[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:it":3,"public-menus:all":38,"post:can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain:it":221,"related:post:can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain:it:1":1606},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","it","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,191,206],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-globe-2","custom",[],{"id":176,"title":177,"url":186,"target":61,"icon":187,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":189,"children":190},"menu-2916d060-8681-482a-af9a-c107a8bc30d8",{"de":178,"en":179,"es":180,"fr":181,"it":182,"ru":183,"sr":184,"zh":185},"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":192,"title":193,"url":202,"target":61,"icon":203,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":204,"children":205},"menu-e2061cd7-ff53-49d5-a0e8-3d49090577c0",{"de":194,"en":195,"es":196,"fr":197,"it":198,"ru":199,"sr":200,"zh":201},"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":207,"title":208,"url":217,"target":61,"icon":218,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":219,"children":220},"menu-a781582e-e90d-418e-b8c9-b070bc37c37e",{"de":209,"en":210,"es":211,"fr":212,"it":213,"ru":214,"sr":215,"zh":216},"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":222,"message":1605},{"id":223,"title":224,"slug":225,"content":226,"contentJson":227,"excerpt":799,"featuredImage":800,"featuredImageAlt":801,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":802,"publishedAt":803,"createdAt":804,"updatedAt":805,"seoLocalePaths":806,"categories":815,"author":824,"translations":828},"498","Può la prova sopravvivere alla pipeline dell'IA? — La catena di conservazione delle prove","can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","\u003Ch2 id=\"section-1\">Domanda\u003C\u002Fh2>\n\u003Cp>Cosa succede quando la fonte originale è corretta, ma la pipeline di IA elimina le informazioni necessarie per comprendere correttamente quella fonte?\u003C\u002Fp>\n\u003Cp>Un documento affidabile può diventare un contesto inaffidabile se il parsing rimuove la sua struttura tabellare, il chunking separa una condizione dalla sua eccezione, la normalizzazione elimina le unità, il retrieval perde i metadati di versione o l'assemblaggio del contesto nasconde un conflitto.\u003C\u002Fp>\n\u003Cp>Questo articolo utilizza il termine Catena di Conservazione delle Prove per indicare il requisito che le prove mantengano le proprietà necessarie per una corretta interpretazione mentre passano dalla fonte al contesto finale dell'IA.\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">Una Catena di Conservazione delle Prove è la sequenza end-to-end di trasformazioni attraverso cui le prove devono mantenere identità, provenienza, ambito, significato e validità prima di poter supportare in modo sicuro un'affermazione generata dall'IA.\u003Ccite class=\"block mt-2 text-sm\">— Definizione operativa — un concetto architetturale introdotto qui, non un termine di ricerca standardizzato.\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"Contenuti\">\u003Cstrong class=\"editorjs-toc__title\">Contenuti\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\">Domanda\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-7\" class=\"editorjs-toc__link\">Cosa Significa Davvero\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-12\" class=\"editorjs-toc__link\">Esempio più Semplice\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-20\" class=\"editorjs-toc__link\">Dove l'esempio smette di funzionare\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-24\" class=\"editorjs-toc__link\">Risposta diretta\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-27\" class=\"editorjs-toc__link\">Perché il RAG può fallire anche con la fonte giusta\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">Gli otto punti di conservazione\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. Identità\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-34\" class=\"editorjs-toc__link\">2. Provenienza\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">3. Struttura\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">4. Unità\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-40\" class=\"editorjs-toc__link\">5. Condizioni ed eccezioni\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-42\" class=\"editorjs-toc__link\">6. Versione e tempo\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-44\" class=\"editorjs-toc__link\">7. Stato di conflitto\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-46\" class=\"editorjs-toc__link\">8. Ambito di accesso\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-48\" class=\"editorjs-toc__link\">Dai dati grezzi alle prove controllate\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-52\" class=\"editorjs-toc__link\">L'ingegneria del contesto fa parte della catena\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">Connessione alla distanza probatoria\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-60\" class=\"editorjs-toc__link\">Connessione al livello di autorità di recupero\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-64\" class=\"editorjs-toc__link\">Modello dati pratico\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-68\" class=\"editorjs-toc__link\">Metodo diagnostico \u002F decisionale\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-72\" class=\"editorjs-toc__link\">Modalità di fallimento: il chunk orfano\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-76\" class=\"editorjs-toc__link\">Modalità di errore: normalizzazione per invenzione\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-80\" class=\"editorjs-toc__link\">Modalità di errore: collasso dei conflitti\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-84\" class=\"editorjs-toc__link\">Modalità di errore: compressione del contesto\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-88\" class=\"editorjs-toc__link\">Caso limite: la compressione con perdita può comunque essere corretta\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-92\" class=\"editorjs-toc__link\">Limitazioni\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-96\" class=\"editorjs-toc__link\">Cosa Cambierebbe Questa Risposta?\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-99\" class=\"editorjs-toc__link\">Conclusione\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-104\" class=\"editorjs-toc__link\">Fonti Primarie\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Ch2 id=\"section-7\">Cosa Significa Davvero\u003C\u002Fh2>\n\u003Cp>La maggior parte delle discussioni sul RAG si concentra sulla possibilità di trovare informazioni rilevanti.\u003C\u002Fp>\n\u003Cp>Ma le prove di solito passano attraverso diverse trasformazioni prima che il modello le veda.\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>Le prove possono degradarsi in qualsiasi punto di quella catena anche quando il documento originale era perfettamente corretto.\u003C\u002Fp>\n\u003Ch2 id=\"section-12\">Esempio più Semplice\u003C\u002Fh2>\n\u003Cp>Immagina che un manuale di prodotto contenga questa affermazione:\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">Temperatura massima di esercizio: 40 °C, tranne quando è installato l'accessorio X, nel qual caso il limite è 35 °C.\u003Ccite class=\"block mt-2 text-sm\">— Affermazione di esempio dalla fonte\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Cp>Un processo di chunking scadente divide la frase.\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>Il retriever restituisce solo il Chunk A.\u003C\u002Fp>\n\u003Cp>Il testo recuperato non è fabbricato. Proviene direttamente dal manuale. Eppure il contesto è incompleto esattamente nel modo che conta.\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\">Dove l'esempio smette di funzionare\u003C\u002Fh2>\n\u003Cp>La conservazione delle prove non significa copiare ogni byte della fonte invariato nel contesto del modello.\u003C\u002Fp>\n\u003Cp>Compressione, normalizzazione, suddivisione in blocchi e riepilogo sono spesso necessari. L'obiettivo è preservare le proprietà richieste per un'interpretazione corretta, non eliminare la trasformazione.\u003C\u002Fp>\n\u003Cp>La difficile questione ingegneristica è quindi: quali proprietà sono semanticamente portanti?\u003C\u002Fp>\n\u003Ch2 id=\"section-24\">Risposta diretta\u003C\u002Fh2>\n\u003Cp>Una prova sopravvive a una pipeline di IA quando le trasformazioni tra fonte e modello preservano sufficienti metadati di identità, provenienza, struttura, unità, condizioni, conflitti e validità affinché l'informazione recuperata significhi in contesto la stessa cosa che significava alla fonte.