[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:zh":3,"public-menus:all":38,"post:from-research-protocol-to-a-general-ai-reasoning-framework:zh":205,"related:post:from-research-protocol-to-a-general-ai-reasoning-framework:zh:1":2661},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","zh","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,181,193],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-folder","custom",[],{"id":176,"title":177,"url":179,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":180},"item-28",{"de":178,"en":178,"es":178,"fr":178,"it":178,"ru":178,"sr":178,"zh":178},"Solr Suggester","\u002Fportfolio\u002Fsolr-fuzzy-suggester-und-solr-infix-suggester-abfrage-ueber-ajax-und-filterung",[],{"id":182,"title":183,"url":191,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":192},"item-27",{"de":184,"en":185,"es":186,"fr":187,"it":188,"ru":189,"sr":190,"zh":185},"Firmenwebseite SEO","Company Website SEO","Sitio web corporativo SEO","Site web d’entreprise SEO","Sito web aziendale SEO","Корпоративный сайт SEO","Пословна веб-страница SEO","\u002Fportfolio\u002Fseo-sem-branding-mobile-webseite-muenchen",[],{"id":194,"title":195,"url":203,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":204},"item-31",{"de":196,"en":197,"es":198,"fr":199,"it":200,"ru":201,"sr":202,"zh":197},"Digitalisierungsportal","Digitalization Portal","Portal de digitalización","Portail de numérisation","Portale di digitalizzazione","Портал цифровизации","Портал за дигитализацију","\u002Fportfolio\u002Fdigitalisierungsportal-archiv-museum-bibliothek-ead-lido-mets-mods",[],{"statusCode":4,"data":206,"message":2660},{"id":207,"title":208,"slug":209,"content":210,"contentJson":211,"excerpt":1237,"featuredImage":1238,"featuredImageAlt":1239,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":1240,"publishedAt":1241,"createdAt":1242,"updatedAt":1243,"seoLocalePaths":1244,"categories":1253,"author":1254,"translations":1258},"465","从研究协议到通用AI推理框架","from-research-protocol-to-a-general-ai-reasoning-framework","\u003Cp>本系列所开发的方法论始于一个研究问题：AI 模型如何帮助调查一个复杂问题，而不是仅仅强化用户提示中已有的假设？\u003C\u002Fp>\n\u003Cp>这个问题最初似乎属于历史或学术研究。实际上，它要广泛得多。\u003C\u002Fp>\n\u003Cp>软件调试、架构决策、技术诊断、安全分析、产品策略以及许多形式的决策支持都具有相同的底层结构。一个问题被提出。一些证据可用。一个或多个解释看似合理。假设进入分析。系统必须确定哪个结论得到最好的支持。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">领域会变化。认识论问题往往不会。\u003C\u002Fblockquote>\n\u003Cp>因此，本文采取下一步：将前几篇文章中开发的研究协议转化为 AI 辅助分析工作的通用推理框架。\u003C\u002Fp>\n\u003Cp>该框架直接建立在四个先前层次之上：\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">超越提示工程：更可靠 AI 推理的方法论\u003C\u002Fa>定义了整体方法论问题；\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">提示是偏见的一部分\u003C\u002Fa>考察了框架和提示依赖；\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">提示不变性：结论能在提示下存活吗？\u003C\u002Fa>引入了针对框架的稳健性测试；以及\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI 推理的证伪：从答案到经过检验的假设\u003C\u002Fa>增加了系统性的反证和竞争性假设。\u003C\u002Fp>\n\u003Ch2>关键泛化\u003C\u002Fh2>\n\u003Cp>这种泛化并不是说每项任务都应像历史研究一样对待。\u003C\u002Fp>\n\u003Cp>那将是一种范畴错误。\u003C\u002Fp>\n\u003Cp>历史研究依赖于年代顺序、来源、传承、史料批判和档案完整性。软件调试依赖于日志、配置、可复现性、执行状态和受控测试。架构依赖于需求、约束、接口、质量属性和运营权衡。\u003C\u002Fp>\n\u003Cp>可以泛化的是围绕这些领域特定证据形式的推理过程。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">该框架在核心上是领域无关的，但从不脱离领域。\u003C\u002Fblockquote>\n\u003Cp>这一区分是根本性的。通用推理方法论不能取代专业知识。相反，它可以定义在领域专业知识产生最终判断之前，应如何处理证据、假设、竞争性假设、矛盾和不确定性。\u003C\u002Fp>\n\u003Ch2>领域无关的推理核心\u003C\u002Fh2>\n\u003Cp>跨分析领域，可以应用相同的高层序列：\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">问题 → 分解 → 证据 → 假设 → 竞争性假设 → 预测 → 反证 → 区分性测试 → 领域验证 → 置信度重新校准 → 结论\u003C\u002Fblockquote>\n\u003Cp>该过程有意防止模型第一个连贯的解释默认成为最终答案。\u003C\u002Fp>\n\u003Cp>相反，第一个解释成为一个候选。\u003C\u002Fp>\n\u003Ch3>1. 定义实际问题\u003C\u002Fh3>\n\u003Cp>在生成解决方案之前，系统应确定实际被问及的是什么。\u003C\u002Fp>\n\u003Cp>用户请求通常同时包含问题和提议的诊断：\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">为什么 nginx 导致我的 API 连接失败？\u003C\u002Fblockquote>\n\u003Cp>实际问题可能反而是：\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">是什么导致 API 连接失败？\u003C\u002Fblockquote>\n\u003Cp>这种差异在语言上很小，在方法论上却巨大。\u003C\u002Fp>\n\u003Cp>第一种表述嵌入了一个因果假设。第二种则将这一假设置于竞争之中。\u003C\u002Fp>\n\u003Cp>这正是 \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">提示词是偏见的一部分\u003C\u002Fa> 中所考察的框架问题。\u003C\u002Fp>\n\u003Ch3>2. 区分事实、假设和未知\u003C\u002Fh3>\n\u003Cp>可靠的推理过程必须防止观察和解释悄然合并。\u003C\u002Fp>\n\u003Cul>\u003Cli>\u003Cb>事实：\u003C\u002Fb>直接由现有证据支持。\u003C\u002Fli>\u003Cli>\u003Cb>解释：\u003C\u002Fb>从事实中推断出的说明。\u003C\u002Fli>\u003Cli>\u003Cb>假设：\u003C\u002Fb>当前推理路径所需但尚未独立确立的命题。\u003C\u002Fli>\u003Cli>\u003Cb>未知：\u003C\u002Fb>进行更强区分所需但目前无法获得的信息。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>这种分类可能看起来很简单，但它解决了 AI 生成分析中的一个常见失败模式：假设被嵌入流畅的文字中，随后又像已经被确立一样重新出现。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">可追溯性始于防止推断伪装成证据。\u003C\u002Fblockquote>\n\u003Ch3>3. 生成相互竞争的解释\u003C\u002Fh3>\n\u003Cp>当存在可信的替代方案时，不应孤立地评估候选解释。\u003C\u002Fp>\n\u003Cp>因此，系统在确定一个假设之前，应生成多个合理的假设。\u003C\u002Fp>\n\u003Cp>目的不是人为的头脑风暴。替代方案应代表能够解释观察结果的真正不同的机制。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">如果没有让任何严肃的竞争者进入测试，那么一个受偏好的假设就尚未胜出。\u003C\u002Fblockquote>\n\u003Cp>这一概念是\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI推理中的证伪\u003C\u002Fa>的核心，其中假设通过预期观察、反证和区分性测试来评估，而不是通过累积的支持性论述。\u003C\u002Fp>\n\u003Ch3>4. 保留证据来源\u003C\u002Fh3>\n\u003Cp>证据应始终可追溯至其来源。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">结论 → 解释 → 观察 → 来源\u003C\u002Fblockquote>\n\u003Cp>日志行、官方技术规范、实验、原始历史文献、专家解释和AI生成的摘要并不具有相同的证据地位。\u003C\u002Fp>\n\u003Cp>因此，该框架不仅仅存储信息。它应保留足够的元数据，以了解某个主张来自何处，以及最终结论距离原始观察有多远。\u003C\u002Fp>\n\u003Ch3>5. 在解释结果之前先推导预测\u003C\u002Fh3>\n\u003Cp>一旦假设存在，每个假设都应产生预期。\u003C\u002Fp>\n\u003Cp>如果该假设正确，我们应该观察到什么？\u003C\u002Fp>\n\u003Cp>如果某个替代假设正确，应该有什么不同？\u003C\u002Fp>\n\u003Cp>在解释所有可用证据之前先推导预期，有助于防止每个观察结果都被事后套入偏好的叙事中。\u003C\u002Fp>\n\u003Ch3>6. 寻找反证\u003C\u002Fh3>\n\u003Cp>生成模型自然擅长为一个看似合理的想法构建连贯的支持。这使得否证尤为重要。\u003C\u002Fp>\n\u003Cp>系统应明确询问：\u003C\u002Fp>\n\u003Cul>\u003Cli>什么观察会显著削弱这一解释？\u003C\u002Fli>\u003Cli>该假设对哪些证据解释得很差？\u003C\u002Fli>\u003Cli>哪个替代假设以更少的假设解释了相同的观察？\u003C\u002Fli>\u003Cli>哪些预期观察缺失了？\u003C\u002Fli>\u003Cli>表面上的支持是否可能由另一种机制解释？\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>这一原则在\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI推理中的证伪：从答案到经过检验的假设\u003C\u002Fa>中有详细展开。\u003C\u002Fp>\n\u003Ch3>7. 测试对提示的依赖性\u003C\u002Fh3>\n\u003Cp>仅靠证据测试并不能揭示模型的分析是否仍然过度依赖于用户的表述。\u003C\u002Fp>\n\u003Cp>因此，对于足够重要的问题，可以在其他合理的表述框架下重新检验结论。\u003C\u002Fp>\n\u003Cul>\u003Cli>\u003Cb>原始：\u003C\u002Fb>保留用户的表述。\u003C\u002Fli>\u003Cli>\u003Cb>盲测：\u003C\u002Fb>移除偏好的结论。\u003C\u002Fli>\u003Cli>\u003Cb>反转：\u003C\u002Fb>突出最强的替代方案。\u003C\u002Fli>\u003Cli>\u003Cb>对抗：\u003C\u002Fb>刻意寻找最强的基于证据的挑战。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>这就是\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">提示不变性：结论能否在提示变化下存活？\u003C\u002Fa>中介绍的提示不变性方法。\u003C\u002Fp>\n\u003Cp>它的作用不是证明一个稳定的答案是真实的。它的作用是揭示那些主要因为表述框架改变而改变的结论。\u003C\u002Fp>\n\u003Ch3>8. 应用领域特定的验证器\u003C\u002Fh3>\n\u003Cp>这就是通用框架有意不再具有普遍性的地方。\u003C\u002Fp>\n\u003Cp>每个领域都需要自己的规则来确定一个解释是否真正可信。\u003C\u002Fp>\n\u003Ctable class=\"w-full border-collapse\">\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">领域\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">典型验证器\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">历史研究\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">年代学、来源、地理合理性、传播渠道、原始资料、史学\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">软件调试\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">日志、复现、运行时状态、配置、受控变更、错误关联\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">软件架构\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">需求、约束、可扩展性、可维护性、安全性、可操作性、成本\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">安全分析\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">遥测、攻击路径、权限、可观察指标、可复现性、威胁模型\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">产品策略\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">客户证据、定价行为、转化、替代方案、市场约束、失败标准\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">项目管理\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">范围、依赖关系、资源、风险、验收标准、里程碑、利益相关者约束\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">科学分析\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">实验设计、测量质量、对照、可复现性、统计证据、替代解释\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftable>\n\u003Cp>框架控制如何处理主张。领域验证器决定这些主张在与现实接触后能否存活。\u003C\u002Fp>\n\u003Ch3>9. 重新校准置信度\u003C\u002Fh3>\n\u003Cp>最终结论不应自动保留初始答案的置信水平。\u003C\u002Fp>\n\u003Cp>在考虑了替代方案、反证、提示变化和领域验证之后，应重新计算置信度。\u003C\u002Fp>\n\u003Cp>因此，有用的输出可以保持不确定性。\u003C\u002Fp>\n\u003Cul>\u003Cli>强烈支持；\u003C\u002Fli>\u003Cli>暂时支持；\u003C\u002Fli>\u003Cli>弱支持；\u003C\u002Fli>\u003Cli>证据不足；\u003C\u002Fli>\u003Cli>实质矛盾；\u003C\u002Fli>\u003Cli>无法用当前证据检验。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">当证据本身不确定时，不确定性不是推理的失败。\u003C\u002Fblockquote>\n\u003Ch2>跨不同领域的同一框架\u003C\u002Fh2>\n\u003Cp>理解通用框架的最简单方法是观察当领域变化时，相同的推理架构如何表现。\u003C\u002Fp>\n\u003Ch3>历史研究\u003C\u002Fh3>\n\u003Cp>问题：确定两种思想传统之间的相似性是否表明存在历史传播。\u003C\u002Fp>\n\u003Cul>\u003Cli>将有文献记载的相似性与解释性类比区分开来。\u003C\u002Fli>\u003Cli>比较直接传播、间接继承、趋同和回溯性解释。\u003C\u002Fli>\u003Cli>检验年代顺序和地理接触。\u003C\u002Fli>\u003Cli>寻找文献或术语方面的痕迹。\u003C\u002Fli>\u003Cli>确定在直接传播假设下预期会出现的证据。\u003C\u002Fli>\u003Cli>寻找会削弱该假设的更早的独立例证。\u003C\u002Fli>\u003Cli>在不预设偏好谱系的情况下重新构建问题。\u003C\u002Fli>\u003Cli>仅在现存证据所支持的置信水平上得出结论。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch3>软件调试\u003C\u002Fh3>\n\u003Cp>问题：确定为什么API连接会间歇性失败。\u003C\u002Fp>\n\u003Cul>\u003Cli>将观察到的故障与开发者怀疑的原因区分开来。\u003C\u002Fli>\u003Cli>生成代理、应用、数据库、网络和客户端假设。\u003C\u002Fli>\u003Cli>确定每个假设所预测的日志和运行时行为。\u003C\u002Fli>\u003Cli>执行能够区分这些假设的测试。\u003C\u002Fli>\u003Cli>在逐层排除的同时尝试复现故障。\u003C\u002Fli>\u003Cli>拒绝与受控观察相矛盾的说明。\u003C\u002Fli>\u003Cli>确定首个被确认的故障机制，而不是最有说服力的叙述。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch3>软件架构\u003C\u002Fh3>\n\u003Cp>问题：为新平台选择一种架构。\u003C\u002Fp>\n\u003Cul>\u003Cli>将需求与实现偏好区分开来。\u003C\u002Fli>\u003Cli>比较模块化单体、服务和混合方案。\u003C\u002Fli>\u003Cli>根据可扩展性、团队结构、部署、运维复杂性和成本测试每个选项。\u003C\u002Fli>\u003Cli>确定哪项需求真正需要架构复杂性。\u003C\u002Fli>\u003Cli>寻找能够满足相同约束的更简单设计。\u003C\u002Fli>\u003Cli>将技术偏好视为假设而非需求。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch3>产品战略\u003C\u002Fh3>\n\u003Cp>问题：确定一项产品能力是否应该商业化。\u003C\u002Fp>\n\u003Cul>\u003Cli>将技术可能性与已证明的需求区分开来。\u003C\u002Fli>\u003Cli>定义替代性的客户问题和替代性解决方案。\u003C\u002Fli>\u003Cli>确定商业论点所预测的可观察市场行为。\u003C\u002Fli>\u003Cli>在解释结果之前定义失败标准。\u003C\u002Fli>\u003Cli>寻找客户以不同方式解决问题的证据。\u003C\u002Fli>\u003Cli>区分兴趣、测试意愿和付费意愿。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch3>项目与交付分析\u003C\u002Fh3>\n\u003Cp>问题：确定为什么项目未能实现计划成果。\u003C\u002Fp>\n\u003Cul>\u003Cli>将延迟等症状与其假定原因区分开来。\u003C\u002Fli>\u003Cli>比较范围、依赖关系、能力、需求、治理和技术风险。\u003C\u002Fli>\u003Cli>通过计划、决策、变更和验收标准追踪证据。\u003C\u002Fli>\u003Cli>检验纠正措施是否针对实际机制而非可见症状。\u003C\u002Fli>\u003Cli>随着新的交付证据出现，重新评估项目假设。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2>为什么这不仅仅是提示工程\u003C\u002Fh2>\n\u003Cp>提示工程改变的是呈现给模型的指令。\u003C\u002Fp>\n\u003Cp>推理框架改变的是允许答案成为结论的过程。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">提示工程优化一次调用。推理框架治理一个推理过程。\u003C\u002Fblockquote>\n\u003Ch2>该框架与现有LLM推理方法相关——但有所不同\u003C\u002Fh2>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">目标不仅仅是让模型搜索更久、反思更多或生成更好的答案。目标是使从证据到结论的路径更能抵抗框架效应、确认偏差和无根据的假设。\u003C\u002Fblockquote>\n\u003Ch2>AI推理质量的分层视角\u003C\u002Fh2>\n\u003Ctable class=\"w-full border-collapse\">\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">层次\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">问题\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">任务理解\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">我们实际要解决的是什么问题？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">证据\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">我们实际知道什么？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">假设\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">我们目前将什么视为理所当然？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">假说\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">哪些可能的机制可以解释证据？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">证伪\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">什么证据可能削弱每种解释？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">提示不变性\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">结论是否过度依赖于表述框架？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">领域验证\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">该解释是否满足相关学科的规则？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">置信度校准\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">幸存的结论应被多大程度地信任？\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftable>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">正确性不是单一属性。它是多个依赖关系同时成立的结果。\u003C\u002Fblockquote>\n\u003Ch2>推理质量与模型能力\u003C\u002Fh2>\n\u003Cp>这又回到了本系列的核心思想之一。\u003C\u002Fp>\n\u003Cp>更强大的模型并不自动意味着每次调用都会以最严谨的方式运用其能力。\u003C\u002Fp>\n\u003Cp>模型可能具备生成替代方案、检查日志、搜索来源、质疑假设并修正结论的能力。这些操作是否实际发生，取决于任务、提示方式、可用工具、上下文以及周围的系统架构。\u003C\u002Fp>\n\u003Cp>因此，该框架区分了两个问题：\u003C\u002Fp>\n\u003Cul>\u003Cli>\u003Cb>能力：\u003C\u002Fb>模型能做什么？\u003C\u002Fli>\u003Cli>\u003Cb>方法论纪律：\u003C\u002Fb>在系统接受结论之前，必须行使哪些能力？\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>这一区分是\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">超越提示工程\u003C\u002Fa>的起点，并且当该方法论从研究领域推广到更广泛场景时，其重要性更加凸显。\u003C\u002Fp>\n\u003Ch2>并非每项任务都需要完整框架\u003C\u002Fh2>\n\u003Cp>该框架不应成为每次AI交互的强制性开销。\u003C\u002Fp>\n\u003Cp>许多任务的认识复杂度较低：\u003C\u002Fp>\n\u003Cul>\u003Cli>格式化此JSON；\u003C\u002Fli>\u003Cli>翻译这句话；\u003C\u002Fli>\u003Cli>转换这些单位；\u003C\u002Fli>\u003Cli>重写这条消息；\u003C\u002Fli>\u003Cli>提取这些字段；\u003C\u002Fli>\u003Cli>总结所提供的文本。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>对此类任务运行四种提示变体、三个竞争假设和一个证伪阶段，会增加成本而不会带来成比例的价值。\u003C\u002Fp>\n\u003Cp>完整方法在以下情况下更有价值：\u003C\u002Fp>\n\u003Cul>\u003Cli>答案取决于解释而非直接检索；\u003C\u002Fli>\u003Cli>用户已有偏好的解释；\u003C\u002Fli>\u003Cli>多种因果机制都看似合理；\u003C\u002Fli>\u003Cli>证据不完整或相互矛盾；\u003C\u002Fli>\u003Cli>决策具有重大的技术、财务或研究后果；\u003C\u002Fli>\u003Cli>AI系统被期望根据结论自主行动。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>因此，方法论深度应随认识风险而扩展。\u003C\u002Fp>\n\u003Ch2>从静态提示到自适应推理策略\u003C\u002Fh2>\n\u003Cp>一旦推广，该框架就不再需要作为一个庞大的提示存在。\u003C\u002Fp>\n\u003Cp>一个实用的系统可以动态地决定一项任务需要哪些控制措施。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">简单任务 → 直接回答\n分析性任务 → 结构化证据检查\n高不确定性任务 → 竞争性假设\n框架敏感任务 → 提示不变性\n高后果假设 → 证伪与领域验证\u003C\u002Fblockquote>\n\u003Cp>这将方法论从固定的提示模板转变为自适应的推理策略。\u003C\u002Fp>\n\u003Cp>系统并不总是进行最大程度的推理。它根据任务所需的严格程度进行推理。\u003C\u002Fp>\n\u003Ch2>一个实用的通用框架\u003C\u002Fh2>\n\u003Cp>完整流程可总结如下：\u003C\u002Fp>\n\u003Col>\u003Cli>\u003Cb>规范化问题。\u003C\u002Fb> 在适当情况下，从任务定义中移除嵌入的结论。\u003C\u002Fli>\u003Cli>\u003Cb>识别证据。\u003C\u002Fb> 将观察与解释分开。\u003C\u002Fli>\u003Cli>\u003Cb>暴露假设。\u003C\u002Fb> 记录尚未确立的命题。\u003C\u002Fli>\u003Cli>\u003Cb>生成严肃的替代方案。\u003C\u002Fb> 不要让第一个假设只与薄弱的稻草人竞争。\u003C\u002Fli>\u003Cli>\u003Cb>保留来源。\u003C\u002Fb> 保持主张可追溯至其来源。\u003C\u002Fli>\u003Cli>\u003Cb>推导预期。\u003C\u002Fb> 定义每个假设所预测的内容。\u003C\u002Fli>\u003Cli>\u003Cb>寻找反证。\u003C\u002Fb> 主动寻找首选假设处理不佳的观察结果。\u003C\u002Fli>\u003Cli>\u003Cb>进行区分性测试。\u003C\u002Fb> 优先选择能区分不同解释的证据。\u003C\u002Fli>\u003Cli>\u003Cb>测试框架依赖性。\u003C\u002Fb> 当用户的框架可能影响结果时，应用提示不变性。\u003C\u002Fli>\u003Cli>\u003Cb>应用领域验证器。\u003C\u002Fb> 使用与问题相关的学科特定标准。\u003C\u002Fli>\u003Cli>\u003Cb>重新校准置信度。\u003C\u002Fb> 允许结论减弱、保持未解决或改变。\u003C\u002Fli>\u003Cli>\u003Cb>保留可追溯性。\u003C\u002Fb> 使从结论回到证据的路径可检查。\u003C\u002Fli>\u003C\u002Fol>\n\u003Ch2>该框架不声称的内容\u003C\u002Fh2>\n\u003Cul>\u003Cli>它不保证输出真实。\u003C\u002Fli>\u003Cli>它不消除幻觉。\u003C\u002Fli>\u003Cli>它不使LLM成为领域专家。\u003C\u002Fli>\u003Cli>它不证明稳定的结论是正确的。\u003C\u002Fli>\u003Cli>它不消除所有推理过程中共有的偏见。\u003C\u002Fli>\u003Cli>它不替代实验、测量或原始来源。\u003C\u002Fli>\u003Cli>它不保证所有相关假设都已生成。\u003C\u002Fli>\u003Cli>它不意味着更多推理总是更好。\u003C\u002Fli>\u003Cli>它不声称完整框架已被验证为标准化的学术方法论。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>其主张更为狭窄且更具辩护性：当假设、框架、竞争性解释、反证和领域验证被明确处理，而不是完全留给一次无约束的生成时，分析性AI系统可以在方法论上变得更强。\u003C\u002Fp>\n\u003Ch2>四篇文章构成一个推理系统\u003C\u002Fh2>\n\u003Cp>之前的文章现在可以理解为不是独立的提示技术，而是一个推理架构的组成部分。\u003C\u002Fp>\n\u003Ctable class=\"w-full border-collapse\">\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">文章\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">在框架中的功能\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">超越提示工程\u003C\u002Fa>\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">定义了从答案生成到方法论推理的转变。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">提示是偏见的一部分\u003C\u002Fa>\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">将用户的表述视为推理偏见的可能来源。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">提示不变性\u003C\u002Fa>\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">测试结论是否在框架的有意义变化中保持不变。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI推理的证伪\u003C\u002Fa>\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">测试假设是否在反证和严肃的替代方案中存活。\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">\u003Cb>从研究协议到通用AI推理框架\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">将控制措施组合成一个与领域无关的核心，并带有领域特定的验证。\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftable>\n\u003Cp>它们共同描述了一个转变：\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">提示 → 答案\n变为\n问题 → 证据 → 假设 → 挑战 → 验证 → 校准结论\u003C\u002Fblockquote>\n\u003Ch2>核心原则\u003C\u002Fh2>\n\u003Cp>该方法论始于一个简单的观察：向一个强大的推理模型提问并不自动意味着模型可用的所有有用推理能力都会被使用。\u003C\u002Fp>\n\u003Cp>解决方案不是将最大推理强加到每一次交互中。\u003C\u002Fp>\n\u003Cp>解决方案是确定哪些推理控制对任务重要，并将其明确化。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">好的AI方法论不会告诉模型应该得出什么结论。它定义了结论在被接受之前必须经受住什么考验。\u003C\u002Fblockquote>\n\u003Cp>这一原则是可迁移的。\u003C\u002Fp>\n\u003Cp>在历史研究中，结论必须经受住年代学和史料考证的检验。在调试中，它必须经受住复现和区分性测试的检验。在架构中，它必须经受住需求和运营约束的检验。在战略中，它必须经受住市场证据和失败标准的检验。\u003C\u002Fp>\n\u003Cp>验证者会变。\u003C\u002Fp>\n\u003Cp>方法论纪律不变。\u003C\u002Fp>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">目标不是让模型总是思考更久。而是一个知道答案何时尚未得到充分论证的系统。\u003C\u002Fblockquote>\n\u003Cdiv class=\"ce-delimiter cdx-block my-8\">\u003C\u002Fdiv>\n\u003Ch2>研究背景\u003C\u002Fh2>\n\u003Cp>本系列中描述的完整框架是一种方法论综合，而非已确立的标准化LLM基准。然而，若干相邻研究方向支持其核心架构前提：当语言模型的问题求解被组织为迭代或多阶段过程而非单次生成时，推理质量可能发生实质性变化。\u003C\u002Fp>\n\u003Cp>ReAct将推理与行动及外部观察相结合，并在问答、事实核查和交互式决策任务中展示了优势。思维树探索多条候选推理路径，通过评估和回溯而非立即锁定单一链条。Self-Refine通过在生成、反馈和精炼之间迭代，在多种任务上展现了改进。Reflexion表明语言智能体可以利用先前尝试的言语反馈来改进后续决策，而无需更新模型权重。\u003C\u002Fp>\n\u003Cp>这些方法并未实现提示不变性或本文提出的以证伪为导向的框架。它们为更广泛的观点提供了独立证据：推理时程序很重要——模型能力与行使该能力的过程并不等同。\u003C\u002Fp>\n\u003Cp>本系列的贡献在于围绕认识论控制来组织这一洞见：证据分离、假设追踪、提示框架分析、竞争性假设、证伪、来源追溯、领域验证和校准不确定性。\u003C\u002Fp>\n\u003Ch3>精选参考文献\u003C\u002Fh3>\n\u003Cul>\u003Cli>Yao, S. et al. — \u003Ci>ReAct: Synergizing Reasoning and Acting in Language Models\u003C\u002Fi>. ICLR, 2023.\u003C\u002Fli>\u003Cli>Yao, S. et al. — \u003Ci>Tree of Thoughts: Deliberate Problem Solving with Large Language Models\u003C\u002Fi>. NeurIPS, 2023.\u003C\u002Fli>\u003Cli>Madaan, A. et al. — \u003Ci>Self-Refine: Iterative Refinement with Self-Feedback\u003C\u002Fi>. NeurIPS, 2023.