel-Quellaussage\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Cp>Ein schlechter Chunking-Prozess teilt den Satz auf.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Chunk A:\nMaximum operating temperature: 40 °C.\n\nChunk B:\nExcept when accessory X is installed, in which case the limit is 35 °C.\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Der Retriever gibt nur Chunk A zurück.\u003C\u002Fp>\n\u003Cp>Der abgerufene Text ist nicht erfunden. Er stammt direkt aus dem Handbuch. Dennoch ist der Kontext genau auf die Weise unvollständig, die entscheidend ist.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Correct source\n        +\nlossy transformation\n        =\nmisleading evidence\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2 id=\"section-20\">Wo das Beispiel nicht mehr funktioniert\u003C\u002Fh2>\n\u003Cp>Beweiserhaltung bedeutet nicht, jedes Quellbyte unverändert in den Modellkontext zu kopieren.\u003C\u002Fp>\n\u003Cp>Komprimierung, Normalisierung, Chunking und Zusammenfassung sind oft notwendig. Das Ziel ist, die für eine korrekte Interpretation erforderlichen Eigenschaften zu bewahren, nicht die Transformation zu eliminieren.\u003C\u002Fp>\n\u003Cp>Die schwierige technische Frage lautet daher: Welche Eigenschaften sind semantisch tragend?\u003C\u002Fp>\n\u003Ch2 id=\"section-24\">Direkte Antwort\u003C\u002Fh2>\n\u003Cp>Beweise überleben eine KI-Pipeline, wenn die Transformationen zwischen Quelle und Modell genügend Identität, Herkunft, Struktur, Einheiten, Bedingungen, Konflikte und Gültigkeitsmetadaten bewahren, damit die abgerufene Information im Kontext dasselbe bedeutet wie an der Quelle.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Evidence Preservation\n=\nmeaning\n+ identity\n+ provenance\n+ scope\n+ units\n+ conditions\n+ validity\n+ traceability\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2 id=\"section-27\">Warum RAG selbst mit der richtigen Quelle fehlschlagen kann\u003C\u002Fh2>\n\u003Cp>Ein Retriever kann das richtige Dokument finden und dennoch die falsche Beweiseinheit liefern.\u003C\u002Fp>\n\u003Cp>Das Problem kann viel früher aufgetreten sein: während OCR, Dokumentenparsing, HTML-Extraktion, Tabellenkonvertierung, Feldzuordnung, Chunk-Erstellung oder Metadatenverlust.\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fde\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python\" 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\">Woher bezieht ein LLM seine Daten? RAG-Datenquellen in Python\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">RAG beginnt mit externen Datenquellen und Abrufmethoden. Beweiserhaltung fragt, was zwischen der Originalquelle und dem abgerufenen Kontext erhalten bleiben muss.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ch2 id=\"section-31\">Die acht Erhaltungspunkte\u003C\u002Fh2>\n\u003Ch3 id=\"section-32\">1. Identität\u003C\u002Fh3>\n\u003Cp>Die Pipeline muss bewahren, welche Entität, welches Produkt, Dokument, welchen Benutzer, Vertrag oder Datensatz der Beweis tatsächlich beschreibt.\u003C\u002Fp>\n\u003Ch3 id=\"section-34\">2. Herkunft\u003C\u002Fh3>\n\u003Cp>Das System sollte in der Lage sein, den abgerufenen Beweis bis zu seiner Originalquelle zurückzuverfolgen und nicht nur bis zu einer anonymen Chunk-Kennung.\u003C\u002Fp>\n\u003Ch3 id=\"section-36\">3. Struktur\u003C\u002Fh3>\n\u003Cp>Tabellen, Überschriften, Listen, Eltern-Kind-Beziehungen und Ausnahmen können Bedeutung tragen, die verschwindet, wenn alles in unabhängige Textfragmente zerlegt wird.\u003C\u002Fp>\n\u003Ch3 id=\"section-38\">4. Einheiten\u003C\u002Fh3>\n\u003Cp>Eine Zahl ohne ihre Einheit kann nutzlos oder gefährlich werden. 100 kann Millisekunden, Kilogramm, Euro, Volt oder Prozent eines Basiswerts bedeuten.\u003C\u002Fp>\n\u003Ch3 id=\"section-40\">5. Bedingungen und Ausnahmen\u003C\u002Fh3>\n\u003Cp>Regeln enthalten häufig Anwendbarkeitsbedingungen. Die Trennung der Regel von der Ausnahme kann die Wörter bewahren, während sie die Bedeutung zerstört.\u003C\u002Fp>\n\u003Ch3 id=\"section-42\">6. Version und Zeit\u003C\u002Fh3>\n\u003Cp>Belege sollten bewahren, wann sie gültig wurden, wann sie beobachtet wurden und ob ein neueres Objekt sie ersetzt.\u003C\u002Fp>\n\u003Ch3 id=\"section-44\">7. Konfliktstatus\u003C\u002Fh3>\n\u003Cp>Wenn zwei Quellen nicht übereinstimmen, sollte die Normalisierung nicht stillschweigend einen Wert auswählen und die Tatsache löschen, dass eine Uneinigkeit bestand.\u003C\u002Fp>\n\u003Ch3 id=\"section-46\">8. Zugriffsbereich\u003C\u002Fh3>\n\u003Cp>Berechtigungs- und Mandantengrenzen müssen während des gesamten Abrufs und der Kontextzusammenstellung an Belege gebunden bleiben.\u003C\u002Fp>\n\u003Ch2 id=\"section-48\">Von Rohdaten zu kontrollierten Belegen\u003C\u002Fh2>\n\u003Cp>Der stärkste Ort zur Bewahrung von Belegen liegt oft vor der Vektorisierung.\u003C\u002Fp>\n\u003Cp>Wenn Quellenidentität, Gültigkeit und Konflikte verloren gehen, bevor Embeddings erstellt werden, kann der Abruf sie später nicht magisch rekonstruieren.\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fde\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information\" 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\">Von Rohdaten zu verifizierbarem KI-Kontext: Eine praktische Architektur für agentenbereite Informationen\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Die Architektur beginnt vor dem Abruf, indem sie Identität, Herkunft, Konflikte, Unbekanntes und Gültigkeit in wiederverwendbaren Informationsobjekten bewahrt.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ch2 id=\"section-52\">Kontext-Engineering ist Teil der Kette\u003C\u002Fh2>\n\u003Cp>Selbst korrekt abgerufene Belege können während der Kontextzusammenstellung beschädigt werden.\u003C\u002Fp>\n\u003Cp>Ein Kontext-Builder kann eine Ausnahme abschneiden, Quellenbezeichnungen entfernen, widersprüchliche Passagen zusammenführen, Abhängigkeiten neu anordnen oder das nützliche Aufmerksamkeitsbudget des Modells überschreiten.\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fde\u002Fblog\u002Fwhat-is-context-engineering-what-the-model-receives-before-it-answers\" 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\">Was ist Context Engineering? Was das Modell erhält, bevor es antwortet\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Context Engineering bestimmt, welche abgerufenen Informationen tatsächlich das Modell erreichen und unter welchen strukturellen, Gültigkeits- und Sicherheitsbeschränkungen.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ch2 id=\"section-56\">Verbindung zur Evidenzdistanz\u003C\u002Fh2>\n\u003Cp>Evidenzerhaltung und Evidenzdistanz beschreiben unterschiedliche Fehlermodi.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Evidence Preservation Chain:\nDid the evidence survive transformation?\n\nEvidence Distance:\nHow far is the final claim from that evidence?\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Ein System kann Evidenz perfekt erhalten und dennoch eine nicht gestützte Schlussfolgerung daraus ziehen. Umgekehrt kann die endgültige Argumentation einfach sein, während die Evidenz selbst bereits früher in der Pipeline beschädigt wurde.\u003C\u002Fp>\n\u003Ch2 id=\"section-60\">Verbindung zur Retrieval-Autoritätsschicht\u003C\u002Fh2>\n\u003Cp>Die Retrieval-Autoritätsschicht wählt aus, welche Quelle Evidenzgewicht verdient.\u003C\u002Fp>\n\u003Cp>Die Evidenzerhaltungskette stellt sicher, dass die Eigenschaften, die diese Quelle autoritativ machen, nach der Auswahl nicht verworfen werden.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Authority at source\n        ↓\nPreservation through pipeline\n        ↓\nAuthority still visible in context\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2 id=\"section-64\">Praktisches Datenmodell\u003C\u002Fh2>\n\u003Cp>Eine Retrieval-Einheit kann mehr als Text tragen.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>{\n  "text": "Maximum operating temperature is 40 °C...",\n  "sourceId": "manual-v4.2",\n  "entityId": "device-184",\n  "section": "Operating limits",\n  "validFrom": "2026-03-01",\n  "unit": "°C",\n  "conditions": ["35 °C when accessory X is installed"],\n  "authority": "manufacturer_manual",\n  "conflicts": [],\n  "accessScope": "public"\n}\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Das genaue Schema hängt von der Anwendung ab. Das architektonische Prinzip ist, dass abrufbarer Text nicht von den Metadaten getrennt werden sollte, die zu seiner Interpretation erforderlich sind.\u003C\u002Fp>\n\u003Ch2 id=\"section-68\">Diagnostische \u002F Entscheidungsmethode\u003C\u002Fh2>\n\u003Cp>Wählen Sie eine wichtige Antwort und verfolgen Sie eine stützende Behauptung rückwärts durch das System.\u003C\u002Fp>\n\u003Col>\u003Cli>Identifizieren Sie die endgültige Behauptung.\u003C\u002Fli>\u003Cli>Identifizieren Sie die genaue Kontextpassage, die sie gestützt hat.\u003C\u002Fli>\u003Cli>Identifizieren Sie den abgerufenen Chunk, der diese Passage erzeugt hat.\u003C\u002Fli>\u003Cli>Identifizieren Sie den normalisierten Datensatz oder das geparste Dokument hinter dem Chunk.\u003C\u002Fli>\u003Cli>Identifizieren Sie die ursprüngliche Quelle.\u003C\u002Fli>\u003Cli>Vergleichen Sie Bedeutung, Umfang, Einheiten, Bedingungen und Version bei jedem Übergang.\u003C\u002Fli>\u003C\u002Fol>\n\u003Cp>Jeder Punkt, an dem wesentliche Bedeutung verschwindet, ist ein Bruch in der Evidenzerhaltungskette.\u003C\u002Fp>\n\u003Ch2 id=\"section-72\">Fehlermodus: Der verwaiste Chunk\u003C\u002Fh2>\n\u003Cp>Ein verwaister Chunk enthält nützlich erscheinenden Text, aber keine verlässliche Beziehung zur Dokumentstruktur, Quellidentität, Version oder den Anwendbarkeitsbedingungen, aus denen er stammt.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>"Refunds are available for 30 days."\n\nMissing:\nWhich product?\nWhich country?\nWhich policy version?\nFrom what effective date?\nWhat exceptions?\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Der Satz kann authentisch sein und dennoch unsichere Evidenz darstellen.\u003C\u002Fp>\n\u003Ch2 id=\"section-76\">Fehlermodus: Normalisierung durch Erfindung\u003C\u002Fh2>\n\u003Cp>Normalisierung sollte Darstellungen angleichen. Sie sollte nicht stillschweigend fehlende Werte erfinden, damit Datensätze vollständig erscheinen.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Unknown\n≠\nFalse\n≠\nZero\n≠\nDefault\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Unbekannte zu bewahren ist Teil davon, Evidenz zu bewahren.\u003C\u002Fp>\n\u003Ch2 id=\"section-80\">Fehlermodus: Konfliktkollaps\u003C\u002Fh2>\n\u003Cp>Angenommen, eine genehmigte Quelle sagt 24 Monate und eine andere sagt 36 Monate.\u003C\u002Fp>\n\u003Cp>Eine Pipeline, die einfach den zuletzt importierten Wert speichert, erzeugt eine saubere Datenbank und zerstört die Evidenz, dass ein Konflikt existierte.\u003C\u002Fp>\n\u003Cp>Ein verifizierbares System sollte beide Werte, ihre Provenienz und den ungelösten Konflikt bewahren, bis eine Regel oder eine menschliche Entscheidung ihn auflöst.\u003C\u002Fp>\n\u003Ch2 id=\"section-84\">Fehlermodus: Kontextkompression\u003C\u002Fh2>\n\u003Cp>Kontextkompression kann nützlich sein, aber eine Zusammenfassung, die Qualifikatoren entfernt, kann die scheinbare Gewissheit erhöhen.