\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\">Perché il RAG può fallire anche con la fonte giusta\u003C\u002Fh2>\n\u003Cp>Un retriever può trovare il documento corretto pur fornendo l'unità di prova sbagliata.\u003C\u002Fp>\n\u003Cp>Il problema potrebbe essersi verificato molto prima: durante l'OCR, il parsing del documento, l'estrazione HTML, la conversione delle tabelle, la mappatura dei campi, la costruzione dei blocchi o la perdita di metadati.\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fit\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\">Da dove ottiene i dati un LLM? Fonti dati RAG in Python\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Il RAG inizia con fonti dati esterne e metodi di recupero. La conservazione delle prove si chiede cosa deve sopravvivere tra la fonte originale e il contesto recuperato.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ch2 id=\"section-31\">Gli otto punti di conservazione\u003C\u002Fh2>\n\u003Ch3 id=\"section-32\">1. Identità\u003C\u002Fh3>\n\u003Cp>La pipeline deve preservare quale entità, prodotto, documento, utente, contratto o record la prova descrive effettivamente.\u003C\u002Fp>\n\u003Ch3 id=\"section-34\">2. Provenienza\u003C\u002Fh3>\n\u003Cp>Il sistema dovrebbe essere in grado di risalire dalla prova recuperata alla sua fonte originale, non solo a un identificatore di blocco anonimo.\u003C\u002Fp>\n\u003Ch3 id=\"section-36\">3. Struttura\u003C\u002Fh3>\n\u003Cp>Tabelle, intestazioni, elenchi, relazioni padre-figlio ed eccezioni possono portare un significato che scompare quando tutto viene appiattito in frammenti di testo indipendenti.\u003C\u002Fp>\n\u003Ch3 id=\"section-38\">4. Unità\u003C\u002Fh3>\n\u003Cp>Un numero senza la sua unità può diventare inutile o pericoloso. 100 può significare millisecondi, chilogrammi, euro, volt o percentuale di una linea di base.\u003C\u002Fp>\n\u003Ch3 id=\"section-40\">5. Condizioni ed eccezioni\u003C\u002Fh3>\n\u003Cp>Le regole contengono frequentemente condizioni di applicabilità. Separare la regola dall'eccezione può preservare le parole ma distruggere il significato.\u003C\u002Fp>\n\u003Ch3 id=\"section-42\">6. Versione e tempo\u003C\u002Fh3>\n\u003Cp>Le prove dovrebbero preservare quando sono diventate valide, quando sono state osservate e se un oggetto più recente le sostituisce.\u003C\u002Fp>\n\u003Ch3 id=\"section-44\">7. Stato di conflitto\u003C\u002Fh3>\n\u003Cp>Se due fonti sono in disaccordo, la normalizzazione non dovrebbe selezionare silenziosamente un valore ed eliminare il fatto che il disaccordo esisteva.\u003C\u002Fp>\n\u003Ch3 id=\"section-46\">8. Ambito di accesso\u003C\u002Fh3>\n\u003Cp>I confini di autorizzazione e tenant devono rimanere collegati alle prove durante il recupero e l'assemblaggio del contesto.\u003C\u002Fp>\n\u003Ch2 id=\"section-48\">Dai dati grezzi alle prove controllate\u003C\u002Fh2>\n\u003Cp>Il luogo più forte per preservare le prove è spesso prima della vettorizzazione.\u003C\u002Fp>\n\u003Cp>Se l'identità della fonte, la validità e i conflitti vengono persi prima che vengano creati gli embedding, il recupero non può ricostruirli magicamente in seguito.\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fit\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\">Dai dati grezzi al contesto AI verificabile: un'architettura pratica per informazioni pronte per gli agenti\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">L'architettura inizia prima del recupero preservando identità, provenienza, conflitti, incognite e validità in oggetti informativi riutilizzabili.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ch2 id=\"section-52\">L'ingegneria del contesto fa parte della catena\u003C\u002Fh2>\n\u003Cp>Anche le prove recuperate correttamente possono essere danneggiate durante l'assemblaggio del contesto.\u003C\u002Fp>\n\u003Cp>Un costruttore di contesto può troncare un'eccezione, rimuovere le etichette di origine, unire passaggi in conflitto, riordinare le dipendenze o superare il budget di attenzione utile del modello.\u003C\u002Fp>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fit\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\">Che cos'è l'ingegneria del contesto? Ciò che il modello riceve prima di rispondere\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">L'ingegneria del contesto determina quali informazioni recuperate raggiungono effettivamente il modello e sotto quali vincoli strutturali, di validità e di sicurezza.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ch2 id=\"section-56\">Connessione alla distanza probatoria\u003C\u002Fh2>\n\u003Cp>La conservazione delle prove e la distanza probatoria descrivono diverse modalità di fallimento.\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>Un sistema può conservare perfettamente le prove e comunque trarne un'inferenza non supportata. Al contrario, il ragionamento finale può essere semplice mentre le prove stesse erano già corrotte in una fase precedente della pipeline.\u003C\u002Fp>\n\u003Ch2 id=\"section-60\">Connessione al livello di autorità di recupero\u003C\u002Fh2>\n\u003Cp>Il livello di autorità di recupero seleziona quale fonte merita peso probatorio.\u003C\u002Fp>\n\u003Cp>La catena di conservazione delle prove garantisce che le proprietà che rendono autorevole quella fonte non vengano scartate dopo la selezione.\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\">Modello dati pratico\u003C\u002Fh2>\n\u003Cp>Un'unità di recupero può trasportare più del semplice testo.\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>Lo schema esatto dipende dall'applicazione. Il principio architetturale è che il testo recuperabile non dovrebbe diventare separato dai metadati necessari per interpretarlo.\u003C\u002Fp>\n\u003Ch2 id=\"section-68\">Metodo diagnostico \u002F decisionale\u003C\u002Fh2>\n\u003Cp>Scegli una risposta importante e traccia un'affermazione di supporto a ritroso attraverso il sistema.\u003C\u002Fp>\n\u003Col>\u003Cli>Identifica l'affermazione finale.\u003C\u002Fli>\u003Cli>Identifica il passaggio esatto del contesto che l'ha supportata.\u003C\u002Fli>\u003Cli>Identifica il chunk recuperato che ha prodotto quel passaggio.\u003C\u002Fli>\u003Cli>Identifica il record normalizzato o il documento analizzato dietro il chunk.\u003C\u002Fli>\u003Cli>Identifica la fonte originale.\u003C\u002Fli>\u003Cli>Confronta significato, ambito, unità, condizioni e versione a ogni transizione.\u003C\u002Fli>\u003C\u002Fol>\n\u003Cp>Qualsiasi punto in cui il significato essenziale scompare è una rottura nella catena di conservazione delle prove.\u003C\u002Fp>\n\u003Ch2 id=\"section-72\">Modalità di fallimento: il chunk orfano\u003C\u002Fh2>\n\u003Cp>Un frammento orfano contiene testo dall'aspetto utile ma nessuna relazione affidabile con la struttura del documento, l'identità della fonte, la versione o le condizioni di applicabilità da cui proviene.\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>La frase può essere autentica e comunque costituire una prova non sicura.\u003C\u002Fp>\n\u003Ch2 id=\"section-76\">Modalità di errore: normalizzazione per invenzione\u003C\u002Fh2>\n\u003Cp>La normalizzazione dovrebbe allineare le rappresentazioni. Non dovrebbe inventare silenziosamente valori mancanti per far sembrare completi i record.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Unknown\n≠\nFalse\n≠\nZero\n≠\nDefault\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Preservare le incognite fa parte della conservazione delle prove.\u003C\u002Fp>\n\u003Ch2 id=\"section-80\">Modalità di errore: collasso dei conflitti\u003C\u002Fh2>\n\u003Cp>Supponiamo che una fonte approvata dica 24 mesi e un'altra dica 36 mesi.\u003C\u002Fp>\n\u003Cp>Una pipeline che si limita a memorizzare l'ultimo valore importato produce un database pulito e distrugge la prova che esisteva un conflitto.\u003C\u002Fp>\n\u003Cp>Un sistema verificabile dovrebbe preservare entrambi i valori, la loro provenienza e il conflitto irrisolto finché una regola o una decisione umana non lo risolve.\u003C\u002Fp>\n\u003Ch2 id=\"section-84\">Modalità di errore: compressione del contesto\u003C\u002Fh2>\n\u003Cp>La compressione del contesto può essere utile, ma un riassunto che rimuove i qualificatori può aumentare la certezza apparente.\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>L'informazione è diventata più breve e meno veritiera allo stesso tempo.\u003C\u002Fp>\n\u003Ch2 id=\"section-88\">Caso limite: la compressione con perdita può comunque essere corretta\u003C\u002Fh2>\n\u003Cp>Non ogni perdita è dannosa.