\u003C\u002Fli>\u003Cli>Shinn, N. et al. — \u003Ci>Reflexion: Language Agents with Verbal Reinforcement Learning\u003C\u002Fi>. NeurIPS, 2023.\u003C\u002Fli>\u003Cli>Jhaveri, A. R., GX-Chen, A., Sucholutsky, I. & Choi, E. — \u003Ci>Failing to Falsify: Evaluating and Mitigating Confirmation Bias in Language Models\u003C\u002Fi>. 2026.\u003C\u002Fli>\u003Cli>Brucks, M. S. & Toubia, O. — \u003Ci>Prompt Architecture Induces Methodological Artifacts in Large Language Models\u003C\u002Fi>. PLOS ONE, 2025.\u003C\u002Fli>\u003C\u002Ful>\n\u003Cdiv class=\"ce-delimiter cdx-block my-8\">\u003C\u002Fdiv>\n\u003Ch2>AI推理方法论系列\u003C\u002Fh2>\n\u003Col>\u003Cli>\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">\u003Cb>超越提示工程：一种更可靠AI推理的方法论\u003C\u002Fb>\u003C\u002Fa> — 为什么仅靠模型能力是不够的。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">\u003Cb>提示是偏见的一部分\u003C\u002Fb>\u003C\u002Fa> — 任务表述如何影响推理环境。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">\u003Cb>提示不变性：结论能否经受住提示的考验？\u003C\u002Fb>\u003C\u002Fa> — 测试结论在替代框架下的稳定性。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">\u003Cb>AI推理的证伪：从答案到经过检验的假设\u003C\u002Fb>\u003C\u002Fa> — 竞争性假设、反证和区分性测试。\u003C\u002Fli>\u003Cli>\u003Cb>从研究协议到通用AI推理框架\u003C\u002Fb> — 将该方法论整合为可复用的领域无关推理核心。\u003C\u002Fli>\u003Cli>同一推理框架在游戏AI中的实际应用——包括游戏助手、自主智能体、补丁敏感知识和实时游戏状态验证——在\u003Ca href=\"https:\u002F\u002Ffigure.rocks\u002Fde\u002Fblog\u002Fwhen-gaming-ai-sounds-right-but-isn-t-the-reasoning-problem-behind-game-assistants-and-agents\">当游戏AI听起来正确但并非如此：游戏助手和智能体背后的推理问题（figure.rocks）\u003C\u002Fa>中进行了探讨。\u003C\u002Fli>\u003C\u002Fol>",{"time":212,"blocks":213,"version":1236},1789803868903,[214,220,225,230,238,243,248,254,259,264,269,274,279,284,289,294,299,304,309,315,321,327,332,338,343,349,355,361,366,372,384,390,395,400,406,412,418,423,428,433,438,443,448,453,458,463,468,473,478,483,488,493,504,509,514,519,524,534,539,544,549,554,559,589,594,599,604,609,614,626,631,636,641,645,650,664,668,673,686,690,695,707,712,717,729,734,739,750,755,760,765,770,775,780,785,790,795,800,805,810,815,820,825,830,835,840,872,877,882,887,892,898,904,910,916,924,930,935,941,947,959,965,971,983,989,994,1000,1005,1010,1015,1020,1025,1030,1050,1055,1070,1075,1080,1085,1108,1113,1118,1123,1128,1133,1138,1143,1148,1153,1158,1163,1168,1173,1178,1183,1188,1193,1198,1203,1215,1219,1224],{"id":215,"data":216,"type":218,"tunes":219},"ESemLsLo8j",{"text":217},"本系列所开发的方法论始于一个研究问题：AI 模型如何帮助调查一个复杂问题，而不是仅仅强化用户提示中已有的假设？","paragraph",{},{"id":221,"data":222,"type":218,"tunes":224},"4Y0K3YWc6x",{"text":223},"这个问题最初似乎属于历史或学术研究。实际上，它要广泛得多。",{},{"id":226,"data":227,"type":218,"tunes":229},"Tdh0-PzDMD",{"text":228},"软件调试、架构决策、技术诊断、安全分析、产品策略以及许多形式的决策支持都具有相同的底层结构。一个问题被提出。一些证据可用。一个或多个解释看似合理。假设进入分析。系统必须确定哪个结论得到最好的支持。",{},{"id":231,"data":232,"type":236,"tunes":237},"y1qR6aL83j",{"text":233,"caption":234,"alignment":235},"领域会变化。认识论问题往往不会。","","left","quote",{},{"id":239,"data":240,"type":218,"tunes":242},"OSYX72lgMi",{"text":241},"因此，本文采取下一步：将前几篇文章中开发的研究协议转化为 AI 辅助分析工作的通用推理框架。",{},{"id":244,"data":245,"type":218,"tunes":247},"0nk7KQPJLF",{"text":246},"该框架直接建立在四个先前层次之上：\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">超越提示工程：更可靠 AI 推理的方法论\u003C\u002Fa>定义了整体方法论问题；\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">提示是偏见的一部分\u003C\u002Fa>考察了框架和提示依赖；\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">提示不变性：结论能在提示下存活吗？\u003C\u002Fa>引入了针对框架的稳健性测试；以及\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI 推理的证伪：从答案到经过检验的假设\u003C\u002Fa>增加了系统性的反证和竞争性假设。",{},{"id":249,"data":250,"type":42,"tunes":253},"8u01PX2pJ_",{"text":251,"level":252},"关键泛化",2,{},{"id":255,"data":256,"type":218,"tunes":258},"KX-3wYIvfo",{"text":257},"这种泛化并不是说每项任务都应像历史研究一样对待。",{},{"id":260,"data":261,"type":218,"tunes":263},"Qlnajp4NIg",{"text":262},"那将是一种范畴错误。",{},{"id":265,"data":266,"type":218,"tunes":268},"K_gM2XnB_E",{"text":267},"历史研究依赖于年代顺序、来源、传承、史料批判和档案完整性。软件调试依赖于日志、配置、可复现性、执行状态和受控测试。架构依赖于需求、约束、接口、质量属性和运营权衡。",{},{"id":270,"data":271,"type":218,"tunes":273},"EQnEh291Ja",{"text":272},"可以泛化的是围绕这些领域特定证据形式的推理过程。",{},{"id":275,"data":276,"type":236,"tunes":278},"041iVGE35A",{"text":277,"caption":234,"alignment":235},"该框架在核心上是领域无关的，但从不脱离领域。",{},{"id":280,"data":281,"type":218,"tunes":283},"GsCAv2bn5D",{"text":282},"这一区分是根本性的。通用推理方法论不能取代专业知识。相反，它可以定义在领域专业知识产生最终判断之前，应如何处理证据、假设、竞争性假设、矛盾和不确定性。",{},{"id":285,"data":286,"type":42,"tunes":288},"iGG8FYrJ2O",{"text":287,"level":252},"领域无关的推理核心",{},{"id":290,"data":291,"type":218,"tunes":293},"GRuYGmzxLe",{"text":292},"跨分析领域，可以应用相同的高层序列：",{},{"id":295,"data":296,"type":236,"tunes":298},"lgjHVx7HZI",{"text":297,"caption":234,"alignment":235},"问题 → 分解 → 证据 → 假设 → 竞争性假设 → 预测 → 反证 → 区分性测试 → 领域验证 → 置信度重新校准 → 结论",{},{"id":300,"data":301,"type":218,"tunes":303},"AKTY6_aOdu",{"text":302},"该过程有意防止模型第一个连贯的解释默认成为最终答案。",{},{"id":305,"data":306,"type":218,"tunes":308},"eipK0zk3WD",{"text":307},"相反，第一个解释成为一个候选。",{},{"id":310,"data":311,"type":42,"tunes":314},"yU78N3Sd2e",{"text":312,"level":313},"1. 定义实际问题",3,{},{"id":316,"data":317,"type":218,"tunes":320},"fDuyj6nSv-",{"text":318,"type":218,"tunes":319},"在生成解决方案之前，系统应确定实际被问及的是什么。",{},{},{"id":322,"data":323,"type":218,"tunes":326},"sctkg30ATQ",{"text":324,"type":218,"tunes":325},"用户请求通常同时包含问题和提议的诊断：",{},{},{"id":328,"data":329,"type":236,"tunes":331},"HvF-G0rwXA",{"text":330,"caption":234,"alignment":235},"为什么 nginx 导致我的 API 连接失败？",{},{"id":333,"data":334,"type":218,"tunes":337},"mFKfHfGIDV",{"text":335,"type":218,"tunes":336},"实际问题可能反而是：",{},{},{"id":339,"data":340,"type":236,"tunes":342},"CbLWis4-Bu",{"text":341,"caption":234,"alignment":235},"是什么导致 API 连接失败？",{},{"id":344,"data":345,"type":218,"tunes":348},"cMcvbsDaMC",{"text":346,"type":218,"tunes":347},"这种差异在语言上很小，在方法论上却巨大。",{},{},{"id":350,"data":351,"type":218,"tunes":354},"E_kQoBwBzK",{"text":352,"type":218,"tunes":353},"第一种表述嵌入了一个因果假设。第二种则将这一假设置于竞争之中。",{},{},{"id":356,"data":357,"type":218,"tunes":360},"M86BF9OUO9",{"text":358,"type":218,"tunes":359},"这正是 \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">提示词是偏见的一部分\u003C\u002Fa> 中所考察的框架问题。",{},{},{"id":362,"data":363,"type":42,"tunes":365},"8jN_GEWnLD",{"text":364,"level":313},"2. 区分事实、假设和未知",{},{"id":367,"data":368,"type":218,"tunes":371},"4-45ArOJIC",{"text":369,"type":218,"tunes":370},"可靠的推理过程必须防止观察和解释悄然合并。",{},{},{"id":373,"data":374,"type":382,"tunes":383},"gn7a0JRFIn",{"meta":375,"items":376,"style":381},{},[377,378,379,380],"\u003Cb>事实：\u003C\u002Fb>直接由现有证据支持。","\u003Cb>解释：\u003C\u002Fb>从事实中推断出的说明。","\u003Cb>假设：\u003C\u002Fb>当前推理路径所需但尚未独立确立的命题。","\u003Cb>未知：\u003C\u002Fb>进行更强区分所需但目前无法获得的信息。","unordered","list",{},{"id":385,"data":386,"type":218,"tunes":389},"7gqnnnAyNH",{"text":387,"type":218,"tunes":388},"这种分类可能看起来很简单，但它解决了 AI 生成分析中的一个常见失败模式：假设被嵌入流畅的文字中，随后又像已经被确立一样重新出现。",{},{},{"id":391,"data":392,"type":236,"tunes":394},"wakPN0U08S",{"text":393,"caption":234,"alignment":235},"可追溯性始于防止推断伪装成证据。",{},{"id":396,"data":397,"type":42,"tunes":399},"UPatPWjaiP",{"text":398,"level":313},"3. 生成相互竞争的解释",{},{"id":401,"data":402,"type":218,"tunes":405},"zuRICKxeRX",{"text":403,"type":218,"tunes":404},"当存在可信的替代方案时，不应孤立地评估候选解释。",{},{},{"id":407,"data":408,"type":218,"tunes":411},"AvfAw-zrKJ",{"text":409,"type":218,"tunes":410},"因此，系统在确定一个假设之前，应生成多个合理的假设。",{},{},{"id":413,"data":414,"type":218,"tunes":417},"5JVkk_l5m9",{"text":415,"type":218,"tunes":416},"目的不是人为的头脑风暴。替代方案应代表能够解释观察结果的真正不同的机制。",{},{},{"id":419,"data":420,"type":236,"tunes":422},"6ZI39jnt9v",{"text":421,"caption":234,"alignment":235},"如果没有让任何严肃的竞争者进入测试，那么一个受偏好的假设就尚未胜出。",{},{"id":424,"data":425,"type":218,"tunes":427},"WGpU3HSfMr",{"text":426},"这一概念是\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI推理中的证伪\u003C\u002Fa>的核心，其中假设通过预期观察、反证和区分性测试来评估，而不是通过累积的支持性论述。",{},{"id":429,"data":430,"type":42,"tunes":432},"mUW1lRBq9a",{"text":431,"level":313},"4. 保留证据来源",{},{"id":434,"data":435,"type":218,"tunes":437},"m2TM2yNjau",{"text":436},"证据应始终可追溯至其来源。",{},{"id":439,"data":440,"type":236,"tunes":442},"3w2fzBzVH4",{"text":441,"caption":234,"alignment":235},"结论 → 解释 → 观察 → 来源",{},{"id":444,"data":445,"type":218,"tunes":447},"rkoICmxwZT",{"text":446},"日志行、官方技术规范、实验、原始历史文献、专家解释和AI生成的摘要并不具有相同的证据地位。",{},{"id":449,"data":450,"type":218,"tunes":452},"8ut4cK6e27",{"text":451},"因此，该框架不仅仅存储信息。它应保留足够的元数据，以了解某个主张来自何处，以及最终结论距离原始观察有多远。",{},{"id":454,"data":455,"type":42,"tunes":457},"hZ6BZYfcIo",{"text":456,"level":313},"5. 在解释结果之前先推导预测",{},{"id":459,"data":460,"type":218,"tunes":462},"aisk7gG2yj",{"text":461},"一旦假设存在，每个假设都应产生预期。",{},{"id":464,"data":465,"type":218,"tunes":467},"IBsZFlw9pf",{"text":466},"如果该假设正确，我们应该观察到什么？",{},{"id":469,"data":470,"type":218,"tunes":472},"Kykmkn9Dtx",{"text":471},"如果某个替代假设正确，应该有什么不同？",{},{"id":474,"data":475,"type":218,"tunes":477},"iS8nPIkYVU",{"text":476},"在解释所有可用证据之前先推导预期，有助于防止每个观察结果都被事后套入偏好的叙事中。",{},{"id":479,"data":480,"type":42,"tunes":482},"1rrToFbYF5",{"text":481,"level":313},"6. 寻找反证",{},{"id":484,"data":485,"type":218,"tunes":487},"NSjS51xlU6",{"text":486},"生成模型自然擅长为一个看似合理的想法构建连贯的支持。这使得否证尤为重要。",{},{"id":489,"data":490,"type":218,"tunes":492},"jPSoTMS13f",{"text":491},"系统应明确询问：",{},{"id":494,"data":495,"type":382,"tunes":503},"DfULqhWZpB",{"meta":496,"items":497,"style":381},{},[498,499,500,501,502],"什么观察会显著削弱这一解释？","该假设对哪些证据解释得很差？","哪个替代假设以更少的假设解释了相同的观察？","哪些预期观察缺失了？","表面上的支持是否可能由另一种机制解释？",{},{"id":505,"data":506,"type":218,"tunes":508},"rM5aFAv4Ch",{"text":507},"这一原则在\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI推理中的证伪：从答案到经过检验的假设\u003C\u002Fa>中有详细展开。",{},{"id":510,"data":511,"type":42,"tunes":513},"sWwUSTmVGW",{"text":512,"level":313},"7. 测试对提示的依赖性",{},{"id":515,"data":516,"type":218,"tunes":518},"wvQwo5fB4W",{"text":517},"仅靠证据测试并不能揭示模型的分析是否仍然过度依赖于用户的表述。",{},{"id":520,"data":521,"type":218,"tunes":523},"Z1AJX_Zx0J",{"text":522},"因此，对于足够重要的问题，可以在其他合理的表述框架下重新检验结论。",{},{"id":525,"data":526,"type":382,"tunes":533},"R9CxBqiZy5",{"meta":527,"items":528,"style":381},{},[529,530,531,532],"\u003Cb>原始：\u003C\u002Fb>保留用户的表述。","\u003Cb>盲测：\u003C\u002Fb>移除偏好的结论。","\u003Cb>反转：\u003C\u002Fb>突出最强的替代方案。","\u003Cb>对抗：\u003C\u002Fb>刻意寻找最强的基于证据的挑战。",{},{"id":535,"data":536,"type":218,"tunes":538},"Z6ReUs6ThD",{"text":537},"这就是\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">提示不变性：结论能否在提示变化下存活？\u003C\u002Fa>中介绍的提示不变性方法。",{},{"id":540,"data":541,"type":218,"tunes":543},"i72azJfoHd",{"text":542},"它的作用不是证明一个稳定的答案是真实的。它的作用是揭示那些主要因为表述框架改变而改变的结论。",{},{"id":545,"data":546,"type":42,"tunes":548},"PQJr3I5KR6",{"text":547,"level":313},"8. 应用领域特定的验证器",{},{"id":550,"data":551,"type":218,"tunes":553},"30EnJRKLgj",{"text":552},"这就是通用框架有意不再具有普遍性的地方。",{},{"id":555,"data":556,"type":218,"tunes":558},"HfkiK88IcE",{"text":557},"每个领域都需要自己的规则来确定一个解释是否真正可信。",{},{"id":560,"data":561,"type":587,"tunes":588},"zinQCXwjug",{"content":562,"stretched":43,"withHeadings":14},[563,566,569,572,575,578,581,584],[564,565],"领域","典型验证器",[567,568],"历史研究","年代学、来源、地理合理性、传播渠道、原始资料、史学",[570,571],"软件调试","日志、复现、运行时状态、配置、受控变更、错误关联",[573,574],"软件架构","需求、约束、可扩展性、可维护性、安全性、可操作性、成本",[576,577],"安全分析","遥测、攻击路径、权限、可观察指标、可复现性、威胁模型",[579,580],"产品策略","客户证据、定价行为、转化、替代方案、市场约束、失败标准",[582,583],"项目管理","范围、依赖关系、资源、风险、验收标准、里程碑、利益相关者约束",[585,586],"科学分析","实验设计、测量质量、对照、可复现性、统计证据、替代解释","table",{},{"id":590,"data":591,"type":218,"tunes":593},"uc0RyUCFOo",{"text":592},"框架控制如何处理主张。领域验证器决定这些主张在与现实接触后能否存活。",{},{"id":595,"data":596,"type":42,"tunes":598},"lAvhdCVN32",{"text":597,"level":313},"9. 重新校准置信度",{},{"id":600,"data":601,"type":218,"tunes":603},"h5PWNd0iA4",{"text":602},"最终结论不应自动保留初始答案的置信水平。",{},{"id":605,"data":606,"type":218,"tunes":608},"PcokJnli0n",{"text":607},"在考虑了替代方案、反证、提示变化和领域验证之后，应重新计算置信度。",{},{"id":610,"data":611,"type":218,"tunes":613},"4-JaOGRNdJ",{"text":612},"因此，有用的输出可以保持不确定性。",{},{"id":615,"data":616,"type":382,"tunes":625},"I6F7TJy23-",{"meta":617,"items":618,"style":381},{},[619,620,621,622,623,624],"强烈支持；","暂时支持；","弱支持；","证据不足；","实质矛盾；","无法用当前证据检验。",{},{"id":627,"data":628,"type":236,"tunes":630},"ozNw_amgCI",{"text":629,"caption":234,"alignment":235},"当证据本身不确定时，不确定性不是推理的失败。",{},{"id":632,"data":633,"type":42,"tunes":635},"LlyeIOmVPC",{"text":634,"level":252},"跨不同领域的同一框架",{},{"id":637,"data":638,"type":218,"tunes":640},"dmsjTGPTp7",{"text":639},"理解通用框架的最简单方法是观察当领域变化时，相同的推理架构如何表现。",{},{"id":642,"data":643,"type":42,"tunes":644},"cU0_n6SCY3",{"text":567,"level":313},{},{"id":646,"data":647,"type":218,"tunes":649},"es1dtKJQlZ",{"text":648},"问题：确定两种思想传统之间的相似性是否表明存在历史传播。",{},{"id":651,"data":652,"type":382,"tunes":663},"Nzl1cThLXK",{"meta":653,"items":654,"style":381},{},[655,656,657,658,659,660,661,662],"将有文献记载的相似性与解释性类比区分开来。","比较直接传播、间接继承、趋同和回溯性解释。","检验年代顺序和地理接触。","寻找文献或术语方面的痕迹。","确定在直接传播假设下预期会出现的证据。","寻找会削弱该假设的更早的独立例证。","在不预设偏好谱系的情况下重新构建问题。","仅在现存证据所支持的置信水平上得出结论。",{},{"id":665,"data":666,"type":42,"tunes":667},"RDf52TTR1X",{"text":570,"level":313},{},{"id":669,"data":670,"type":218,"tunes":672},"V_aQmUWc8S",{"text":671},"问题：确定为什么API连接会间歇性失败。",{},{"id":674,"data":675,"type":382,"tunes":685},"YOUlOo_WTe",{"meta":676,"items":677,"style":381},{},[678,679,680,681,682,683,684],"将观察到的故障与开发者怀疑的原因区分开来。","生成代理、应用、数据库、网络和客户端假设。","确定每个假设所预测的日志和运行时行为。","执行能够区分这些假设的测试。","在逐层排除的同时尝试复现故障。","拒绝与受控观察相矛盾的说明。","确定首个被确认的故障机制，而不是最有说服力的叙述。",{},{"id":687,"data":688,"type":42,"tunes":689},"tG7q0LkXG8",{"text":573,"level":313},{},{"id":691,"data":692,"type":218,"tunes":694},"Siq7TARR4r",{"text":693},"问题：为新平台选择一种架构。",{},{"id":696,"data":697,"type":382,"tunes":706},"PYBaGFi2Ol",{"meta":698,"items":699,"style":381},{},[700,701,702,703,704,705],"将需求与实现偏好区分开来。","比较模块化单体、服务和混合方案。","根据可扩展性、团队结构、部署、运维复杂性和成本测试每个选项。","确定哪项需求真正需要架构复杂性。","寻找能够满足相同约束的更简单设计。","将技术偏好视为假设而非需求。",{},{"id":708,"data":709,"type":42,"tunes":711},"-kikosSmX1",{"text":710,"level":313},"产品战略",{},{"id":713,"data":714,"type":218,"tunes":716},"RW8gaj5yQo",{"text":715},"问题：确定一项产品能力是否应该商业化。",{},{"id":718,"data":719,"type":382,"tunes":728},"JZBAxRuw9D",{"meta":720,"items":721,"style":381},{},[722,723,724,725,726,727],"将技术可能性与已证明的需求区分开来。","定义替代性的客户问题和替代性解决方案。","确定商业论点所预测的可观察市场行为。","在解释结果之前定义失败标准。","寻找客户以不同方式解决问题的证据。","区分兴趣、测试意愿和付费意愿。",{},{"id":730,"data":731,"type":42,"tunes":733},"NfuFu8q13_",{"text":732,"level":313},"项目与交付分析",{},{"id":735,"data":736,"type":218,"tunes":738},"GaM4feM9nZ",{"text":737},"问题：确定为什么项目未能实现计划成果。",{},{"id":740,"data":741,"type":382,"tunes":749},"-G7BLfDmdU",{"meta":742,"items":743,"style":381},{},[744,745,746,747,748],"将延迟等症状与其假定原因区分开来。","比较范围、依赖关系、能力、需求、治理和技术风险。","通过计划、决策、变更和验收标准追踪证据。","检验纠正措施是否针对实际机制而非可见症状。","随着新的交付证据出现，重新评估项目假设。",{},{"id":751,"data":752,"type":42,"tunes":754},"5iAhrZD3wL",{"text":753,"level":252},"为什么这不仅仅是提示工程",{},{"id":756,"data":757,"type":218,"tunes":759},"dm4MhgnM3L",{"text":758},"提示工程改变的是呈现给模型的指令。",{},{"id":761,"data":762,"type":218,"tunes":764},"87q5CSIEmb",{"text":763},"推理框架改变的是允许答案成为结论的过程。",{},{"id":766,"data":767,"type":236,"tunes":769},"7C_ZI5IHle",{"text":768,"caption":234,"alignment":235},"提示工程优化一次调用。推理框架治理一个推理过程。",{},{"id":771,"data":772},"SFq4pwo4Gx",{"text":773,"type":218,"tunes":774},"随着AI系统从聊天界面转向智能体、RAG流水线、自动化研究、开发工具和决策支持系统，这一差异变得尤为重要。",{},{"id":776,"data":777},"WItU2n6hS2",{"text":778,"type":218,"tunes":779},"一个精心设计的单一提示可能改善一次模型调用。而一个推理框架可以定义在系统给出答案之前，多次调用、检索到的证据、工具结果和验证阶段如何相互作用。",{},{"id":781,"data":782,"type":42,"tunes":784},"99vYubKOKi",{"text":783,"level":252},"该框架与现有LLM推理方法相关——但有所不同",{},{"id":786,"data":787},"y8Zf5V4OPJ",{"text":788,"type":218,"tunes":789},"更广泛的研究领域已经包含若干重要方法，表明当推理被结构化为一个过程而非单次生成时，LLM的性能可以得到提升。",{},{"id":791,"data":792},"gEIVPfVGEd",{"text":793,"type":218,"tunes":794},"\u003Cb>ReAct\u003C\u002Fb>将推理与来自外部环境的行动和观察相结合。模型不再完全依赖内部模型知识，而是可以行动、观察新信息并更新其下一步。",{},{"id":796,"data":797},"kKS8VaCYmR",{"text":798,"type":218,"tunes":799},"\u003Cb>Self-Refine\u003C\u002Fb>使用迭代生成、反馈和精炼，表明初始LLM输出通常可以通过显式的自我反馈得到改进，而无需额外的模型训练。",{},{"id":801,"data":802},"Puh4hZgRfK",{"text":803,"type":218,"tunes":804},"\u003Cb>Reflexion\u003C\u002Fb>使用言语反馈和情景记忆，使语言智能体能够从先前的尝试中学习并修改后续行为。",{},{"id":806,"data":807},"nAB7uNC8AX",{"text":808,"type":218,"tunes":809},"\u003Cb>思维树\u003C\u002Fb>探索多条推理路径，而不是立即确定一条从左到右的单一链，允许对候选方法进行搜索、评估和回溯。",{},{"id":811,"data":812},"B5Q7GKEqmh",{"text":813,"type":218,"tunes":814},"这些方法在目的和实现上差异很大，但它们确立了一个重要的一般性观点：推理时的结构可以实质性地改变模型性能。",{},{"id":816,"data":817},"Hj82H7bI5T",{"text":818,"type":218,"tunes":819},"本系列提出的框架针对的是一个不同的主要目标。",{},{"id":821,"data":822,"type":236,"tunes":824},"q8bQMqN2Ir",{"text":823,"caption":234,"alignment":235},"目标不仅仅是让模型搜索更久、反思更多或生成更好的答案。目标是使从证据到结论的路径更能抵抗框架效应、确认偏差和无根据的假设。",{},{"id":826,"data":827},"d1bIpPAnpB",{"text":828,"type":218,"tunes":829},"因此，提示不变性、面向证伪的测试和领域特定验证充当的是认识论控制手段，而非通用的推理扩展技术。",{},{"id":831,"data":832,"type":42,"tunes":834},"eXXYr_pAms",{"text":833,"level":252},"AI推理质量的分层视角",{},{"id":836,"data":837},"DqLa7MmX58",{"text":838,"type":218,"tunes":839},"由此产生的系统可以理解为若干层次。",{},{"id":841,"data":842,"type":587,"tunes":871},"0NzL-zFxBY",{"content":843,"stretched":43,"withHeadings":14},[844,847,850,853,856,859,862,865,868],[845,846],"层次","问题",[848,849],"任务理解","我们实际要解决的是什么问题？",[851,852],"证据","我们实际知道什么？",[854,855],"假设","我们目前将什么视为理所当然？",[857,858],"假说","哪些可能的机制可以解释证据？",[860,861],"证伪","什么证据可能削弱每种解释？",[863,864],"提示不变性","结论是否过度依赖于表述框架？",[866,867],"领域验证","该解释是否满足相关学科的规则？",[869,870],"置信度校准","幸存的结论应被多大程度地信任？",{},{"id":873,"data":874},"xm0-PFgN-i",{"text":875,"type":218,"tunes":876},"一个答案可能在任一层次上失败。",{},{"id":878,"data":879},"eCszY4dwX1",{"text":880,"type":218,"tunes":881},"它可能正确地回答了错误的问题。它可能从虚假证据出发进行正确推理。它可能检索到正确的事实，但选择了错误的因果解释。它可能产生一个强有力的解释，但当提示被重新表述时该解释就消失了。或者它可能通过所有一般推理检查，却仍然违反某个领域特定的约束。",{},{"id":883,"data":884,"type":236,"tunes":886},"hSmlX6vujv",{"text":885,"caption":234,"alignment":235},"正确性不是单一属性。它是多个依赖关系同时成立的结果。",{},{"id":888,"data":889,"type":42,"tunes":891},"EydNhsxtzy",{"text":890,"level":252},"推理质量与模型能力",{},{"id":893,"data":894,"type":218,"tunes":897},"s4Km9d-Rxy",{"text":895,"type":218,"tunes":896},"这又回到了本系列的核心思想之一。",{},{},{"id":899,"data":900,"type":218,"tunes":903},"39JzRxeSgF",{"text":901,"type":218,"tunes":902},"更强大的模型并不自动意味着每次调用都会以最严谨的方式运用其能力。",{},{},{"id":905,"data":906,"type":218,"tunes":909},"KEw35qMKGc",{"text":907,"type":218,"tunes":908},"模型可能具备生成替代方案、检查日志、搜索来源、质疑假设并修正结论的能力。