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Source:\n"Evidence suggests X under conditions A and B, but the sample is small."\n\nBad compression:\n"X is proven."\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Die Information wurde kürzer und gleichzeitig weniger wahrheitsgemäß.\u003C\u002Fp>\n\u003Ch2 id=\"section-88\">Randfall: Verlustbehaftete Kompression kann dennoch korrekt sein\u003C\u002Fh2>\n\u003Cp>Nicht jeder Verlust ist schädlich.\u003C\u002Fp>\n\u003Cp>Das Entfernen visueller Formatierung aus einem Absatz kann keine semantischen Konsequenzen haben. Die Umrechnung von Celsius in Kelvin kann die Bedeutung bewahren, wenn die Transformation explizit und korrekt ist. Das Deduplizieren exakter Kopien kann den Abruf verbessern, ohne Evidenz zu verlieren.\u003C\u002Fp>\n\u003Cp>Das Ziel ist daher nicht die verlustfreie Speicherung in jeder Phase. Es ist die semantisch ausreichende Bewahrung.\u003C\u002Fp>\n\u003Ch2 id=\"section-92\">Einschränkungen\u003C\u002Fh2>\n\u003Cp>Die Beweiserhaltungskette ist ein architektonischer Rahmen, keine Garantie dafür, dass erhaltene Beweise korrekt interpretiert werden.\u003C\u002Fp>\n\u003Cp>Ein Modell kann perfekt erhaltene Beweise erhalten und trotzdem schlecht schlussfolgern. Ein Retriever kann die Herkunft bewahren und trotzdem die falsche Quelle einordnen. Eine Quelle selbst kann auch falsch sein.\u003C\u002Fp>\n\u003Cp>Die Bewahrung schützt die Beweispipeline. Sie ersetzt nicht die Quellenbewertung oder das Schlussfolgern.\u003C\u002Fp>\n\u003Ch2 id=\"section-96\">Was würde diese Antwort ändern?\u003C\u002Fh2>\n\u003Cp>Zukünftige Systeme könnten strukturierte Beweisobjekte anstelle von Textabschnitten abrufen, Herkunftsdiagramme automatisch bewahren und maschinenlesbare Gültigkeitsbeschränkungen direkt für Modelle und Agenten bereitstellen.\u003C\u002Fp>\n\u003Cp>Das würde die Beweiserhaltung einfacher umzusetzen machen, aber es würde das Prinzip wichtiger statt weniger wichtig machen.\u003C\u002Fp>\n\u003Ch2 id=\"section-99\">Fazit\u003C\u002Fh2>\n\u003Cp>Zuverlässige KI erfordert mehr als das Finden einer korrekten Quelle.\u003C\u002Fp>\n\u003Cp>Die Eigenschaften, die die Quelle nützlich machen, müssen Parsing, Normalisierung, Chunking, Retrieval und Kontextzusammenstellung überstehen.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Source Authority\n      ↓\nEvidence Preservation Chain\n      ↓\nRetrievable Evidence\n      ↓\nEvidence Distance\n      ↓\nReasoning\n      ↓\nAnswer Validity Boundary\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Wenn die Kette bricht, kann das Modell Informationen erhalten, die technisch authentisch, aber nicht mehr sicher als die ursprünglichen Beweise zu interpretieren sind.\u003C\u002Fp>\n\u003Ch2 id=\"section-104\">Primärquellen\u003C\u002Fh2>\n\u003Cp>Patrick Lewis et al., \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\" target=\"_blank\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa> (2020).\u003C\u002Fp>\n\u003Cp>Nelson F. Liu et al., \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\" target=\"_blank\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa> (2023).\u003C\u002Fp>\n\u003Cp>W3C, \u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\" target=\"_blank\">PROV Overview\u003C\u002Fa>.\u003C\u002Fp>",{"time":227,"blocks":228,"version":796},1792096456868,[229,235,241,246,251,259,266,271,276,281,287,292,297,302,308,313,318,323,328,333,338,343,348,353,358,363,368,373,378,383,394,399,404,409,414,419,424,429,434,439,444,449,454,459,464,469,474,479,484,489,494,502,507,512,517,526,531,536,541,546,551,556,561,566,571,576,581,586,591,596,611,616,621,626,631,636,641,646,651,656,661,666,671,676,681,686,691,696,701,706,711,716,721,726,731,736,741,746,751,756,761,766,771,776,781,786,791],{"id":230,"data":231,"type":41,"tunes":234},"tP10jlMV2G",{"text":232,"level":233},"Frage",2,{},{"id":236,"data":237,"type":239,"tunes":240},"zD8a-5x7r9",{"text":238},"Was passiert, wenn die ursprüngliche Quelle korrekt ist, aber die KI-Pipeline die Informationen entfernt, die erforderlich sind, um diese Quelle korrekt zu verstehen?","paragraph",{},{"id":242,"data":243,"type":239,"tunes":245},"LJjTjSV74R",{"text":244},"Ein zuverlässiges Dokument kann zu einem unzuverlässigen Kontext werden, wenn das Parsing seine Tabellenstruktur entfernt, das Chunking eine Bedingung von ihrer Ausnahme trennt, die Normalisierung Einheiten fallen lässt, das Retrieval Versionsmetadaten verliert oder die Kontextzusammenstellung einen Konflikt verbirgt.",{},{"id":247,"data":248,"type":239,"tunes":250},"kOzPNiKvzu",{"text":249},"Dieser Artikel verwendet den Begriff Evidence Preservation Chain für die Anforderung, dass Beweise die für eine korrekte Interpretation notwendigen Eigenschaften behalten müssen, während sie von der Quelle zum endgültigen KI-Kontext wandern.",{},{"id":252,"data":253,"type":257,"tunes":258},"xUf9Qt8UHn",{"text":254,"caption":255,"alignment":256},"Eine Evidence Preservation Chain ist die End-to-End-Sequenz von Transformationen, durch die Beweise Identität, Herkunft, Umfang, Bedeutung und Gültigkeit behalten müssen, bevor sie sicher eine KI-generierte Behauptung stützen können.","Arbeitsdefinition — ein hier eingeführtes architektonisches Konzept, kein standardisierter Forschungsbegriff.","left","quote",{},{"id":260,"data":261,"type":264,"tunes":265},"RxQsaJsNo6",{"title":262,"maxLevel":263,"minLevel":233},"Inhalt",3,"tableOfContents",{},{"id":267,"data":268,"type":41,"tunes":270},"NBMzNZ_qHB",{"text":269,"level":233},"Was das wirklich bedeutet",{},{"id":272,"data":273,"type":239,"tunes":275},"zAEKGXCEmK",{"text":274},"Die meisten Diskussionen über RAG konzentrieren sich darauf, ob relevante Informationen gefunden werden können.",{},{"id":277,"data":278,"type":239,"tunes":280},"i2gu8IM3j1",{"text":279},"Aber Beweise durchlaufen normalerweise mehrere Transformationen, bevor das Modell sie sieht.",{},{"id":282,"data":283,"type":285,"tunes":286},"lN52dh3dxK",{"code":284},"Original Source\n   ↓\nParsing\n   ↓\nNormalization\n   ↓\nChunking\n   ↓\nIndexing\n   ↓\nRetrieval\n   ↓\nReranking\n   ↓\nContext Assembly\n   ↓\nModel\n   ↓\nClaim","code",{},{"id":288,"data":289,"type":239,"tunes":291},"G90TdnX4WC",{"text":290},"Die Beweise können an jedem Punkt dieser Kette verfallen, selbst wenn das Originaldokument vollkommen korrekt war.",{},{"id":293,"data":294,"type":41,"tunes":296},"mC_OsF2ji3",{"text":295,"level":233},"Einfachstes Beispiel",{},{"id":298,"data":299,"type":239,"tunes":301},"P8XUivVa7g",{"text":300},"Stellen Sie sich vor, ein Produkthandbuch enthält diese Aussage:",{},{"id":303,"data":304,"type":257,"tunes":307},"2LamDXESHK",{"text":305,"caption":306,"alignment":256},"Maximale Betriebstemperatur: 40 °C, außer wenn Zubehör X installiert ist, in diesem Fall beträgt die Grenze 35 °C.","Beispiel-Quellaussage",{},{"id":309,"data":310,"type":239,"tunes":312},"BNY1TVLg3y",{"text":311},"Ein schlechter Chunking-Prozess teilt den Satz auf.",{},{"id":314,"data":315,"type":285,"tunes":317},"xaAZ_aJazc",{"code":316},"Chunk A:\nMaximum operating temperature: 40 °C.\n\nChunk B:\nExcept when accessory X is installed, in which case the limit is 35 °C.",{},{"id":319,"data":320,"type":239,"tunes":322},"q0ORvmk__K",{"text":321},"Der Retriever gibt nur Chunk A zurück.",{},{"id":324,"data":325,"type":239,"tunes":327},"MupQI-_xIz",{"text":326},"Der abgerufene Text ist nicht erfunden. Er stammt direkt aus dem Handbuch. Dennoch ist der Kontext genau auf die Weise unvollständig, die entscheidend ist.",{},{"id":329,"data":330,"type":285,"tunes":332},"cYEvuH7MR6",{"code":331},"Correct source\n        +\nlossy transformation\n        =\nmisleading evidence",{},{"id":334,"data":335,"type":41,"tunes":337},"F_ZvT1AfTN",{"text":336,"level":233},"Wo das Beispiel nicht mehr funktioniert",{},{"id":339,"data":340,"type":239,"tunes":342},"flP5_wVcvt",{"text":341},"Beweiserhaltung bedeutet nicht, jedes Quellbyte unverändert in den Modellkontext zu kopieren.",{},{"id":344,"data":345,"type":239,"tunes":347},"rrlavifFLP",{"text":346},"Komprimierung, Normalisierung, Chunking und Zusammenfassung sind oft notwendig. Das Ziel ist, die für eine korrekte Interpretation erforderlichen Eigenschaften zu bewahren, nicht die Transformation zu eliminieren.",{},{"id":349,"data":350,"type":239,"tunes":352},"CM_pAZ8ozk",{"text":351},"Die schwierige technische Frage lautet daher: Welche Eigenschaften sind semantisch tragend?",{},{"id":354,"data":355,"type":41,"tunes":357},"U8VCT8z_Nf",{"text":356,"level":233},"Direkte Antwort",{},{"id":359,"data":360,"type":239,"tunes":362},"Hs1rUljfaq",{"text":361},"Beweise überleben eine KI-Pipeline, wenn die Transformationen zwischen Quelle und Modell genügend Identität, Herkunft, Struktur, Einheiten, Bedingungen, Konflikte und Gültigkeitsmetadaten bewahren, damit die abgerufene Information im Kontext dasselbe bedeutet wie an der Quelle.",{},{"id":364,"data":365,"type":285,"tunes":367},"2r2HxbzQFg",{"code":366},"Evidence Preservation\n=\nmeaning\n+ identity\n+ provenance\n+ scope\n+ units\n+ conditions\n+ validity\n+ traceability",{},{"id":369,"data":370,"type":41,"tunes":372},"ka_HV02MYw",{"text":371,"level":233},"Warum RAG selbst mit der richtigen Quelle fehlschlagen kann",{},{"id":374,"data":375,"type":239,"tunes":377},"eKv9aiSIC9",{"text":376},"Ein Retriever kann das richtige Dokument finden und dennoch die falsche Beweiseinheit liefern.",{},{"id":379,"data":380,"type":239,"tunes":382},"-EAX8UkdxQ",{"text":381},"Das Problem kann viel früher aufgetreten sein: während OCR, Dokumentenparsing, HTML-Extraktion, Tabellenkonvertierung, Feldzuordnung, Chunk-Erstellung oder Metadatenverlust.",{},{"id":384,"data":385,"type":392,"tunes":393},"oYCdWaadX3",{"link":386,"meta":387},"https:\u002F\u002Fstajic.de\u002Fde\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python",{"image":388,"title":390,"description":391},{"url":389},"","Woher bezieht ein LLM seine Daten? RAG-Datenquellen in Python","RAG beginnt mit externen Datenquellen und Abrufmethoden. Beweiserhaltung fragt, was zwischen der Originalquelle und dem abgerufenen Kontext erhalten bleiben muss.","linkTool",{},{"id":395,"data":396,"type":41,"tunes":398},"A_6idqO75O",{"text":397,"level":233},"Die acht Erhaltungspunkte",{},{"id":400,"data":401,"type":41,"tunes":403},"NaU6o_qWBJ",{"text":402,"level":263},"1. Identität",{},{"id":405,"data":406,"type":239,"tunes":408},"5mR31I7W1M",{"text":407},"Die Pipeline muss bewahren, welche Entität, welches Produkt, Dokument, welchen Benutzer, Vertrag oder Datensatz der Beweis tatsächlich beschreibt.",{},{"id":410,"data":411,"type":41,"tunes":413},"UIaEdeVESf",{"text":412,"level":263},"2. Herkunft",{},{"id":415,"data":416,"type":239,"tunes":418},"vAtWlAr_L5",{"text":417},"Das System sollte in der Lage sein, den abgerufenen Beweis bis zu seiner Originalquelle zurückzuverfolgen und nicht nur bis zu einer anonymen Chunk-Kennung.",{},{"id":420,"data":421,"type":41,"tunes":423},"6EGnXzdt_j",{"text":422,"level":263},"3. Struktur",{},{"id":425,"data":426,"type":239,"tunes":428},"9NSFX7Dq4k",{"text":427},"Tabellen, Überschriften, Listen, Eltern-Kind-Beziehungen und Ausnahmen können Bedeutung tragen, die verschwindet, wenn alles in unabhängige Textfragmente zerlegt wird.",{},{"id":430,"data":431,"type":41,"tunes":433},"87QZ9m4_q7",{"text":432,"level":263},"4. Einheiten",{},{"id":435,"data":436,"type":239,"tunes":438},"0e7PrlT5MU",{"text":437},"Eine Zahl ohne ihre Einheit kann nutzlos oder gefährlich werden. 