\u003C\u002Fp>\n\u003Cp>Rimuovere la formattazione visiva da un paragrafo può non avere alcuna conseguenza semantica. Convertire Celsius in Kelvin può preservare il significato se la trasformazione è esplicita e corretta. Deduplicare copie esatte può migliorare il recupero senza perdere prove.\u003C\u002Fp>\n\u003Cp>L'obiettivo non è quindi la conservazione senza perdite in ogni fase. È una conservazione semanticamente sufficiente.\u003C\u002Fp>\n\u003Ch2 id=\"section-92\">Limitazioni\u003C\u002Fh2>\n\u003Cp>La Catena di Conservazione delle Prove è un quadro architetturale, non una garanzia che le prove conservate vengano interpretate correttamente.\u003C\u002Fp>\n\u003Cp>Un modello può ricevere prove perfettamente conservate e ragionare comunque male. Un sistema di recupero può preservare la provenienza e classificare comunque la fonte sbagliata. Anche la fonte stessa può essere errata.\u003C\u002Fp>\n\u003Cp>La conservazione protegge la pipeline delle prove. Non sostituisce la valutazione della fonte o il ragionamento.\u003C\u002Fp>\n\u003Ch2 id=\"section-96\">Cosa Cambierebbe Questa Risposta?\u003C\u002Fh2>\n\u003Cp>I sistemi futuri potrebbero recuperare oggetti di prova strutturati anziché frammenti di testo, preservare automaticamente i grafi di provenienza ed esporre vincoli di validità leggibili dalla macchina direttamente a modelli e agenti.\u003C\u002Fp>\n\u003Cp>Ciò renderebbe la conservazione delle prove più facile da implementare, ma renderebbe il principio più importante anziché meno importante.\u003C\u002Fp>\n\u003Ch2 id=\"section-99\">Conclusione\u003C\u002Fh2>\n\u003Cp>Un'IA affidabile richiede più che trovare una fonte corretta.\u003C\u002Fp>\n\u003Cp>Le proprietà che rendono utile la fonte devono sopravvivere a parsing, normalizzazione, chunking, recupero e assemblaggio del contesto.\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>Se la catena si rompe, il modello può ricevere informazioni tecnicamente autentiche ma non più sicure da interpretare come le prove originali.\u003C\u002Fp>\n\u003Ch2 id=\"section-104\">Fonti Primarie\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":228,"blocks":229,"version":798},1792096580764,[230,236,242,247,252,260,267,272,277,282,288,293,298,303,309,314,319,324,329,334,339,344,349,354,359,364,369,374,379,384,395,400,405,410,415,420,425,430,435,440,445,450,455,460,465,470,475,480,485,490,495,504,509,514,519,528,533,538,543,548,553,558,563,568,573,578,583,588,593,598,613,618,623,628,633,638,643,648,653,658,663,668,673,678,683,688,693,698,703,708,713,718,723,728,733,738,743,748,753,758,763,768,773,778,783,788,793],{"id":231,"data":232,"type":42,"tunes":235},"tP10jlMV2G",{"text":233,"level":234},"Domanda",2,{},{"id":237,"data":238,"type":240,"tunes":241},"zD8a-5x7r9",{"text":239},"Cosa succede quando la fonte originale è corretta, ma la pipeline di IA elimina le informazioni necessarie per comprendere correttamente quella fonte?","paragraph",{},{"id":243,"data":244,"type":240,"tunes":246},"LJjTjSV74R",{"text":245},"Un documento affidabile può diventare un contesto inaffidabile se il parsing rimuove la sua struttura tabellare, il chunking separa una condizione dalla sua eccezione, la normalizzazione elimina le unità, il retrieval perde i metadati di versione o l'assemblaggio del contesto nasconde un conflitto.",{},{"id":248,"data":249,"type":240,"tunes":251},"kOzPNiKvzu",{"text":250},"Questo articolo utilizza il termine Catena di Conservazione delle Prove per indicare il requisito che le prove mantengano le proprietà necessarie per una corretta interpretazione mentre passano dalla fonte al contesto finale dell'IA.",{},{"id":253,"data":254,"type":258,"tunes":259},"xUf9Qt8UHn",{"text":255,"caption":256,"alignment":257},"Una Catena di Conservazione delle Prove è la sequenza end-to-end di trasformazioni attraverso cui le prove devono mantenere identità, provenienza, ambito, significato e validità prima di poter supportare in modo sicuro un'affermazione generata dall'IA.","Definizione operativa — un concetto architetturale introdotto qui, non un termine di ricerca standardizzato.","left","quote",{},{"id":261,"data":262,"type":265,"tunes":266},"RxQsaJsNo6",{"title":263,"maxLevel":264,"minLevel":234},"Contenuti",3,"tableOfContents",{},{"id":268,"data":269,"type":42,"tunes":271},"NBMzNZ_qHB",{"text":270,"level":234},"Cosa Significa Davvero",{},{"id":273,"data":274,"type":240,"tunes":276},"zAEKGXCEmK",{"text":275},"La maggior parte delle discussioni sul RAG si concentra sulla possibilità di trovare informazioni rilevanti.",{},{"id":278,"data":279,"type":240,"tunes":281},"i2gu8IM3j1",{"text":280},"Ma le prove di solito passano attraverso diverse trasformazioni prima che il modello le veda.",{},{"id":283,"data":284,"type":286,"tunes":287},"lN52dh3dxK",{"code":285},"Original Source\n   ↓\nParsing\n   ↓\nNormalization\n   ↓\nChunking\n   ↓\nIndexing\n   ↓\nRetrieval\n   ↓\nReranking\n   ↓\nContext Assembly\n   ↓\nModel\n   ↓\nClaim","code",{},{"id":289,"data":290,"type":240,"tunes":292},"G90TdnX4WC",{"text":291},"Le prove possono degradarsi in qualsiasi punto di quella catena anche quando il documento originale era perfettamente corretto.",{},{"id":294,"data":295,"type":42,"tunes":297},"mC_OsF2ji3",{"text":296,"level":234},"Esempio più Semplice",{},{"id":299,"data":300,"type":240,"tunes":302},"P8XUivVa7g",{"text":301},"Immagina che un manuale di prodotto contenga questa affermazione:",{},{"id":304,"data":305,"type":258,"tunes":308},"2LamDXESHK",{"text":306,"caption":307,"alignment":257},"Temperatura massima di esercizio: 40 °C, tranne quando è installato l'accessorio X, nel qual caso il limite è 35 °C.","Affermazione di esempio dalla fonte",{},{"id":310,"data":311,"type":240,"tunes":313},"BNY1TVLg3y",{"text":312},"Un processo di chunking scadente divide la frase.",{},{"id":315,"data":316,"type":286,"tunes":318},"xaAZ_aJazc",{"code":317},"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":320,"data":321,"type":240,"tunes":323},"q0ORvmk__K",{"text":322},"Il retriever restituisce solo il Chunk A.",{},{"id":325,"data":326,"type":240,"tunes":328},"MupQI-_xIz",{"text":327},"Il testo recuperato non è fabbricato. Proviene direttamente dal manuale. Eppure il contesto è incompleto esattamente nel modo che conta.",{},{"id":330,"data":331,"type":286,"tunes":333},"cYEvuH7MR6",{"code":332},"Correct source\n        +\nlossy transformation\n        =\nmisleading evidence",{},{"id":335,"data":336,"type":42,"tunes":338},"F_ZvT1AfTN",{"text":337,"level":234},"Dove l'esempio smette di funzionare",{},{"id":340,"data":341,"type":240,"tunes":343},"flP5_wVcvt",{"text":342},"La conservazione delle prove non significa copiare ogni byte della fonte invariato nel contesto del modello.",{},{"id":345,"data":346,"type":240,"tunes":348},"rrlavifFLP",{"text":347},"Compressione, normalizzazione, suddivisione in blocchi e riepilogo sono spesso necessari. L'obiettivo è preservare le proprietà richieste per un'interpretazione corretta, non eliminare la trasformazione.",{},{"id":350,"data":351,"type":240,"tunes":353},"CM_pAZ8ozk",{"text":352},"La difficile questione ingegneristica è quindi: quali proprietà sono semanticamente portanti?",{},{"id":355,"data":356,"type":42,"tunes":358},"U8VCT8z_Nf",{"text":357,"level":234},"Risposta diretta",{},{"id":360,"data":361,"type":240,"tunes":363},"Hs1rUljfaq",{"text":362},"Una prova sopravvive a una pipeline di IA quando le trasformazioni tra fonte e modello preservano sufficienti metadati di identità, provenienza, struttura, unità, condizioni, conflitti e validità affinché l'informazione recuperata significhi in contesto la stessa cosa che significava alla fonte.",{},{"id":365,"data":366,"type":286,"tunes":368},"2r2HxbzQFg",{"code":367},"Evidence Preservation\n=\nmeaning\n+ identity\n+ provenance\n+ scope\n+ units\n+ conditions\n+ validity\n+ traceability",{},{"id":370,"data":371,"type":42,"tunes":373},"ka_HV02MYw",{"text":372,"level":234},"Perché il RAG può fallire anche con la fonte giusta",{},{"id":375,"data":376,"type":240,"tunes":378},"eKv9aiSIC9",{"text":377},"Un retriever può trovare il documento corretto pur fornendo l'unità di prova sbagliata.",{},{"id":380,"data":381,"type":240,"tunes":383},"-EAX8UkdxQ",{"text":382},"Il problema potrebbe essersi verificato molto prima: durante l'OCR, il parsing del documento, l'estrazione HTML, la conversione delle tabelle, la mappatura dei campi, la costruzione dei blocchi o la perdita di metadati.",{},{"id":385,"data":386,"type":393,"tunes":394},"oYCdWaadX3",{"link":387,"meta":388},"https:\u002F\u002Fstajic.de\u002Fit\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python",{"image":389,"title":391,"description":392},{"url":390},"","Da