这些操作是否实际发生，取决于任务、提示方式、可用工具、上下文以及周围的系统架构。",{},{},{"id":911,"data":912,"type":218,"tunes":915},"qijWfW7iDp",{"text":913,"type":218,"tunes":914},"因此，该框架区分了两个问题：",{},{},{"id":917,"data":918,"type":382,"tunes":923},"9TvQxkmlHA",{"meta":919,"items":920,"style":381},{},[921,922],"\u003Cb>能力：\u003C\u002Fb>模型能做什么？","\u003Cb>方法论纪律：\u003C\u002Fb>在系统接受结论之前，必须行使哪些能力？",{},{"id":925,"data":926,"type":218,"tunes":929},"rmrusg3t3h",{"text":927,"type":218,"tunes":928},"这一区分是\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">超越提示工程\u003C\u002Fa>的起点，并且当该方法论从研究领域推广到更广泛场景时，其重要性更加凸显。",{},{},{"id":931,"data":932,"type":42,"tunes":934},"W321yYXV97",{"text":933,"level":252},"并非每项任务都需要完整框架",{},{"id":936,"data":937,"type":218,"tunes":940},"QwEqB6r20x",{"text":938,"type":218,"tunes":939},"该框架不应成为每次AI交互的强制性开销。",{},{},{"id":942,"data":943,"type":218,"tunes":946},"XxxtP1e8oN",{"text":944,"type":218,"tunes":945},"许多任务的认识复杂度较低：",{},{},{"id":948,"data":949,"type":382,"tunes":958},"yJw3vKdi9z",{"meta":950,"items":951,"style":381},{},[952,953,954,955,956,957],"格式化此JSON；","翻译这句话；","转换这些单位；","重写这条消息；","提取这些字段；","总结所提供的文本。",{},{"id":960,"data":961,"type":218,"tunes":964},"c_3UzzXOlF",{"text":962,"type":218,"tunes":963},"对此类任务运行四种提示变体、三个竞争假设和一个证伪阶段，会增加成本而不会带来成比例的价值。",{},{},{"id":966,"data":967,"type":218,"tunes":970},"89oDA_RK5n",{"text":968,"type":218,"tunes":969},"完整方法在以下情况下更有价值：",{},{},{"id":972,"data":973,"type":382,"tunes":982},"T9axudVitU",{"meta":974,"items":975,"style":381},{},[976,977,978,979,980,981],"答案取决于解释而非直接检索；","用户已有偏好的解释；","多种因果机制都看似合理；","证据不完整或相互矛盾；","决策具有重大的技术、财务或研究后果；","AI系统被期望根据结论自主行动。",{},{"id":984,"data":985,"type":218,"tunes":988},"voWYwXjAQH",{"text":986,"type":218,"tunes":987},"因此，方法论深度应随认识风险而扩展。",{},{},{"id":990,"data":991,"type":42,"tunes":993},"j_hLO6152C",{"text":992,"level":252},"从静态提示到自适应推理策略",{},{"id":995,"data":996,"type":218,"tunes":999},"qS-EwzQq6d",{"text":997,"type":218,"tunes":998},"一旦推广，该框架就不再需要作为一个庞大的提示存在。",{},{},{"id":1001,"data":1002,"type":218,"tunes":1004},"hU3rHmIWSs",{"text":1003},"一个实用的系统可以动态地决定一项任务需要哪些控制措施。",{},{"id":1006,"data":1007,"type":236,"tunes":1009},"syItl8JEDM",{"text":1008,"caption":234,"alignment":235},"简单任务 → 直接回答\n分析性任务 → 结构化证据检查\n高不确定性任务 → 竞争性假设\n框架敏感任务 → 提示不变性\n高后果假设 → 证伪与领域验证",{},{"id":1011,"data":1012,"type":218,"tunes":1014},"Kl_PJTQeZR",{"text":1013},"这将方法论从固定的提示模板转变为自适应的推理策略。",{},{"id":1016,"data":1017,"type":218,"tunes":1019},"x8-TRyTmoY",{"text":1018},"系统并不总是进行最大程度的推理。它根据任务所需的严格程度进行推理。",{},{"id":1021,"data":1022,"type":42,"tunes":1024},"HN_GUpAvOU",{"text":1023,"level":252},"一个实用的通用框架",{},{"id":1026,"data":1027,"type":218,"tunes":1029},"uJr0pPsy7a",{"text":1028},"完整流程可总结如下：",{},{"id":1031,"data":1032,"type":382,"tunes":1049},"hWyrOdZeoR",{"meta":1033,"items":1035,"style":1048},{"counterType":1034},"numeric",[1036,1037,1038,1039,1040,1041,1042,1043,1044,1045,1046,1047],"\u003Cb>规范化问题。\u003C\u002Fb> 在适当情况下，从任务定义中移除嵌入的结论。","\u003Cb>识别证据。\u003C\u002Fb> 将观察与解释分开。","\u003Cb>暴露假设。\u003C\u002Fb> 记录尚未确立的命题。","\u003Cb>生成严肃的替代方案。\u003C\u002Fb> 不要让第一个假设只与薄弱的稻草人竞争。","\u003Cb>保留来源。\u003C\u002Fb> 保持主张可追溯至其来源。","\u003Cb>推导预期。\u003C\u002Fb> 定义每个假设所预测的内容。","\u003Cb>寻找反证。\u003C\u002Fb> 主动寻找首选假设处理不佳的观察结果。","\u003Cb>进行区分性测试。\u003C\u002Fb> 优先选择能区分不同解释的证据。","\u003Cb>测试框架依赖性。\u003C\u002Fb> 当用户的框架可能影响结果时，应用提示不变性。","\u003Cb>应用领域验证器。\u003C\u002Fb> 使用与问题相关的学科特定标准。","\u003Cb>重新校准置信度。\u003C\u002Fb> 允许结论减弱、保持未解决或改变。","\u003Cb>保留可追溯性。\u003C\u002Fb> 使从结论回到证据的路径可检查。","ordered",{},{"id":1051,"data":1052,"type":42,"tunes":1054},"j37GdesnVv",{"text":1053,"level":252},"该框架不声称的内容",{},{"id":1056,"data":1057,"type":382,"tunes":1069},"we5tXuRA71",{"meta":1058,"items":1059,"style":381},{},[1060,1061,1062,1063,1064,1065,1066,1067,1068],"它不保证输出真实。","它不消除幻觉。","它不使LLM成为领域专家。","它不证明稳定的结论是正确的。","它不消除所有推理过程中共有的偏见。","它不替代实验、测量或原始来源。","它不保证所有相关假设都已生成。","它不意味着更多推理总是更好。","它不声称完整框架已被验证为标准化的学术方法论。",{},{"id":1071,"data":1072,"type":218,"tunes":1074},"71CdxmGzXp",{"text":1073},"其主张更为狭窄且更具辩护性：当假设、框架、竞争性解释、反证和领域验证被明确处理，而不是完全留给一次无约束的生成时，分析性AI系统可以在方法论上变得更强。",{},{"id":1076,"data":1077,"type":42,"tunes":1079},"aGiLyuXQib",{"text":1078,"level":252},"四篇文章构成一个推理系统",{},{"id":1081,"data":1082,"type":218,"tunes":1084},"6k1TB7nciF",{"text":1083},"之前的文章现在可以理解为不是独立的提示技术，而是一个推理架构的组成部分。",{},{"id":1086,"data":1087,"type":587,"tunes":1107},"1wFRu7dDCM",{"content":1088,"stretched":43,"withHeadings":14},[1089,1092,1095,1098,1101,1104],[1090,1091],"文章","在框架中的功能",[1093,1094],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">超越提示工程\u003C\u002Fa>","定义了从答案生成到方法论推理的转变。",[1096,1097],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">提示是偏见的一部分\u003C\u002Fa>","将用户的表述视为推理偏见的可能来源。",[1099,1100],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">提示不变性\u003C\u002Fa>","测试结论是否在框架的有意义变化中保持不变。",[1102,1103],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">AI推理的证伪\u003C\u002Fa>","测试假设是否在反证和严肃的替代方案中存活。",[1105,1106],"\u003Cb>从研究协议到通用AI推理框架\u003C\u002Fb>","将控制措施组合成一个与领域无关的核心，并带有领域特定的验证。",{},{"id":1109,"data":1110,"type":218,"tunes":1112},"zTMfJ0udpF",{"text":1111},"它们共同描述了一个转变：",{},{"id":1114,"data":1115,"type":236,"tunes":1117},"zGNT1eU0mo",{"text":1116,"caption":234,"alignment":235},"提示 → 答案\n变为\n问题 → 证据 → 假设 → 挑战 → 验证 → 校准结论",{},{"id":1119,"data":1120,"type":42,"tunes":1122},"SAvL1hIiBl",{"text":1121,"level":252},"核心原则",{},{"id":1124,"data":1125,"type":218,"tunes":1127},"NGbjz4ORtZ",{"text":1126},"该方法论始于一个简单的观察：向一个强大的推理模型提问并不自动意味着模型可用的所有有用推理能力都会被使用。",{},{"id":1129,"data":1130,"type":218,"tunes":1132},"kyuBsra2Ng",{"text":1131},"解决方案不是将最大推理强加到每一次交互中。",{},{"id":1134,"data":1135,"type":218,"tunes":1137},"JWk6uFnvRu",{"text":1136},"解决方案是确定哪些推理控制对任务重要，并将其明确化。",{},{"id":1139,"data":1140,"type":236,"tunes":1142},"LMbGMcE3ij",{"text":1141,"caption":234,"alignment":235},"好的AI方法论不会告诉模型应该得出什么结论。它定义了结论在被接受之前必须经受住什么考验。",{},{"id":1144,"data":1145,"type":218,"tunes":1147},"1cjcDV1NCT",{"text":1146},"这一原则是可迁移的。",{},{"id":1149,"data":1150,"type":218,"tunes":1152},"TnWIF9A5Le",{"text":1151},"在历史研究中，结论必须经受住年代学和史料考证的检验。在调试中，它必须经受住复现和区分性测试的检验。在架构中，它必须经受住需求和运营约束的检验。在战略中，它必须经受住市场证据和失败标准的检验。",{},{"id":1154,"data":1155,"type":218,"tunes":1157},"fTl3br5YpI",{"text":1156},"验证者会变。",{},{"id":1159,"data":1160,"type":218,"tunes":1162},"50R9xCfQF6",{"text":1161},"方法论纪律不变。",{},{"id":1164,"data":1165,"type":236,"tunes":1167},"S9NuFbc7vE",{"text":1166,"caption":234,"alignment":235},"目标不是让模型总是思考更久。而是一个知道答案何时尚未得到充分论证的系统。",{},{"id":1169,"data":1170,"type":1171,"tunes":1172},"tyPyGw9pPq",{},"delimiter",{},{"id":1174,"data":1175,"type":42,"tunes":1177},"w8KrYoGhp7",{"text":1176,"level":252},"研究背景",{},{"id":1179,"data":1180,"type":218,"tunes":1182},"DbP2a862nC",{"text":1181},"本系列中描述的完整框架是一种方法论综合，而非已确立的标准化LLM基准。然而，若干相邻研究方向支持其核心架构前提：当语言模型的问题求解被组织为迭代或多阶段过程而非单次生成时，推理质量可能发生实质性变化。",{},{"id":1184,"data":1185,"type":218,"tunes":1187},"URbMK6Bkg2",{"text":1186},"ReAct将推理与行动及外部观察相结合，并在问答、事实核查和交互式决策任务中展示了优势。思维树探索多条候选推理路径，通过评估和回溯而非立即锁定单一链条。Self-Refine通过在生成、反馈和精炼之间迭代，在多种任务上展现了改进。Reflexion表明语言智能体可以利用先前尝试的言语反馈来改进后续决策，而无需更新模型权重。",{},{"id":1189,"data":1190,"type":218,"tunes":1192},"zlf21zoZ1q",{"text":1191},"这些方法并未实现提示不变性或本文提出的以证伪为导向的框架。它们为更广泛的观点提供了独立证据：推理时程序很重要——模型能力与行使该能力的过程并不等同。",{},{"id":1194,"data":1195,"type":218,"tunes":1197},"LeMaefilCi",{"text":1196},"本系列的贡献在于围绕认识论控制来组织这一洞见：证据分离、假设追踪、提示框架分析、竞争性假设、证伪、来源追溯、领域验证和校准不确定性。",{},{"id":1199,"data":1200,"type":42,"tunes":1202},"Py8YVojnNq",{"text":1201,"level":313},"精选参考文献",{},{"id":1204,"data":1205,"type":382,"tunes":1214},"Cg1YGO8RNK",{"meta":1206,"items":1207,"style":381},{},[1208,1209,1210,1211,1212,1213],"Yao, S. et al. — \u003Ci>ReAct: Synergizing Reasoning and Acting in Language Models\u003C\u002Fi>. ICLR, 2023.","Yao, S. et al. — \u003Ci>Tree of Thoughts: Deliberate Problem Solving with Large Language Models\u003C\u002Fi>. NeurIPS, 2023.","Madaan, A. et al. — \u003Ci>Self-Refine: Iterative Refinement with Self-Feedback\u003C\u002Fi>. NeurIPS, 2023.","Shinn, N. et al. — \u003Ci>Reflexion: Language Agents with Verbal Reinforcement Learning\u003C\u002Fi>. NeurIPS, 2023.","Jhaveri, A. R., GX-Chen, A., Sucholutsky, I. & Choi, E. — \u003Ci>Failing to Falsify: Evaluating and Mitigating Confirmation Bias in Language Models\u003C\u002Fi>. 2026.","Brucks, M. S. & Toubia, O. — \u003Ci>Prompt Architecture Induces Methodological Artifacts in Large Language Models\u003C\u002Fi>. PLOS ONE, 2025.",{},{"id":1216,"data":1217,"type":1171,"tunes":1218},"_6koyQ-al5",{},{},{"id":1220,"data":1221,"type":42,"tunes":1223},"-CU6TfVSPg",{"text":1222,"level":252},"AI推理方法论系列",{},{"id":1225,"data":1226,"type":382,"tunes":1235},"LiQ_OM7vV7",{"meta":1227,"items":1228,"style":1048},{"counterType":1034},[1229,1230,1231,1232,1233,1234],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">\u003Cb>超越提示工程：一种更可靠AI推理的方法论\u003C\u002Fb>\u003C\u002Fa> — 为什么仅靠模型能力是不够的。","\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">\u003Cb>提示是偏见的一部分\u003C\u002Fb>\u003C\u002Fa> — 任务表述如何影响推理环境。","\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">\u003Cb>提示不变性：结论能否经受住提示的考验？\u003C\u002Fb>\u003C\u002Fa> — 测试结论在替代框架下的稳定性。","\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">\u003Cb>AI推理的证伪：从答案到经过检验的假设\u003C\u002Fb>\u003C\u002Fa> — 竞争性假设、反证和区分性测试。","\u003Cb>从研究协议到通用AI推理框架\u003C\u002Fb> — 将该方法论整合为可复用的领域无关推理核心。","同一推理框架在游戏AI中的实际应用——包括游戏助手、自主智能体、补丁敏感知识和实时游戏状态验证——在\u003Ca href=\"https:\u002F\u002Ffigure.rocks\u002Fde\u002Fblog\u002Fwhen-gaming-ai-sounds-right-but-isn-t-the-reasoning-problem-behind-game-assistants-and-agents\">当游戏AI听起来正确但并非如此：游戏助手和智能体背后的推理问题（figure.rocks）\u003C\u002Fa>中进行了探讨。",{},"2.31","为严谨的AI辅助研究而开发的一套方法论，其可推广范围远超研究本身。通过将证据与假设分离、检验相互竞争的假设、控制提示框架、搜寻反证并应用领域特定的验证器，同一推理架构可以改进调试、软件设计、战略、技术分析以及AI辅助决策支持。","\u002Fuploads\u002F2026\u002F09\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework-1789802635691-hhf78v.webp","from-research-protocol-to-a-general-ai-reasoning-framework-1789802635691-hhf78v","PUBLISHED","2026-09-19T01:16:00.000Z","2026-09-19T07:16:24.024Z","2026-09-19T07:46:41.108Z",{"en":1245,"de":1246,"sr":1247,"es":1248,"fr":1249,"it":1250,"ru":1251,"zh":1252},"\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Fde\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Fsr\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Fes\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Ffr\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Fit\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Fru\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework","\u002Fzh\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework",[],{"id":1255,"login":1256,"email":234,"displayName":1257},"20","rooth8233","Aleksandar Stajić",[1259,2091],{"lang":1260,"title":1261,"content":1262,"contentJson":1263,"excerpt":2090},"en","From Research Protocol to a General AI Reasoning Framework","{\"time\":1789803202269,\"blocks\":[{\"id\":\"ESemLsLo8j\",\"type\":\"paragraph\",\"data\":{\"text\":\"The methodology developed across this series began with a research problem: how can an AI model help investigate a complex question without simply reinforcing the assumptions already present in the user's prompt?\"},\"tunes\":{}},{\"id\":\"4Y0K3YWc6x\",\"type\":\"paragraph\",\"data\":{\"text\":\"That problem initially appears to belong to historical or academic research. In reality, it is much broader.\"},\"tunes\":{}},{\"id\":\"Tdh0-PzDMD\",\"type\":\"paragraph\",\"data\":{\"text\":\"Software debugging, architecture decisions, technical diagnosis, security analysis, product strategy and many forms of decision support share the same underlying structure. A problem is presented. Some evidence is available. One or more explanations appear plausible. Assumptions enter the analysis. The system must determine which conclusion is best supported.\"},\"tunes\":{}},{\"id\":\"y1qR6aL83j\",\"type\":\"quote\",\"data\":{\"text\":\"The domain changes. The epistemic problem often does not.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"OSYX72lgMi\",\"type\":\"paragraph\",\"data\":{\"text\":\"This article therefore takes the next step: converting the research protocol developed in the previous articles into a general reasoning framework for AI-assisted analytical work.\"},\"tunes\":{}},{\"id\":\"0nk7KQPJLF\",\"type\":\"paragraph\",\"data\":{\"text\":\"The framework builds directly on four previous layers: \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\\\">Beyond Prompt Engineering: A Methodology for More Reliable AI Reasoning\u003C\u002Fa> defines the overall methodological problem; \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\\\">The Prompt Is Part of the Bias\u003C\u002Fa> examines framing and prompt dependency; \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\\\">Prompt Invariance: Does the Conclusion Survive the Prompt?\u003C\u002Fa> introduces a robustness test against framing; and \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\\\">Falsification for AI Reasoning: From Answers to Tested Hypotheses\u003C\u002Fa> adds systematic disconfirmation and competing hypotheses.\"},\"tunes\":{}},{\"id\":\"8u01PX2pJ_\",\"type\":\"header\",\"data\":{\"text\":\"The Key Generalization\",\"level\":2},\"tunes\":{}},{\"id\":\"KX-3wYIvfo\",\"type\":\"paragraph\",\"data\":{\"text\":\"The generalization is not that every task should be treated like historical research.\"},\"tunes\":{}},{\"id\":\"Qlnajp4NIg\",\"type\":\"paragraph\",\"data\":{\"text\":\"That would be a category error.\"},\"tunes\":{}},{\"id\":\"K_gM2XnB_E\",\"type\":\"paragraph\",\"data\":{\"text\":\"Historical research depends on chronology, provenance, transmission, source criticism and archival completeness. Software debugging depends on logs, configuration, reproducibility, execution state and controlled tests. Architecture depends on requirements, constraints, interfaces, quality attributes and operational trade-offs.\"},\"tunes\":{}},{\"id\":\"EQnEh291Ja\",\"type\":\"paragraph\",\"data\":{\"text\":\"What can be generalized is the reasoning process surrounding those domain-specific forms of evidence.\"},\"tunes\":{}},{\"id\":\"041iVGE35A\",\"type\":\"quote\",\"data\":{\"text\":\"The framework is domain-independent at its core, but never domain-free.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"GsCAv2bn5D\",\"type\":\"paragraph\",\"data\":{\"text\":\"This distinction is fundamental. A universal reasoning methodology cannot replace expertise. It can instead define how evidence, assumptions, hypotheses, contradictions and uncertainty should be handled before domain expertise produces the final judgment.\"},\"tunes\":{}},{\"id\":\"iGG8FYrJ2O\",\"type\":\"header\",\"data\":{\"text\":\"A Domain-Independent Reasoning Core\",\"level\":2},\"tunes\":{}},{\"id\":\"GRuYGmzxLe\",\"type\":\"paragraph\",\"data\":{\"text\":\"Across analytical domains, the same high-level sequence can be applied:\"},\"tunes\":{}},{\"id\":\"lgjHVx7HZI\",\"type\":\"quote\",\"data\":{\"text\":\"Problem → decomposition → evidence → assumptions → competing hypotheses → predictions → counter-evidence → discriminating tests → domain validation → confidence recalibration → conclusion\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"AKTY6_aOdu\",\"type\":\"paragraph\",\"data\":{\"text\":\"The process deliberately prevents the model's first coherent explanation from becoming the final answer by default.\"},\"tunes\":{}},{\"id\":\"eipK0zk3WD\",\"type\":\"paragraph\",\"data\":{\"text\":\"Instead, the first explanation becomes a candidate.\"},\"tunes\":{}},{\"id\":\"yU78N3Sd2e\",\"type\":\"header\",\"data\":{\"text\":\"1. Define the Actual Problem\",\"level\":3},\"tunes\":{}},{\"id\":\"fDuyj6nSv-\",\"type\":\"paragraph\",\"data\":{\"text\":\"Before generating a solution, the system should determine what is actually being asked.\"},\"tunes\":{}},{\"id\":\"sctkg30ATQ\",\"type\":\"paragraph\",\"data\":{\"text\":\"User requests frequently contain both a problem and a proposed diagnosis:\"},\"tunes\":{}},{\"id\":\"HvF-G0rwXA\",\"type\":\"quote\",\"data\":{\"text\":\"Why is nginx causing my API connection failures?\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"mFKfHfGIDV\",\"type\":\"paragraph\",\"data\":{\"text\":\"The actual problem may instead be:\"},\"tunes\":{}},{\"id\":\"CbLWis4-Bu\",\"type\":\"quote\",\"data\":{\"text\":\"What is causing the API connection failures?\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"cMcvbsDaMC\",\"type\":\"paragraph\",\"data\":{\"text\":\"The difference is small linguistically and enormous methodologically.\"},\"tunes\":{}},{\"id\":\"E_kQoBwBzK\",\"type\":\"paragraph\",\"data\":{\"text\":\"The first formulation embeds a causal hypothesis. The second exposes that hypothesis to competition.\"},\"tunes\":{}},{\"id\":\"M86BF9OUO9\",\"type\":\"paragraph\",\"data\":{\"text\":\"This is precisely the framing problem examined in \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\\\">The Prompt Is Part of the Bias\u003C\u002Fa>.\"},\"tunes\":{}},{\"id\":\"8jN_GEWnLD\",\"type\":\"header\",\"data\":{\"text\":\"2. Separate Facts, Assumptions and Unknowns\",\"level\":3},\"tunes\":{}},{\"id\":\"4-45ArOJIC\",\"type\":\"paragraph\",\"data\":{\"text\":\"A reliable reasoning process must prevent observations and interpretations from silently merging.\"},\"tunes\":{}},{\"id\":\"gn7a0JRFIn\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"\u003Cb>Fact:\u003C\u002Fb> directly supported by available evidence.\",\"\u003Cb>Interpretation:\u003C\u002Fb> an explanation inferred from facts.\",\"\u003Cb>Assumption:\u003C\u002Fb> a proposition required by the current reasoning path but not independently established.\",\"\u003Cb>Unknown:\u003C\u002Fb> information necessary for stronger discrimination but currently unavailable.\"]},\"tunes\":{}},{\"id\":\"7gqnnnAyNH\",\"type\":\"paragraph\",\"data\":{\"text\":\"This classification may appear elementary, but it solves a common failure mode in AI-generated analysis: assumptions become embedded in fluent prose and later reappear as if they had already been established.\"},\"tunes\":{}},{\"id\":\"wakPN0U08S\",\"type\":\"quote\",\"data\":{\"text\":\"Traceability begins by preventing inference from disguising itself as evidence.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"UPatPWjaiP\",\"type\":\"header\",\"data\":{\"text\":\"3. Generate Competing Explanations\",\"level\":3},\"tunes\":{}},{\"id\":\"zuRICKxeRX\",\"type\":\"paragraph\",\"data\":{\"text\":\"A candidate explanation should not be evaluated in isolation when credible alternatives exist.\"},\"tunes\":{}},{\"id\":\"AvfAw-zrKJ\",\"type\":\"paragraph\",\"data\":{\"text\":\"The system should therefore generate several plausible hypotheses before committing to one.\"},\"tunes\":{}},{\"id\":\"5JVkk_l5m9\",\"type\":\"paragraph\",\"data\":{\"text\":\"The purpose is not artificial brainstorming. Alternatives should represent genuinely different mechanisms capable of explaining the observations.\"},\"tunes\":{}},{\"id\":\"6ZI39jnt9v\",\"type\":\"quote\",\"data\":{\"text\":\"A preferred hypothesis has not won if no serious competitor was allowed to enter the test.