100 kann Millisekunden, Kilogramm, Euro, Volt oder Prozent eines Basiswerts bedeuten.",{},{"id":440,"data":441,"type":41,"tunes":443},"Bid-hGyRbP",{"text":442,"level":263},"5. Bedingungen und Ausnahmen",{},{"id":445,"data":446,"type":239,"tunes":448},"Td_5jAm0ME",{"text":447},"Regeln enthalten häufig Anwendbarkeitsbedingungen. Die Trennung der Regel von der Ausnahme kann die Wörter bewahren, während sie die Bedeutung zerstört.",{},{"id":450,"data":451,"type":41,"tunes":453},"y5CedEdzCc",{"text":452,"level":263},"6. Version und Zeit",{},{"id":455,"data":456,"type":239,"tunes":458},"K9nDZI5yQf",{"text":457},"Belege sollten bewahren, wann sie gültig wurden, wann sie beobachtet wurden und ob ein neueres Objekt sie ersetzt.",{},{"id":460,"data":461,"type":41,"tunes":463},"UtnbpFacpC",{"text":462,"level":263},"7. Konfliktstatus",{},{"id":465,"data":466,"type":239,"tunes":468},"rQMwxCKxIE",{"text":467},"Wenn zwei Quellen nicht übereinstimmen, sollte die Normalisierung nicht stillschweigend einen Wert auswählen und die Tatsache löschen, dass eine Uneinigkeit bestand.",{},{"id":470,"data":471,"type":41,"tunes":473},"615Imi6TRP",{"text":472,"level":263},"8. Zugriffsbereich",{},{"id":475,"data":476,"type":239,"tunes":478},"WZuieLPrus",{"text":477},"Berechtigungs- und Mandantengrenzen müssen während des gesamten Abrufs und der Kontextzusammenstellung an Belege gebunden bleiben.",{},{"id":480,"data":481,"type":41,"tunes":483},"je6wCH54vs",{"text":482,"level":233},"Von Rohdaten zu kontrollierten Belegen",{},{"id":485,"data":486,"type":239,"tunes":488},"fqi7FxVIcZ",{"text":487},"Der stärkste Ort zur Bewahrung von Belegen liegt oft vor der Vektorisierung.",{},{"id":490,"data":491,"type":239,"tunes":493},"iA9_KWjiwC",{"text":492},"Wenn Quellenidentität, Gültigkeit und Konflikte verloren gehen, bevor Embeddings erstellt werden, kann der Abruf sie später nicht magisch rekonstruieren.",{},{"id":495,"data":496,"type":392,"tunes":501},"y6cwyGR2PK",{"link":497,"meta":498},"https:\u002F\u002Fstajic.de\u002Fde\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information",{"image":499,"title":177,"description":500},{"url":389},"Die Architektur beginnt vor dem Abruf, indem sie Identität, Herkunft, Konflikte, Unbekanntes und Gültigkeit in wiederverwendbaren Informationsobjekten bewahrt.",{},{"id":503,"data":504,"type":41,"tunes":506},"piEK0zmvxT",{"text":505,"level":233},"Kontext-Engineering ist Teil der Kette",{},{"id":508,"data":509,"type":239,"tunes":511},"Hs-hKf-7cy",{"text":510},"Selbst korrekt abgerufene Belege können während der Kontextzusammenstellung beschädigt werden.",{},{"id":513,"data":514,"type":239,"tunes":516},"kdWLXPTg4h",{"text":515},"Ein Kontext-Builder kann eine Ausnahme abschneiden, Quellenbezeichnungen entfernen, widersprüchliche Passagen zusammenführen, Abhängigkeiten neu anordnen oder das nützliche Aufmerksamkeitsbudget des Modells überschreiten.",{},{"id":518,"data":519,"type":392,"tunes":525},"vQ5Jya2Rv8",{"link":520,"meta":521},"https:\u002F\u002Fstajic.de\u002Fde\u002Fblog\u002Fwhat-is-context-engineering-what-the-model-receives-before-it-answers",{"image":522,"title":523,"description":524},{"url":389},"Was ist Context Engineering? Was das Modell erhält, bevor es antwortet","Context Engineering bestimmt, welche abgerufenen Informationen tatsächlich das Modell erreichen und unter welchen strukturellen, Gültigkeits- und Sicherheitsbeschränkungen.",{},{"id":527,"data":528,"type":41,"tunes":530},"oC18nhSD3Y",{"text":529,"level":233},"Verbindung zur Evidenzdistanz",{},{"id":532,"data":533,"type":239,"tunes":535},"tbewzxhkkL",{"text":534},"Evidenzerhaltung und Evidenzdistanz beschreiben unterschiedliche Fehlermodi.",{},{"id":537,"data":538,"type":285,"tunes":540},"XzYMjN1nNf",{"code":539},"Evidence Preservation Chain:\nDid the evidence survive transformation?\n\nEvidence Distance:\nHow far is the final claim from that evidence?",{},{"id":542,"data":543,"type":239,"tunes":545},"013RZ4lijY",{"text":544},"Ein System kann Evidenz perfekt erhalten und dennoch eine nicht gestützte Schlussfolgerung daraus ziehen. Umgekehrt kann die endgültige Argumentation einfach sein, während die Evidenz selbst bereits früher in der Pipeline beschädigt wurde.",{},{"id":547,"data":548,"type":41,"tunes":550},"cdXu2rD4d-",{"text":549,"level":233},"Verbindung zur Retrieval-Autoritätsschicht",{},{"id":552,"data":553,"type":239,"tunes":555},"asRwQ_zGye",{"text":554},"Die Retrieval-Autoritätsschicht wählt aus, welche Quelle Evidenzgewicht verdient.",{},{"id":557,"data":558,"type":239,"tunes":560},"dMwV7G7LC4",{"text":559},"Die Evidenzerhaltungskette stellt sicher, dass die Eigenschaften, die diese Quelle autoritativ machen, nach der Auswahl nicht verworfen werden.",{},{"id":562,"data":563,"type":285,"tunes":565},"ky0NtD4IAl",{"code":564},"Authority at source\n        ↓\nPreservation through pipeline\n        ↓\nAuthority still visible in context",{},{"id":567,"data":568,"type":41,"tunes":570},"Em0D8AL_b7",{"text":569,"level":233},"Praktisches Datenmodell",{},{"id":572,"data":573,"type":239,"tunes":575},"iDWVeCI3US",{"text":574},"Eine Retrieval-Einheit kann mehr als Text tragen.",{},{"id":577,"data":578,"type":285,"tunes":580},"Ybc3rCbvY0",{"code":579},"{\n  \"text\": \"Maximum operating temperature is 40 °C...\",\n  \"sourceId\": \"manual-v4.2\",\n  \"entityId\": \"device-184\",\n  \"section\": \"Operating limits\",\n  \"validFrom\": \"2026-03-01\",\n  \"unit\": \"°C\",\n  \"conditions\": [\"35 °C when accessory X is installed\"],\n  \"authority\": \"manufacturer_manual\",\n  \"conflicts\": [],\n  \"accessScope\": \"public\"\n}",{},{"id":582,"data":583,"type":239,"tunes":585},"g007D00k4V",{"text":584},"Das genaue Schema hängt von der Anwendung ab. Das architektonische Prinzip ist, dass abrufbarer Text nicht von den Metadaten getrennt werden sollte, die zu seiner Interpretation erforderlich sind.",{},{"id":587,"data":588,"type":41,"tunes":590},"Wi-cP9HpKU",{"text":589,"level":233},"Diagnostische \u002F Entscheidungsmethode",{},{"id":592,"data":593,"type":239,"tunes":595},"yUPTwfB9uP",{"text":594},"Wählen Sie eine wichtige Antwort und verfolgen Sie eine stützende Behauptung rückwärts durch das System.",{},{"id":597,"data":598,"type":609,"tunes":610},"IRrMfVVINa",{"meta":599,"items":601,"style":608},{"counterType":600},"numeric",[602,603,604,605,606,607],"Identifizieren Sie die endgültige Behauptung.","Identifizieren Sie die genaue Kontextpassage, die sie gestützt hat.","Identifizieren Sie den abgerufenen Chunk, der diese Passage erzeugt hat.","Identifizieren Sie den normalisierten Datensatz oder das geparste Dokument hinter dem Chunk.","Identifizieren Sie die ursprüngliche Quelle.","Vergleichen Sie Bedeutung, Umfang, Einheiten, Bedingungen und Version bei jedem Übergang.","ordered","list",{},{"id":612,"data":613,"type":239,"tunes":615},"05LsNWXcjf",{"text":614},"Jeder Punkt, an dem wesentliche Bedeutung verschwindet, ist ein Bruch in der Evidenzerhaltungskette.",{},{"id":617,"data":618,"type":41,"tunes":620},"raBFHAylDf",{"text":619,"level":233},"Fehlermodus: Der verwaiste Chunk",{},{"id":622,"data":623,"type":239,"tunes":625},"iniLj3zXLh",{"text":624},"Ein verwaister Chunk enthält nützlich erscheinenden Text, aber keine verlässliche Beziehung zur Dokumentstruktur, Quellidentität, Version oder den Anwendbarkeitsbedingungen, aus denen er stammt.",{},{"id":627,"data":628,"type":285,"tunes":630},"1zagZ01tls",{"code":629},"\"Refunds are available for 30 days.\"\n\nMissing:\nWhich product?\nWhich country?\nWhich policy version?\nFrom what effective date?\nWhat exceptions?",{},{"id":632,"data":633,"type":239,"tunes":635},"sZ0-pxHB2K",{"text":634},"Der Satz kann authentisch sein und dennoch unsichere Evidenz darstellen.",{},{"id":637,"data":638,"type":41,"tunes":640},"wCo24MdrUE",{"text":639,"level":233},"Fehlermodus: Normalisierung durch Erfindung",{},{"id":642,"data":643,"type":239,"tunes":645},"_8tEy420r3",{"text":644},"Normalisierung sollte Darstellungen angleichen. Sie sollte nicht stillschweigend fehlende Werte erfinden, damit Datensätze vollständig erscheinen.",{},{"id":647,"data":648,"type":285,"tunes":650},"DvCye0QKKX",{"code":649},"Unknown\n≠\nFalse\n≠\nZero\n≠\nDefault",{},{"id":652,"data":653,"type":239,"tunes":655},"gTRNnqfBbR",{"text":654},"Unbekannte zu bewahren ist Teil davon, Evidenz zu bewahren.",{},{"id":657,"data":658,"type":41,"tunes":660},"Azp57iMou6",{"text":659,"level":233},"Fehlermodus: Konfliktkollaps",{},{"id":662,"data":663,"type":239,"tunes":665},"HXmXzrSkqL",{"text":664},"Angenommen, eine genehmigte Quelle sagt 24 Monate und eine andere sagt 36 Monate.",{},{"id":667,"data":668,"type":239,"tunes":670},"qDsZWYHjhc",{"text":669},"Eine Pipeline, die einfach den zuletzt importierten Wert speichert, erzeugt eine saubere Datenbank und zerstört die Evidenz, dass ein Konflikt existierte.",{},{"id":672,"data":673,"type":239,"tunes":675},"NkYagndI9y",{"text":674},"Ein verifizierbares System sollte beide Werte, ihre Provenienz und den ungelösten Konflikt bewahren, bis eine Regel oder eine menschliche Entscheidung ihn auflöst.",{},{"id":677,"data":678,"type":41,"tunes":680},"KrRJTl84Dq",{"text":679,"level":233},"Fehlermodus: Kontextkompression",{},{"id":682,"data":683,"type":239,"tunes":685},"VVy7pkhXMG",{"text":684},"Kontextkompression kann nützlich sein, aber eine Zusammenfassung, die Qualifikatoren entfernt, kann die scheinbare Gewissheit erhöhen.",{},{"id":687,"data":688,"type":285,"tunes":690},"MTqe1-sAHt",{"code":689},"Source:\n\"Evidence suggests X under conditions A and B, but the sample is small.\"\n\nBad compression:\n\"X is proven.