dove ottiene i dati un LLM? Fonti dati RAG in Python","Il RAG inizia con fonti dati esterne e metodi di recupero. La conservazione delle prove si chiede cosa deve sopravvivere tra la fonte originale e il contesto recuperato.","linkTool",{},{"id":396,"data":397,"type":42,"tunes":399},"A_6idqO75O",{"text":398,"level":234},"Gli otto punti di conservazione",{},{"id":401,"data":402,"type":42,"tunes":404},"NaU6o_qWBJ",{"text":403,"level":264},"1. Identità",{},{"id":406,"data":407,"type":240,"tunes":409},"5mR31I7W1M",{"text":408},"La pipeline deve preservare quale entità, prodotto, documento, utente, contratto o record la prova descrive effettivamente.",{},{"id":411,"data":412,"type":42,"tunes":414},"UIaEdeVESf",{"text":413,"level":264},"2. Provenienza",{},{"id":416,"data":417,"type":240,"tunes":419},"vAtWlAr_L5",{"text":418},"Il sistema dovrebbe essere in grado di risalire dalla prova recuperata alla sua fonte originale, non solo a un identificatore di blocco anonimo.",{},{"id":421,"data":422,"type":42,"tunes":424},"6EGnXzdt_j",{"text":423,"level":264},"3. Struttura",{},{"id":426,"data":427,"type":240,"tunes":429},"9NSFX7Dq4k",{"text":428},"Tabelle, intestazioni, elenchi, relazioni padre-figlio ed eccezioni possono portare un significato che scompare quando tutto viene appiattito in frammenti di testo indipendenti.",{},{"id":431,"data":432,"type":42,"tunes":434},"87QZ9m4_q7",{"text":433,"level":264},"4. Unità",{},{"id":436,"data":437,"type":240,"tunes":439},"0e7PrlT5MU",{"text":438},"Un numero senza la sua unità può diventare inutile o pericoloso. 100 può significare millisecondi, chilogrammi, euro, volt o percentuale di una linea di base.",{},{"id":441,"data":442,"type":42,"tunes":444},"Bid-hGyRbP",{"text":443,"level":264},"5. Condizioni ed eccezioni",{},{"id":446,"data":447,"type":240,"tunes":449},"Td_5jAm0ME",{"text":448},"Le regole contengono frequentemente condizioni di applicabilità. Separare la regola dall'eccezione può preservare le parole ma distruggere il significato.",{},{"id":451,"data":452,"type":42,"tunes":454},"y5CedEdzCc",{"text":453,"level":264},"6. Versione e tempo",{},{"id":456,"data":457,"type":240,"tunes":459},"K9nDZI5yQf",{"text":458},"Le prove dovrebbero preservare quando sono diventate valide, quando sono state osservate e se un oggetto più recente le sostituisce.",{},{"id":461,"data":462,"type":42,"tunes":464},"UtnbpFacpC",{"text":463,"level":264},"7. Stato di conflitto",{},{"id":466,"data":467,"type":240,"tunes":469},"rQMwxCKxIE",{"text":468},"Se due fonti sono in disaccordo, la normalizzazione non dovrebbe selezionare silenziosamente un valore ed eliminare il fatto che il disaccordo esisteva.",{},{"id":471,"data":472,"type":42,"tunes":474},"615Imi6TRP",{"text":473,"level":264},"8. Ambito di accesso",{},{"id":476,"data":477,"type":240,"tunes":479},"WZuieLPrus",{"text":478},"I confini di autorizzazione e tenant devono rimanere collegati alle prove durante il recupero e l'assemblaggio del contesto.",{},{"id":481,"data":482,"type":42,"tunes":484},"je6wCH54vs",{"text":483,"level":234},"Dai dati grezzi alle prove controllate",{},{"id":486,"data":487,"type":240,"tunes":489},"fqi7FxVIcZ",{"text":488},"Il luogo più forte per preservare le prove è spesso prima della vettorizzazione.",{},{"id":491,"data":492,"type":240,"tunes":494},"iA9_KWjiwC",{"text":493},"Se l'identità della fonte, la validità e i conflitti vengono persi prima che vengano creati gli embedding, il recupero non può ricostruirli magicamente in seguito.",{},{"id":496,"data":497,"type":393,"tunes":503},"y6cwyGR2PK",{"link":498,"meta":499},"https:\u002F\u002Fstajic.de\u002Fit\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information",{"image":500,"title":501,"description":502},{"url":390},"Dai dati grezzi al contesto AI verificabile: un'architettura pratica per informazioni pronte per gli agenti","L'architettura inizia prima del recupero preservando identità, provenienza, conflitti, incognite e validità in oggetti informativi riutilizzabili.",{},{"id":505,"data":506,"type":42,"tunes":508},"piEK0zmvxT",{"text":507,"level":234},"L'ingegneria del contesto fa parte della catena",{},{"id":510,"data":511,"type":240,"tunes":513},"Hs-hKf-7cy",{"text":512},"Anche le prove recuperate correttamente possono essere danneggiate durante l'assemblaggio del contesto.",{},{"id":515,"data":516,"type":240,"tunes":518},"kdWLXPTg4h",{"text":517},"Un costruttore di contesto può troncare un'eccezione, rimuovere le etichette di origine, unire passaggi in conflitto, riordinare le dipendenze o superare il budget di attenzione utile del modello.",{},{"id":520,"data":521,"type":393,"tunes":527},"vQ5Jya2Rv8",{"link":522,"meta":523},"https:\u002F\u002Fstajic.de\u002Fit\u002Fblog\u002Fwhat-is-context-engineering-what-the-model-receives-before-it-answers",{"image":524,"title":525,"description":526},{"url":390},"Che cos'è l'ingegneria del contesto? Ciò che il modello riceve prima di rispondere","L'ingegneria del contesto determina quali informazioni recuperate raggiungono effettivamente il modello e sotto quali vincoli strutturali, di validità e di sicurezza.",{},{"id":529,"data":530,"type":42,"tunes":532},"oC18nhSD3Y",{"text":531,"level":234},"Connessione alla distanza probatoria",{},{"id":534,"data":535,"type":240,"tunes":537},"tbewzxhkkL",{"text":536},"La conservazione delle prove e la distanza probatoria descrivono diverse modalità di fallimento.",{},{"id":539,"data":540,"type":286,"tunes":542},"XzYMjN1nNf",{"code":541},"Evidence Preservation Chain:\nDid the evidence survive transformation?\n\nEvidence Distance:\nHow far is the final claim from that evidence?",{},{"id":544,"data":545,"type":240,"tunes":547},"013RZ4lijY",{"text":546},"Un sistema può conservare perfettamente le prove e comunque trarne un'inferenza non supportata. Al contrario, il ragionamento finale può essere semplice mentre le prove stesse erano già corrotte in una fase precedente della pipeline.",{},{"id":549,"data":550,"type":42,"tunes":552},"cdXu2rD4d-",{"text":551,"level":234},"Connessione al livello di autorità di recupero",{},{"id":554,"data":555,"type":240,"tunes":557},"asRwQ_zGye",{"text":556},"Il livello di autorità di recupero seleziona quale fonte merita peso probatorio.",{},{"id":559,"data":560,"type":240,"tunes":562},"dMwV7G7LC4",{"text":561},"La catena di conservazione delle prove garantisce che le proprietà che rendono autorevole quella fonte non vengano scartate dopo la selezione.",{},{"id":564,"data":565,"type":286,"tunes":567},"ky0NtD4IAl",{"code":566},"Authority at source\n        ↓\nPreservation through pipeline\n        ↓\nAuthority still visible in context",{},{"id":569,"data":570,"type":42,"tunes":572},"Em0D8AL_b7",{"text":571,"level":234},"Modello dati pratico",{},{"id":574,"data":575,"type":240,"tunes":577},"iDWVeCI3US",{"text":576},"Un'unità di recupero può trasportare più del semplice testo.",{},{"id":579,"data":580,"type":286,"tunes":582},"Ybc3rCbvY0",{"code":581},"{\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":584,"data":585,"type":240,"tunes":587},"g007D00k4V",{"text":586},"Lo schema esatto dipende dall'applicazione. Il principio architetturale è che il testo recuperabile non dovrebbe diventare separato dai metadati necessari per interpretarlo.",{},{"id":589,"data":590,"type":42,"tunes":592},"Wi-cP9HpKU",{"text":591,"level":234},"Metodo diagnostico \u002F decisionale",{},{"id":594,"data":595,"type":240,"tunes":597},"yUPTwfB9uP",{"text":596},"Scegli una risposta importante e traccia un'affermazione di supporto a ritroso attraverso il sistema.",{},{"id":599,"data":600,"type":611,"tunes":612},"IRrMfVVINa",{"meta":601,"items":603,"style":610},{"counterType":602},"numeric",[604,605,606,607,608,609],"Identifica l'affermazione finale.","Identifica il passaggio esatto del contesto che l'ha supportata.","Identifica il chunk recuperato che ha prodotto quel passaggio.","Identifica il record normalizzato o il documento analizzato dietro il chunk.","Identifica la fonte originale.","Confronta significato, ambito, unità, condizioni e versione a ogni transizione.","ordered","list",{},{"id":614,"data":615,"type":240,"tunes":617},"05LsNWXcjf",{"text":616},"Qualsiasi punto in cui il significato essenziale scompare è una rottura nella catena di conservazione delle prove.",{},{"id":619,"data":620,"type":42,"tunes":622},"raBFHAylDf",{"text":621,"level":234},"Modalità di fallimento: il chunk orfano",{},{"id":624,"data":625,"type":240,"tunes":627},"iniLj3zXLh",{"text":626},"Un frammento orfano contiene testo dall'aspetto utile ma nessuna relazione affidabile con la struttura del documento, l'identità della fonte, la versione o le condizioni di applicabilità da cui proviene.",{},{"id":629,"data":630,"type":286,"tunes":632},"1zagZ01tls",{"code":631},"\"Refunds are available for 30 days.