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"WGpU3HSfMr\",\"type\":\"paragraph\",\"data\":{\"text\":\"This concept is central to \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\\\">Falsification for AI Reasoning\u003C\u002Fa>, where hypotheses are evaluated through expected observations, counter-evidence and discriminating tests rather than accumulated supportive prose.\"},\"tunes\":{}},{\"id\":\"mUW1lRBq9a\",\"type\":\"header\",\"data\":{\"text\":\"4. Preserve Evidence Provenance\",\"level\":3},\"tunes\":{}},{\"id\":\"m2TM2yNjau\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence should remain traceable to its origin.\"},\"tunes\":{}},{\"id\":\"3w2fzBzVH4\",\"type\":\"quote\",\"data\":{\"text\":\"Conclusion → interpretation → observation → source\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"rkoICmxwZT\",\"type\":\"paragraph\",\"data\":{\"text\":\"A log line, an official technical specification, an experiment, a primary historical document, an expert interpretation and an AI-generated summary do not have identical evidential status.\"},\"tunes\":{}},{\"id\":\"8ut4cK6e27\",\"type\":\"paragraph\",\"data\":{\"text\":\"The framework therefore does not merely store information. It should preserve enough metadata to know where a claim came from and how far the final conclusion is from the original observation.\"},\"tunes\":{}},{\"id\":\"hZ6BZYfcIo\",\"type\":\"header\",\"data\":{\"text\":\"5. Derive Predictions Before Explaining Results\",\"level\":3},\"tunes\":{}},{\"id\":\"aisk7gG2yj\",\"type\":\"paragraph\",\"data\":{\"text\":\"Once hypotheses exist, each should generate expectations.\"},\"tunes\":{}},{\"id\":\"IBsZFlw9pf\",\"type\":\"paragraph\",\"data\":{\"text\":\"If the hypothesis is correct, what should we observe?\"},\"tunes\":{}},{\"id\":\"Kykmkn9Dtx\",\"type\":\"paragraph\",\"data\":{\"text\":\"If an alternative is correct, what should be different?\"},\"tunes\":{}},{\"id\":\"iS8nPIkYVU\",\"type\":\"paragraph\",\"data\":{\"text\":\"Deriving expectations before interpreting all available evidence helps prevent every observation from being retrofitted into the preferred narrative.\"},\"tunes\":{}},{\"id\":\"1rrToFbYF5\",\"type\":\"header\",\"data\":{\"text\":\"6. Search for Counter-Evidence\",\"level\":3},\"tunes\":{}},{\"id\":\"NSjS51xlU6\",\"type\":\"paragraph\",\"data\":{\"text\":\"A generative model is naturally effective at constructing coherent support for a plausible idea. This makes disconfirmation particularly important.\"},\"tunes\":{}},{\"id\":\"jPSoTMS13f\",\"type\":\"paragraph\",\"data\":{\"text\":\"The system should explicitly ask:\"},\"tunes\":{}},{\"id\":\"DfULqhWZpB\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"What observation would substantially weaken this explanation?\",\"What evidence does the hypothesis explain poorly?\",\"Which alternative explains the same observations with fewer assumptions?\",\"Which expected observations are missing?\",\"Could the apparent support be explained by another mechanism?\"]},\"tunes\":{}},{\"id\":\"rM5aFAv4Ch\",\"type\":\"paragraph\",\"data\":{\"text\":\"The principle is developed in detail in \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\\\">Falsification for AI Reasoning: From Answers to Tested Hypotheses\u003C\u002Fa>.\"},\"tunes\":{}},{\"id\":\"sWwUSTmVGW\",\"type\":\"header\",\"data\":{\"text\":\"7. Test Dependence on the Prompt\",\"level\":3},\"tunes\":{}},{\"id\":\"wvQwo5fB4W\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence testing alone does not reveal whether the model's analysis remains excessively dependent on the user's formulation.\"},\"tunes\":{}},{\"id\":\"Z1AJX_Zx0J\",\"type\":\"paragraph\",\"data\":{\"text\":\"For sufficiently important problems, the conclusion can therefore be retested under alternative legitimate framings.\"},\"tunes\":{}},{\"id\":\"R9CxBqiZy5\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"\u003Cb>Original:\u003C\u002Fb> preserve the user's formulation.\",\"\u003Cb>Blind:\u003C\u002Fb> remove the preferred conclusion.\",\"\u003Cb>Inverted:\u003C\u002Fb> foreground the strongest alternative.\",\"\u003Cb>Adversarial:\u003C\u002Fb> deliberately search for the strongest evidence-based challenge.\"]},\"tunes\":{}},{\"id\":\"Z6ReUs6ThD\",\"type\":\"paragraph\",\"data\":{\"text\":\"This is the Prompt Invariance method introduced in \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\\\">Prompt Invariance: Does the Conclusion Survive the Prompt?\u003C\u002Fa>.\"},\"tunes\":{}},{\"id\":\"i72azJfoHd\",\"type\":\"paragraph\",\"data\":{\"text\":\"Its function is not to prove that a stable answer is true. Its function is to expose conclusions that change primarily because the framing changed.\"},\"tunes\":{}},{\"id\":\"PQJr3I5KR6\",\"type\":\"header\",\"data\":{\"text\":\"8. Apply Domain-Specific Validators\",\"level\":3},\"tunes\":{}},{\"id\":\"30EnJRKLgj\",\"type\":\"paragraph\",\"data\":{\"text\":\"This is where the general framework deliberately stops being universal.\"},\"tunes\":{}},{\"id\":\"HfkiK88IcE\",\"type\":\"paragraph\",\"data\":{\"text\":\"Every domain requires its own rules for determining whether an explanation is actually credible.\"},\"tunes\":{}},{\"id\":\"zinQCXwjug\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Domain\",\"Typical Validators\"],[\"Historical research\",\"Chronology, provenance, geographical plausibility, transmission channels, primary sources, historiography\"],[\"Software debugging\",\"Logs, reproduction, runtime state, configuration, controlled changes, error correlation\"],[\"Software architecture\",\"Requirements, constraints, scalability, maintainability, security, operability, cost\"],[\"Security analysis\",\"Telemetry, attack path, permissions, observable indicators, reproducibility, threat model\"],[\"Product strategy\",\"Customer evidence, pricing behaviour, conversion, alternatives, market constraints, failure criteria\"],[\"Project management\",\"Scope, dependencies, resources, risks, acceptance criteria, milestones, stakeholder constraints\"],[\"Scientific analysis\",\"Experimental design, measurement quality, controls, reproducibility, statistical evidence, alternative explanations\"]]},\"tunes\":{}},{\"id\":\"uc0RyUCFOo\",\"type\":\"paragraph\",\"data\":{\"text\":\"The framework controls how claims are processed. Domain validators determine whether those claims survive contact with reality.\"},\"tunes\":{}},{\"id\":\"lAvhdCVN32\",\"type\":\"header\",\"data\":{\"text\":\"9. Recalibrate Confidence\",\"level\":3},\"tunes\":{}},{\"id\":\"h5PWNd0iA4\",\"type\":\"paragraph\",\"data\":{\"text\":\"The final conclusion should not preserve the confidence level of the initial answer automatically.\"},\"tunes\":{}},{\"id\":\"PcokJnli0n\",\"type\":\"paragraph\",\"data\":{\"text\":\"Confidence should be recalculated after alternatives, counter-evidence, prompt variation and domain validation have been considered.\"},\"tunes\":{}},{\"id\":\"4-JaOGRNdJ\",\"type\":\"paragraph\",\"data\":{\"text\":\"A useful output can therefore remain uncertain.\"},\"tunes\":{}},{\"id\":\"I6F7TJy23-\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"strongly supported;\",\"provisionally supported;\",\"weakly supported;\",\"underdetermined;\",\"substantially contradicted;\",\"not testable with current evidence.\"]},\"tunes\":{}},{\"id\":\"ozNw_amgCI\",\"type\":\"quote\",\"data\":{\"text\":\"Uncertainty is not a failure of reasoning when the evidence itself is uncertain.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"LlyeIOmVPC\",\"type\":\"header\",\"data\":{\"text\":\"The Same Framework Across Different Domains\",\"level\":2},\"tunes\":{}},{\"id\":\"dmsjTGPTp7\",\"type\":\"paragraph\",\"data\":{\"text\":\"The easiest way to understand the general framework is to observe how the same reasoning architecture behaves when the domain changes.\"},\"tunes\":{}},{\"id\":\"cU0_n6SCY3\",\"type\":\"header\",\"data\":{\"text\":\"Historical Research\",\"level\":3},\"tunes\":{}},{\"id\":\"es1dtKJQlZ\",\"type\":\"paragraph\",\"data\":{\"text\":\"Problem: determine whether similarities between two intellectual traditions indicate historical transmission.\"},\"tunes\":{}},{\"id\":\"Nzl1cThLXK\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Separate documented similarities from interpretive parallels.\",\"Compare direct transmission, indirect inheritance, convergence and retrospective interpretation.\",\"Test chronology and geographical contact.\",\"Search for documentary or terminological traces.\",\"Identify evidence expected under direct transmission.\",\"Search for earlier independent examples that weaken that hypothesis.\",\"Reframe the question without the preferred genealogy.\",\"Conclude only at the confidence level supported by the surviving evidence.\"]},\"tunes\":{}},{\"id\":\"RDf52TTR1X\",\"type\":\"header\",\"data\":{\"text\":\"Software Debugging\",\"level\":3},\"tunes\":{}},{\"id\":\"V_aQmUWc8S\",\"type\":\"paragraph\",\"data\":{\"text\":\"Problem: determine why an API connection fails intermittently.\"},\"tunes\":{}},{\"id\":\"YOUlOo_WTe\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Separate observed failures from the developer's suspected cause.\",\"Generate proxy, application, database, network and client hypotheses.\",\"Determine what logs and runtime behaviour each hypothesis predicts.\",\"Perform tests capable of distinguishing among them.\",\"Attempt to reproduce the failure while eliminating layers.\",\"Reject explanations contradicted by controlled observations.\",\"Identify the first confirmed failure mechanism rather than the most persuasive narrative.\"]},\"tunes\":{}},{\"id\":\"tG7q0LkXG8\",\"type\":\"header\",\"data\":{\"text\":\"Software Architecture\",\"level\":3},\"tunes\":{}},{\"id\":\"Siq7TARR4r\",\"type\":\"paragraph\",\"data\":{\"text\":\"Problem: choose an architecture for a new platform.\"},\"tunes\":{}},{\"id\":\"PYBaGFi2Ol\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Separate requirements from implementation preferences.\",\"Compare modular monolith, services and hybrid alternatives.\",\"Test each option against scalability, team structure, deployment, operational complexity and cost.\",\"Identify which requirement genuinely requires architectural complexity.\",\"Search for simpler designs capable of satisfying the same constraints.\",\"Treat technology preference as an assumption rather than a requirement.\"]},\"tunes\":{}},{\"id\":\"-kikosSmX1\",\"type\":\"header\",\"data\":{\"text\":\"Product Strategy\",\"level\":3},\"tunes\":{}},{\"id\":\"RW8gaj5yQo\",\"type\":\"paragraph\",\"data\":{\"text\":\"Problem: determine whether a product capability should be commercialized.\"},\"tunes\":{}},{\"id\":\"JZBAxRuw9D\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Separate technical possibility from demonstrated demand.\",\"Define alternative customer problems and alternative solutions.\",\"Identify observable market behaviour predicted by the business thesis.\",\"Define failure criteria before interpreting results.\",\"Search for evidence that customers solve the problem differently.\",\"Distinguish interest, willingness to test and willingness to pay.\"]},\"tunes\":{}},{\"id\":\"NfuFu8q13_\",\"type\":\"header\",\"data\":{\"text\":\"Project and Delivery Analysis\",\"level\":3},\"tunes\":{}},{\"id\":\"GaM4feM9nZ\",\"type\":\"paragraph\",\"data\":{\"text\":\"Problem: determine why a project is missing planned outcomes.\"},\"tunes\":{}},{\"id\":\"-G7BLfDmdU\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Separate symptoms such as delay from their assumed causes.\",\"Compare scope, dependencies, capacity, requirements, governance and technical risk.\",\"Trace evidence through plans, decisions, changes and acceptance criteria.\",\"Test whether corrective action addresses the actual mechanism rather than the visible symptom.\",\"Reassess the project hypothesis as new delivery evidence appears.\"]},\"tunes\":{}},{\"id\":\"5iAhrZD3wL\",\"type\":\"header\",\"data\":{\"text\":\"Why This Is More Than Prompt Engineering\",\"level\":2},\"tunes\":{}},{\"id\":\"dm4MhgnM3L\",\"type\":\"paragraph\",\"data\":{\"text\":\"Prompt engineering changes the instruction presented to a model.\"},\"tunes\":{}},{\"id\":\"87q5CSIEmb\",\"type\":\"paragraph\",\"data\":{\"text\":\"A reasoning framework changes the process through which an answer is allowed to become a conclusion.\"},\"tunes\":{}},{\"id\":\"7C_ZI5IHle\",\"type\":\"quote\",\"data\":{\"text\":\"Prompt engineering optimizes an invocation. A reasoning framework governs an inference process.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"SFq4pwo4Gx\",\"type\":\"paragraph\",\"data\":{\"text\":\"This difference becomes important as AI systems move from chat interfaces toward agents, RAG pipelines, automated research, development tools and decision-support systems.\"},\"tunes\":{}},{\"id\":\"WItU2n6hS2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A single carefully engineered prompt may improve one model call. A reasoning framework can define how multiple calls, retrieved evidence, tool results and validation stages interact before the system commits to an answer.\"},\"tunes\":{}},{\"id\":\"99vYubKOKi\",\"type\":\"header\",\"data\":{\"text\":\"The Framework Is Related to — but Different from — Existing LLM Reasoning Methods\",\"level\":2},\"tunes\":{}},{\"id\":\"y8Zf5V4OPJ\",\"type\":\"paragraph\",\"data\":{\"text\":\"The broader research landscape already contains several important approaches showing that LLM performance can improve when inference is structured as a process rather than a single generation.\"},\"tunes\":{}},{\"id\":\"gEIVPfVGEd\",\"type\":\"paragraph\",\"data\":{\"text\":\"\u003Cb>ReAct\u003C\u002Fb> combines reasoning with actions and observations from an external environment. Instead of relying exclusively on internal model knowledge, the model can act, observe new information and update its next step.\"},\"tunes\":{}},{\"id\":\"kKS8VaCYmR\",\"type\":\"paragraph\",\"data\":{\"text\":\"\u003Cb>Self-Refine\u003C\u002Fb> uses iterative generation, feedback and refinement, demonstrating that an initial LLM output can often be improved through explicit self-feedback without additional model training.\"},\"tunes\":{}},{\"id\":\"Puh4hZgRfK\",\"type\":\"paragraph\",\"data\":{\"text\":\"\u003Cb>Reflexion\u003C\u002Fb> uses verbal feedback and episodic memory to allow language agents to learn from previous attempts and modify later behaviour.\"},\"tunes\":{}},{\"id\":\"nAB7uNC8AX\",\"type\":\"paragraph\",\"data\":{\"text\":\"\u003Cb>Tree of Thoughts\u003C\u002Fb> explores multiple reasoning paths rather than committing immediately to one left-to-right chain, allowing search, evaluation and backtracking over candidate approaches.\"},\"tunes\":{}},{\"id\":\"B5Q7GKEqmh\",\"type\":\"paragraph\",\"data\":{\"text\":\"These methods differ substantially in purpose and implementation, but they establish an important general point: inference-time structure can materially change model performance.\"},\"tunes\":{}},{\"id\":\"Hj82H7bI5T\",\"type\":\"paragraph\",\"data\":{\"text\":\"The framework proposed in this series addresses a different primary objective.\"},\"tunes\":{}},{\"id\":\"q8bQMqN2Ir\",\"type\":\"quote\",\"data\":{\"text\":\"The goal is not simply to make the model search longer, reflect more or generate a better answer. The goal is to make the path from evidence to conclusion more resistant to framing, confirmation and unsupported assumptions.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"d1bIpPAnpB\",\"type\":\"paragraph\",\"data\":{\"text\":\"Prompt Invariance, falsification-oriented testing and domain-specific validation therefore serve as epistemic controls rather than generic inference-expansion techniques.\"},\"tunes\":{}},{\"id\":\"eXXYr_pAms\",\"type\":\"header\",\"data\":{\"text\":\"A Layered View of AI Reasoning Quality\",\"level\":2},\"tunes\":{}},{\"id\":\"DqLa7MmX58\",\"type\":\"paragraph\",\"data\":{\"text\":\"The resulting system can be understood as several layers.\"},\"tunes\":{}},{\"id\":\"0NzL-zFxBY\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Layer\",\"Question\"],[\"Task understanding\",\"What problem are we actually solving?\"],[\"Evidence\",\"What do we actually know?\"],[\"Assumptions\",\"What are we currently taking for granted?\"],[\"Hypotheses\",\"What plausible mechanisms could explain the evidence?\"],[\"Falsification\",\"What evidence could weaken each explanation?\"],[\"Prompt Invariance\",\"Does the conclusion depend excessively on framing?\"],[\"Domain validation\",\"Does the explanation satisfy the rules of the relevant discipline?\"],[\"Confidence calibration\",\"How strongly should the surviving conclusion be trusted?\"]]},\"tunes\":{}},{\"id\":\"xm0-PFgN-i\",\"type\":\"paragraph\",\"data\":{\"text\":\"An answer can fail at any layer.\"},\"tunes\":{}},{\"id\":\"eCszY4dwX1\",\"type\":\"paragraph\",\"data\":{\"text\":\"It can answer the wrong problem correctly. It can reason correctly from false evidence. It can retrieve correct facts but choose the wrong causal explanation. It can produce a strong explanation that disappears when the prompt is reframed. Or it can survive all general reasoning checks and still violate a domain-specific constraint.\"},\"tunes\":{}},{\"id\":\"hSmlX6vujv\",\"type\":\"quote\",\"data\":{\"text\":\"Correctness is not one property. It is the result of several dependencies surviving simultaneously.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"EydNhsxtzy\",\"type\":\"header\",\"data\":{\"text\":\"Reasoning Quality Versus Model Capability\",\"level\":2},\"tunes\":{}},{\"id\":\"s4Km9d-Rxy\",\"type\":\"paragraph\",\"data\":{\"text\":\"This leads back to one of the central ideas of the series.\"},\"tunes\":{}},{\"id\":\"39JzRxeSgF\",\"type\":\"paragraph\",\"data\":{\"text\":\"A stronger model does not automatically imply that every invocation will use its capabilities in the most rigorous possible way.\"},\"tunes\":{}},{\"id\":\"KEw35qMKGc\",\"type\":\"paragraph\",\"data\":{\"text\":\"A model may possess the capability to generate alternatives, inspect logs, search sources, challenge assumptions and revise a conclusion. Whether all of those operations actually occur depends on the task, prompting, available tools, context and surrounding system architecture.\"},\"tunes\":{}},{\"id\":\"qijWfW7iDp\",\"type\":\"paragraph\",\"data\":{\"text\":\"The framework therefore separates two questions:\"},\"tunes\":{}},{\"id\":\"9TvQxkmlHA\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"\u003Cb>Capability:\u003C\u002Fb> what can the model do?