\"",{},{"id":692,"data":693,"type":239,"tunes":695},"d61VdY7Z5d",{"text":694},"Die Information wurde kürzer und gleichzeitig weniger wahrheitsgemäß.",{},{"id":697,"data":698,"type":41,"tunes":700},"k4jHwZaoLl",{"text":699,"level":233},"Randfall: Verlustbehaftete Kompression kann dennoch korrekt sein",{},{"id":702,"data":703,"type":239,"tunes":705},"92uTDyGplP",{"text":704},"Nicht jeder Verlust ist schädlich.",{},{"id":707,"data":708,"type":239,"tunes":710},"U0cu-bjBmq",{"text":709},"Das Entfernen visueller Formatierung aus einem Absatz kann keine semantischen Konsequenzen haben. Die Umrechnung von Celsius in Kelvin kann die Bedeutung bewahren, wenn die Transformation explizit und korrekt ist. Das Deduplizieren exakter Kopien kann den Abruf verbessern, ohne Evidenz zu verlieren.",{},{"id":712,"data":713,"type":239,"tunes":715},"GJpi4mOxtH",{"text":714},"Das Ziel ist daher nicht die verlustfreie Speicherung in jeder Phase. Es ist die semantisch ausreichende Bewahrung.",{},{"id":717,"data":718,"type":41,"tunes":720},"7XXF7EHC5m",{"text":719,"level":233},"Einschränkungen",{},{"id":722,"data":723,"type":239,"tunes":725},"FW8TcQldR-",{"text":724},"Die Beweiserhaltungskette ist ein architektonischer Rahmen, keine Garantie dafür, dass erhaltene Beweise korrekt interpretiert werden.",{},{"id":727,"data":728,"type":239,"tunes":730},"V1rP2gB1C8",{"text":729},"Ein Modell kann perfekt erhaltene Beweise erhalten und trotzdem schlecht schlussfolgern. Ein Retriever kann die Herkunft bewahren und trotzdem die falsche Quelle einordnen. Eine Quelle selbst kann auch falsch sein.",{},{"id":732,"data":733,"type":239,"tunes":735},"GCfe1Di90D",{"text":734},"Die Bewahrung schützt die Beweispipeline. Sie ersetzt nicht die Quellenbewertung oder das Schlussfolgern.",{},{"id":737,"data":738,"type":41,"tunes":740},"654Bx39EsI",{"text":739,"level":233},"Was würde diese Antwort ändern?",{},{"id":742,"data":743,"type":239,"tunes":745},"UFx-k4mGz6",{"text":744},"Zukünftige Systeme könnten strukturierte Beweisobjekte anstelle von Textabschnitten abrufen, Herkunftsdiagramme automatisch bewahren und maschinenlesbare Gültigkeitsbeschränkungen direkt für Modelle und Agenten bereitstellen.",{},{"id":747,"data":748,"type":239,"tunes":750},"lQMoO7JEMG",{"text":749},"Das würde die Beweiserhaltung einfacher umzusetzen machen, aber es würde das Prinzip wichtiger statt weniger wichtig machen.",{},{"id":752,"data":753,"type":41,"tunes":755},"gL1c_bEYfC",{"text":754,"level":233},"Fazit",{},{"id":757,"data":758,"type":239,"tunes":760},"xElsMHA7R-",{"text":759},"Zuverlässige KI erfordert mehr als das Finden einer korrekten Quelle.",{},{"id":762,"data":763,"type":239,"tunes":765},"Heg2LJJSzg",{"text":764},"Die Eigenschaften, die die Quelle nützlich machen, müssen Parsing, Normalisierung, Chunking, Retrieval und Kontextzusammenstellung überstehen.",{},{"id":767,"data":768,"type":285,"tunes":770},"MzylWKf1Jm",{"code":769},"Source Authority\n      ↓\nEvidence Preservation Chain\n      ↓\nRetrievable Evidence\n      ↓\nEvidence Distance\n      ↓\nReasoning\n      ↓\nAnswer Validity Boundary",{},{"id":772,"data":773,"type":239,"tunes":775},"m_RMMC7dVx",{"text":774},"Wenn die Kette bricht, kann das Modell Informationen erhalten, die technisch authentisch, aber nicht mehr sicher als die ursprünglichen Beweise zu interpretieren sind.",{},{"id":777,"data":778,"type":41,"tunes":780},"IkNMvsnAJq",{"text":779,"level":233},"Primärquellen",{},{"id":782,"data":783,"type":239,"tunes":785},"QqTaKgzQ8k",{"text":784},"Patrick Lewis et al., \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\" target=\"_blank\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa> (2020).",{},{"id":787,"data":788,"type":239,"tunes":790},"UturEzvxw2",{"text":789},"Nelson F. Liu et al., \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\" target=\"_blank\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa> (2023).",{},{"id":792,"data":793,"type":239,"tunes":795},"I2fpdupo7S",{"text":794},"W3C, \u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\" target=\"_blank\">PROV Overview\u003C\u002Fa>.",{},"2.31","Belege können an der Quelle korrekt sein und nach Parsing, Normalisierung, Chunking, Retrieval und Kontextzusammenstellung dennoch unzuverlässig werden. Die Evidence Preservation Chain verfolgt, ob Identität, Provenienz, Geltungsbereich und Bedeutung die gesamte KI-Pipeline überstehen.","\u002Fuploads\u002F2026\u002F10\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain-1792096396174-urdg9z.webp","can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain-1792096396174-urdg9z","PUBLISHED","2026-10-15T16:31:00.000Z","2026-10-15T20:31:59.026Z","2026-10-15T20:38:15.936Z",{"en":805,"de":806,"sr":807,"es":808,"fr":809,"it":810,"ru":811,"zh":812},"\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Fde\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Fsr\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Fes\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Ffr\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Fit\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Fru\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u002Fzh\u002Fblog\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain",[814,818],{"id":815,"name":816,"slug":817},68,"Risiken, Kontrollen & Nachweise","risks-and-controls",{"id":819,"name":820,"slug":821},49,"Kontrollen & Nachweise","controls",{"id":823,"login":824,"email":389,"displayName":825},"20","rooth8233","Aleksandar Stajić",[827,1159],{"lang":7,"title":223,"content":225,"contentJson":828,"excerpt":797},{"time":227,"blocks":829,"version":796},[830,833,836,839,842,845,848,851,854,857,860,863,866,869,872,875,878,881,884,887,890,893,896,899,902,905,908,911,914,917,922,925,928,931,934,937,940,943,946,949,952,955,958,961,964,967,970,973,976,979,982,987,990,993,996,1001,1004,1007,1010,1013,1016,1019,1022,1025,1028,1031,1034,1037,1040,1043,1048,1051,1054,1057,1060,1063,1066,1069,1072,1075,1078,1081,1084,1087,1090,1093,1096,1099,1102,1105,1108,1111,1114,1117,1120,1123,1126,1129,1132,1135,1138,1141,1144,1147,1150,1153,1156],{"id":230,"data":831,"type":41,"tunes":832},{"text":232,"level":233},{},{"id":236,"data":834,"type":239,"tunes":835},{"text":238},{},{"id":242,"data":837,"type":239,"tunes":838},{"text":244},{},{"id":247,"data":840,"type":239,"tunes":841},{"text":249},{},{"id":252,"data":843,"type":257,"tunes":844},{"text":254,"caption":255,"alignment":256},{},{"id":260,"data":846,"type":264,"tunes":847},{"title":262,"maxLevel":263,"minLevel":233},{},{"id":267,"data":849,"type":41,"tunes":850},{"text":269,"level":233},{},{"id":272,"data":852,"type":239,"tunes":853},{"text":274},{},{"id":277,"data":855,"type":239,"tunes":856},{"text":279},{},{"id":282,"data":858,"type":285,"tunes":859},{"code":284},{},{"id":288,"data":861,"type":239,"tunes":862},{"text":290},{},{"id":293,"data":864,"type":41,"tunes":865},{"text":295,"level":233},{},{"id":298,"data":867,"type":239,"tunes":868},{"text":300},{},{"id":303,"data":870,"type":257,"tunes":871},{"text":305,"caption":306,"alignment":256},{},{"id":309,"data":873,"type":239,"tunes":874},{"text":311},{},{"id":314,"data":876,"type":285,"tunes":877},{"code":316},{},{"id":319,"data":879,"type":239,"tunes":880},{"text":321},{},{"id":324,"data":882,"type":239,"tunes":883},{"text":326},{},{"id":329,"data":885,"type":285,"tunes":886},{"code":331},{},{"id":334,"data":888,"type":41,"tunes":889},{"text":336,"level":233},{},{"id":339,"data":891,"type":239,"tunes":892},{"text":341},{},{"id":344,"data":894,"type":239,"tunes":895},{"text":346},{},{"id":349,"data":897,"type":239,"tunes":898},{"text":351},{},{"id":354,"data":900,"type":41,"tunes":901},{"text":356,"level":233},{},{"id":359,"data":903,"type":239,"tunes":904},{"text":361},{},{"id":364,"data":906,"type":285,"tunes":907},{"code":366},{},{"id":369,"data":909,"type":41,"tunes":910},{"text":371,"level":233},{},{"id":374,"data":912,"type":239,"tunes":913},{"text":376},{},{"id":379,"data":915,"type":239,"tunes":916},{"text":381},{},{"id":384,"data":918,"type":392,"tunes":921},{"link":386,"meta":919},{"image":920,"title":390,"description":391},{"url":389},{},{"id":395,"data":923,"type":41,"tunes":924},{"text":397,"level":233},{},{"id":400,"data":926,"type":41,"tunes":927},{"text":402,"level":263},{},{"id":405,"data":929,"type":239,"tunes":930},{"text":407},{},{"id":410,"data":932,"type":41,"tunes":933},{"text":412,"level":263},{},{"id":415,"data":935,"type":239,"tunes":936},{"text":417},{},{"id":420,"data":938,"type":41,"tunes":939},{"text":422,"level":263},{},{"id":425,"data":941,"type":239,"tunes":942},{"text":427},{},{"id":430,"data":944,"type":41,"tunes":945},{"text":432,"level":263},{},{"id":435,"data":947,"type":239,"tunes":948},{"text":437},{},{"id":440,"data":950,"type":41,"tunes":951},{"text":442,"level":263},{},{"id":445,"data":953,"type":239,"tunes":954},{"text":447},{},{"id":450,"data":956,"type":41,"tunes":957},{"text":452,"level":263},{},{"id":455,"data":959,"type":239,"tunes":960},{"text":457},{},{"id":460,"data":962,"type":41,"tunes":963},{"text":462,"level":263},{},{"id":465,"data":965,"type":239,"tunes":966},{"text":467},{},{"id":470,"data":968,"type":41,"tunes":969},{"text":472,"level":263},{},{"id":475,"data":971,"type":239,"tunes":972},{"text":477},{},{"id":480,"data":974,"type":41,"tunes":975},{"text":482,"level":233},{},{"id":485,"data":977,"type":239,"tunes":978},{"text":487},{},{"id":490,"data":980,"type":239,"tunes":981},{"text":492},{},{"id":495,"data":983,"type":392,"tunes":986},{"link":497,"meta":984},{"image":985,"title":177,"description":500},{"url":389},{},{"id":503,"data":988,"type":41,"tunes":989},{"text":505,"level":233},{},{"id":508,"data":991,"type":239,"tunes":992},{"text":510},{},{"id":513,"data":994,"type":239,"tunes":995},{"text":515},{},{"id":518,"data":997,"type":392,"tunes":1000},{"link":520,"meta":998},{"image":999,"title":523,"description":524},{"url":389},{},{"id":527,"data":1002,"type":41,"tunes":1003},{"text":529,"level":233},{},{"id":532,"data":1005,"type":239,"tunes":1006},{"text":534},{},{"id":537,"data":1008,"type":285,"tunes":1009},{"code":539},{},{"id":542,"data":1011,"type":239,"tunes":1012},{"text":544},{},{"id":547,"data":1014,"type":41,"tunes":1015},{"text":549,"level":233},{},{"id":552,"data":1017,"type":239,"tunes":1018},{"text":554},{},{"id":557,"data":1020,"type":239,"tunes":1021},{"text":559},{},{"id":562,"data":1023,"type":285,"tunes":1024},{"code":564},{},{"id":567,"data":1026,"type":41,"tunes":1027},{"text":569,"level":233},{},{"id":572,"data":1029,"type":239,"tunes":1030},{"text":574},{},{"id":577,"data":1032,"type":285,"tunes":1033},{"code":579},{},{"id":582,"data":1035,"type":239,"tunes":1036},{"text":584},{},{"id":587,"data":1038,"type":41,"tunes":1039},{"text":589,"level":233},{},{"id":592,"data":1041,"type":239,"tunes":1042},{"text":594},{},{"id":597,"data":1044,"type":609,"tunes":1047},{"meta":1045,"items":1046,"style":608},{"counterType":600},[602,603,604,605,606,607],{},{"id":612,"data":1049,"type":239,"tunes":1050},{"text":614},{},{"id":617,"data":1052,"type":41,"tunes":1053},{"text":619,"level":233},{},{"id":622,"data":1055,"type":239,"tunes":1056},{"text":624},{},{"id":627,"data":1058,"type":285,"tunes":1059},{"code":629},{},{"id":632,"data":1061,"type":239,"tunes":1062},{"text":634},{},{"id":637,"data":1064,"type":41,"tunes":1065},{"text":639,"level":233},{},{"id":642,"data":1067,"type":239,"tunes":1068},{"text":644},{},{"id":647,"data":1070,"type":285,"tunes":1071},{"code":649},{},{"id":652,"data":1073,"type":239,"tunes":1074},{"text":654},{},{"id":657,"data":1076,"type":41,"tunes":1077},{"text":659,"level":233},{},{"id":662,"data":1079,"type":239,"tunes":1080},{"text":664},{},{"id":667,"data":1082,"type":239,"tunes":1083},{"text":669},{},{"id":672,"data":1085,"type":239,"tunes":1086},{"text":674},{},{"id":677,"data":1088,"type":41,"tunes":1089},{"text":679,"level":233},{},{"id":682,"data":1091,"type":239,"tunes":1092},{"text":684},{},{"id":687,"data":1094,"type":285,"tunes":1095},{"code":689},{},{"id":692,"data":1097,"type":239,"tunes":1098},{"text":694},{},{"id":697,"data":1100,"type":41,"tunes":1101},{"text":699,"level":233},{},{"id":702,"data":1103,"type":239,"tunes":1104},{"text":704},{},{"id":707,"data":1106,"type":239,"tunes":1107},{"text":709},{},{"id":712,"data":1109,"type":239,"tunes":1110},{"text":714},{},{"id":717,"data":1112,"type":41,"tunes":1113},{"text":719,"level":233},{},{"id":722,"data":1115,"type":239,"tunes":1116},{"text":724},{},{"id":727,"data":1118,"type":239,"tunes":1119},{"text":729},{},{"id":732,"data":1121,"type":239,"tunes":1122},{"text":734},{},{"id":737,"data":1124,"type":41,"tunes":1125},{"text":739,"level":233},{},{"id":742,"data":1127,"type":239,"tunes":1128},{"text":744},{},{"id":747,"data":1130,"type":239,"tunes":1131},{"text":749},{},{"id":752,"data":1133,"type":41,"tunes":1134},{"text":754,"level":233},{},{"id":757,"data":1136,"type":239,"tunes":1137},{"text":759},{},{"id":762,"data":1139,"type":239,"tunes":1140},{"text":764},{},{"id":767,"data":1142,"type":285,"tunes":1143},{"code":769},{},{"id":772,"data":1145,"type":239,"tunes":1146},{"text":774},{},{"id":777,"data":1148,"type":41,"tunes":1149},{"text":779,"level":233},{},{"id":782,"data":1151,"type":239,"tunes":1152},{"text":784},{},{"id":787,"data":1154,"type":239,"tunes":1155},{"text":789},{},{"id":792,"data":1157,"type":239,"tunes":1158},{"text":794},{},{"lang":1160,"title":1161,"content":1162,"contentJson":1163,"excerpt":1602},"en","Can Evidence Survive the AI Pipeline? — The Evidence Preservation Chain","{\"time\":1792096415735,\"blocks\":[{\"id\":\"tP10jlMV2G\",\"type\":\"header\",\"data\":{\"text\":\"Question\",\"level\":2},\"tunes\":{}},{\"id\":\"zD8a-5x7r9\",\"type\":\"paragraph\",\"data\":{\"text\":\"What happens when the original source is correct, but the AI pipeline strips away the information required to understand that source correctly?