\"\n\nMissing:\nWhich product?\nWhich country?\nWhich policy version?\nFrom what effective date?\nWhat exceptions?",{},{"id":634,"data":635,"type":240,"tunes":637},"sZ0-pxHB2K",{"text":636},"La frase può essere autentica e comunque costituire una prova non sicura.",{},{"id":639,"data":640,"type":42,"tunes":642},"wCo24MdrUE",{"text":641,"level":234},"Modalità di errore: normalizzazione per invenzione",{},{"id":644,"data":645,"type":240,"tunes":647},"_8tEy420r3",{"text":646},"La normalizzazione dovrebbe allineare le rappresentazioni. Non dovrebbe inventare silenziosamente valori mancanti per far sembrare completi i record.",{},{"id":649,"data":650,"type":286,"tunes":652},"DvCye0QKKX",{"code":651},"Unknown\n≠\nFalse\n≠\nZero\n≠\nDefault",{},{"id":654,"data":655,"type":240,"tunes":657},"gTRNnqfBbR",{"text":656},"Preservare le incognite fa parte della conservazione delle prove.",{},{"id":659,"data":660,"type":42,"tunes":662},"Azp57iMou6",{"text":661,"level":234},"Modalità di errore: collasso dei conflitti",{},{"id":664,"data":665,"type":240,"tunes":667},"HXmXzrSkqL",{"text":666},"Supponiamo che una fonte approvata dica 24 mesi e un'altra dica 36 mesi.",{},{"id":669,"data":670,"type":240,"tunes":672},"qDsZWYHjhc",{"text":671},"Una pipeline che si limita a memorizzare l'ultimo valore importato produce un database pulito e distrugge la prova che esisteva un conflitto.",{},{"id":674,"data":675,"type":240,"tunes":677},"NkYagndI9y",{"text":676},"Un sistema verificabile dovrebbe preservare entrambi i valori, la loro provenienza e il conflitto irrisolto finché una regola o una decisione umana non lo risolve.",{},{"id":679,"data":680,"type":42,"tunes":682},"KrRJTl84Dq",{"text":681,"level":234},"Modalità di errore: compressione del contesto",{},{"id":684,"data":685,"type":240,"tunes":687},"VVy7pkhXMG",{"text":686},"La compressione del contesto può essere utile, ma un riassunto che rimuove i qualificatori può aumentare la certezza apparente.",{},{"id":689,"data":690,"type":286,"tunes":692},"MTqe1-sAHt",{"code":691},"Source:\n\"Evidence suggests X under conditions A and B, but the sample is small.\"\n\nBad compression:\n\"X is proven.\"",{},{"id":694,"data":695,"type":240,"tunes":697},"d61VdY7Z5d",{"text":696},"L'informazione è diventata più breve e meno veritiera allo stesso tempo.",{},{"id":699,"data":700,"type":42,"tunes":702},"k4jHwZaoLl",{"text":701,"level":234},"Caso limite: la compressione con perdita può comunque essere corretta",{},{"id":704,"data":705,"type":240,"tunes":707},"92uTDyGplP",{"text":706},"Non ogni perdita è dannosa.",{},{"id":709,"data":710,"type":240,"tunes":712},"U0cu-bjBmq",{"text":711},"Rimuovere la formattazione visiva da un paragrafo può non avere alcuna conseguenza semantica. Convertire Celsius in Kelvin può preservare il significato se la trasformazione è esplicita e corretta. Deduplicare copie esatte può migliorare il recupero senza perdere prove.",{},{"id":714,"data":715,"type":240,"tunes":717},"GJpi4mOxtH",{"text":716},"L'obiettivo non è quindi la conservazione senza perdite in ogni fase. È una conservazione semanticamente sufficiente.",{},{"id":719,"data":720,"type":42,"tunes":722},"7XXF7EHC5m",{"text":721,"level":234},"Limitazioni",{},{"id":724,"data":725,"type":240,"tunes":727},"FW8TcQldR-",{"text":726},"La Catena di Conservazione delle Prove è un quadro architetturale, non una garanzia che le prove conservate vengano interpretate correttamente.",{},{"id":729,"data":730,"type":240,"tunes":732},"V1rP2gB1C8",{"text":731},"Un modello può ricevere prove perfettamente conservate e ragionare comunque male. Un sistema di recupero può preservare la provenienza e classificare comunque la fonte sbagliata. Anche la fonte stessa può essere errata.",{},{"id":734,"data":735,"type":240,"tunes":737},"GCfe1Di90D",{"text":736},"La conservazione protegge la pipeline delle prove. Non sostituisce la valutazione della fonte o il ragionamento.",{},{"id":739,"data":740,"type":42,"tunes":742},"654Bx39EsI",{"text":741,"level":234},"Cosa Cambierebbe Questa Risposta?",{},{"id":744,"data":745,"type":240,"tunes":747},"UFx-k4mGz6",{"text":746},"I sistemi futuri potrebbero recuperare oggetti di prova strutturati anziché frammenti di testo, preservare automaticamente i grafi di provenienza ed esporre vincoli di validità leggibili dalla macchina direttamente a modelli e agenti.",{},{"id":749,"data":750,"type":240,"tunes":752},"lQMoO7JEMG",{"text":751},"Ciò renderebbe la conservazione delle prove più facile da implementare, ma renderebbe il principio più importante anziché meno importante.",{},{"id":754,"data":755,"type":42,"tunes":757},"gL1c_bEYfC",{"text":756,"level":234},"Conclusione",{},{"id":759,"data":760,"type":240,"tunes":762},"xElsMHA7R-",{"text":761},"Un'IA affidabile richiede più che trovare una fonte corretta.",{},{"id":764,"data":765,"type":240,"tunes":767},"Heg2LJJSzg",{"text":766},"Le proprietà che rendono utile la fonte devono sopravvivere a parsing, normalizzazione, chunking, recupero e assemblaggio del contesto.",{},{"id":769,"data":770,"type":286,"tunes":772},"MzylWKf1Jm",{"code":771},"Source Authority\n      ↓\nEvidence Preservation Chain\n      ↓\nRetrievable Evidence\n      ↓\nEvidence Distance\n      ↓\nReasoning\n      ↓\nAnswer Validity Boundary",{},{"id":774,"data":775,"type":240,"tunes":777},"m_RMMC7dVx",{"text":776},"Se la catena si rompe, il modello può ricevere informazioni tecnicamente autentiche ma non più sicure da interpretare come le prove originali.",{},{"id":779,"data":780,"type":42,"tunes":782},"IkNMvsnAJq",{"text":781,"level":234},"Fonti Primarie",{},{"id":784,"data":785,"type":240,"tunes":787},"QqTaKgzQ8k",{"text":786},"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":789,"data":790,"type":240,"tunes":792},"UturEzvxw2",{"text":791},"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":794,"data":795,"type":240,"tunes":797},"I2fpdupo7S",{"text":796},"W3C, \u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\" target=\"_blank\">PROV Overview\u003C\u002Fa>.",{},"2.31","Le prove possono essere corrette alla fonte e diventare comunque inaffidabili dopo l'analisi, la normalizzazione, la suddivisione in blocchi, il recupero e l'assemblaggio del contesto. La Catena di Preservazione delle Prove tiene traccia se identità, provenienza, ambito e significato sopravvivono all'intera pipeline di IA.","\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":807,"de":808,"sr":809,"es":810,"fr":811,"it":812,"ru":813,"zh":814},"\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",[816,820],{"id":817,"name":818,"slug":819},68,"Rischi, controlli e prove","risks-and-controls",{"id":821,"name":822,"slug":823},49,"Controlli e prove","controls",{"id":825,"login":826,"email":390,"displayName":827},"20","rooth8233","Aleksandar Stajić",[829,1273],{"lang":830,"title":831,"content":832,"contentJson":833,"excerpt":1272},"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":834,"blocks":835,"version":1271},1792096415735,[836,840,844,848,852,857,861,865,869,873,876,880,884,888,893,897,900,904,908,911,915,919,923,927,931,935,938,942,946,950,958,962,966,970,974,978,982,986,990,994,998,1002,1006,1010,1014,1018,1022,1026,1030,1034,1038,1046,1050,1054,1058,1066,1070,1074,1077,1081,1085,1089,1093,1096,1100,1104,1107,1111,1115,1119,1130,1134,1138,1142,1145,1149,1153,1157,1160,1164,1168,1172,1176,1180,1184,1188,1191,1195,1199,1203,1207,1211,1215,1219,1223,1227,1231,1235,1239,1243,1247,1251,1254,1258,1262,1265,1268],{"id":231,"data":837,"type":42,"tunes":839},{"text":838,"level":234},"Question",{},{"id":237,"data":841,"type":240,"tunes":843},{"text":842},"What happens when the original source is correct, but the AI pipeline strips away the information required to understand that source correctly?",{},{"id":243,"data":845,"type":240,"tunes":847},{"text":846},"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":248,"data":849,"type":240,"tunes":851},{"text":850},"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":253,"data":853,"type":258,"tunes":856},{"text":854,"caption":855,"alignment":257},"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":261,"data":858,"type":265,"tunes":860},{"title":859,"maxLevel":264,"minLevel":234},"Contents",{},{"id":268,"data":862,"type":42,"tunes":864},{"text":863,"level":234},"What This Really Means",{},{"id":273,"data":866,"type":240,"tunes":868},{"text":867},"Most discussions about RAG focus on whether relevant information can be found.",{},{"id":278,"data":870,"type":240,"tunes":872},{"text":871},"But evidence usually passes through several