\",\"\u003Cb>Methodological discipline:\u003C\u002Fb> which of those capabilities must be exercised before the system accepts a conclusion?\"]},\"tunes\":{}},{\"id\":\"rmrusg3t3h\",\"type\":\"paragraph\",\"data\":{\"text\":\"The distinction was the starting point of \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\\\">Beyond Prompt Engineering\u003C\u002Fa> and becomes even more important once the methodology is generalized beyond research.\"},\"tunes\":{}},{\"id\":\"W321yYXV97\",\"type\":\"header\",\"data\":{\"text\":\"Not Every Task Needs the Full Framework\",\"level\":2},\"tunes\":{}},{\"id\":\"QwEqB6r20x\",\"type\":\"paragraph\",\"data\":{\"text\":\"The framework should not become compulsory overhead for every AI interaction.\"},\"tunes\":{}},{\"id\":\"XxxtP1e8oN\",\"type\":\"paragraph\",\"data\":{\"text\":\"Many tasks have low epistemic complexity:\"},\"tunes\":{}},{\"id\":\"yJw3vKdi9z\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"format this JSON;\",\"translate this sentence;\",\"convert these units;\",\"rewrite this message;\",\"extract these fields;\",\"summarize this supplied text.\"]},\"tunes\":{}},{\"id\":\"c_3UzzXOlF\",\"type\":\"paragraph\",\"data\":{\"text\":\"Running four prompt variants, three competing hypotheses and a falsification stage for such tasks would add cost without proportional value.\"},\"tunes\":{}},{\"id\":\"89oDA_RK5n\",\"type\":\"paragraph\",\"data\":{\"text\":\"The full method becomes more valuable when:\"},\"tunes\":{}},{\"id\":\"T9axudVitU\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"the answer depends on interpretation rather than direct retrieval;\",\"the user already has a preferred explanation;\",\"multiple causal mechanisms are plausible;\",\"the evidence is incomplete or contradictory;\",\"the decision has significant technical, financial or research consequences;\",\"the AI system is expected to act autonomously on the conclusion.\"]},\"tunes\":{}},{\"id\":\"voWYwXjAQH\",\"type\":\"paragraph\",\"data\":{\"text\":\"Methodological depth should therefore scale with epistemic risk.\"},\"tunes\":{}},{\"id\":\"j_hLO6152C\",\"type\":\"header\",\"data\":{\"text\":\"From Static Prompt to Adaptive Reasoning Policy\",\"level\":2},\"tunes\":{}},{\"id\":\"qS-EwzQq6d\",\"type\":\"paragraph\",\"data\":{\"text\":\"Once generalized, the framework no longer needs to exist as one enormous prompt.\"},\"tunes\":{}},{\"id\":\"hU3rHmIWSs\",\"type\":\"paragraph\",\"data\":{\"text\":\"A practical system can decide dynamically which controls a task requires.\"},\"tunes\":{}},{\"id\":\"syItl8JEDM\",\"type\":\"quote\",\"data\":{\"text\":\"Simple task → direct answer\\nAnalytical task → structured evidence check\\nHigh-uncertainty task → competing hypotheses\\nFraming-sensitive task → Prompt Invariance\\nHigh-consequence hypothesis → falsification and domain validation\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"Kl_PJTQeZR\",\"type\":\"paragraph\",\"data\":{\"text\":\"This transforms the methodology from a fixed prompt template into an adaptive reasoning policy.\"},\"tunes\":{}},{\"id\":\"x8-TRyTmoY\",\"type\":\"paragraph\",\"data\":{\"text\":\"The system does not always reason maximally. It reasons as rigorously as the task requires.\"},\"tunes\":{}},{\"id\":\"HN_GUpAvOU\",\"type\":\"header\",\"data\":{\"text\":\"A Practical General Framework\",\"level\":2},\"tunes\":{}},{\"id\":\"uJr0pPsy7a\",\"type\":\"paragraph\",\"data\":{\"text\":\"The complete process can be summarized as follows:\"},\"tunes\":{}},{\"id\":\"hWyrOdZeoR\",\"type\":\"list\",\"data\":{\"style\":\"ordered\",\"meta\":{\"counterType\":\"numeric\"},\"items\":[\"\u003Cb>Normalize the problem.\u003C\u002Fb> Remove embedded conclusions from the task definition where appropriate.\",\"\u003Cb>Identify evidence.\u003C\u002Fb> Separate observations from interpretations.\",\"\u003Cb>Expose assumptions.\u003C\u002Fb> Record propositions that have not yet been established.\",\"\u003Cb>Generate serious alternatives.\u003C\u002Fb> Do not allow the first hypothesis to compete only against weak strawmen.\",\"\u003Cb>Preserve provenance.\u003C\u002Fb> Keep claims traceable to their sources.\",\"\u003Cb>Derive expectations.\u003C\u002Fb> Define what each hypothesis predicts.\",\"\u003Cb>Search for counter-evidence.\u003C\u002Fb> Actively look for observations the preferred hypothesis handles poorly.\",\"\u003Cb>Run discriminating tests.\u003C\u002Fb> Prefer evidence that differentiates among explanations.\",\"\u003Cb>Test framing dependence.\u003C\u002Fb> Apply Prompt Invariance when the user's framing may influence the result.\",\"\u003Cb>Apply domain validators.\u003C\u002Fb> Use the discipline-specific criteria relevant to the problem.\",\"\u003Cb>Recalibrate confidence.\u003C\u002Fb> Allow the conclusion to weaken, remain unresolved or change.\",\"\u003Cb>Preserve traceability.\u003C\u002Fb> Make the path from conclusion back to evidence inspectable.\"]},\"tunes\":{}},{\"id\":\"j37GdesnVv\",\"type\":\"header\",\"data\":{\"text\":\"What the Framework Does Not Claim\",\"level\":2},\"tunes\":{}},{\"id\":\"we5tXuRA71\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"It does not guarantee truthful output.\",\"It does not eliminate hallucinations.\",\"It does not make an LLM a domain expert.\",\"It does not prove that stable conclusions are correct.\",\"It does not eliminate biases shared across all reasoning passes.\",\"It does not replace experiments, measurements or primary sources.\",\"It does not guarantee that all relevant hypotheses have been generated.\",\"It does not imply that more reasoning is always better.\",\"It does not claim that the complete framework has already been validated as a standardized academic methodology.\"]},\"tunes\":{}},{\"id\":\"71CdxmGzXp\",\"type\":\"paragraph\",\"data\":{\"text\":\"Its claim is narrower and more defensible: analytical AI systems can be made methodologically stronger when assumptions, framing, competing explanations, counter-evidence and domain validation are handled explicitly rather than left entirely to one unconstrained generation.\"},\"tunes\":{}},{\"id\":\"aGiLyuXQib\",\"type\":\"header\",\"data\":{\"text\":\"The Four Articles Become One Reasoning System\",\"level\":2},\"tunes\":{}},{\"id\":\"6k1TB7nciF\",\"type\":\"paragraph\",\"data\":{\"text\":\"The previous articles can now be understood not as separate prompting techniques but as components of one reasoning architecture.\"},\"tunes\":{}},{\"id\":\"1wFRu7dDCM\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Article\",\"Function in the Framework\"],[\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\\\">Beyond Prompt Engineering\u003C\u002Fa>\",\"Defines the shift from answer generation to methodological reasoning.\"],[\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\\\">The Prompt Is Part of the Bias\u003C\u002Fa>\",\"Treats the user's formulation as a possible source of reasoning bias.\"],[\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\\\">Prompt Invariance\u003C\u002Fa>\",\"Tests whether the conclusion survives meaningful changes in framing.\"],[\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\\\">Falsification for AI Reasoning\u003C\u002Fa>\",\"Tests whether hypotheses survive disconfirming evidence and serious alternatives.\"],[\"\u003Cb>From Research Protocol to a General AI Reasoning Framework\u003C\u002Fb>\",\"Combines the controls into a domain-independent core with domain-specific validation.\"]]},\"tunes\":{}},{\"id\":\"zTMfJ0udpF\",\"type\":\"paragraph\",\"data\":{\"text\":\"Together they describe a transition:\"},\"tunes\":{}},{\"id\":\"zGNT1eU0mo\",\"type\":\"quote\",\"data\":{\"text\":\"Prompt → answer\\nbecomes\\nProblem → evidence → hypotheses → challenge → validation → calibrated conclusion\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"SAvL1hIiBl\",\"type\":\"header\",\"data\":{\"text\":\"The Central Principle\",\"level\":2},\"tunes\":{}},{\"id\":\"NGbjz4ORtZ\",\"type\":\"paragraph\",\"data\":{\"text\":\"The methodology began with a simple observation: asking a powerful reasoning model a question does not automatically mean that every useful reasoning capability available to the model will be used.\"},\"tunes\":{}},{\"id\":\"kyuBsra2Ng\",\"type\":\"paragraph\",\"data\":{\"text\":\"The solution is not to force maximal reasoning into every interaction.\"},\"tunes\":{}},{\"id\":\"JWk6uFnvRu\",\"type\":\"paragraph\",\"data\":{\"text\":\"The solution is to identify which reasoning controls matter for the task and make them explicit.\"},\"tunes\":{}},{\"id\":\"LMbGMcE3ij\",\"type\":\"quote\",\"data\":{\"text\":\"A good AI methodology does not tell the model what conclusion to reach. It defines what a conclusion must survive before it is accepted.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"1cjcDV1NCT\",\"type\":\"paragraph\",\"data\":{\"text\":\"That principle is transferable.\"},\"tunes\":{}},{\"id\":\"TnWIF9A5Le\",\"type\":\"paragraph\",\"data\":{\"text\":\"In historical research, the conclusion must survive chronology and source criticism. In debugging, it must survive reproduction and discriminating tests. In architecture, it must survive requirements and operational constraints. In strategy, it must survive market evidence and failure criteria.\"},\"tunes\":{}},{\"id\":\"fTl3br5YpI\",\"type\":\"paragraph\",\"data\":{\"text\":\"The validators change.\"},\"tunes\":{}},{\"id\":\"50R9xCfQF6\",\"type\":\"paragraph\",\"data\":{\"text\":\"The methodological discipline remains.\"},\"tunes\":{}},{\"id\":\"S9NuFbc7vE\",\"type\":\"quote\",\"data\":{\"text\":\"The objective is not a model that always thinks longer. It is a system that knows when an answer is not yet justified.\",\"caption\":\"\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"tyPyGw9pPq\",\"type\":\"delimiter\",\"data\":{},\"tunes\":{}},{\"id\":\"w8KrYoGhp7\",\"type\":\"header\",\"data\":{\"text\":\"Research Context\",\"level\":2},\"tunes\":{}},{\"id\":\"DbP2a862nC\",\"type\":\"paragraph\",\"data\":{\"text\":\"The complete framework described in this series is a methodological synthesis rather than an established standardized LLM benchmark. Several adjacent lines of research nevertheless support its central architectural premise: inference quality can change substantially when language-model problem solving is organized as an iterative or multi-stage process rather than a single generation.\"},\"tunes\":{}},{\"id\":\"URbMK6Bkg2\",\"type\":\"paragraph\",\"data\":{\"text\":\"ReAct combines reasoning with actions and external observations and demonstrated benefits across question answering, fact verification and interactive decision tasks. Tree of Thoughts explores multiple candidate reasoning paths with evaluation and backtracking rather than committing immediately to one chain. Self-Refine showed improvements across diverse tasks by iterating between generation, feedback and refinement. Reflexion demonstrated that language agents can use verbal feedback from previous attempts to improve later decisions without updating model weights.\"},\"tunes\":{}},{\"id\":\"zlf21zoZ1q\",\"type\":\"paragraph\",\"data\":{\"text\":\"These approaches do not implement Prompt Invariance or the falsification-oriented framework proposed here. They provide independent evidence for the broader idea that inference-time procedure matters: model capability and the process through which that capability is exercised are not identical.\"},\"tunes\":{}},{\"id\":\"LeMaefilCi\",\"type\":\"paragraph\",\"data\":{\"text\":\"The contribution of this series is to organize that insight around epistemic controls: evidence separation, assumption tracking, prompt-framing analysis, competing hypotheses, falsification, provenance, domain validation and calibrated uncertainty.\"},\"tunes\":{}},{\"id\":\"Py8YVojnNq\",\"type\":\"header\",\"data\":{\"text\":\"Selected References\",\"level\":3},\"tunes\":{}},{\"id\":\"Cg1YGO8RNK\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Yao, S. et al. — \u003Ci>ReAct: Synergizing Reasoning and Acting in Language Models\u003C\u002Fi>. ICLR, 2023.\",\"Yao, S. et al. — \u003Ci>Tree of Thoughts: Deliberate Problem Solving with Large Language Models\u003C\u002Fi>. NeurIPS, 2023.\",\"Madaan, A. et al. — \u003Ci>Self-Refine: Iterative Refinement with Self-Feedback\u003C\u002Fi>. NeurIPS, 2023.\",\"Shinn, N. et al. — \u003Ci>Reflexion: Language Agents with Verbal Reinforcement Learning\u003C\u002Fi>. NeurIPS, 2023.\",\"Jhaveri, A. R., GX-Chen, A., Sucholutsky, I. & Choi, E. — \u003Ci>Failing to Falsify: Evaluating and Mitigating Confirmation Bias in Language Models\u003C\u002Fi>. 2026.\",\"Brucks, M. S. & Toubia, O. — \u003Ci>Prompt Architecture Induces Methodological Artifacts in Large Language Models\u003C\u002Fi>. PLOS ONE, 2025.\"]},\"tunes\":{}},{\"id\":\"_6koyQ-al5\",\"type\":\"delimiter\",\"data\":{},\"tunes\":{}},{\"id\":\"-CU6TfVSPg\",\"type\":\"header\",\"data\":{\"text\":\"The AI Reasoning Methodology Series\",\"level\":2},\"tunes\":{}},{\"id\":\"LiQ_OM7vV7\",\"type\":\"list\",\"data\":{\"style\":\"ordered\",\"meta\":{\"counterType\":\"numeric\"},\"items\":[\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\\\">\u003Cb>Beyond Prompt Engineering: A Methodology for More Reliable AI Reasoning\u003C\u002Fb>\u003C\u002Fa> — why model capability alone is not enough.\",\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\\\">\u003Cb>The Prompt Is Part of the Bias\u003C\u002Fb>\u003C\u002Fa> — how task formulation can influence the reasoning environment.\",\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\\\">\u003Cb>Prompt Invariance: Does the Conclusion Survive the Prompt?\u003C\u002Fb>\u003C\u002Fa> — testing conclusion stability across alternative framings.\",\"\u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\\\">\u003Cb>Falsification for AI Reasoning: From Answers to Tested Hypotheses\u003C\u002Fb>\u003C\u002Fa> — competing hypotheses, counter-evidence and discriminating tests.\",\"\u003Cb>From Research Protocol to a General AI Reasoning Framework\u003C\u002Fb> — integrating the methodology into a reusable domain-independent reasoning core.\",\"A practical application of the same reasoning framework in gaming AI — including game assistants, autonomous agents, patch-sensitive knowledge and live game-state validation — is explored in \u003Ca href=\\\"https:\u002F\u002Ffigure.rocks\u002Fde\u002Fblog\u002Fwhen-gaming-ai-sounds-right-but-isn-t-the-reasoning-problem-behind-game-assistants-and-agents\\\">When Gaming AI Sounds Right but Isn't: The Reasoning Problem Behind Game Assistants and Agents on figure.rocks.\u003C\u002Fa>\"]},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":1264,"blocks":1265,"version":2089},1789803202269,[1266,1270,1274,1278,1282,1286,1290,1294,1298,1302,1306,1310,1314,1318,1322,1326,1330,1334,1338,1342,1346,1350,1354,1358,1362,1366,1370,1374,1378,1382,1391,1395,1399,1403,1407,1411,1415,1419,1423,1427,1431,1435,1439,1443,1447,1451,1455,1459,1463,1467,1471,1475,1485,1489,1493,1497,1501,1510,1514,1518,1522,1526,1530,1558,1562,1566,1570,1574,1578,1589,1593,1597,1601,1605,1609,1622,1626,1630,1642,1646,1650,1661,1665,1669,1680,1684,1688,1698,1702,1706,1710,1714,1718,1722,1726,1730,1734,1738,1742,1746,1750,1754,1758,1762,1766,1770,1801,1805,1809,1813,1817,1821,1825,1829,1833,1840,1844,1848,1852,1856,1867,1871,1875,1886,1890,1894,1898,1902,1906,1910,1914,1918,1922,1939,1943,1957,1961,1965,1969,1991,1995,1999,2003,2007,2011,2015,2019,2023,2027,2031,2035,2039,2042,2046,2050,2054,2058,2062,2066,2071,2074,2078],{"id":215,"data":1267,"type":218,"tunes":1269},{"text":1268},"The methodology developed across this series began with a research problem: how can an AI model help investigate a complex question without simply reinforcing the assumptions already present in the user's prompt?",{},{"id":221,"data":1271,"type":218,"tunes":1273},{"text":1272},"That problem initially appears to belong to historical or academic research. In reality, it is much broader.",{},{"id":226,"data":1275,"type":218,"tunes":1277},{"text":1276},"Software debugging, architecture decisions, technical diagnosis, security analysis, product strategy and many forms of decision support share the same underlying structure. A problem is presented. Some evidence is available. One or more explanations appear plausible. Assumptions enter the analysis. The system must determine which conclusion is best supported.",{},{"id":231,"data":1279,"type":236,"tunes":1281},{"text":1280,"caption":234,"alignment":235},"The domain changes. The epistemic problem often does not.",{},{"id":239,"data":1283,"type":218,"tunes":1285},{"text":1284},"This article therefore takes the next step: converting the research protocol developed in the previous articles into a general reasoning framework for AI-assisted analytical work.",{},{"id":244,"data":1287,"type":218,"tunes":1289},{"text":1288},"The framework builds directly on four previous layers: \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">Beyond Prompt Engineering: A Methodology for More Reliable AI Reasoning\u003C\u002Fa> defines the overall methodological problem; \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">The Prompt Is Part of the Bias\u003C\u002Fa> examines framing and prompt dependency; \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">Prompt Invariance: Does the Conclusion Survive the Prompt?