\"},\"tunes\":{}},{\"id\":\"LJjTjSV74R\",\"type\":\"paragraph\",\"data\":{\"text\":\"A reliable document can become unreliable context if parsing removes its table structure, chunking separates a condition from its exception, normalization drops units, retrieval loses version metadata or context assembly hides a conflict.\"},\"tunes\":{}},{\"id\":\"kOzPNiKvzu\",\"type\":\"paragraph\",\"data\":{\"text\":\"This article uses the term Evidence Preservation Chain for the requirement that evidence retain the properties necessary for correct interpretation as it moves from source to final AI context.\"},\"tunes\":{}},{\"id\":\"xUf9Qt8UHn\",\"type\":\"quote\",\"data\":{\"text\":\"An Evidence Preservation Chain is the end-to-end sequence of transformations through which evidence must retain identity, provenance, scope, meaning and validity before it can safely support an AI-generated claim.\",\"caption\":\"Working definition — an architectural concept introduced here, not a standardized research term.\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"RxQsaJsNo6\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"NBMzNZ_qHB\",\"type\":\"header\",\"data\":{\"text\":\"What This Really Means\",\"level\":2},\"tunes\":{}},{\"id\":\"zAEKGXCEmK\",\"type\":\"paragraph\",\"data\":{\"text\":\"Most discussions about RAG focus on whether relevant information can be found.\"},\"tunes\":{}},{\"id\":\"i2gu8IM3j1\",\"type\":\"paragraph\",\"data\":{\"text\":\"But evidence usually passes through several transformations before the model sees it.\"},\"tunes\":{}},{\"id\":\"lN52dh3dxK\",\"type\":\"code\",\"data\":{\"code\":\"Original Source\\n   ↓\\nParsing\\n   ↓\\nNormalization\\n   ↓\\nChunking\\n   ↓\\nIndexing\\n   ↓\\nRetrieval\\n   ↓\\nReranking\\n   ↓\\nContext Assembly\\n   ↓\\nModel\\n   ↓\\nClaim\"},\"tunes\":{}},{\"id\":\"G90TdnX4WC\",\"type\":\"paragraph\",\"data\":{\"text\":\"The evidence can degrade at any point in that chain even when the original document was perfectly correct.\"},\"tunes\":{}},{\"id\":\"mC_OsF2ji3\",\"type\":\"header\",\"data\":{\"text\":\"Simplest Example\",\"level\":2},\"tunes\":{}},{\"id\":\"P8XUivVa7g\",\"type\":\"paragraph\",\"data\":{\"text\":\"Imagine a product manual contains this statement:\"},\"tunes\":{}},{\"id\":\"2LamDXESHK\",\"type\":\"quote\",\"data\":{\"text\":\"Maximum operating temperature: 40 °C, except when accessory X is installed, in which case the limit is 35 °C.\",\"caption\":\"Example source statement\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"BNY1TVLg3y\",\"type\":\"paragraph\",\"data\":{\"text\":\"A poor chunking process splits the sentence.\"},\"tunes\":{}},{\"id\":\"xaAZ_aJazc\",\"type\":\"code\",\"data\":{\"code\":\"Chunk A:\\nMaximum operating temperature: 40 °C.\\n\\nChunk B:\\nExcept when accessory X is installed, in which case the limit is 35 °C.\"},\"tunes\":{}},{\"id\":\"q0ORvmk__K\",\"type\":\"paragraph\",\"data\":{\"text\":\"The retriever returns only Chunk A.\"},\"tunes\":{}},{\"id\":\"MupQI-_xIz\",\"type\":\"paragraph\",\"data\":{\"text\":\"The retrieved text is not fabricated. It came directly from the manual. Yet the context is incomplete in exactly the way that matters.\"},\"tunes\":{}},{\"id\":\"cYEvuH7MR6\",\"type\":\"code\",\"data\":{\"code\":\"Correct source\\n        +\\nlossy transformation\\n        =\\nmisleading evidence\"},\"tunes\":{}},{\"id\":\"F_ZvT1AfTN\",\"type\":\"header\",\"data\":{\"text\":\"Where the Example Stops Working\",\"level\":2},\"tunes\":{}},{\"id\":\"flP5_wVcvt\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence preservation does not mean copying every source byte unchanged into the model context.\"},\"tunes\":{}},{\"id\":\"rrlavifFLP\",\"type\":\"paragraph\",\"data\":{\"text\":\"Compression, normalization, chunking and summarization are often necessary. The goal is to preserve the properties required for correct interpretation, not to eliminate transformation.\"},\"tunes\":{}},{\"id\":\"CM_pAZ8ozk\",\"type\":\"paragraph\",\"data\":{\"text\":\"The difficult engineering question is therefore: Which properties are semantically load-bearing?\"},\"tunes\":{}},{\"id\":\"U8VCT8z_Nf\",\"type\":\"header\",\"data\":{\"text\":\"Direct Answer\",\"level\":2},\"tunes\":{}},{\"id\":\"Hs1rUljfaq\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence survives an AI pipeline when the transformations between source and model preserve enough identity, provenance, structure, units, conditions, conflicts and validity metadata for the retrieved information to mean the same thing in context that it meant at the source.\"},\"tunes\":{}},{\"id\":\"2r2HxbzQFg\",\"type\":\"code\",\"data\":{\"code\":\"Evidence Preservation\\n=\\nmeaning\\n+ identity\\n+ provenance\\n+ scope\\n+ units\\n+ conditions\\n+ validity\\n+ traceability\"},\"tunes\":{}},{\"id\":\"ka_HV02MYw\",\"type\":\"header\",\"data\":{\"text\":\"Why RAG Can Fail Even With the Right Source\",\"level\":2},\"tunes\":{}},{\"id\":\"eKv9aiSIC9\",\"type\":\"paragraph\",\"data\":{\"text\":\"A retriever can find the correct document while still supplying the wrong evidence unit.\"},\"tunes\":{}},{\"id\":\"-EAX8UkdxQ\",\"type\":\"paragraph\",\"data\":{\"text\":\"The problem may have occurred much earlier: during OCR, document parsing, HTML extraction, table conversion, field mapping, chunk construction or metadata loss.\"},\"tunes\":{}},{\"id\":\"oYCdWaadX3\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"Where Does an LLM Get Its Data? RAG Data Sources in Python\",\"description\":\"RAG begins with external data sources and retrieval methods. Evidence preservation asks what must survive between the original source and the retrieved context.\"}},\"tunes\":{}},{\"id\":\"A_6idqO75O\",\"type\":\"header\",\"data\":{\"text\":\"The Eight Preservation Points\",\"level\":2},\"tunes\":{}},{\"id\":\"NaU6o_qWBJ\",\"type\":\"header\",\"data\":{\"text\":\"1. Identity\",\"level\":3},\"tunes\":{}},{\"id\":\"5mR31I7W1M\",\"type\":\"paragraph\",\"data\":{\"text\":\"The pipeline must preserve what entity, product, document, user, contract or record the evidence actually describes.\"},\"tunes\":{}},{\"id\":\"UIaEdeVESf\",\"type\":\"header\",\"data\":{\"text\":\"2. Provenance\",\"level\":3},\"tunes\":{}},{\"id\":\"vAtWlAr_L5\",\"type\":\"paragraph\",\"data\":{\"text\":\"The system should be able to trace the retrieved evidence back to its original source rather than only to an anonymous chunk identifier.\"},\"tunes\":{}},{\"id\":\"6EGnXzdt_j\",\"type\":\"header\",\"data\":{\"text\":\"3. Structure\",\"level\":3},\"tunes\":{}},{\"id\":\"9NSFX7Dq4k\",\"type\":\"paragraph\",\"data\":{\"text\":\"Tables, headings, lists, parent-child relationships and exceptions can carry meaning that disappears when everything is flattened into independent text fragments.\"},\"tunes\":{}},{\"id\":\"87QZ9m4_q7\",\"type\":\"header\",\"data\":{\"text\":\"4. Units\",\"level\":3},\"tunes\":{}},{\"id\":\"0e7PrlT5MU\",\"type\":\"paragraph\",\"data\":{\"text\":\"A number without its unit can become useless or dangerous. 100 may mean milliseconds, kilograms, euros, volts or percent of a baseline.\"},\"tunes\":{}},{\"id\":\"Bid-hGyRbP\",\"type\":\"header\",\"data\":{\"text\":\"5. Conditions and Exceptions\",\"level\":3},\"tunes\":{}},{\"id\":\"Td_5jAm0ME\",\"type\":\"paragraph\",\"data\":{\"text\":\"Rules frequently contain applicability conditions. Separating the rule from the exception can preserve the words while destroying the meaning.\"},\"tunes\":{}},{\"id\":\"y5CedEdzCc\",\"type\":\"header\",\"data\":{\"text\":\"6. Version and Time\",\"level\":3},\"tunes\":{}},{\"id\":\"K9nDZI5yQf\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence should preserve when it became valid, when it was observed and whether a newer object supersedes it.\"},\"tunes\":{}},{\"id\":\"UtnbpFacpC\",\"type\":\"header\",\"data\":{\"text\":\"7. Conflict State\",\"level\":3},\"tunes\":{}},{\"id\":\"rQMwxCKxIE\",\"type\":\"paragraph\",\"data\":{\"text\":\"If two sources disagree, normalization should not silently select one value and erase the fact that disagreement existed.\"},\"tunes\":{}},{\"id\":\"615Imi6TRP\",\"type\":\"header\",\"data\":{\"text\":\"8. Access Scope\",\"level\":3},\"tunes\":{}},{\"id\":\"WZuieLPrus\",\"type\":\"paragraph\",\"data\":{\"text\":\"Permission and tenant boundaries must remain attached to evidence throughout retrieval and context assembly.\"},\"tunes\":{}},{\"id\":\"je6wCH54vs\",\"type\":\"header\",\"data\":{\"text\":\"From Raw Data to Controlled Evidence\",\"level\":2},\"tunes\":{}},{\"id\":\"fqi7FxVIcZ\",\"type\":\"paragraph\",\"data\":{\"text\":\"The strongest place to preserve evidence is often before vectorization.\"},\"tunes\":{}},{\"id\":\"iA9_KWjiwC\",\"type\":\"paragraph\",\"data\":{\"text\":\"If source identity, validity and conflicts are lost before embeddings are created, retrieval cannot magically reconstruct them later.\"},\"tunes\":{}},{\"id\":\"y6cwyGR2PK\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"From Raw Data to Verifiable AI Context: A Practical Architecture for Agent-Ready Information\",\"description\":\"The architecture starts before retrieval by preserving identity, provenance, conflicts, unknowns and validity in reusable information objects.\"}},\"tunes\":{}},{\"id\":\"piEK0zmvxT\",\"type\":\"header\",\"data\":{\"text\":\"Context Engineering Is Part of the Chain\",\"level\":2},\"tunes\":{}},{\"id\":\"Hs-hKf-7cy\",\"type\":\"paragraph\",\"data\":{\"text\":\"Even correctly retrieved evidence can be damaged during context assembly.\"},\"tunes\":{}},{\"id\":\"kdWLXPTg4h\",\"type\":\"paragraph\",\"data\":{\"text\":\"A context builder may truncate an exception, remove source labels, merge conflicting passages, reorder dependencies or exceed the model's useful attention budget.