transformations before the model sees it.",{},{"id":283,"data":874,"type":286,"tunes":875},{"code":285},{},{"id":289,"data":877,"type":240,"tunes":879},{"text":878},"The evidence can degrade at any point in that chain even when the original document was perfectly correct.",{},{"id":294,"data":881,"type":42,"tunes":883},{"text":882,"level":234},"Simplest Example",{},{"id":299,"data":885,"type":240,"tunes":887},{"text":886},"Imagine a product manual contains this statement:",{},{"id":304,"data":889,"type":258,"tunes":892},{"text":890,"caption":891,"alignment":257},"Maximum operating temperature: 40 °C, except when accessory X is installed, in which case the limit is 35 °C.","Example source statement",{},{"id":310,"data":894,"type":240,"tunes":896},{"text":895},"A poor chunking process splits the sentence.",{},{"id":315,"data":898,"type":286,"tunes":899},{"code":317},{},{"id":320,"data":901,"type":240,"tunes":903},{"text":902},"The retriever returns only Chunk A.",{},{"id":325,"data":905,"type":240,"tunes":907},{"text":906},"The retrieved text is not fabricated. It came directly from the manual. Yet the context is incomplete in exactly the way that matters.",{},{"id":330,"data":909,"type":286,"tunes":910},{"code":332},{},{"id":335,"data":912,"type":42,"tunes":914},{"text":913,"level":234},"Where the Example Stops Working",{},{"id":340,"data":916,"type":240,"tunes":918},{"text":917},"Evidence preservation does not mean copying every source byte unchanged into the model context.",{},{"id":345,"data":920,"type":240,"tunes":922},{"text":921},"Compression, normalization, chunking and summarization are often necessary. The goal is to preserve the properties required for correct interpretation, not to eliminate transformation.",{},{"id":350,"data":924,"type":240,"tunes":926},{"text":925},"The difficult engineering question is therefore: Which properties are semantically load-bearing?",{},{"id":355,"data":928,"type":42,"tunes":930},{"text":929,"level":234},"Direct Answer",{},{"id":360,"data":932,"type":240,"tunes":934},{"text":933},"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":365,"data":936,"type":286,"tunes":937},{"code":367},{},{"id":370,"data":939,"type":42,"tunes":941},{"text":940,"level":234},"Why RAG Can Fail Even With the Right Source",{},{"id":375,"data":943,"type":240,"tunes":945},{"text":944},"A retriever can find the correct document while still supplying the wrong evidence unit.",{},{"id":380,"data":947,"type":240,"tunes":949},{"text":948},"The problem may have occurred much earlier: during OCR, document parsing, HTML extraction, table conversion, field mapping, chunk construction or metadata loss.",{},{"id":385,"data":951,"type":393,"tunes":957},{"link":952,"meta":953},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python",{"image":954,"title":955,"description":956},{"url":390},"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":396,"data":959,"type":42,"tunes":961},{"text":960,"level":234},"The Eight Preservation Points",{},{"id":401,"data":963,"type":42,"tunes":965},{"text":964,"level":264},"1. Identity",{},{"id":406,"data":967,"type":240,"tunes":969},{"text":968},"The pipeline must preserve what entity, product, document, user, contract or record the evidence actually describes.",{},{"id":411,"data":971,"type":42,"tunes":973},{"text":972,"level":264},"2. Provenance",{},{"id":416,"data":975,"type":240,"tunes":977},{"text":976},"The system should be able to trace the retrieved evidence back to its original source rather than only to an anonymous chunk identifier.",{},{"id":421,"data":979,"type":42,"tunes":981},{"text":980,"level":264},"3. Structure",{},{"id":426,"data":983,"type":240,"tunes":985},{"text":984},"Tables, headings, lists, parent-child relationships and exceptions can carry meaning that disappears when everything is flattened into independent text fragments.",{},{"id":431,"data":987,"type":42,"tunes":989},{"text":988,"level":264},"4. Units",{},{"id":436,"data":991,"type":240,"tunes":993},{"text":992},"A number without its unit can become useless or dangerous. 100 may mean milliseconds, kilograms, euros, volts or percent of a baseline.",{},{"id":441,"data":995,"type":42,"tunes":997},{"text":996,"level":264},"5. Conditions and Exceptions",{},{"id":446,"data":999,"type":240,"tunes":1001},{"text":1000},"Rules frequently contain applicability conditions. Separating the rule from the exception can preserve the words while destroying the meaning.",{},{"id":451,"data":1003,"type":42,"tunes":1005},{"text":1004,"level":264},"6. Version and Time",{},{"id":456,"data":1007,"type":240,"tunes":1009},{"text":1008},"Evidence should preserve when it became valid, when it was observed and whether a newer object supersedes it.",{},{"id":461,"data":1011,"type":42,"tunes":1013},{"text":1012,"level":264},"7. Conflict State",{},{"id":466,"data":1015,"type":240,"tunes":1017},{"text":1016},"If two sources disagree, normalization should not silently select one value and erase the fact that disagreement existed.",{},{"id":471,"data":1019,"type":42,"tunes":1021},{"text":1020,"level":264},"8. Access Scope",{},{"id":476,"data":1023,"type":240,"tunes":1025},{"text":1024},"Permission and tenant boundaries must remain attached to evidence throughout retrieval and context assembly.",{},{"id":481,"data":1027,"type":42,"tunes":1029},{"text":1028,"level":234},"From Raw Data to Controlled Evidence",{},{"id":486,"data":1031,"type":240,"tunes":1033},{"text":1032},"The strongest place to preserve evidence is often before vectorization.",{},{"id":491,"data":1035,"type":240,"tunes":1037},{"text":1036},"If source identity, validity and conflicts are lost before embeddings are created, retrieval cannot magically reconstruct them later.",{},{"id":496,"data":1039,"type":393,"tunes":1045},{"link":1040,"meta":1041},"https:\u002F\u002Fstajic.de\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information",{"image":1042,"title":1043,"description":1044},{"url":390},"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":505,"data":1047,"type":42,"tunes":1049},{"text":1048,"level":234},"Context Engineering Is Part of the Chain",{},{"id":510,"data":1051,"type":240,"tunes":1053},{"text":1052},"Even correctly retrieved evidence can be damaged during context assembly.",{},{"id":515,"data":1055,"type":240,"tunes":1057},{"text":1056},"A context builder may truncate an exception, remove source labels, merge conflicting passages, reorder dependencies or exceed the model's useful attention budget.",{},{"id":520,"data":1059,"type":393,"tunes":1065},{"link":1060,"meta":1061},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhat-is-context-engineering-what-the-model-receives-before-it-answers",{"image":1062,"title":1063,"description":1064},{"url":390},"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":529,"data":1067,"type":42,"tunes":1069},{"text":1068,"level":234},"Connection to Evidence Distance",{},{"id":534,"data":1071,"type":240,"tunes":1073},{"text":1072},"Evidence Preservation and Evidence Distance describe different failure modes.",{},{"id":539,"data":1075,"type":286,"tunes":1076},{"code":541},{},{"id":544,"data":1078,"type":240,"tunes":1080},{"text":1079},"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":549,"data":1082,"type":42,"tunes":1084},{"text":1083,"level":234},"Connection to the Retrieval Authority Layer",{},{"id":554,"data":1086,"type":240,"tunes":1088},{"text":1087},"The Retrieval Authority Layer selects which source deserves evidential weight.",{},{"id":559,"data":1090,"type":240,"tunes":1092},{"text":1091},"The Evidence Preservation Chain ensures that the properties making that source authoritative are not discarded after selection.",{},{"id":564,"data":1094,"type":286,"tunes":1095},{"code":566},{},{"id":569,"data":1097,"type":42,"tunes":1099},{"text":1098,"level":234},"Practical Data Model",{},{"id":574,"data":1101,"type":240,"tunes":1103},{"text":1102},"A retrieval unit can carry more than text.",{},{"id":579,"data":1105,"type":286,"tunes":1106},{"code":581},{},{"id":584,"data":1108,"type":240,"tunes":1110},{"text":1109},"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":589,"data":1112,"type":42,"tunes":1114},{"text":1113,"level":234},"Diagnostic \u002F Decision Method",{},{"id":594,"data":1116,"type":240,"tunes":1118},{"text":1117},"Choose one important answer and trace one supporting claim backwards through the system.",{},{"id":599,"data":1120,"type":611,"tunes":1129},{"meta":1121,"items":1122,"style":610},{"counterType":602},[1123,1124,1125,1126,1127,1128],"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":614,"data":1131,"type":240,"tunes":1133},{"text":1132},"Any point where essential