\u003C\u002Fa> introduces a robustness test against framing; and \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">Falsification for AI Reasoning: From Answers to Tested Hypotheses\u003C\u002Fa> adds systematic disconfirmation and competing hypotheses.",{},{"id":249,"data":1291,"type":42,"tunes":1293},{"text":1292,"level":252},"The Key Generalization",{},{"id":255,"data":1295,"type":218,"tunes":1297},{"text":1296},"The generalization is not that every task should be treated like historical research.",{},{"id":260,"data":1299,"type":218,"tunes":1301},{"text":1300},"That would be a category error.",{},{"id":265,"data":1303,"type":218,"tunes":1305},{"text":1304},"Historical research depends on chronology, provenance, transmission, source criticism and archival completeness. Software debugging depends on logs, configuration, reproducibility, execution state and controlled tests. Architecture depends on requirements, constraints, interfaces, quality attributes and operational trade-offs.",{},{"id":270,"data":1307,"type":218,"tunes":1309},{"text":1308},"What can be generalized is the reasoning process surrounding those domain-specific forms of evidence.",{},{"id":275,"data":1311,"type":236,"tunes":1313},{"text":1312,"caption":234,"alignment":235},"The framework is domain-independent at its core, but never domain-free.",{},{"id":280,"data":1315,"type":218,"tunes":1317},{"text":1316},"This distinction is fundamental. A universal reasoning methodology cannot replace expertise. It can instead define how evidence, assumptions, hypotheses, contradictions and uncertainty should be handled before domain expertise produces the final judgment.",{},{"id":285,"data":1319,"type":42,"tunes":1321},{"text":1320,"level":252},"A Domain-Independent Reasoning Core",{},{"id":290,"data":1323,"type":218,"tunes":1325},{"text":1324},"Across analytical domains, the same high-level sequence can be applied:",{},{"id":295,"data":1327,"type":236,"tunes":1329},{"text":1328,"caption":234,"alignment":235},"Problem → decomposition → evidence → assumptions → competing hypotheses → predictions → counter-evidence → discriminating tests → domain validation → confidence recalibration → conclusion",{},{"id":300,"data":1331,"type":218,"tunes":1333},{"text":1332},"The process deliberately prevents the model's first coherent explanation from becoming the final answer by default.",{},{"id":305,"data":1335,"type":218,"tunes":1337},{"text":1336},"Instead, the first explanation becomes a candidate.",{},{"id":310,"data":1339,"type":42,"tunes":1341},{"text":1340,"level":313},"1. Define the Actual Problem",{},{"id":316,"data":1343,"type":218,"tunes":1345},{"text":1344},"Before generating a solution, the system should determine what is actually being asked.",{},{"id":322,"data":1347,"type":218,"tunes":1349},{"text":1348},"User requests frequently contain both a problem and a proposed diagnosis:",{},{"id":328,"data":1351,"type":236,"tunes":1353},{"text":1352,"caption":234,"alignment":235},"Why is nginx causing my API connection failures?",{},{"id":333,"data":1355,"type":218,"tunes":1357},{"text":1356},"The actual problem may instead be:",{},{"id":339,"data":1359,"type":236,"tunes":1361},{"text":1360,"caption":234,"alignment":235},"What is causing the API connection failures?",{},{"id":344,"data":1363,"type":218,"tunes":1365},{"text":1364},"The difference is small linguistically and enormous methodologically.",{},{"id":350,"data":1367,"type":218,"tunes":1369},{"text":1368},"The first formulation embeds a causal hypothesis. The second exposes that hypothesis to competition.",{},{"id":356,"data":1371,"type":218,"tunes":1373},{"text":1372},"This is precisely the framing problem examined in \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">The Prompt Is Part of the Bias\u003C\u002Fa>.",{},{"id":362,"data":1375,"type":42,"tunes":1377},{"text":1376,"level":313},"2. Separate Facts, Assumptions and Unknowns",{},{"id":367,"data":1379,"type":218,"tunes":1381},{"text":1380},"A reliable reasoning process must prevent observations and interpretations from silently merging.",{},{"id":373,"data":1383,"type":382,"tunes":1390},{"meta":1384,"items":1385,"style":381},{},[1386,1387,1388,1389],"\u003Cb>Fact:\u003C\u002Fb> directly supported by available evidence.","\u003Cb>Interpretation:\u003C\u002Fb> an explanation inferred from facts.","\u003Cb>Assumption:\u003C\u002Fb> a proposition required by the current reasoning path but not independently established.","\u003Cb>Unknown:\u003C\u002Fb> information necessary for stronger discrimination but currently unavailable.",{},{"id":385,"data":1392,"type":218,"tunes":1394},{"text":1393},"This classification may appear elementary, but it solves a common failure mode in AI-generated analysis: assumptions become embedded in fluent prose and later reappear as if they had already been established.",{},{"id":391,"data":1396,"type":236,"tunes":1398},{"text":1397,"caption":234,"alignment":235},"Traceability begins by preventing inference from disguising itself as evidence.",{},{"id":396,"data":1400,"type":42,"tunes":1402},{"text":1401,"level":313},"3. Generate Competing Explanations",{},{"id":401,"data":1404,"type":218,"tunes":1406},{"text":1405},"A candidate explanation should not be evaluated in isolation when credible alternatives exist.",{},{"id":407,"data":1408,"type":218,"tunes":1410},{"text":1409},"The system should therefore generate several plausible hypotheses before committing to one.",{},{"id":413,"data":1412,"type":218,"tunes":1414},{"text":1413},"The purpose is not artificial brainstorming. Alternatives should represent genuinely different mechanisms capable of explaining the observations.",{},{"id":419,"data":1416,"type":236,"tunes":1418},{"text":1417,"caption":234,"alignment":235},"A preferred hypothesis has not won if no serious competitor was allowed to enter the test.",{},{"id":424,"data":1420,"type":218,"tunes":1422},{"text":1421},"This concept is central to \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">Falsification for AI Reasoning\u003C\u002Fa>, where hypotheses are evaluated through expected observations, counter-evidence and discriminating tests rather than accumulated supportive prose.",{},{"id":429,"data":1424,"type":42,"tunes":1426},{"text":1425,"level":313},"4. Preserve Evidence Provenance",{},{"id":434,"data":1428,"type":218,"tunes":1430},{"text":1429},"Evidence should remain traceable to its origin.",{},{"id":439,"data":1432,"type":236,"tunes":1434},{"text":1433,"caption":234,"alignment":235},"Conclusion → interpretation → observation → source",{},{"id":444,"data":1436,"type":218,"tunes":1438},{"text":1437},"A log line, an official technical specification, an experiment, a primary historical document, an expert interpretation and an AI-generated summary do not have identical evidential status.",{},{"id":449,"data":1440,"type":218,"tunes":1442},{"text":1441},"The framework therefore does not merely store information. It should preserve enough metadata to know where a claim came from and how far the final conclusion is from the original observation.",{},{"id":454,"data":1444,"type":42,"tunes":1446},{"text":1445,"level":313},"5. Derive Predictions Before Explaining Results",{},{"id":459,"data":1448,"type":218,"tunes":1450},{"text":1449},"Once hypotheses exist, each should generate expectations.",{},{"id":464,"data":1452,"type":218,"tunes":1454},{"text":1453},"If the hypothesis is correct, what should we observe?",{},{"id":469,"data":1456,"type":218,"tunes":1458},{"text":1457},"If an alternative is correct, what should be different?",{},{"id":474,"data":1460,"type":218,"tunes":1462},{"text":1461},"Deriving expectations before interpreting all available evidence helps prevent every observation from being retrofitted into the preferred narrative.",{},{"id":479,"data":1464,"type":42,"tunes":1466},{"text":1465,"level":313},"6. Search for Counter-Evidence",{},{"id":484,"data":1468,"type":218,"tunes":1470},{"text":1469},"A generative model is naturally effective at constructing coherent support for a plausible idea. This makes disconfirmation particularly important.",{},{"id":489,"data":1472,"type":218,"tunes":1474},{"text":1473},"The system should explicitly ask:",{},{"id":494,"data":1476,"type":382,"tunes":1484},{"meta":1477,"items":1478,"style":381},{},[1479,1480,1481,1482,1483],"What observation would substantially weaken this explanation?","What evidence does the hypothesis explain poorly?","Which alternative explains the same observations with fewer assumptions?","Which expected observations are missing?","Could the apparent support be explained by another mechanism?",{},{"id":505,"data":1486,"type":218,"tunes":1488},{"text":1487},"The principle is developed in detail in \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">Falsification for AI Reasoning: From Answers to Tested Hypotheses\u003C\u002Fa>.",{},{"id":510,"data":1490,"type":42,"tunes":1492},{"text":1491,"level":313},"7. Test Dependence on the Prompt",{},{"id":515,"data":1494,"type":218,"tunes":1496},{"text":1495},"Evidence testing alone does not reveal whether the model's analysis remains excessively dependent on the user's formulation.",{},{"id":520,"data":1498,"type":218,"tunes":1500},{"text":1499},"For sufficiently important problems, the conclusion can therefore be retested under alternative legitimate framings.",{},{"id":525,"data":1502,"type":382,"tunes":1509},{"meta":1503,"items":1504,"style":381},{},[1505,1506,1507,1508],"\u003Cb>Original:\u003C\u002Fb> preserve the user's formulation.","\u003Cb>Blind:\u003C\u002Fb> remove the preferred conclusion.","\u003Cb>Inverted:\u003C\u002Fb> foreground the strongest alternative.","\u003Cb>Adversarial:\u003C\u002Fb> deliberately search for the strongest evidence-based challenge.",{},{"id":535,"data":1511,"type":218,"tunes":1513},{"text":1512},"This is the Prompt Invariance method introduced in \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">Prompt Invariance: Does the Conclusion Survive the Prompt?\u003C\u002Fa>.",{},{"id":540,"data":1515,"type":218,"tunes":1517},{"text":1516},"Its function is not to prove that a stable answer is true. Its function is to expose conclusions that change primarily because the framing changed.",{},{"id":545,"data":1519,"type":42,"tunes":1521},{"text":1520,"level":313},"8. Apply Domain-Specific Validators",{},{"id":550,"data":1523,"type":218,"tunes":1525},{"text":1524},"This is where the general framework deliberately stops being universal.",{},{"id":555,"data":1527,"type":218,"tunes":1529},{"text":1528},"Every domain requires its own rules for determining whether an explanation is actually credible.",{},{"id":560,"data":1531,"type":587,"tunes":1557},{"content":1532,"stretched":43,"withHeadings":14},[1533,1536,1539,1542,1545,1548,1551,1554],[1534,1535],"Domain","Typical Validators",[1537,1538],"Historical research","Chronology, provenance, geographical plausibility, transmission channels, primary sources, historiography",[1540,1541],"Software debugging","Logs, reproduction, runtime state, configuration, controlled changes, error correlation",[1543,1544],"Software architecture","Requirements, constraints, scalability, maintainability, security, operability, cost",[1546,1547],"Security analysis","Telemetry, attack path, permissions, observable indicators, reproducibility, threat model",[1549,1550],"Product strategy","Customer evidence, pricing behaviour, conversion, alternatives, market constraints, failure criteria",[1552,1553],"Project management","Scope, dependencies, resources, risks, acceptance criteria, milestones, stakeholder constraints",[1555,1556],"Scientific analysis","Experimental design, measurement quality, controls, reproducibility, statistical evidence, alternative explanations",{},{"id":590,"data":1559,"type":218,"tunes":1561},{"text":1560},"The framework controls how claims are processed. Domain validators determine whether those claims survive contact with reality.",{},{"id":595,"data":1563,"type":42,"tunes":1565},{"text":1564,"level":313},"9. Recalibrate Confidence",{},{"id":600,"data":1567,"type":218,"tunes":1569},{"text":1568},"The final conclusion should not preserve the confidence level of the initial answer automatically.",{},{"id":605,"data":1571,"type":218,"tunes":1573},{"text":1572},"Confidence should be recalculated after alternatives, counter-evidence, prompt variation and domain validation have been considered.",{},{"id":610,"data":1575,"type":218,"tunes":1577},{"text":1576},"A useful output can therefore remain uncertain.",{},{"id":615,"data":1579,"type":382,"tunes":1588},{"meta":1580,"items":1581,"style":381},{},[1582,1583,1584,1585,1586,1587],"strongly supported;","provisionally supported;","weakly supported;","underdetermined;","substantially contradicted;","not testable with current evidence.",{},{"id":627,"data":1590,"type":236,"tunes":1592},{"text":1591,"caption":234,"alignment":235},"Uncertainty is not a failure of reasoning when the evidence itself is uncertain.",{},{"id":632,"data":1594,"type":42,"tunes":1596},{"text":1595,"level":252},"The Same Framework Across Different Domains",{},{"id":637,"data":1598,"type":218,"tunes":1600},{"text":1599},"The easiest way to understand the general framework is to observe how the same reasoning architecture behaves when the domain changes.",{},{"id":642,"data":1602,"type":42,"tunes":1604},{"text":1603,"level":313},"Historical Research",{},{"id":646,"data":1606,"type":218,"tunes":1608},{"text":1607},"Problem: determine whether similarities between two intellectual traditions indicate historical transmission.",{},{"id":651,"data":1610,"type":382,"tunes":1621},{"meta":1611,"items":1612,"style":381},{},[1613,1614,1615,1616,1617,1618,1619,1620],"Separate documented similarities from interpretive parallels.","Compare direct transmission, indirect inheritance, convergence and retrospective interpretation.","Test chronology and geographical contact.","Search for documentary or terminological traces.","Identify evidence expected under direct transmission.","Search for earlier independent examples that weaken that hypothesis.","Reframe the question without the preferred genealogy.","Conclude only at the confidence level supported by the surviving evidence.",{},{"id":665,"data":1623,"type":42,"tunes":1625},{"text":1624,"level":313},"Software Debugging",{},{"id":669,"data":1627,"type":218,"tunes":1629},{"text":1628},"Problem: determine why an API connection fails intermittently.",{},{"id":674,"data":1631,"type":382,"tunes":1641},{"meta":1632,"items":1633,"style":381},{},[1634,1635,1636,1637,1638,1639,1640],"Separate observed failures from the developer's suspected cause.","Generate proxy, application, database, network and client hypotheses.","Determine what logs and runtime behaviour each hypothesis predicts.","Perform tests capable of distinguishing among them.","Attempt to reproduce the failure while eliminating layers.","Reject explanations contradicted by controlled observations.","Identify the first confirmed failure mechanism rather than the most persuasive narrative.",{},{"id":687,"data":1643,"type":42,"tunes":1645},{"text":1644,"level":313},"Software Architecture",{},{"id":691,"data":1647,"type":218,"tunes":1649},{"text":1648},"Problem: choose an architecture for a new platform.",{},{"id":696,"data":1651,"type":382,"tunes":1660},{"meta":1652,"items":1653,"style":381},{},[1654,1655,1656,1657,1658,1659],"Separate requirements from implementation preferences.","Compare modular monolith, services and hybrid alternatives.","Test each option against scalability, team structure, deployment, operational complexity and cost.","Identify which requirement genuinely requires architectural complexity.","Search for simpler designs capable of satisfying the same constraints.","Treat technology preference as an assumption rather than a requirement.",{},{"id":708,"data":1662,"type":42,"tunes":1664},{"text":1663,"level":313},"Product Strategy",{},{"id":713,"data":1666,"type":218,"tunes":1668},{"text":1667},"Problem: determine whether a product capability should be commercialized.",{},{"id":718,"data":1670,"type":382,"tunes":1679},{"meta":1671,"items":1672,"style":381},{},[1673,1674,1675,1676,1677,1678],"Separate technical possibility from demonstrated demand.","Define alternative customer problems and alternative solutions.","Identify observable market behaviour predicted by the business thesis.","Define failure criteria before interpreting results.","Search for evidence that customers solve the problem differently.","Distinguish interest, willingness to test and willingness to pay.",{},{"id":730,"data":1681,"type":42,"tunes":1683},{"text":1682,"level":313},"Project and Delivery Analysis",{},{"id":735,"data":1685,"type":218,"tunes":1687},{"text":1686},"Problem: determine why a project is missing planned outcomes.",{},{"id":740,"data":1689,"type":382,"tunes":1697},{"meta":1690,"items":1691,"style":381},{},[1692,1693,1694,1695,1696],"Separate symptoms such as delay from their assumed causes.","Compare scope, dependencies, capacity, requirements, governance and technical risk.","Trace evidence through plans, decisions, changes and acceptance criteria.","Test whether corrective action addresses the actual mechanism rather than the visible symptom.","Reassess the project hypothesis as new delivery evidence appears.",{},{"id":751,"data":1699,"type":42,"tunes":1701},{"text":1700,"level":252},"Why This Is More Than Prompt Engineering",{},{"id":756,"data":1703,"type":218,"tunes":1705},{"text":1704},"Prompt engineering changes the instruction presented to a model.",{},{"id":761,"data":1707,"type":218,"tunes":1709},{"text":1708},"A reasoning framework changes the process through which an answer is allowed to become a conclusion.",{},{"id":766,"data":1711,"type":236,"tunes":1713},{"text":1712,"caption":234,"alignment":235},"Prompt engineering optimizes an invocation. A reasoning framework governs an inference process.",{},{"id":771,"data":1715,"type":218,"tunes":1717},{"text":1716},"This difference becomes important as AI systems move from chat interfaces toward agents, RAG pipelines, automated research, development tools and decision-support systems.",{},{"id":776,"data":1719,"type":218,"tunes":1721},{"text":1720},"A single carefully engineered prompt may improve one model call. A reasoning framework can define how multiple calls, retrieved evidence, tool results and validation stages interact before the system commits to an answer.",{},{"id":781,"data":1723,"type":42,"tunes":1725},{"text":1724,"level":252},"The Framework Is Related to — but Different from — Existing LLM Reasoning Methods",{},{"id":786,"data":1727,"type":218,"tunes":1729},{"text":1728},"The broader research landscape already contains several important approaches showing that LLM performance can improve when inference is structured as a process rather than a single generation.",{},{"id":791,"data":1731,"type":218,"tunes":1733},{"text":1732},"\u003Cb>ReAct\u003C\u002Fb> combines reasoning with actions and observations from an external environment. Instead of relying exclusively on internal model knowledge, the model can act, observe new information and update its next step.",{},{"id":796,"data":1735,"type":218,"tunes":1737},{"text":1736},"\u003Cb>Self-Refine\u003C\u002Fb> uses iterative generation, feedback and refinement, demonstrating that an initial LLM output can often be improved through explicit self-feedback without additional model training.",{},{"id":801,"data":1739,"type":218,"tunes":1741},{"text":1740},"\u003Cb>Reflexion\u003C\u002Fb> uses verbal feedback and episodic memory to allow language agents to learn from previous attempts and modify later behaviour.",{},{"id":806,"data":1743,"type":218,"tunes":1745},{"text":1744},"\u003Cb>Tree of