\"},\"tunes\":{}},{\"id\":\"vQ5Jya2Rv8\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhat-is-context-engineering-what-the-model-receives-before-it-answers\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"What Is Context Engineering? What the Model Receives Before It Answers\",\"description\":\"Context engineering determines which retrieved information actually reaches the model and under which structural, validity and security constraints.\"}},\"tunes\":{}},{\"id\":\"oC18nhSD3Y\",\"type\":\"header\",\"data\":{\"text\":\"Connection to Evidence Distance\",\"level\":2},\"tunes\":{}},{\"id\":\"tbewzxhkkL\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Preservation and Evidence Distance describe different failure modes.\"},\"tunes\":{}},{\"id\":\"XzYMjN1nNf\",\"type\":\"code\",\"data\":{\"code\":\"Evidence Preservation Chain:\\nDid the evidence survive transformation?\\n\\nEvidence Distance:\\nHow far is the final claim from that evidence?\"},\"tunes\":{}},{\"id\":\"013RZ4lijY\",\"type\":\"paragraph\",\"data\":{\"text\":\"A system can preserve evidence perfectly and still make an unsupported inference from it. Conversely, the final reasoning can be simple while the evidence itself was already corrupted earlier in the pipeline.\"},\"tunes\":{}},{\"id\":\"cdXu2rD4d-\",\"type\":\"header\",\"data\":{\"text\":\"Connection to the Retrieval Authority Layer\",\"level\":2},\"tunes\":{}},{\"id\":\"asRwQ_zGye\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Retrieval Authority Layer selects which source deserves evidential weight.\"},\"tunes\":{}},{\"id\":\"dMwV7G7LC4\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Evidence Preservation Chain ensures that the properties making that source authoritative are not discarded after selection.\"},\"tunes\":{}},{\"id\":\"ky0NtD4IAl\",\"type\":\"code\",\"data\":{\"code\":\"Authority at source\\n        ↓\\nPreservation through pipeline\\n        ↓\\nAuthority still visible in context\"},\"tunes\":{}},{\"id\":\"Em0D8AL_b7\",\"type\":\"header\",\"data\":{\"text\":\"Practical Data Model\",\"level\":2},\"tunes\":{}},{\"id\":\"iDWVeCI3US\",\"type\":\"paragraph\",\"data\":{\"text\":\"A retrieval unit can carry more than text.\"},\"tunes\":{}},{\"id\":\"Ybc3rCbvY0\",\"type\":\"code\",\"data\":{\"code\":\"{\\n  \\\"text\\\": \\\"Maximum operating temperature is 40 °C...\\\",\\n  \\\"sourceId\\\": \\\"manual-v4.2\\\",\\n  \\\"entityId\\\": \\\"device-184\\\",\\n  \\\"section\\\": \\\"Operating limits\\\",\\n  \\\"validFrom\\\": \\\"2026-03-01\\\",\\n  \\\"unit\\\": \\\"°C\\\",\\n  \\\"conditions\\\": [\\\"35 °C when accessory X is installed\\\"],\\n  \\\"authority\\\": \\\"manufacturer_manual\\\",\\n  \\\"conflicts\\\": [],\\n  \\\"accessScope\\\": \\\"public\\\"\\n}\"},\"tunes\":{}},{\"id\":\"g007D00k4V\",\"type\":\"paragraph\",\"data\":{\"text\":\"The exact schema depends on the application. The architectural principle is that retrievable text should not become detached from the metadata required to interpret it.\"},\"tunes\":{}},{\"id\":\"Wi-cP9HpKU\",\"type\":\"header\",\"data\":{\"text\":\"Diagnostic \u002F Decision Method\",\"level\":2},\"tunes\":{}},{\"id\":\"yUPTwfB9uP\",\"type\":\"paragraph\",\"data\":{\"text\":\"Choose one important answer and trace one supporting claim backwards through the system.\"},\"tunes\":{}},{\"id\":\"IRrMfVVINa\",\"type\":\"list\",\"data\":{\"style\":\"ordered\",\"meta\":{\"counterType\":\"numeric\"},\"items\":[\"Identify the final claim.\",\"Identify the exact context passage that supported it.\",\"Identify the retrieved chunk that produced that passage.\",\"Identify the normalized record or parsed document behind the chunk.\",\"Identify the original source.\",\"Compare meaning, scope, units, conditions and version at every transition.\"]},\"tunes\":{}},{\"id\":\"05LsNWXcjf\",\"type\":\"paragraph\",\"data\":{\"text\":\"Any point where essential meaning disappears is a break in the Evidence Preservation Chain.\"},\"tunes\":{}},{\"id\":\"raBFHAylDf\",\"type\":\"header\",\"data\":{\"text\":\"Failure Mode: The Orphan Chunk\",\"level\":2},\"tunes\":{}},{\"id\":\"iniLj3zXLh\",\"type\":\"paragraph\",\"data\":{\"text\":\"An orphan chunk contains useful-looking text but no reliable relationship to the document structure, source identity, version or applicability conditions from which it came.\"},\"tunes\":{}},{\"id\":\"1zagZ01tls\",\"type\":\"code\",\"data\":{\"code\":\"\\\"Refunds are available for 30 days.\\\"\\n\\nMissing:\\nWhich product?\\nWhich country?\\nWhich policy version?\\nFrom what effective date?\\nWhat exceptions?\"},\"tunes\":{}},{\"id\":\"sZ0-pxHB2K\",\"type\":\"paragraph\",\"data\":{\"text\":\"The sentence may be authentic and still be unsafe evidence.\"},\"tunes\":{}},{\"id\":\"wCo24MdrUE\",\"type\":\"header\",\"data\":{\"text\":\"Failure Mode: Normalization by Invention\",\"level\":2},\"tunes\":{}},{\"id\":\"_8tEy420r3\",\"type\":\"paragraph\",\"data\":{\"text\":\"Normalization should align representations. It should not silently invent missing values to make records look complete.\"},\"tunes\":{}},{\"id\":\"DvCye0QKKX\",\"type\":\"code\",\"data\":{\"code\":\"Unknown\\n≠\\nFalse\\n≠\\nZero\\n≠\\nDefault\"},\"tunes\":{}},{\"id\":\"gTRNnqfBbR\",\"type\":\"paragraph\",\"data\":{\"text\":\"Preserving unknowns is part of preserving evidence.\"},\"tunes\":{}},{\"id\":\"Azp57iMou6\",\"type\":\"header\",\"data\":{\"text\":\"Failure Mode: Conflict Collapse\",\"level\":2},\"tunes\":{}},{\"id\":\"HXmXzrSkqL\",\"type\":\"paragraph\",\"data\":{\"text\":\"Suppose one approved source says 24 months and another says 36 months.\"},\"tunes\":{}},{\"id\":\"qDsZWYHjhc\",\"type\":\"paragraph\",\"data\":{\"text\":\"A pipeline that simply stores the last imported value produces a clean database and destroys the evidence that a conflict existed.\"},\"tunes\":{}},{\"id\":\"NkYagndI9y\",\"type\":\"paragraph\",\"data\":{\"text\":\"A verifiable system should preserve both values, their provenance and the unresolved conflict until a rule or human decision resolves it.\"},\"tunes\":{}},{\"id\":\"KrRJTl84Dq\",\"type\":\"header\",\"data\":{\"text\":\"Failure Mode: Context Compression\",\"level\":2},\"tunes\":{}},{\"id\":\"VVy7pkhXMG\",\"type\":\"paragraph\",\"data\":{\"text\":\"Context compression can be useful, but a summary that removes qualifiers can increase apparent certainty.\"},\"tunes\":{}},{\"id\":\"MTqe1-sAHt\",\"type\":\"code\",\"data\":{\"code\":\"Source:\\n\\\"Evidence suggests X under conditions A and B, but the sample is small.\\\"\\n\\nBad compression:\\n\\\"X is proven.\\\"\"},\"tunes\":{}},{\"id\":\"d61VdY7Z5d\",\"type\":\"paragraph\",\"data\":{\"text\":\"The information became shorter and less truthful at the same time.\"},\"tunes\":{}},{\"id\":\"k4jHwZaoLl\",\"type\":\"header\",\"data\":{\"text\":\"Edge Case: Lossy Compression Can Still Be Correct\",\"level\":2},\"tunes\":{}},{\"id\":\"92uTDyGplP\",\"type\":\"paragraph\",\"data\":{\"text\":\"Not every loss is harmful.\"},\"tunes\":{}},{\"id\":\"U0cu-bjBmq\",\"type\":\"paragraph\",\"data\":{\"text\":\"Removing visual formatting from a paragraph may have no semantic consequence. Converting Celsius to Kelvin can preserve meaning if the transformation is explicit and correct. Deduplicating exact copies can improve retrieval without losing evidence.\"},\"tunes\":{}},{\"id\":\"GJpi4mOxtH\",\"type\":\"paragraph\",\"data\":{\"text\":\"The objective is therefore not lossless storage at every stage. It is semantically sufficient preservation.\"},\"tunes\":{}},{\"id\":\"7XXF7EHC5m\",\"type\":\"header\",\"data\":{\"text\":\"Limitations\",\"level\":2},\"tunes\":{}},{\"id\":\"FW8TcQldR-\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Evidence Preservation Chain is an architectural framework, not a guarantee that preserved evidence will be interpreted correctly.\"},\"tunes\":{}},{\"id\":\"V1rP2gB1C8\",\"type\":\"paragraph\",\"data\":{\"text\":\"A model can receive perfectly preserved evidence and still reason badly. A retriever can preserve provenance and still rank the wrong source. A source itself can also be wrong.\"},\"tunes\":{}},{\"id\":\"GCfe1Di90D\",\"type\":\"paragraph\",\"data\":{\"text\":\"Preservation protects the evidence pipeline. It does not replace source evaluation or reasoning.\"},\"tunes\":{}},{\"id\":\"654Bx39EsI\",\"type\":\"header\",\"data\":{\"text\":\"What Would Change This Answer?\",\"level\":2},\"tunes\":{}},{\"id\":\"UFx-k4mGz6\",\"type\":\"paragraph\",\"data\":{\"text\":\"Future systems may retrieve structured evidence objects rather than text chunks, preserve provenance graphs automatically and expose machine-readable validity constraints directly to models and agents.\"},\"tunes\":{}},{\"id\":\"lQMoO7JEMG\",\"type\":\"paragraph\",\"data\":{\"text\":\"That would make evidence preservation easier to implement, but it would make the principle more important rather than less important.\"},\"tunes\":{}},{\"id\":\"gL1c_bEYfC\",\"type\":\"header\",\"data\":{\"text\":\"Conclusion\",\"level\":2},\"tunes\":{}},{\"id\":\"xElsMHA7R-\",\"type\":\"paragraph\",\"data\":{\"text\":\"Reliable AI requires more than finding a correct source.\"},\"tunes\":{}},{\"id\":\"Heg2LJJSzg\",\"type\":\"paragraph\",\"data\":{\"text\":\"The properties that make the source useful must survive parsing, normalization, chunking, retrieval and context assembly.\"},\"tunes\":{}},{\"id\":\"MzylWKf1Jm\",\"type\":\"code\",\"data\":{\"code\":\"Source Authority\\n      ↓\\nEvidence Preservation Chain\\n      ↓\\nRetrievable Evidence\\n      ↓\\nEvidence Distance\\n      ↓\\nReasoning\\n      ↓\\nAnswer Validity Boundary\"},\"tunes\":{}},{\"id\":\"m_RMMC7dVx\",\"type\":\"paragraph\",\"data\":{\"text\":\"If the chain breaks, the model may receive information that is technically authentic but no longer safe to interpret as the original evidence.\"},\"tunes\":{}},{\"id\":\"IkNMvsnAJq\",\"type\":\"header\",\"data\":{\"text\":\"Primary Sources\",\"level\":2},\"tunes\":{}},{\"id\":\"QqTaKgzQ8k\",\"type\":\"paragraph\",\"data\":{\"text\":\"Patrick Lewis et al., \u003Ca href=\\\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\\\" target=\\\"_blank\\\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa> (2020).\"},\"tunes\":{}},{\"id\":\"UturEzvxw2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Nelson F. Liu et al., \u003Ca href=\\\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\\\" target=\\\"_blank\\\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa> (2023).\"},\"tunes\":{}},{\"id\":\"I2fpdupo7S\",\"type\":\"paragraph\",\"data\":{\"text\":\"W3C, \u003Ca href=\\\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\\\" target=\\\"_blank\\\">PROV Overview\u003C\u002Fa>.