meaning disappears is a break in the Evidence Preservation Chain.",{},{"id":619,"data":1135,"type":42,"tunes":1137},{"text":1136,"level":234},"Failure Mode: The Orphan Chunk",{},{"id":624,"data":1139,"type":240,"tunes":1141},{"text":1140},"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":629,"data":1143,"type":286,"tunes":1144},{"code":631},{},{"id":634,"data":1146,"type":240,"tunes":1148},{"text":1147},"The sentence may be authentic and still be unsafe evidence.",{},{"id":639,"data":1150,"type":42,"tunes":1152},{"text":1151,"level":234},"Failure Mode: Normalization by Invention",{},{"id":644,"data":1154,"type":240,"tunes":1156},{"text":1155},"Normalization should align representations. It should not silently invent missing values to make records look complete.",{},{"id":649,"data":1158,"type":286,"tunes":1159},{"code":651},{},{"id":654,"data":1161,"type":240,"tunes":1163},{"text":1162},"Preserving unknowns is part of preserving evidence.",{},{"id":659,"data":1165,"type":42,"tunes":1167},{"text":1166,"level":234},"Failure Mode: Conflict Collapse",{},{"id":664,"data":1169,"type":240,"tunes":1171},{"text":1170},"Suppose one approved source says 24 months and another says 36 months.",{},{"id":669,"data":1173,"type":240,"tunes":1175},{"text":1174},"A pipeline that simply stores the last imported value produces a clean database and destroys the evidence that a conflict existed.",{},{"id":674,"data":1177,"type":240,"tunes":1179},{"text":1178},"A verifiable system should preserve both values, their provenance and the unresolved conflict until a rule or human decision resolves it.",{},{"id":679,"data":1181,"type":42,"tunes":1183},{"text":1182,"level":234},"Failure Mode: Context Compression",{},{"id":684,"data":1185,"type":240,"tunes":1187},{"text":1186},"Context compression can be useful, but a summary that removes qualifiers can increase apparent certainty.",{},{"id":689,"data":1189,"type":286,"tunes":1190},{"code":691},{},{"id":694,"data":1192,"type":240,"tunes":1194},{"text":1193},"The information became shorter and less truthful at the same time.",{},{"id":699,"data":1196,"type":42,"tunes":1198},{"text":1197,"level":234},"Edge Case: Lossy Compression Can Still Be Correct",{},{"id":704,"data":1200,"type":240,"tunes":1202},{"text":1201},"Not every loss is harmful.",{},{"id":709,"data":1204,"type":240,"tunes":1206},{"text":1205},"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":714,"data":1208,"type":240,"tunes":1210},{"text":1209},"The objective is therefore not lossless storage at every stage. It is semantically sufficient preservation.",{},{"id":719,"data":1212,"type":42,"tunes":1214},{"text":1213,"level":234},"Limitations",{},{"id":724,"data":1216,"type":240,"tunes":1218},{"text":1217},"The Evidence Preservation Chain is an architectural framework, not a guarantee that preserved evidence will be interpreted correctly.",{},{"id":729,"data":1220,"type":240,"tunes":1222},{"text":1221},"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":734,"data":1224,"type":240,"tunes":1226},{"text":1225},"Preservation protects the evidence pipeline. It does not replace source evaluation or reasoning.",{},{"id":739,"data":1228,"type":42,"tunes":1230},{"text":1229,"level":234},"What Would Change This Answer?",{},{"id":744,"data":1232,"type":240,"tunes":1234},{"text":1233},"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":749,"data":1236,"type":240,"tunes":1238},{"text":1237},"That would make evidence preservation easier to implement, but it would make the principle more important rather than less important.",{},{"id":754,"data":1240,"type":42,"tunes":1242},{"text":1241,"level":234},"Conclusion",{},{"id":759,"data":1244,"type":240,"tunes":1246},{"text":1245},"Reliable AI requires more than finding a correct source.",{},{"id":764,"data":1248,"type":240,"tunes":1250},{"text":1249},"The properties that make the source useful must survive parsing, normalization, chunking, retrieval and context assembly.",{},{"id":769,"data":1252,"type":286,"tunes":1253},{"code":771},{},{"id":774,"data":1255,"type":240,"tunes":1257},{"text":1256},"If the chain breaks, the model may receive information that is technically authentic but no longer safe to interpret as the original evidence.",{},{"id":779,"data":1259,"type":42,"tunes":1261},{"text":1260,"level":234},"Primary Sources",{},{"id":784,"data":1263,"type":240,"tunes":1264},{"text":786},{},{"id":789,"data":1266,"type":240,"tunes":1267},{"text":791},{},{"id":794,"data":1269,"type":240,"tunes":1270},{"text":796},{},"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.",{"lang":7,"title":224,"content":226,"contentJson":1274,"excerpt":799},{"time":228,"blocks":1275,"version":798},[1276,1279,1282,1285,1288,1291,1294,1297,1300,1303,1306,1309,1312,1315,1318,1321,1324,1327,1330,1333,1336,1339,1342,1345,1348,1351,1354,1357,1360,1363,1368,1371,1374,1377,1380,1383,1386,1389,1392,1395,1398,1401,1404,1407,1410,1413,1416,1419,1422,1425,1428,1433,1436,1439,1442,1447,1450,1453,1456,1459,1462,1465,1468,1471,1474,1477,1480,1483,1486,1489,1494,1497,1500,1503,1506,1509,1512,1515,1518,1521,1524,1527,1530,1533,1536,1539,1542,1545,1548,1551,1554,1557,1560,1563,1566,1569,1572,1575,1578,1581,1584,1587,1590,1593,1596,1599,1602],{"id":231,"data":1277,"type":42,"tunes":1278},{"text":233,"level":234},{},{"id":237,"data":1280,"type":240,"tunes":1281},{"text":239},{},{"id":243,"data":1283,"type":240,"tunes":1284},{"text":245},{},{"id":248,"data":1286,"type":240,"tunes":1287},{"text":250},{},{"id":253,"data":1289,"type":258,"tunes":1290},{"text":255,"caption":256,"alignment":257},{},{"id":261,"data":1292,"type":265,"tunes":1293},{"title":263,"maxLevel":264,"minLevel":234},{},{"id":268,"data":1295,"type":42,"tunes":1296},{"text":270,"level":234},{},{"id":273,"data":1298,"type":240,"tunes":1299},{"text":275},{},{"id":278,"data":1301,"type":240,"tunes":1302},{"text":280},{},{"id":283,"data":1304,"type":286,"tunes":1305},{"code":285},{},{"id":289,"data":1307,"type":240,"tunes":1308},{"text":291},{},{"id":294,"data":1310,"type":42,"tunes":1311},{"text":296,"level":234},{},{"id":299,"data":1313,"type":240,"tunes":1314},{"text":301},{},{"id":304,"data":1316,"type":258,"tunes":1317},{"text":306,"caption":307,"alignment":257},{},{"id":310,"data":1319,"type":240,"tunes":1320},{"text":312},{},{"id":315,"data":1322,"type":286,"tunes":1323},{"code":317},{},{"id":320,"data":1325,"type":240,"tunes":1326},{"text":322},{},{"id":325,"data":1328,"type":240,"tunes":1329},{"text":327},{},{"id":330,"data":1331,"type":286,"tunes":1332},{"code":332},{},{"id":335,"data":1334,"type":42,"tunes":1335},{"text":337,"level":234},{},{"id":340,"data":1337,"type":240,"tunes":1338},{"text":342},{},{"id":345,"data":1340,"type":240,"tunes":1341},{"text":347},{},{"id":350,"data":1343,"type":240,"tunes":1344},{"text":352},{},{"id":355,"data":1346,"type":42,"tunes":1347},{"text":357,"level":234},{},{"id":360,"data":1349,"type":240,"tunes":1350},{"text":362},{},{"id":365,"data":1352,"type":286,"tunes":1353},{"code":367},{},{"id":370,"data":1355,"type":42,"tunes":1356},{"text":372,"level":234},{},{"id":375,"data":1358,"type":240,"tunes":1359},{"text":377},{},{"id":380,"data":1361,"type":240,"tunes":1362},{"text":382},{},{"id":385,"data":1364,"type":393,"tunes":1367},{"link":387,"meta":1365},{"image":1366,"title":391,"description":392},{"url":390},{},{"id":396,"data":1369,"type":42,"tunes":1370},{"text":398,"level":234},{},{"id":401,"data":1372,"type":42,"tunes":1373},{"text":403,"level":264},{},{"id":406,"data":1375,"type":240,"tunes":1376},{"text":408},{},{"id":411,"data":1378,"type":42,"tunes":1379},{"text":413,"level":264},{},{"id":416,"data":1381,"type":240,"tunes":1382},{"text":418},{},{"id":421,"data":1384,"type":42,"tunes":1385},{"text":423,"level":264},{},{"id":426,"data":1387,"type":240,"tunes":1388},{"text":428},{},{"id":431,"data":1390,"type":42,"tunes":1391},{"text":433,"level":264},{},{"id":436,"data":1393,"type":240,"tunes":1394},{"text":438},{},{"id":441,"data":1396,"type":42,"tunes":1397},{"text":443,"level":264},{},{"id":446,"data":1399,"type":240,"tunes":1400},{"text":448},{},{"id":451,"data":1402,"type":42,"tunes":1403},{"text":453,"level":264},{},{"id":456,"data":1405,"type":240,"tunes":1406},{"text":458},{},{"id":461,"data":1408,"type":42,"tunes":1409},{"text":463,"level":264},{},{"id":466,"data":1411,"type":240,"tunes":1412},{"text":468},{},{"id":471,"data":1414,"type":42,"tunes":1415},{"text":473,"level":264},{},{"id":476,"data":1417,"type":240,"tunes":1418},{"text":478},{},{"id":481,"data":1420,"type":42,"tunes":1421},{"text":483,"level":234},{},{"id":486,"data":1423,"type":240,"tunes":1424},{"text":488},{},{"id":491,"data":1426,