Thoughts\u003C\u002Fb> explores multiple reasoning paths rather than committing immediately to one left-to-right chain, allowing search, evaluation and backtracking over candidate approaches.",{},{"id":811,"data":1747,"type":218,"tunes":1749},{"text":1748},"These methods differ substantially in purpose and implementation, but they establish an important general point: inference-time structure can materially change model performance.",{},{"id":816,"data":1751,"type":218,"tunes":1753},{"text":1752},"The framework proposed in this series addresses a different primary objective.",{},{"id":821,"data":1755,"type":236,"tunes":1757},{"text":1756,"caption":234,"alignment":235},"The goal is not simply to make the model search longer, reflect more or generate a better answer. The goal is to make the path from evidence to conclusion more resistant to framing, confirmation and unsupported assumptions.",{},{"id":826,"data":1759,"type":218,"tunes":1761},{"text":1760},"Prompt Invariance, falsification-oriented testing and domain-specific validation therefore serve as epistemic controls rather than generic inference-expansion techniques.",{},{"id":831,"data":1763,"type":42,"tunes":1765},{"text":1764,"level":252},"A Layered View of AI Reasoning Quality",{},{"id":836,"data":1767,"type":218,"tunes":1769},{"text":1768},"The resulting system can be understood as several layers.",{},{"id":841,"data":1771,"type":587,"tunes":1800},{"content":1772,"stretched":43,"withHeadings":14},[1773,1776,1779,1782,1785,1788,1791,1794,1797],[1774,1775],"Layer","Question",[1777,1778],"Task understanding","What problem are we actually solving?",[1780,1781],"Evidence","What do we actually know?",[1783,1784],"Assumptions","What are we currently taking for granted?",[1786,1787],"Hypotheses","What plausible mechanisms could explain the evidence?",[1789,1790],"Falsification","What evidence could weaken each explanation?",[1792,1793],"Prompt Invariance","Does the conclusion depend excessively on framing?",[1795,1796],"Domain validation","Does the explanation satisfy the rules of the relevant discipline?",[1798,1799],"Confidence calibration","How strongly should the surviving conclusion be trusted?",{},{"id":873,"data":1802,"type":218,"tunes":1804},{"text":1803},"An answer can fail at any layer.",{},{"id":878,"data":1806,"type":218,"tunes":1808},{"text":1807},"It can answer the wrong problem correctly. It can reason correctly from false evidence. It can retrieve correct facts but choose the wrong causal explanation. It can produce a strong explanation that disappears when the prompt is reframed. Or it can survive all general reasoning checks and still violate a domain-specific constraint.",{},{"id":883,"data":1810,"type":236,"tunes":1812},{"text":1811,"caption":234,"alignment":235},"Correctness is not one property. It is the result of several dependencies surviving simultaneously.",{},{"id":888,"data":1814,"type":42,"tunes":1816},{"text":1815,"level":252},"Reasoning Quality Versus Model Capability",{},{"id":893,"data":1818,"type":218,"tunes":1820},{"text":1819},"This leads back to one of the central ideas of the series.",{},{"id":899,"data":1822,"type":218,"tunes":1824},{"text":1823},"A stronger model does not automatically imply that every invocation will use its capabilities in the most rigorous possible way.",{},{"id":905,"data":1826,"type":218,"tunes":1828},{"text":1827},"A model may possess the capability to generate alternatives, inspect logs, search sources, challenge assumptions and revise a conclusion. Whether all of those operations actually occur depends on the task, prompting, available tools, context and surrounding system architecture.",{},{"id":911,"data":1830,"type":218,"tunes":1832},{"text":1831},"The framework therefore separates two questions:",{},{"id":917,"data":1834,"type":382,"tunes":1839},{"meta":1835,"items":1836,"style":381},{},[1837,1838],"\u003Cb>Capability:\u003C\u002Fb> what can the model do?","\u003Cb>Methodological discipline:\u003C\u002Fb> which of those capabilities must be exercised before the system accepts a conclusion?",{},{"id":925,"data":1841,"type":218,"tunes":1843},{"text":1842},"The distinction was the starting point of \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">Beyond Prompt Engineering\u003C\u002Fa> and becomes even more important once the methodology is generalized beyond research.",{},{"id":931,"data":1845,"type":42,"tunes":1847},{"text":1846,"level":252},"Not Every Task Needs the Full Framework",{},{"id":936,"data":1849,"type":218,"tunes":1851},{"text":1850},"The framework should not become compulsory overhead for every AI interaction.",{},{"id":942,"data":1853,"type":218,"tunes":1855},{"text":1854},"Many tasks have low epistemic complexity:",{},{"id":948,"data":1857,"type":382,"tunes":1866},{"meta":1858,"items":1859,"style":381},{},[1860,1861,1862,1863,1864,1865],"format this JSON;","translate this sentence;","convert these units;","rewrite this message;","extract these fields;","summarize this supplied text.",{},{"id":960,"data":1868,"type":218,"tunes":1870},{"text":1869},"Running four prompt variants, three competing hypotheses and a falsification stage for such tasks would add cost without proportional value.",{},{"id":966,"data":1872,"type":218,"tunes":1874},{"text":1873},"The full method becomes more valuable when:",{},{"id":972,"data":1876,"type":382,"tunes":1885},{"meta":1877,"items":1878,"style":381},{},[1879,1880,1881,1882,1883,1884],"the answer depends on interpretation rather than direct retrieval;","the user already has a preferred explanation;","multiple causal mechanisms are plausible;","the evidence is incomplete or contradictory;","the decision has significant technical, financial or research consequences;","the AI system is expected to act autonomously on the conclusion.",{},{"id":984,"data":1887,"type":218,"tunes":1889},{"text":1888},"Methodological depth should therefore scale with epistemic risk.",{},{"id":990,"data":1891,"type":42,"tunes":1893},{"text":1892,"level":252},"From Static Prompt to Adaptive Reasoning Policy",{},{"id":995,"data":1895,"type":218,"tunes":1897},{"text":1896},"Once generalized, the framework no longer needs to exist as one enormous prompt.",{},{"id":1001,"data":1899,"type":218,"tunes":1901},{"text":1900},"A practical system can decide dynamically which controls a task requires.",{},{"id":1006,"data":1903,"type":236,"tunes":1905},{"text":1904,"caption":234,"alignment":235},"Simple task → direct answer\nAnalytical task → structured evidence check\nHigh-uncertainty task → competing hypotheses\nFraming-sensitive task → Prompt Invariance\nHigh-consequence hypothesis → falsification and domain validation",{},{"id":1011,"data":1907,"type":218,"tunes":1909},{"text":1908},"This transforms the methodology from a fixed prompt template into an adaptive reasoning policy.",{},{"id":1016,"data":1911,"type":218,"tunes":1913},{"text":1912},"The system does not always reason maximally. It reasons as rigorously as the task requires.",{},{"id":1021,"data":1915,"type":42,"tunes":1917},{"text":1916,"level":252},"A Practical General Framework",{},{"id":1026,"data":1919,"type":218,"tunes":1921},{"text":1920},"The complete process can be summarized as follows:",{},{"id":1031,"data":1923,"type":382,"tunes":1938},{"meta":1924,"items":1925,"style":1048},{"counterType":1034},[1926,1927,1928,1929,1930,1931,1932,1933,1934,1935,1936,1937],"\u003Cb>Normalize the problem.\u003C\u002Fb> Remove embedded conclusions from the task definition where appropriate.","\u003Cb>Identify evidence.\u003C\u002Fb> Separate observations from interpretations.","\u003Cb>Expose assumptions.\u003C\u002Fb> Record propositions that have not yet been established.","\u003Cb>Generate serious alternatives.\u003C\u002Fb> Do not allow the first hypothesis to compete only against weak strawmen.","\u003Cb>Preserve provenance.\u003C\u002Fb> Keep claims traceable to their sources.","\u003Cb>Derive expectations.\u003C\u002Fb> Define what each hypothesis predicts.","\u003Cb>Search for counter-evidence.\u003C\u002Fb> Actively look for observations the preferred hypothesis handles poorly.","\u003Cb>Run discriminating tests.\u003C\u002Fb> Prefer evidence that differentiates among explanations.","\u003Cb>Test framing dependence.\u003C\u002Fb> Apply Prompt Invariance when the user's framing may influence the result.","\u003Cb>Apply domain validators.\u003C\u002Fb> Use the discipline-specific criteria relevant to the problem.","\u003Cb>Recalibrate confidence.\u003C\u002Fb> Allow the conclusion to weaken, remain unresolved or change.","\u003Cb>Preserve traceability.\u003C\u002Fb> Make the path from conclusion back to evidence inspectable.",{},{"id":1051,"data":1940,"type":42,"tunes":1942},{"text":1941,"level":252},"What the Framework Does Not Claim",{},{"id":1056,"data":1944,"type":382,"tunes":1956},{"meta":1945,"items":1946,"style":381},{},[1947,1948,1949,1950,1951,1952,1953,1954,1955],"It does not guarantee truthful output.","It does not eliminate hallucinations.","It does not make an LLM a domain expert.","It does not prove that stable conclusions are correct.","It does not eliminate biases shared across all reasoning passes.","It does not replace experiments, measurements or primary sources.","It does not guarantee that all relevant hypotheses have been generated.","It does not imply that more reasoning is always better.","It does not claim that the complete framework has already been validated as a standardized academic methodology.",{},{"id":1071,"data":1958,"type":218,"tunes":1960},{"text":1959},"Its claim is narrower and more defensible: analytical AI systems can be made methodologically stronger when assumptions, framing, competing explanations, counter-evidence and domain validation are handled explicitly rather than left entirely to one unconstrained generation.",{},{"id":1076,"data":1962,"type":42,"tunes":1964},{"text":1963,"level":252},"The Four Articles Become One Reasoning System",{},{"id":1081,"data":1966,"type":218,"tunes":1968},{"text":1967},"The previous articles can now be understood not as separate prompting techniques but as components of one reasoning architecture.",{},{"id":1086,"data":1970,"type":587,"tunes":1990},{"content":1971,"stretched":43,"withHeadings":14},[1972,1975,1978,1981,1984,1987],[1973,1974],"Article","Function in the Framework",[1976,1977],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">Beyond Prompt Engineering\u003C\u002Fa>","Defines the shift from answer generation to methodological reasoning.",[1979,1980],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">The Prompt Is Part of the Bias\u003C\u002Fa>","Treats the user's formulation as a possible source of reasoning bias.",[1982,1983],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">Prompt Invariance\u003C\u002Fa>","Tests whether the conclusion survives meaningful changes in framing.",[1985,1986],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">Falsification for AI Reasoning\u003C\u002Fa>","Tests whether hypotheses survive disconfirming evidence and serious alternatives.",[1988,1989],"\u003Cb>From Research Protocol to a General AI Reasoning Framework\u003C\u002Fb>","Combines the controls into a domain-independent core with domain-specific validation.",{},{"id":1109,"data":1992,"type":218,"tunes":1994},{"text":1993},"Together they describe a transition:",{},{"id":1114,"data":1996,"type":236,"tunes":1998},{"text":1997,"caption":234,"alignment":235},"Prompt → answer\nbecomes\nProblem → evidence → hypotheses → challenge → validation → calibrated conclusion",{},{"id":1119,"data":2000,"type":42,"tunes":2002},{"text":2001,"level":252},"The Central Principle",{},{"id":1124,"data":2004,"type":218,"tunes":2006},{"text":2005},"The methodology began with a simple observation: asking a powerful reasoning model a question does not automatically mean that every useful reasoning capability available to the model will be used.",{},{"id":1129,"data":2008,"type":218,"tunes":2010},{"text":2009},"The solution is not to force maximal reasoning into every interaction.",{},{"id":1134,"data":2012,"type":218,"tunes":2014},{"text":2013},"The solution is to identify which reasoning controls matter for the task and make them explicit.",{},{"id":1139,"data":2016,"type":236,"tunes":2018},{"text":2017,"caption":234,"alignment":235},"A good AI methodology does not tell the model what conclusion to reach. It defines what a conclusion must survive before it is accepted.",{},{"id":1144,"data":2020,"type":218,"tunes":2022},{"text":2021},"That principle is transferable.",{},{"id":1149,"data":2024,"type":218,"tunes":2026},{"text":2025},"In historical research, the conclusion must survive chronology and source criticism. In debugging, it must survive reproduction and discriminating tests. In architecture, it must survive requirements and operational constraints. In strategy, it must survive market evidence and failure criteria.",{},{"id":1154,"data":2028,"type":218,"tunes":2030},{"text":2029},"The validators change.",{},{"id":1159,"data":2032,"type":218,"tunes":2034},{"text":2033},"The methodological discipline remains.",{},{"id":1164,"data":2036,"type":236,"tunes":2038},{"text":2037,"caption":234,"alignment":235},"The objective is not a model that always thinks longer. It is a system that knows when an answer is not yet justified.",{},{"id":1169,"data":2040,"type":1171,"tunes":2041},{},{},{"id":1174,"data":2043,"type":42,"tunes":2045},{"text":2044,"level":252},"Research Context",{},{"id":1179,"data":2047,"type":218,"tunes":2049},{"text":2048},"The complete framework described in this series is a methodological synthesis rather than an established standardized LLM benchmark. Several adjacent lines of research nevertheless support its central architectural premise: inference quality can change substantially when language-model problem solving is organized as an iterative or multi-stage process rather than a single generation.",{},{"id":1184,"data":2051,"type":218,"tunes":2053},{"text":2052},"ReAct combines reasoning with actions and external observations and demonstrated benefits across question answering, fact verification and interactive decision tasks. Tree of Thoughts explores multiple candidate reasoning paths with evaluation and backtracking rather than committing immediately to one chain. Self-Refine showed improvements across diverse tasks by iterating between generation, feedback and refinement. Reflexion demonstrated that language agents can use verbal feedback from previous attempts to improve later decisions without updating model weights.",{},{"id":1189,"data":2055,"type":218,"tunes":2057},{"text":2056},"These approaches do not implement Prompt Invariance or the falsification-oriented framework proposed here. They provide independent evidence for the broader idea that inference-time procedure matters: model capability and the process through which that capability is exercised are not identical.",{},{"id":1194,"data":2059,"type":218,"tunes":2061},{"text":2060},"The contribution of this series is to organize that insight around epistemic controls: evidence separation, assumption tracking, prompt-framing analysis, competing hypotheses, falsification, provenance, domain validation and calibrated uncertainty.",{},{"id":1199,"data":2063,"type":42,"tunes":2065},{"text":2064,"level":313},"Selected References",{},{"id":1204,"data":2067,"type":382,"tunes":2070},{"meta":2068,"items":2069,"style":381},{},[1208,1209,1210,1211,1212,1213],{},{"id":1216,"data":2072,"type":1171,"tunes":2073},{},{},{"id":1220,"data":2075,"type":42,"tunes":2077},{"text":2076,"level":252},"The AI Reasoning Methodology Series",{},{"id":1225,"data":2079,"type":382,"tunes":2088},{"meta":2080,"items":2081,"style":1048},{"counterType":1034},[2082,2083,2084,2085,2086,2087],"\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning\">\u003Cb>Beyond Prompt Engineering: A Methodology for More Reliable AI Reasoning\u003C\u002Fb>\u003C\u002Fa> — why model capability alone is not enough.","\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fthe-prompt-is-part-of-the-bias-how-ai-framing-shapes-reasoning\">\u003Cb>The Prompt Is Part of the Bias\u003C\u002Fb>\u003C\u002Fa> — how task formulation can influence the reasoning environment.","\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt\">\u003Cb>Prompt Invariance: Does the Conclusion Survive the Prompt?\u003C\u002Fb>\u003C\u002Fa> — testing conclusion stability across alternative framings.","\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffalsification-for-ai-reasoning-from-answers-to-tested-hypotheses\">\u003Cb>Falsification for AI Reasoning: From Answers to Tested Hypotheses\u003C\u002Fb>\u003C\u002Fa> — competing hypotheses, counter-evidence and discriminating tests.","\u003Cb>From Research Protocol to a General AI Reasoning Framework\u003C\u002Fb> — integrating the methodology into a reusable domain-independent reasoning core.","A practical application of the same reasoning framework in gaming AI — including game assistants, autonomous agents, patch-sensitive knowledge and live game-state validation — is explored in \u003Ca href=\"https:\u002F\u002Ffigure.rocks\u002Fde\u002Fblog\u002Fwhen-gaming-ai-sounds-right-but-isn-t-the-reasoning-problem-behind-game-assistants-and-agents\">When Gaming AI Sounds Right but Isn't: The Reasoning Problem Behind Game Assistants and Agents on figure.rocks.