\"},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":1164,"blocks":1165,"version":1601},1792096415735,[1166,1170,1174,1178,1182,1187,1191,1195,1199,1203,1206,1210,1214,1218,1223,1227,1230,1234,1238,1241,1245,1249,1253,1257,1261,1265,1268,1272,1276,1280,1288,1292,1296,1300,1304,1308,1312,1316,1320,1324,1328,1332,1336,1340,1344,1348,1352,1356,1360,1364,1368,1376,1380,1384,1388,1396,1400,1404,1407,1411,1415,1419,1423,1426,1430,1434,1437,1441,1445,1449,1460,1464,1468,1472,1475,1479,1483,1487,1490,1494,1498,1502,1506,1510,1514,1518,1521,1525,1529,1533,1537,1541,1545,1549,1553,1557,1561,1565,1569,1573,1577,1581,1584,1588,1592,1595,1598],{"id":230,"data":1167,"type":41,"tunes":1169},{"text":1168,"level":233},"Question",{},{"id":236,"data":1171,"type":239,"tunes":1173},{"text":1172},"What happens when the original source is correct, but the AI pipeline strips away the information required to understand that source correctly?",{},{"id":242,"data":1175,"type":239,"tunes":1177},{"text":1176},"A reliable document can become unreliable context if parsing removes its table structure, chunking separates a condition from its exception, normalization drops units, retrieval loses version metadata or context assembly hides a conflict.",{},{"id":247,"data":1179,"type":239,"tunes":1181},{"text":1180},"This article uses the term Evidence Preservation Chain for the requirement that evidence retain the properties necessary for correct interpretation as it moves from source to final AI context.",{},{"id":252,"data":1183,"type":257,"tunes":1186},{"text":1184,"caption":1185,"alignment":256},"An Evidence Preservation Chain is the end-to-end sequence of transformations through which evidence must retain identity, provenance, scope, meaning and validity before it can safely support an AI-generated claim.","Working definition — an architectural concept introduced here, not a standardized research term.",{},{"id":260,"data":1188,"type":264,"tunes":1190},{"title":1189,"maxLevel":263,"minLevel":233},"Contents",{},{"id":267,"data":1192,"type":41,"tunes":1194},{"text":1193,"level":233},"What This Really Means",{},{"id":272,"data":1196,"type":239,"tunes":1198},{"text":1197},"Most discussions about RAG focus on whether relevant information can be found.",{},{"id":277,"data":1200,"type":239,"tunes":1202},{"text":1201},"But evidence usually passes through several transformations before the model sees it.",{},{"id":282,"data":1204,"type":285,"tunes":1205},{"code":284},{},{"id":288,"data":1207,"type":239,"tunes":1209},{"text":1208},"The evidence can degrade at any point in that chain even when the original document was perfectly correct.",{},{"id":293,"data":1211,"type":41,"tunes":1213},{"text":1212,"level":233},"Simplest Example",{},{"id":298,"data":1215,"type":239,"tunes":1217},{"text":1216},"Imagine a product manual contains this statement:",{},{"id":303,"data":1219,"type":257,"tunes":1222},{"text":1220,"caption":1221,"alignment":256},"Maximum operating temperature: 40 °C, except when accessory X is installed, in which case the limit is 35 °C.","Example source statement",{},{"id":309,"data":1224,"type":239,"tunes":1226},{"text":1225},"A poor chunking process splits the sentence.",{},{"id":314,"data":1228,"type":285,"tunes":1229},{"code":316},{},{"id":319,"data":1231,"type":239,"tunes":1233},{"text":1232},"The retriever returns only Chunk A.",{},{"id":324,"data":1235,"type":239,"tunes":1237},{"text":1236},"The retrieved text is not fabricated. It came directly from the manual. Yet the context is incomplete in exactly the way that matters.",{},{"id":329,"data":1239,"type":285,"tunes":1240},{"code":331},{},{"id":334,"data":1242,"type":41,"tunes":1244},{"text":1243,"level":233},"Where the Example Stops Working",{},{"id":339,"data":1246,"type":239,"tunes":1248},{"text":1247},"Evidence preservation does not mean copying every source byte unchanged into the model context.",{},{"id":344,"data":1250,"type":239,"tunes":1252},{"text":1251},"Compression, normalization, chunking and summarization are often necessary. The goal is to preserve the properties required for correct interpretation, not to eliminate transformation.",{},{"id":349,"data":1254,"type":239,"tunes":1256},{"text":1255},"The difficult engineering question is therefore: Which properties are semantically load-bearing?",{},{"id":354,"data":1258,"type":41,"tunes":1260},{"text":1259,"level":233},"Direct Answer",{},{"id":359,"data":1262,"type":239,"tunes":1264},{"text":1263},"Evidence survives an AI pipeline when the transformations between source and model preserve enough identity, provenance, structure, units, conditions, conflicts and validity metadata for the retrieved information to mean the same thing in context that it meant at the source.",{},{"id":364,"data":1266,"type":285,"tunes":1267},{"code":366},{},{"id":369,"data":1269,"type":41,"tunes":1271},{"text":1270,"level":233},"Why RAG Can Fail Even With the Right Source",{},{"id":374,"data":1273,"type":239,"tunes":1275},{"text":1274},"A retriever can find the correct document while still supplying the wrong evidence unit.",{},{"id":379,"data":1277,"type":239,"tunes":1279},{"text":1278},"The problem may have occurred much earlier: during OCR, document parsing, HTML extraction, table conversion, field mapping, chunk construction or metadata loss.",{},{"id":384,"data":1281,"type":392,"tunes":1287},{"link":1282,"meta":1283},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python",{"image":1284,"title":1285,"description":1286},{"url":389},"Where Does an LLM Get Its Data? RAG Data Sources in Python","RAG begins with external data sources and retrieval methods. Evidence preservation asks what must survive between the original source and the retrieved context.",{},{"id":395,"data":1289,"type":41,"tunes":1291},{"text":1290,"level":233},"The Eight Preservation Points",{},{"id":400,"data":1293,"type":41,"tunes":1295},{"text":1294,"level":263},"1. Identity",{},{"id":405,"data":1297,"type":239,"tunes":1299},{"text":1298},"The pipeline must preserve what entity, product, document, user, contract or record the evidence actually describes.",{},{"id":410,"data":1301,"type":41,"tunes":1303},{"text":1302,"level":263},"2. Provenance",{},{"id":415,"data":1305,"type":239,"tunes":1307},{"text":1306},"The system should be able to trace the retrieved evidence back to its original source rather than only to an anonymous chunk identifier.",{},{"id":420,"data":1309,"type":41,"tunes":1311},{"text":1310,"level":263},"3. Structure",{},{"id":425,"data":1313,"type":239,"tunes":1315},{"text":1314},"Tables, headings, lists, parent-child relationships and exceptions can carry meaning that disappears when everything is flattened into independent text fragments.",{},{"id":430,"data":1317,"type":41,"tunes":1319},{"text":1318,"level":263},"4. Units",{},{"id":435,"data":1321,"type":239,"tunes":1323},{"text":1322},"A number without its unit can become useless or dangerous. 100 may mean milliseconds, kilograms, euros, volts or percent of a baseline.",{},{"id":440,"data":1325,"type":41,"tunes":1327},{"text":1326,"level":263},"5. Conditions and Exceptions",{},{"id":445,"data":1329,"type":239,"tunes":1331},{"text":1330},"Rules frequently contain applicability conditions. Separating the rule from the exception can preserve the words while destroying the meaning.",{},{"id":450,"data":1333,"type":41,"tunes":1335},{"text":1334,"level":263},"6. Version and Time",{},{"id":455,"data":1337,"type":239,"tunes":1339},{"text":1338},"Evidence should preserve when it became valid, when it was observed and whether a newer object supersedes it.",{},{"id":460,"data":1341,"type":41,"tunes":1343},{"text":1342,"level":263},"7. Conflict State",{},{"id":465,"data":1345,"type":239,"tunes":1347},{"text":1346},"If two sources disagree, normalization should not silently select one value and erase the fact that disagreement existed.",{},{"id":470,"data":1349,"type":41,"tunes":1351},{"text":1350,"level":263},"8. Access Scope",{},{"id":475,"data":1353,"type":239,"tunes":1355},{"text":1354},"Permission and tenant boundaries must remain attached to evidence throughout retrieval and context assembly.",{},{"id":480,"data":1357,"type":41,"tunes":1359},{"text":1358,"level":233},"From Raw Data to Controlled Evidence",{},{"id":485,"data":1361,"type":239,"tunes":1363},{"text":1362},"The strongest place to preserve evidence is often before vectorization.",{},{"id":490,"data":1365,"type":239,"tunes":1367},{"text":1366},"If source identity, validity and conflicts are lost before embeddings are created, retrieval cannot magically reconstruct them later.",{},{"id":495,"data":1369,"type":392,"tunes":1375},{"link":1370,"meta":1371},"https:\u002F\u002Fstajic.de\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information",{"image":1372,"title":1373,"description":1374},{"url":389},"From Raw Data to Verifiable AI Context: A Practical Architecture for Agent-Ready Information","The architecture starts before retrieval by preserving identity, provenance, conflicts, unknowns and validity in reusable information objects.",{},{"id":503,"data":1377,"type":41,"tunes":1379},{"text":1378,"level":233},"Context Engineering Is Part of the Chain",{},{"id":508,"data":1381,"type":239,"tunes":1383},{"text":1382},"Even correctly retrieved evidence can be damaged during context assembly.",{},{"id":513,"data":1385,"type":239,"tunes":1387},{"text":1386},"A context builder may truncate an exception, remove source labels, merge conflicting passages, reorder dependencies or exceed the model's useful attention budget.",{},{"id":518,"data":1389,"type":392,"tunes":1395},{"link":1390,"meta":1391},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhat-is-context-engineering-what-the-model-receives-before-it-answers",{"image":1392,"title":1393,"description":1394},{"url":389},"What Is Context Engineering? What the Model Receives Before It Answers","Context engineering determines which retrieved information actually reaches the model and under which structural, validity and security constraints.",{},{"id":527,"data":1397,"type":41,"tunes":1399},{"text":1398,"level":233},"Connection to Evidence Distance",{},{"id":532,"data":1401,"type":239,"tunes":1403},{"text":1402},"Evidence Preservation and Evidence Distance describe different failure modes.",{},{"id":537,"data":1405,"type":285,"tunes":1406},{"code":539},{},{"id":542,"data":1408,"type":239,"tunes":1410},{"text":1409},"A system can preserve evidence perfectly and still make an unsupported inference from it. Conversely, the final reasoning can be simple while the evidence itself was already corrupted earlier in the pipeline.",{},{"id":547,"data":1412,"type":41,"tunes":1414},{"text":1413,"level":233},"Connection to the Retrieval Authority Layer",{},{"id":552,"data":1416,"type":239,"tunes":1418},{"text":1417},"The Retrieval Authority Layer selects which source deserves evidential weight.",{},{"id":557,"data":1420,"type":239,"tunes":1422},{"text":1421},"The Evidence Preservation Chain ensures that the properties making that source authoritative are not discarded after selection.",{},{"id":562,"data":1424,"type":285,"tunes":1425},{"code":564},{},{"id":567,"data":1427,"type":41,"tunes":1429},{"text":1428,"level":233},"Practical Data Model",{},{"id":572,"data":1431,"type":239,"tunes":1433},{"text":1432},"A retrieval unit can carry more than text.",{},{"id":577,"data":1435,"type":285,"tunes":1436},{"code":579},{},{"id":582,"data":1438,"type":239,"tunes":1440},{"text":1439},"The exact schema depends on the application. The architectural principle is that retrievable text should not become detached from the metadata required to interpret it.",{},{"id":587,"data":1442,"type":41,"tunes":1444},{"text":1443,"level":233},"Diagnostic \u002F Decision Method",{},{"id":592,"data":1446,"type":239,"tunes":1448},{"text":1447},"Choose one important answer and trace one supporting claim backwards through the system.",{},{"id":597,"data":1450,"type":609,"tunes":1459},{"meta":1451,"items":1452,"style":608},{"counterType":600},[1453,1454,1455,1456,1457,1458],"Identify the final claim.","Identify the exact context passage that supported it.","Identify the retrieved chunk that produced that passage.","Identify the normalized record or parsed document behind the chunk.","Identify the original source.","Compare meaning, scope, units, conditions and version at every transition.",{},{"id":612,"data":1461,"type":239,"tunes":1463},{"text":1462},"Any point where essential meaning disappears is a break in the Evidence Preservation Chain.",{},{"id":617,"data":1465,"type":41,"tunes":1467},{"text":1466,"level":233},"Failure Mode: The Orphan Chunk",{},{"id":622,"data":1469,"type":239,"tunes":1471},{"text":1470},"An orphan chunk contains useful-looking text but no reliable relationship to the document structure, source identity, version or applicability conditions from which it came.",{},{"id":627,"data":1473,"type":285,"tunes":1474},{"code":629},{},{"id":632,"data":1476,"type":239,"tunes":1478},{"text":1477},"The sentence may be authentic and still be unsafe evidence.",{},{"id":637,"data":1480,"type":41,"tunes":1482},{"text":1481,"level":233},"Failure Mode: Normalization by Invention",{},{"id":642,"data":1484,"type":239,"tunes":1486},{"text":1485},"Normalization should align representations. It should not silently invent missing values to make records look complete.",{},{"id":647,"data":1488,"type":285,"tunes":1489},{"code":649},{},{"id":652,"data":1491,"type":239,"tunes":1493},{"text":1492},"Preserving unknowns is part of preserving evidence.",{},{"id":657,"data":1495,"type":41,"tunes":1497},{"text":1496,"level":233},"Failure Mode: Conflict Collapse",{},{"id":662,"data":1499,"type":239,"tunes":1501},{"text":1500},"Suppose one approved source says 24 months and another says 36 months.",{},{"id":667,"data":1503,"type":239,"tunes":1505},{"text":1504},"A pipeline that simply stores the last imported value produces a clean database and destroys the evidence that a conflict existed.",{},{"id":672,"data":1507,"type":239,"tunes":1509},{"text":1508},"A verifiable system should preserve both values, their provenance and the unresolved conflict until a rule or human decision resolves it.",{},{"id":677,"data":1511,"type":41,"tunes":1513},{"text":1512,"level":233},"Failure Mode: Context Compression",{},{"id":682,"data":1515,"type":239,"tunes":1517},{"text":1516},"Context compression can be useful, but a summary that removes qualifiers can increase apparent certainty.",{},{"id":687,"data":1519,"type":285,"tunes":1520},{"code":689},{},{"id":692,"data":1522,"type":239,"tunes":1524},{"text":1523},"The information became shorter and less truthful at the same time.",{},{"id":697,"data":1526,"type":41,"tunes":1528},{"text":1527,"level":233},"Edge Case: Lossy Compression Can Still Be Correct",{},{"id":702,"data":1530,"type":239,"tunes":1532},{"text":1531},"Not every loss is harmful.",{},{"id":707,"data":1534,"type":239,"tunes":1536},{"text":1535},"Removing visual formatting from a paragraph may have no semantic consequence. Converting Celsius to Kelvin can preserve meaning if the transformation is explicit and correct. Deduplicating exact copies can improve retrieval without losing evidence.",{},{"id":712,"data":1538,"type":239,"tunes":1540},{"text":1539},"The objective is therefore not lossless storage at every stage. It is semantically sufficient preservation.",{},{"id":717,"data":1542,"type":41,"tunes":1544},{"text":1543,"level":233},"Limitations",{},{"id":722,"data":1546,"type":239,"tunes":1548},{"text":1547},"The Evidence Preservation Chain is an architectural framework, not a guarantee that preserved evidence will be interpreted correctly.",{},{"id":727,"data":1550,"type":239,"tunes":1552},{"text":1551},"A model can receive perfectly preserved evidence and still reason badly. A retriever can preserve provenance and still rank the wrong source. A source itself can also be wrong.",{},{"id":732,"data":1554,"type":239,"tunes":1556},{"text":1555},"Preservation protects the evidence pipeline. It does not replace source evaluation or reasoning.",{},{"id":737,"data":1558,"type":41,"tunes":1560},{"text":1559,"level":233},"What Would Change This Answer?",{},{"id":742,"data":1562,"type":239,"tunes":1564},{"text":1563},"Future systems may retrieve structured evidence objects rather than text chunks, preserve provenance graphs automatically and expose machine-readable validity constraints directly to models and agents.",{},{"id":747,"data":1566,"type":239,"tunes":1568},{"text":1567},"That would make evidence preservation easier to implement, but it would make the principle more important rather than less important.",{},{"id":752,"data":1570,"type":41,"tunes":1572},{"text":1571,"level":233},"Conclusion",{},{"id":757,"data":1574,"type":239,"tunes":1576},{"text":1575},"Reliable AI requires more than finding a correct source.",{},{"id":762,"data":1578,"type":239,"tunes":1580},{"text":1579},"The properties that make the source useful must survive parsing, normalization, chunking, retrieval and context assembly.",{},{"id":767,"data":1582,"type":285,"tunes":1583},{"code":769},{},{"id":772,"data":1585,"type":239,"tunes":1587},{"text":1586},"If the chain breaks, the model may receive information that is technically authentic but no longer safe to interpret as the original evidence.",{},{"id":777,"data":1589,"type":41,"tunes":1591},{"text":1590,"level":233},"Primary Sources",{},{"id":782,"data":1593,"type":239,"tunes":1594},{"text":784},{},{"id":787,"data":1596,"type":239,"tunes":1597},{"text":789},{},{"id":792,"data":1599,"type":239,"tunes":1600},{"text":794},{},"2.31.6","Evidence can be correct at the source and still become unreliable after parsing, normalization, chunking, retrieval and context assembly. The Evidence Preservation Chain tracks whether identity, provenance, scope and meaning survive the entire AI pipeline.","Post erfolgreich abgerufen",{"items":1605,"source":1662,"manualIds":1663,"manualMatchedIds":1664},[1606,1613,1620,1627,1634,1641,1648,1655],{"id":1607,"slug":1608,"title":1609,"excerpt":1610,"featuredImage":1611,"publishedAt":1612},"481","generative-ai-explained-models-retrieval-tools-and-applications-are-not-the-same-thing","Generative KI erklärt: Modelle, Retrieval, Tools und Anwendungen sind nicht dasselbe","Generative KI ist mehr als ein Modell. Erfahren Sie, wie Modelle, Retrieval, Tools, Kontext, Runtimes und Anwendungen in Produktions-KI-Systemen zusammenwirken.","\u002Fuploads\u002F2026\u002F10\u002Fgenerative-ai-explained-models-retrieval-tools-and-applications-are-not-the-same-thing-1791475411822-pp0dvz.webp","2026-10-08T12:00:00.000Z",{"id":1614,"slug":1615,"title":1616,"excerpt":1617,"featuredImage":1618,"publishedAt":1619},"381","enterprise-grade-multi-tenant-architecture-for-an-international-platform","Unternehmensfähige mandantenfähige Architektur für eine internationale Plattform","Loving Rocks ist eine Hochzeitsplattform auf Unternehmensniveau, konzipiert mit einer echten Mehrmandantenarchitektur, isolierten Datenbanken pro Mandant und integrierter Internationalisierung für globale Skalierbarkeit, Sicherheit und langfristige Betriebsstabilität.","\u002Fuploads\u002F2026\u002F01\u002Fenterprise-grade-multi-tenant-architecture-for-an-international-platform-1769789121298-b6v7ak.webp","2026-01-30T12:04:00.000Z",{"id":1621,"slug":1622,"title":1623,"excerpt":1624,"featuredImage":1625,"publishedAt":1626},"495","sovereign-ai-control-of-models-data-infrastructure-and-dependencies","Souveräne KI: Kontrolle über Modelle, Daten, Infrastruktur und Abhängigkeiten","Souveräne KI bedeutet wirksame Kontrolle über Modelle, Daten, Infrastruktur, Software, Betrieb und strategische Abhängigkeiten – nicht einfach, wo ein KI-Modell gehostet wird.","\u002Fuploads\u002F2026\u002F10\u002Fsovereign-ai-control-of-models-data-infrastructure-and-dependencies-1791488833132-niy85x.webp","2026-10-08T15:45:00.000Z",{"id":1628,"slug":1629,"title":1630,"excerpt":1631,"featuredImage":1632,"publishedAt":1633},"492","mcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits","MCP erklärt: Was es verbindet, was es nicht tut und wo es passt","Das Model Context Protocol verbindet KI-Anwendungen über eine standardisierte Client-Server-Grenze mit externen Tools, Ressourcen und Prompts. Erfahren Sie, was MCP tut, was es nicht tut und wo es in der Agentenarchitektur einzuordnen ist.","\u002Fuploads\u002F2026\u002F10\u002Fmcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits-1791486640275-7ub1cq.webp","2026-10-08T15:09:00.000Z",{"id":1635,"slug":1636,"title":1637,"excerpt":1638,"featuredImage":1639,"publishedAt":1640},"482","adr-vs-nfr-architecture-decisions-and-system-quality-are-not-the-same-thing","ADR vs. NFR: Architekturentscheidungen und Systemqualität sind nicht dasselbe","ADR vs. NFR erklärt: Erfahren Sie, wie Systemqualitätsanforderungen Architekturentscheidungen beeinflussen, wie ADRs Abwägungen dokumentieren und warum die Validierung getrennt bleibt.","\u002Fuploads\u002F2026\u002F10\u002Fadr-vs-nfr-architecture-decisions-and-system-quality-are-not-the-same-thing-1791475921511-6zgen1.webp","2026-10-08T12:11:00.000Z",{"id":1642,"slug":1643,"title":1644,"excerpt":1645,"featuredImage":1646,"publishedAt":1647},"494","air-gapped-ai-how-ai-systems-work-without-internet-or-cloud-access","Luftgetrennte KI: Wie KI-Systeme ohne Internet- oder Cloud-Zugriff funktionieren","Air-gapped AI führt Modelle, RAG und KI-Anwendungen innerhalb einer isolierten Sicherheitsdomäne ohne Internet- oder Cloud-Abhängigkeiten aus. Erfahren Sie, wie Modelle, Daten, Updates und Tools offline funktionieren.","\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":1649,"slug":1650,"title":1651,"excerpt":1652,"featuredImage":1653,"publishedAt":1654},"497","how-far-is-an-ai-claim-from-the-evidence-evidence-distance","Wie weit ist eine KI-Behauptung von den Beweisen entfernt? — Evidenzabstand","Nicht alle unterstützten Antworten stehen ihren Belegen gleich nahe. Die Evidenzdistanz ist eine praktische Möglichkeit zu beschreiben, wie viele Transformationen, Zusammenfassungen, Annahmen und Inferenzschritte eine KI-Behauptung von direkt überprüfbaren Belegen trennen.","\u002Fuploads\u002F2026\u002F10\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance-1792095863209-3c1mog.webp","2026-10-15T00:23:00.000Z",{"id":1656,"slug":1657,"title":1658,"excerpt":1659,"featuredImage":1660,"publishedAt":1661},"491","ai-governance-models-data-permissions-risk-and-auditability","KI-Governance: Modelle, Daten, Berechtigungen, Risiko und Auditierbarkeit","KI-Governance definiert, wer KI-Systeme über Modelle, Anbieter, Daten, Berechtigungen, Risiken, Evaluierung und den gesamten Lebenszyklus hinweg genehmigen, betreiben, ändern und prüfen darf.","\u002Fuploads\u002F2026\u002F10\u002Fai-governance-models-data-permissions-risk-and-auditability-1791485901301-g60xyu.webp","2026-10-08T12:56:00.000Z","fallback",[],[]]