"type":240,"tunes":1427},{"text":493},{},{"id":496,"data":1429,"type":393,"tunes":1432},{"link":498,"meta":1430},{"image":1431,"title":501,"description":502},{"url":390},{},{"id":505,"data":1434,"type":42,"tunes":1435},{"text":507,"level":234},{},{"id":510,"data":1437,"type":240,"tunes":1438},{"text":512},{},{"id":515,"data":1440,"type":240,"tunes":1441},{"text":517},{},{"id":520,"data":1443,"type":393,"tunes":1446},{"link":522,"meta":1444},{"image":1445,"title":525,"description":526},{"url":390},{},{"id":529,"data":1448,"type":42,"tunes":1449},{"text":531,"level":234},{},{"id":534,"data":1451,"type":240,"tunes":1452},{"text":536},{},{"id":539,"data":1454,"type":286,"tunes":1455},{"code":541},{},{"id":544,"data":1457,"type":240,"tunes":1458},{"text":546},{},{"id":549,"data":1460,"type":42,"tunes":1461},{"text":551,"level":234},{},{"id":554,"data":1463,"type":240,"tunes":1464},{"text":556},{},{"id":559,"data":1466,"type":240,"tunes":1467},{"text":561},{},{"id":564,"data":1469,"type":286,"tunes":1470},{"code":566},{},{"id":569,"data":1472,"type":42,"tunes":1473},{"text":571,"level":234},{},{"id":574,"data":1475,"type":240,"tunes":1476},{"text":576},{},{"id":579,"data":1478,"type":286,"tunes":1479},{"code":581},{},{"id":584,"data":1481,"type":240,"tunes":1482},{"text":586},{},{"id":589,"data":1484,"type":42,"tunes":1485},{"text":591,"level":234},{},{"id":594,"data":1487,"type":240,"tunes":1488},{"text":596},{},{"id":599,"data":1490,"type":611,"tunes":1493},{"meta":1491,"items":1492,"style":610},{"counterType":602},[604,605,606,607,608,609],{},{"id":614,"data":1495,"type":240,"tunes":1496},{"text":616},{},{"id":619,"data":1498,"type":42,"tunes":1499},{"text":621,"level":234},{},{"id":624,"data":1501,"type":240,"tunes":1502},{"text":626},{},{"id":629,"data":1504,"type":286,"tunes":1505},{"code":631},{},{"id":634,"data":1507,"type":240,"tunes":1508},{"text":636},{},{"id":639,"data":1510,"type":42,"tunes":1511},{"text":641,"level":234},{},{"id":644,"data":1513,"type":240,"tunes":1514},{"text":646},{},{"id":649,"data":1516,"type":286,"tunes":1517},{"code":651},{},{"id":654,"data":1519,"type":240,"tunes":1520},{"text":656},{},{"id":659,"data":1522,"type":42,"tunes":1523},{"text":661,"level":234},{},{"id":664,"data":1525,"type":240,"tunes":1526},{"text":666},{},{"id":669,"data":1528,"type":240,"tunes":1529},{"text":671},{},{"id":674,"data":1531,"type":240,"tunes":1532},{"text":676},{},{"id":679,"data":1534,"type":42,"tunes":1535},{"text":681,"level":234},{},{"id":684,"data":1537,"type":240,"tunes":1538},{"text":686},{},{"id":689,"data":1540,"type":286,"tunes":1541},{"code":691},{},{"id":694,"data":1543,"type":240,"tunes":1544},{"text":696},{},{"id":699,"data":1546,"type":42,"tunes":1547},{"text":701,"level":234},{},{"id":704,"data":1549,"type":240,"tunes":1550},{"text":706},{},{"id":709,"data":1552,"type":240,"tunes":1553},{"text":711},{},{"id":714,"data":1555,"type":240,"tunes":1556},{"text":716},{},{"id":719,"data":1558,"type":42,"tunes":1559},{"text":721,"level":234},{},{"id":724,"data":1561,"type":240,"tunes":1562},{"text":726},{},{"id":729,"data":1564,"type":240,"tunes":1565},{"text":731},{},{"id":734,"data":1567,"type":240,"tunes":1568},{"text":736},{},{"id":739,"data":1570,"type":42,"tunes":1571},{"text":741,"level":234},{},{"id":744,"data":1573,"type":240,"tunes":1574},{"text":746},{},{"id":749,"data":1576,"type":240,"tunes":1577},{"text":751},{},{"id":754,"data":1579,"type":42,"tunes":1580},{"text":756,"level":234},{},{"id":759,"data":1582,"type":240,"tunes":1583},{"text":761},{},{"id":764,"data":1585,"type":240,"tunes":1586},{"text":766},{},{"id":769,"data":1588,"type":286,"tunes":1589},{"code":771},{},{"id":774,"data":1591,"type":240,"tunes":1592},{"text":776},{},{"id":779,"data":1594,"type":42,"tunes":1595},{"text":781,"level":234},{},{"id":784,"data":1597,"type":240,"tunes":1598},{"text":786},{},{"id":789,"data":1600,"type":240,"tunes":1601},{"text":791},{},{"id":794,"data":1603,"type":240,"tunes":1604},{"text":796},{},"Post erfolgreich abgerufen",{"items":1607,"source":1664,"manualIds":1665,"manualMatchedIds":1666},[1608,1615,1622,1629,1636,1643,1650,1657],{"id":1609,"slug":1610,"title":1611,"excerpt":1612,"featuredImage":1613,"publishedAt":1614},"492","mcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits","MCP spiegato: cosa collega, cosa non fa e dove si colloca","Il Model Context Protocol collega le applicazioni di intelligenza artificiale a strumenti, risorse e prompt esterni attraverso un confine standard client-server. Scopri cosa fa MCP, cosa non fa e dove si colloca nell'architettura degli agenti.","\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":1616,"slug":1617,"title":1618,"excerpt":1619,"featuredImage":1620,"publishedAt":1621},"381","enterprise-grade-multi-tenant-architecture-for-an-international-platform","Architettura Multi-Tenant di Livello Enterprise per una Piattaforma Internazionale","Loving Rocks è una piattaforma per matrimoni di livello enterprise progettata con una vera architettura multi-tenant, database isolati per tenant e internazionalizzazione integrata per scalabilità globale, sicurezza e stabilità operativa a lungo termine.","\u002Fuploads\u002F2026\u002F01\u002Fenterprise-grade-multi-tenant-architecture-for-an-international-platform-1769789121298-b6v7ak.webp","2026-01-30T12:04:00.000Z",{"id":1623,"slug":1624,"title":1625,"excerpt":1626,"featuredImage":1627,"publishedAt":1628},"491","ai-governance-models-data-permissions-risk-and-auditability","Governance dell'IA: modelli, dati, autorizzazioni, rischio e verificabilità","La governance dell'IA definisce chi può approvare, gestire, modificare e verificare i sistemi di IA attraverso modelli, fornitori, dati, autorizzazioni, rischi, valutazione e l'intero ciclo di vita.","\u002Fuploads\u002F2026\u002F10\u002Fai-governance-models-data-permissions-risk-and-auditability-1791485901301-g60xyu.webp","2026-10-08T12:56:00.000Z",{"id":1630,"slug":1631,"title":1632,"excerpt":1633,"featuredImage":1634,"publishedAt":1635},"494","air-gapped-ai-how-ai-systems-work-without-internet-or-cloud-access","AI Air-Gapped: Come funzionano i sistemi di IA senza Internet o accesso al cloud","L'IA air-gapped esegue modelli, RAG e applicazioni di IA all'interno di un dominio di sicurezza isolato senza dipendenze da internet o dal cloud. Scopri come modelli, dati, aggiornamenti e strumenti operano offline.","\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":1637,"slug":1638,"title":1639,"excerpt":1640,"featuredImage":1641,"publishedAt":1642},"482","adr-vs-nfr-architecture-decisions-and-system-quality-are-not-the-same-thing","ADR vs NFR: decisioni architetturali e qualità del sistema non sono la stessa cosa","ADR vs NFR spiegati: scopri come i requisiti di qualità del sistema guidano le decisioni architetturali, come gli ADR registrano i compromessi e perché la validazione rimane separata.","\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":1644,"slug":1645,"title":1646,"excerpt":1647,"featuredImage":1648,"publishedAt":1649},"497","how-far-is-an-ai-claim-from-the-evidence-evidence-distance","Quanto dista un'affermazione dell'IA dalle prove? — Distanza probatoria","Non tutte le risposte supportate sono ugualmente vicine alle loro prove. La Distanza dalle Prove è un modo pratico per descrivere quante trasformazioni, sintesi, assunzioni e passaggi inferenziali separano un'affermazione dell'IA da prove direttamente ispezionabili.","\u002Fuploads\u002F2026\u002F10\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance-1792095863209-3c1mog.webp","2026-10-15T00:23:00.000Z",{"id":1651,"slug":1652,"title":1653,"excerpt":1654,"featuredImage":1655,"publishedAt":1656},"481","generative-ai-explained-models-retrieval-tools-and-applications-are-not-the-same-thing","L'IA generativa spiegata: modelli, recupero, strumenti e applicazioni non sono la stessa cosa","L'IA generativa è più di un modello. Scopri come modelli, recupero, strumenti, contesto, runtime e applicazioni si integrano nei sistemi di IA in produzione.","\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":1658,"slug":1659,"title":1660,"excerpt":1661,"featuredImage":1662,"publishedAt":1663},"495","sovereign-ai-control-of-models-data-infrastructure-and-dependencies","IA sovrana: controllo di modelli, dati, infrastrutture e dipendenze","L'IA sovrana riguarda il controllo effettivo su modelli, dati, infrastrutture, software, operazioni e dipendenze strategiche — non semplicemente dove è ospitato un modello di IA.","\u002Fuploads\u002F2026\u002F10\u002Fsovereign-ai-control-of-models-data-infrastructure-and-dependencies-1791488833132-niy85x.webp","2026-10-08T15:45:00.000Z","fallback",[],[]]