\u003C\u002Fa>",{},"2.31.6","A methodology developed for rigorous AI-assisted research can be generalized far beyond research itself. By separating evidence from assumptions, testing competing hypotheses, controlling prompt framing, searching for counter-evidence and applying domain-specific validators, the same reasoning architecture can improve debugging, software design, strategy, technical analysis and AI-assisted decision support.",{"lang":7,"title":208,"content":210,"contentJson":2092,"excerpt":1237},{"time":212,"blocks":2093,"version":1236},[2094,2097,2100,2103,2106,2109,2112,2115,2118,2121,2124,2127,2130,2133,2136,2139,2142,2145,2148,2151,2155,2159,2162,2166,2169,2173,2177,2181,2184,2188,2193,2197,2200,2203,2207,2211,2215,2218,2221,2224,2227,2230,2233,2236,2239,2242,2245,2248,2251,2254,2257,2260,2265,2268,2271,2274,2277,2282,2285,2288,2291,2294,2297,2309,2312,2315,2318,2321,2324,2329,2332,2335,2338,2341,2344,2349,2352,2355,2360,2363,2366,2371,2374,2377,2382,2385,2388,2393,2396,2399,2402,2405,2408,2411,2414,2417,2420,2423,2426,2429,2432,2435,2438,2441,2444,2447,2460,2463,2466,2469,2472,2476,2480,2484,2488,2493,2497,2500,2504,2508,2513,2517,2521,2526,2530,2533,2537,2540,2543,2546,2549,2552,2555,2560,2563,2568,2571,2574,2577,2587,2590,2593,2596,2599,2602,2605,2608,2611,2614,2617,2620,2623,2626,2629,2632,2635,2638,2641,2644,2649,2652,2655],{"id":215,"data":2095,"type":218,"tunes":2096},{"text":217},{},{"id":221,"data":2098,"type":218,"tunes":2099},{"text":223},{},{"id":226,"data":2101,"type":218,"tunes":2102},{"text":228},{},{"id":231,"data":2104,"type":236,"tunes":2105},{"text":233,"caption":234,"alignment":235},{},{"id":239,"data":2107,"type":218,"tunes":2108},{"text":241},{},{"id":244,"data":2110,"type":218,"tunes":2111},{"text":246},{},{"id":249,"data":2113,"type":42,"tunes":2114},{"text":251,"level":252},{},{"id":255,"data":2116,"type":218,"tunes":2117},{"text":257},{},{"id":260,"data":2119,"type":218,"tunes":2120},{"text":262},{},{"id":265,"data":2122,"type":218,"tunes":2123},{"text":267},{},{"id":270,"data":2125,"type":218,"tunes":2126},{"text":272},{},{"id":275,"data":2128,"type":236,"tunes":2129},{"text":277,"caption":234,"alignment":235},{},{"id":280,"data":2131,"type":218,"tunes":2132},{"text":282},{},{"id":285,"data":2134,"type":42,"tunes":2135},{"text":287,"level":252},{},{"id":290,"data":2137,"type":218,"tunes":2138},{"text":292},{},{"id":295,"data":2140,"type":236,"tunes":2141},{"text":297,"caption":234,"alignment":235},{},{"id":300,"data":2143,"type":218,"tunes":2144},{"text":302},{},{"id":305,"data":2146,"type":218,"tunes":2147},{"text":307},{},{"id":310,"data":2149,"type":42,"tunes":2150},{"text":312,"level":313},{},{"id":316,"data":2152,"type":218,"tunes":2154},{"text":318,"type":218,"tunes":2153},{},{},{"id":322,"data":2156,"type":218,"tunes":2158},{"text":324,"type":218,"tunes":2157},{},{},{"id":328,"data":2160,"type":236,"tunes":2161},{"text":330,"caption":234,"alignment":235},{},{"id":333,"data":2163,"type":218,"tunes":2165},{"text":335,"type":218,"tunes":2164},{},{},{"id":339,"data":2167,"type":236,"tunes":2168},{"text":341,"caption":234,"alignment":235},{},{"id":344,"data":2170,"type":218,"tunes":2172},{"text":346,"type":218,"tunes":2171},{},{},{"id":350,"data":2174,"type":218,"tunes":2176},{"text":352,"type":218,"tunes":2175},{},{},{"id":356,"data":2178,"type":218,"tunes":2180},{"text":358,"type":218,"tunes":2179},{},{},{"id":362,"data":2182,"type":42,"tunes":2183},{"text":364,"level":313},{},{"id":367,"data":2185,"type":218,"tunes":2187},{"text":369,"type":218,"tunes":2186},{},{},{"id":373,"data":2189,"type":382,"tunes":2192},{"meta":2190,"items":2191,"style":381},{},[377,378,379,380],{},{"id":385,"data":2194,"type":218,"tunes":2196},{"text":387,"type":218,"tunes":2195},{},{},{"id":391,"data":2198,"type":236,"tunes":2199},{"text":393,"caption":234,"alignment":235},{},{"id":396,"data":2201,"type":42,"tunes":2202},{"text":398,"level":313},{},{"id":401,"data":2204,"type":218,"tunes":2206},{"text":403,"type":218,"tunes":2205},{},{},{"id":407,"data":2208,"type":218,"tunes":2210},{"text":409,"type":218,"tunes":2209},{},{},{"id":413,"data":2212,"type":218,"tunes":2214},{"text":415,"type":218,"tunes":2213},{},{},{"id":419,"data":2216,"type":236,"tunes":2217},{"text":421,"caption":234,"alignment":235},{},{"id":424,"data":2219,"type":218,"tunes":2220},{"text":426},{},{"id":429,"data":2222,"type":42,"tunes":2223},{"text":431,"level":313},{},{"id":434,"data":2225,"type":218,"tunes":2226},{"text":436},{},{"id":439,"data":2228,"type":236,"tunes":2229},{"text":441,"caption":234,"alignment":235},{},{"id":444,"data":2231,"type":218,"tunes":2232},{"text":446},{},{"id":449,"data":2234,"type":218,"tunes":2235},{"text":451},{},{"id":454,"data":2237,"type":42,"tunes":2238},{"text":456,"level":313},{},{"id":459,"data":2240,"type":218,"tunes":2241},{"text":461},{},{"id":464,"data":2243,"type":218,"tunes":2244},{"text":466},{},{"id":469,"data":2246,"type":218,"tunes":2247},{"text":471},{},{"id":474,"data":2249,"type":218,"tunes":2250},{"text":476},{},{"id":479,"data":2252,"type":42,"tunes":2253},{"text":481,"level":313},{},{"id":484,"data":2255,"type":218,"tunes":2256},{"text":486},{},{"id":489,"data":2258,"type":218,"tunes":2259},{"text":491},{},{"id":494,"data":2261,"type":382,"tunes":2264},{"meta":2262,"items":2263,"style":381},{},[498,499,500,501,502],{},{"id":505,"data":2266,"type":218,"tunes":2267},{"text":507},{},{"id":510,"data":2269,"type":42,"tunes":2270},{"text":512,"level":313},{},{"id":515,"data":2272,"type":218,"tunes":2273},{"text":517},{},{"id":520,"data":2275,"type":218,"tunes":2276},{"text":522},{},{"id":525,"data":2278,"type":382,"tunes":2281},{"meta":2279,"items":2280,"style":381},{},[529,530,531,532],{},{"id":535,"data":2283,"type":218,"tunes":2284},{"text":537},{},{"id":540,"data":2286,"type":218,"tunes":2287},{"text":542},{},{"id":545,"data":2289,"type":42,"tunes":2290},{"text":547,"level":313},{},{"id":550,"data":2292,"type":218,"tunes":2293},{"text":552},{},{"id":555,"data":2295,"type":218,"tunes":2296},{"text":557},{},{"id":560,"data":2298,"type":587,"tunes":2308},{"content":2299,"stretched":43,"withHeadings":14},[2300,2301,2302,2303,2304,2305,2306,2307],[564,565],[567,568],[570,571],[573,574],[576,577],[579,580],[582,583],[585,586],{},{"id":590,"data":2310,"type":218,"tunes":2311},{"text":592},{},{"id":595,"data":2313,"type":42,"tunes":2314},{"text":597,"level":313},{},{"id":600,"data":2316,"type":218,"tunes":2317},{"text":602},{},{"id":605,"data":2319,"type":218,"tunes":2320},{"text":607},{},{"id":610,"data":2322,"type":218,"tunes":2323},{"text":612},{},{"id":615,"data":2325,"type":382,"tunes":2328},{"meta":2326,"items":2327,"style":381},{},[619,620,621,622,623,624],{},{"id":627,"data":2330,"type":236,"tunes":2331},{"text":629,"caption":234,"alignment":235},{},{"id":632,"data":2333,"type":42,"tunes":2334},{"text":634,"level":252},{},{"id":637,"data":2336,"type":218,"tunes":2337},{"text":639},{},{"id":642,"data":2339,"type":42,"tunes":2340},{"text":567,"level":313},{},{"id":646,"data":2342,"type":218,"tunes":2343},{"text":648},{},{"id":651,"data":2345,"type":382,"tunes":2348},{"meta":2346,"items":2347,"style":381},{},[655,656,657,658,659,660,661,662],{},{"id":665,"data":2350,"type":42,"tunes":2351},{"text":570,"level":313},{},{"id":669,"data":2353,"type":218,"tunes":2354},{"text":671},{},{"id":674,"data":2356,"type":382,"tunes":2359},{"meta":2357,"items":2358,"style":381},{},[678,679,680,681,682,683,684],{},{"id":687,"data":2361,"type":42,"tunes":2362},{"text":573,"level":313},{},{"id":691,"data":2364,"type":218,"tunes":2365},{"text":693},{},{"id":696,"data":2367,"type":382,"tunes":2370},{"meta":2368,"items":2369,"style":381},{},[700,701,702,703,704,705],{},{"id":708,"data":2372,"type":42,"tunes":2373},{"text":710,"level":313},{},{"id":713,"data":2375,"type":218,"tunes":2376},{"text":715},{},{"id":718,"data":2378,"type":382,"tunes":2381},{"meta":2379,"items":2380,"style":381},{},[722,723,724,725,726,727],{},{"id":730,"data":2383,"type":42,"tunes":2384},{"text":732,"level":313},{},{"id":735,"data":2386,"type":218,"tunes":2387},{"text":737},{},{"id":740,"data":2389,"type":382,"tunes":2392},{"meta":2390,"items":2391,"style":381},{},[744,745,746,747,748],{},{"id":751,"data":2394,"type":42,"tunes":2395},{"text":753,"level":252},{},{"id":756,"data":2397,"type":218,"tunes":2398},{"text":758},{},{"id":761,"data":2400,"type":218,"tunes":2401},{"text":763},{},{"id":766,"data":2403,"type":236,"tunes":2404},{"text":768,"caption":234,"alignment":235},{},{"id":771,"data":2406},{"text":773,"type":218,"tunes":2407},{},{"id":776,"data":2409},{"text":778,"type":218,"tunes":2410},{},{"id":781,"data":2412,"type":42,"tunes":2413},{"text":783,"level":252},{},{"id":786,"data":2415},{"text":788,"type":218,"tunes":2416},{},{"id":791,"data":2418},{"text":793,"type":218,"tunes":2419},{},{"id":796,"data":2421},{"text":798,"type":218,"tunes":2422},{},{"id":801,"data":2424},{"text":803,"type":218,"tunes":2425},{},{"id":806,"data":2427},{"text":808,"type":218,"tunes":2428},{},{"id":811,"data":2430},{"text":813,"type":218,"tunes":2431},{},{"id":816,"data":2433},{"text":818,"type":218,"tunes":2434},{},{"id":821,"data":2436,"type":236,"tunes":2437},{"text":823,"caption":234,"alignment":235},{},{"id":826,"data":2439},{"text":828,"type":218,"tunes":2440},{},{"id":831,"data":2442,"type":42,"tunes":2443},{"text":833,"level":252},{},{"id":836,"data":2445},{"text":838,"type":218,"tunes":2446},{},{"id":841,"data":2448,"type":587,"tunes":2459},{"content":2449,"stretched":43,"withHeadings":14},[2450,2451,2452,2453,2454,2455,2456,2457,2458],[845,846],[848,849],[851,852],[854,855],[857,858],[860,861],[863,864],[866,867],[869,870],{},{"id":873,"data":2461},{"text":875,"type":218,"tunes":2462},{},{"id":878,"data":2464},{"text":880,"type":218,"tunes":2465},{},{"id":883,"data":2467,"type":236,"tunes":2468},{"text":885,"caption":234,"alignment":235},{},{"id":888,"data":2470,"type":42,"tunes":2471},{"text":890,"level":252},{},{"id":893,"data":2473,"type":218,"tunes":2475},{"text":895,"type":218,"tunes":2474},{},{},{"id":899,"data":2477,"type":218,"tunes":2479},{"text":901,"type":218,"tunes":2478},{},{},{"id":905,"data":2481,"type":218,"tunes":2483},{"text":907,"type":218,"tunes":2482},{},{},{"id":911,"data":2485,"type":218,"tunes":2487},{"text":913,"type":218,"tunes":2486},{},{},{"id":917,"data":2489,"type":382,"tunes":2492},{"meta":2490,"items":2491,"style":381},{},[921,922],{},{"id":925,"data":2494,"type":218,"tunes":2496},{"text":927,"type":218,"tunes":2495},{},{},{"id":931,"data":2498,"type":42,"tunes":2499},{"text":933,"level":252},{},{"id":936,"data":2501,"type":218,"tunes":2503},{"text":938,"type":218,"tunes":2502},{},{},{"id":942,"data":2505,"type":218,"tunes":2507},{"text":944,"type":218,"tunes":2506},{},{},{"id":948,"data":2509,"type":382,"tunes":2512},{"meta":2510,"items":2511,"style":381},{},[952,953,954,955,956,957],{},{"id":960,"data":2514,"type":218,"tunes":2516},{"text":962,"type":218,"tunes":2515},{},{},{"id":966,"data":2518,"type":218,"tunes":2520},{"text":968,"type":218,"tunes":2519},{},{},{"id":972,"data":2522,"type":382,"tunes":2525},{"meta":2523,"items":2524,"style":381},{},[976,977,978,979,980,981],{},{"id":984,"data":2527,"type":218,"tunes":2529},{"text":986,"type":218,"tunes":2528},{},{},{"id":990,"data":2531,"type":42,"tunes":2532},{"text":992,"level":252},{},{"id":995,"data":2534,"type":218,"tunes":2536},{"text":997,"type":218,"tunes":2535},{},{},{"id":1001,"data":2538,"type":218,"tunes":2539},{"text":1003},{},{"id":1006,"data":2541,"type":236,"tunes":2542},{"text":1008,"caption":234,"alignment":235},{},{"id":1011,"data":2544,"type":218,"tunes":2545},{"text":1013},{},{"id":1016,"data":2547,"type":218,"tunes":2548},{"text":1018},{},{"id":1021,"data":2550,"type":42,"tunes":2551},{"text":1023,"level":252},{},{"id":1026,"data":2553,"type":218,"tunes":2554},{"text":1028},{},{"id":1031,"data":2556,"type":382,"tunes":2559},{"meta":2557,"items":2558,"style":1048},{"counterType":1034},[1036,1037,1038,1039,1040,1041,1042,1043,1044,1045,1046,1047],{},{"id":1051,"data":2561,"type":42,"tunes":2562},{"text":1053,"level":252},{},{"id":1056,"data":2564,"type":382,"tunes":2567},{"meta":2565,"items":2566,"style":381},{},[1060,1061,1062,1063,1064,1065,1066,1067,1068],{},{"id":1071,"data":2569,"type":218,"tunes":2570},{"text":1073},{},{"id":1076,"data":2572,"type":42,"tunes":2573},{"text":1078,"level":252},{},{"id":1081,"data":2575,"type":218,"tunes":2576},{"text":1083},{},{"id":1086,"data":2578,"type":587,"tunes":2586},{"content":2579,"stretched":43,"withHeadings":14},[2580,2581,2582,2583,2584,2585],[1090,1091],[1093,1094],[1096,1097],[1099,1100],[1102,1103],[1105,1106],{},{"id":1109,"data":2588,"type":218,"tunes":2589},{"text":1111},{},{"id":1114,"data":2591,"type":236,"tunes":2592},{"text":1116,"caption":234,"alignment":235},{},{"id":1119,"data":2594,"type":42,"tunes":2595},{"text":1121,"level":252},{},{"id":1124,"data":2597,"type":218,"tunes":2598},{"text":1126},{},{"id":1129,"data":2600,"type":218,"tunes":2601},{"text":1131},{},{"id":1134,"data":2603,"type":218,"tunes":2604},{"text":1136},{},{"id":1139,"data":2606,"type":236,"tunes":2607},{"text":1141,"caption":234,"alignment":235},{},{"id":1144,"data":2609,"type":218,"tunes":2610},{"text":1146},{},{"id":1149,"data":2612,"type":218,"tunes":2613},{"text":1151},{},{"id":1154,"data":2615,"type":218,"tunes":2616},{"text":1156},{},{"id":1159,"data":2618,"type":218,"tunes":2619},{"text":1161},{},{"id":1164,"data":2621,"type":236,"tunes":2622},{"text":1166,"caption":234,"alignment":235},{},{"id":1169,"data":2624,"type":1171,"tunes":2625},{},{},{"id":1174,"data":2627,"type":42,"tunes":2628},{"text":1176,"level":252},{},{"id":1179,"data":2630,"type":218,"tunes":2631},{"text":1181},{},{"id":1184,"data":2633,"type":218,"tunes":2634},{"text":1186},{},{"id":1189,"data":2636,"type":218,"tunes":2637},{"text":1191},{},{"id":1194,"data":2639,"type":218,"tunes":2640},{"text":1196},{},{"id":1199,"data":2642,"type":42,"tunes":2643},{"text":1201,"level":313},{},{"id":1204,"data":2645,"type":382,"tunes":2648},{"meta":2646,"items":2647,"style":381},{},[1208,1209,1210,1211,1212,1213],{},{"id":1216,"data":2650,"type":1171,"tunes":2651},{},{},{"id":1220,"data":2653,"type":42,"tunes":2654},{"text":1222,"level":252},{},{"id":1225,"data":2656,"type":382,"tunes":2659},{"meta":2657,"items":2658,"style":1048},{"counterType":1034},[1229,1230,1231,1232,1233,1234],{},"Post erfolgreich abgerufen",{"items":2662,"source":2747,"manualIds":2748,"manualMatchedIds":2749},[2663,2670,2677,2684,2691,2698,2705,2712,2719,2726,2733,2740],{"id":2664,"slug":2665,"title":2666,"excerpt":2667,"featuredImage":2668,"publishedAt":2669},"435","ultimate-guide-to-acceptance-criteria-for-llm-adoption-in-enterprise-playbooks","企业剧本中采用大语言模型的验收标准终极指南","掌握定义精确验收标准的艺术，确保大型语言模型在企业环境中成功集成。本全面指南提供可操作的框架、实例及最佳实践，专为剧本驱动式应用量身定制。","\u002Fuploads\u002F2026\u002F09\u002Fultimate-guide-to-acceptance-criteria-for-llm-adoption-in-enterprise-playbooks-1788540267775-zgr6mm.webp","2026-09-06T11:50:00.000Z",{"id":2671,"slug":2672,"title":2673,"excerpt":2674,"featuredImage":2675,"publishedAt":2676},"463","prompt-invariance-does-the-conclusion-survive-the-prompt","提示不变性：结论在提示变化后是否依然成立？","一种用于检验AI结论是否依赖于问题表述方式的实用方法。提示不变性比较原始、盲测、反向和对抗性表述，同时保持证据结构受控。","\u002Fuploads\u002F2026\u002F09\u002Fprompt-invariance-does-the-conclusion-survive-the-prompt-1789809799910-s0vbcb.webp","2026-09-19T01:09:00.000Z",{"id":2678,"slug":2679,"title":2680,"excerpt":2681,"featuredImage":2682,"publishedAt":2683},"483","what-is-an-ai-solution-architect-system-boundaries-responsibilities-and-trade-offs","什么是AI解决方案架构师？系统边界、职责与权衡","AI解决方案架构师将业务需求转化为生产就绪的AI系统，涵盖数据、模型、工具、安全、运行时、评估和运维。","\u002Fuploads\u002F2026\u002F10\u002Fwhat-is-an-ai-solution-architect-system-boundaries-responsibilities-and-trade-offs-1791476643267-1st5xz.webp","2026-10-08T12:23:00.000Z",{"id":2685,"slug":2686,"title":2687,"excerpt":2688,"featuredImage":2689,"publishedAt":2690},"385","comfyui-on-fedora-43-two-virtual-environments-one-click-start-march-2026","Fedora 43上的ComfyUI：双虚拟环境 + 一键启动（2026年3月）","目标：保留两个Python虚拟环境（例如3.12和3.14）以确保兼容性，但通过一个简洁、轻量的配置自动启动ComfyUI。","\u002Fuploads\u002F2026\u002F03\u002Fcomfyui-on-fedora-43-two-virtual-environments-one-click-start-march-2026-1772356102271-qqqlsm.webp","2026-03-01T04:57:00.000Z",{"id":2692,"slug":2693,"title":2694,"excerpt":2695,"featuredImage":2696,"publishedAt":2697},"461","beyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning","超越提示工程：一种更可靠的人工智能推理方法","大型语言模型未必是因为缺乏推理能力而失败。它们常常失败，是因为推理过程没有得到充分的约束、质疑或验证。本文提出一种与领域无关的方法论，将提示转化为一种结构化的认知过程：将事实与假设分离，生成相互竞争的假说，检验反证，应用证伪，并检查结论在替代性框架下是否仍然稳定。目标不是让模型“更少同意”，而是让它的结论更少依赖于用户最初的框架。","\u002Fuploads\u002F2026\u002F09\u002Fbeyond-prompt-engineering-a-methodology-for-more-reliable-ai-reasoning-1789804466431-qba1zb.webp","2026-09-19T00:55:00.000Z",{"id":2699,"slug":2700,"title":2701,"excerpt":2702,"featuredImage":2703,"publishedAt":2704},"482","adr-vs-nfr-architecture-decisions-and-system-quality-are-not-the-same-thing","ADR 与 NFR：架构决策与系统质量并非同一回事","ADR 与 NFR 解析：了解系统质量需求如何驱动架构决策、ADR 如何记录权衡取舍，以及为什么验证保持独立。","\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":2706,"slug":2707,"title":2708,"excerpt":2709,"featuredImage":2710,"publishedAt":2711},"467","the-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers","答案有效性边界：相关性到可靠AI答案之间缺失的层级","一个来源可能相关、权威，但对于所提出的问题仍然是错误的。缺失的层次是适用性：答案成立的条件，以及迫使其被重新考虑的变化。本文介绍了“答案有效性边界”这一面向人类、AI搜索和RAG系统的来源设计模式。","\u002Fuploads\u002F2026\u002F09\u002Fthe-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers-1790272901306-1g5jly.webp","2026-09-24T11:59:00.000Z",{"id":2713,"slug":2714,"title":2715,"excerpt":2716,"featuredImage":2717,"publishedAt":2718},"492","mcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits","MCP 解析：它连接什么、不做什么以及它适用于何处","模型上下文协议通过标准的客户端-服务器边界，将AI应用程序连接到外部工具、资源和提示。了解MCP能做什么、不能做什么，以及它在智能体架构中的定位。","\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":2720,"slug":2721,"title":2722,"excerpt":2723,"featuredImage":2724,"publishedAt":2725},"448","google-io-2026-android-xr-and-intelligent-eyewear","Google I\u002FO 2026：Android XR、智能眼镜与环境AI界面","Google I\u002FO 2026 将 Android XR 和智能眼镜从概念推向实际平台方向。本文解析了音频眼镜、显示眼镜、Gemini 驱动的上下文感知、开发者影响、隐私风险，以及为何可穿戴 AI 更关乎创造环境辅助界面，而非取代手机。","\u002Fuploads\u002F2026\u002F05\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear-1779227942270-dtsm9y.webp","2026-05-21T11:05:00.000Z",{"id":2727,"slug":2728,"title":2729,"excerpt":2730,"featuredImage":2731,"publishedAt":2732},"479","where-does-an-llm-get-its-data-rag-data-sources-in-python","LLM从哪里获取数据？Python中的RAG数据源","LLM 并不会神奇地知道你的文件、数据库或 API。这个 RAG 系列的实用续篇用简单的 Python 展示了外部数据如何变成可检索的证据：从文本文件和 SQL 到全文搜索、嵌入、上下文组装以及最终的 LLM 调用。","\u002Fuploads\u002F2026\u002F09\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python-1790517200521-nfsi5i.webp","2026-09-27T05:51:00.000Z",{"id":2734,"slug":2735,"title":2736,"excerpt":2737,"featuredImage":2738,"publishedAt":2739},"484","what-is-an-ai-platform-architect-models-data-runtime-security-and-operations","什么是AI平台架构师？模型、数据、运行时、安全与运维","AI平台架构师负责跨模型、提供商、检索、智能体、身份、安全、评估、可观测性和运营设计可复用的AI基础。","\u002Fuploads\u002F2026\u002F10\u002Fwhat-is-an-ai-platform-architect-models-data-runtime-security-and-operations-1791477229171-ou3zcc.webp","2026-10-08T12:32:00.000Z",{"id":2741,"slug":2742,"title":2743,"excerpt":2744,"featuredImage":2745,"publishedAt":2746},"380","streamlining-code-quality-testing-with-eslint-and-prettier","优化代码质量：使用ESLint与Prettier进行测试","在现代软件开发中，保持一致的代码质量和风格至关重要。ESLint与Prettier提供了强大的组合方案，能够自动化这些关键环节，确保代码库的整洁性、可读性，并遵循既定标准。本文将深入探讨这两款工具如何无缝融入测试工作流，从而提升开发者的工作效率与项目的可维护性。","\u002Fuploads\u002F2026\u002F01\u002Ftesting-with-eslint-and-prettier-1769204102989-ezczs0-1769376472926-hwqqkt.webp","2026-01-25T10:26:00.000Z","fallback",[],[]]